The eruption of Soufrière Hills volcano, Montserrat, from 1995 to 1999

The Eruption of Soufriere Hills Volcano, Montserrat, From 1995 to 1999 Dedicated to Peter Francis Geological Society...

Author: Timothy H. Druitt | B. Peter Kokelaar

31 downloads 1307 Views 85MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

The Eruption of Soufriere Hills Volcano, Montserrat, From 1995 to 1999

Dedicated to Peter Francis

Geological Society Publications Society Book Editors A. J. FLEET (CHIEF EDITOR) P. DOYLE F. J. GREGORY J. S. GRIFFITHS A. J. HARTLEY R. E. HOLDSWORTH A. C. MORTON N. S. ROBINS M. S. STOKER J. P. TURNER

Society books reviewing procedures The Society makes every effort to ensure that the scientific and production quality of its books matches that of its journals. Since 1997. all book proposals have been refereed by specialist reviewers as well as by the Society's Books Editorial Committee. If the referees identify weaknesses in the proposal, these must be addressed before the proposal is accepted. Once the book is accepted, the Society has a team of Book Editors (listed above) who ensure that the volume editors follow strict guidelines on refereeing and quality control. We insist that individual papers can only be accepted after satisfactory review by two independent referees. The questions on the review forms are similar to those for Journal of the Geological Society. The referees' forms and comments must be available to the Society's Book Editors on request. Although many of the books result from meetings, the editors are expected to commission papers that were not presented at the meeting to ensure that the book provides a balanced coverage of the subject. Being accepted for presentation at the meeting does not guarantee inclusion in the book. Geological Society Special Publications and some other books series are included in the ISI Index of Scientific Book Contents, but they do not have an impact factor, the latter being applicable only to journals. More information about submitting a proposal and producing a Society Publication can be found on the Society's web site: www.geolsoc.org.uk.

It is recommended that reference to all or part of this book should be made in one of the following ways: DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London, Memoirs, 21. LUCKETT, R., BAPTIE, B. & NEUBERG, J. 2002. The relationship between degassing and rockfall signals at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 595-602.

GEOLOGICAL SOCIETY MEMOIRS NO. 21

The Eruption of Soufriere Hills Volcano, Montserrat From 1995 to 1999

EDITED BY

T. H. DRUITT Universite Blaise Pascal, Clermont-Ferrand, France and

B. P. KOKELAAR University of Liverpool, UK

2002 Published by The Geological Society London

THE GEOLOGICAL SOCIETY The Geological Society of London (GSL) was founded in 1807. It is the oldest national geological society in the world and the largest in Europe. It w;as incorporated under Royal Charter in 1825 and is Registered Charity 210161. The Society is the UK national learned and professional society for geology with a worldwide Fellowship (FGS) of 9000. The Society has the power to confer Chartered status on suitably qualified Fellows, and about 2000 of the Fellowship carry the title (CGeol). Chartered Geologists may also obtain the equivalent European title, European Geologist (EurGeol). One fifth of the Society's fellowship resides outside the UK. To find out more about the Society. log on to www.geolsoc.org.uk. The Geological Society Publishing House (Bath, UK) produces the Society's international journals and books. and acts as European distributor for selected publications of the American Association of Petroleum Geologists (AAPG), the American Geological Institute (AGI). the Indonesian Petroleum Association£ (IPA), the Geological Society of America (GSA), the Society for Sedimentary Geology (SEPM) and the Geologists' Association (GA). Joint marketing agree ments ensure that GSL Fellows may purchase these societies' publications at a discount. The Society's online bookshop (accessible from www.geolsoc.org.uk) offers secure book purchasing with your credit or debit card. To find out about joining the Society and benefiting from substantial discounts on publications of GSL and other societies worldwide, consult www.geolsoc.org.uk, or contact the Fellowship Department at: The Geological Society. Burlington House. Piccadilly. London W1J OBG: Tel. —44 (0)20 7434 9944; Fax +44 (0)20 7439 8975; E-mail: enquiries@ geolsoc.org.uk. For information about the Society's meetings, consult Events on www.geolsoc.org.uk. To find out more about the Society's Corporate Affiliates Scheme, write to [email protected].

Published by The Geological Society from: The Geological Society Publishing House Unit 7 Brassmill Enterprise Centre Brassmill Lane Bath BA1 3JN, UK (Orders'. Tel. +44 (0)1225 445046 Fax +44 (0)1225 442836) Online bookshop: http://bookshop.geolsoc.org.uk The publishers make no representation, express or implied, with regard to the accuracy of the information contained in this book and cannot accept any legal responsibility for any errors or omissions that may be made. @ The Geological Society 2002. All rights reserved. No reproduction, copy or transmission of this publication may be made without written permission. No paragraph of this publication may be reproduced, copied or transmitted save with the provisions of the Copyright Licensing Agency, 90 Tottenham Court Road, London W1P 9HE. Users registered with the Copyright Clearance Center, 27 Congress Street, Salem, MA 01970, USA: the item-fee code for this publication is 0435-4052/02/S 15.00. British LibraryCCataolologuingin Publication Data A catalogue record for this book is available from the British Library. ISBN 1-86239-098-3 ISSN: 0435-4052 Typeset by Aarontype Ltd, Easton, Bristol, UK Printed by Cambrian Press, Aberystwyth, UK.

Distributors USA AAPG Bookstore PO Box 979 Tulsa OK 74101-0979 USA Orders: Tel. +1 918 584-2555 Fax +1 918 560-2652 E-mail bookshop@ aapg.org Affiliated East-West Press PVT Ltd G-l/16 Ansari Road. Daryaganj. New Delhi 110002 India Orders: Tel. +91 11 327-9113 Fax +91 11 326-0538 E-mail affiliat@ nda.vsnl.net.in Japan Kanda Book Trading Co. Cityhouse Tama 204 Tsurumaki 1-3-10 Tama-shi Tokyo 206-0034 Japan Orders: Tel. +81 (0)423 57-7650 Fax +81 (0)423 57-7651

Contents

Foreword Preface Acknowledgements In Memorium Peter Francis Background and overview of the eruption Setting, chronology and consequences of the eruption of Soufriere Hills Volcano, Montserrat (1995-1999): KOKELAAR, B. P. The eruption of Soufriere Hills Volcano, Montserrat (1995-1998): overview of scientific results: SPARKS, R. S. J. & YOUNG, S. R. The Montserrat Volcano Observatory: its evolution, organization, role and activities: ASPINALL, W. P., LOUGHLIN, S. C, MICHAEL, F. V., MILLER, A. D., NORTON, G. E., ROWLEY, K. C., SPARKS, R. S. J. & YOUNG, S. R. The volcanic evolution of Montserrat using 40Ar/39Ar geochronology: HARFORD, C. L., PRINGLE, M. S., SPARKS, R. S. J. & YOUNG, S. R. Volcanic processes, products and hazards Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano, Montserrat: WATTS, R. B., HERD, R. A., SPARKS, R. S. J. & YOUNG, S. R. Dynamics of magma ascent and lava extrusion at Soufriere Hills Volcano, Montserrat: MELNIK, O. & SPARKS, R. S. J. Mechanisms of lava dome instability and generation of rockfalls and pyroclastic flows at Soufriere Hills Volcano, Montserrat: CALDER, E. S., LUCKETT, R., SPARKS, R. S. J. & VOIGHT, B. Pyroclastic flows and surges generated by the 25 June 1997 dome collapse, Soufriere Hills Volcano, Montserrat: LOUGHLIN, S. C., CALDER, E. S., CLARKE, A., COLE, P. D., LUCKETT, R., MANGAN, M. T., PYLE, D. M., SPARKS, R. S. J., VOIGHT, B. & WATTS, R. B. Eyewitness accounts of the 25 June 1997 pyroclastic flows and surges at Soufriere Hills Volcano, Montserrat, and implications for disaster mitigation: LOUGHLIN, S. C., BAXTER, P. J., ASPINALL, W. P., DARROUX, B., HARFORD, C. L. & MILLER, A. D. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano, Montserrat: COLE, P. D., CALDER, E. S., SPARKS, R. S. J., CLARKE, A. B., DRUITT, T. H., YOUNG, S. R., HERD, R. A., HARFORD, C. L. & NORTON, G. E. Small-volume, highly mobile pyroclastic flows formed by rapid sedimentation from pyroclastic surges at Soufriere Hills Volcano, Montserrat: an important volcanic hazard: DRUITT, T. H., CALDER, E. S., COLE, P. D., HOBLITT, R. P., LOUGHLIN, S. C., NORTON, G. E., RITCHIE, L. J., SPARKS, R. S. J. & VOIGHT, B. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat: DRUITT, T. H., YOUNG, S. R., BAPTIE, B., BONADONNA, C., CALDER, E. S., CLARKE, A .B., COLE, P. D., HARFORD, C .L., HERD, R. A., LUCKETT, R., RYAN, G. & VOIGHT, B. Modelling of conduit flow dynamics during explosive activity at Soufriere Hills Volcano, Montserrat: MELNIK, O. & SPARKS, R. S. J. Computational modelling of the transient dynamics of the August 1997 Vulcanian explosions at Soufriere Hills Volcano, Montserrat: influence of initial conduit conditions on near-vent pyroclastic dispersal: CLARKE, A. B., NERI, A., VOIGHT, B., MACEDONIO, G. & DRUITT, T. H. Hazard implications of small-scale edifice instability and sector collapse: a case history from Soufriere Hills Volcano, Montserrat: YOUNG, S. R., VOIGHT, B., BARCLAY, J., HERD, R. A., KOMOROWSKI, J.-C, MILLER, A. D., SPARKS, R. S. J. & STEWART, R. C. The 26 December (Boxing Day) 1997 sector collapse and debris avalanche at Soufriere Hills Volcano, Montserrat: VOIGHT, B., KOMOROWSKI, J.-C., NORTON, G. E., BELOUSOV, A. B., BELOUSOVA, M., BOUDON, G., FRANCIS, P. W., FRANZ, W., HEINRICH, P., SPARKS, R. S. J. & YOUNG, S. R. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat: SPARKS, R. S. J., BARCLAY, J., CALDER, E. S., HERD, R. A., KOMOROWSKI, J.-C., LUCKETT, R., NORTON, G. E., RITCHIE, L. J., VOIGHT, B. & WOODS, A. W. Sedimentology of deposits from the pyroclastic density current of 26 December 1997 at Soufriere Hills Volcano, Montserrat: RITCHIE, L. J., COLE, P. D. & SPARKS, R. S. J. The explosive decompression of a pressurised volcanic dome: the 26 December 1997 collapse and explosion of Soufriere Hills Volcano, Montserrat: WOODS, A. W., SPARKS, R. S. J., RITCHIE, L. J., BATEY, J., GLADSTONE, C. & BURSIK, M. I. Pyroclastic flow and explosive activity of the lava dome of Soufriere Hills volcano, Montserrat, during a period of no magma extrusion (March 1998-November 1999): NORTON, G. E., WATTS, R. B., VOIGHT, B., MATTIOLI, G. S., HERD, R. A., YOUNG, S. R., DEVINE, J. D., ASPINALL, W. P., BONADONNA, C., BAPTIE, B. J., EDMONDS, M., HARFORD, C. L., JOLLY, A. D., LOUGHLIN, S. C., LUCKETT, R. & SPARKS, R. S. J. Tephra fallout in the eruption of Soufriere Hills Volcano, Montserrat: BONADONNA, C., MAYBERRY, G. C., CALDER, E. S., SPARKS, R. S. J., CHOUX, C., JACKSON, P., LEJEUNE, A. M., LOUGHLIN, S. C., NORTON, G. E., ROSE, W.I., RYAN, G. & YOUNG, S. R. Numerical modelling of tephra fallout associated with dome collapses and Vulcanian explosions: application to hazard assessment on Montserrat: BONADONNA, C., MACEDONIO, G. & SPARKS, R. S. J.

vii ix xi xiii

1 45 71 93

115 153 173 191 211 231 263 281 307 319 349 363 409 435 457 467

483 517

vi

CONTENTS

Dynamics of volcanic and meteorological clouds produced on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano. Montserrat: MAYBERRY, G. C, ROSE, W. I. & BLUTH, G. J. S. Monitoring of airborne particulate matter during the eruption of Soufriere Hills Volcano, Montserrat: MOORE. K.R.. DUFFELL. H.. NICHOLL, A. & SEARL. A. Geophysical and gas studies Seismicity, gas emission and deformation from 18 July to 25 September 1995 during the initial phreatic phase of the eruption of Soufriere Hills Volcano, Montserrat: GARDNER, C. A. & WHITE. R. A. Spaceborne radar measurements of the eruption of Soufriere Hills Volcano, Montserrat: WADGE, G., SCHEUCHL, B. & STEVENS. N. F. The relationship between degassing and rockfall signals at Soufriere Hills Volcano, Montserrat: LUCKETT. R.. BAPTIE. B. & NEUBERG, J. A model of the seismic wavefield in gas-charged magma: application to Soufriere Hills Volcano. Montserrat: NEUBERG. J. & O'GORMAN, C. Observations of low-frequency earthquakes and volcanic tremor at Soufriere Hills Volcano. Montserrat: BAPTIE. B.. LUCKETT. R. & NEUBERG, J. Variation in HC1/SO2 gas ratios observed by Fourier transform spectroscopy at Soufriere Hills Volcano. Montserrat: OPPENHEIMER, C., EDMONDS, M., FRANCIS, P. & BURTON, M. Index

539 557

567 583 595 603 611 621 640

Foreword Volcanoes are the most violent surface expression of the Earth's internal energy. Only impacts of large extra-terrestrial bodies can match the explosive release and devastation of the largest volcanoes. Indeed for some of the most dramatic events the Earth has seen - the large terrestrial extinctions of animal life - the jury is still out as to whether they were brought about by meteoritic impact or by wide-scale effects of volcanic activity. Volcanoes have it too when it comes to sustained visual impact. Earthquakes, tsunamis and avalanches all cause massive devastation, but it is accomplished in the blink of an eye, and floods rise with a progressive and depressing inevitability. Volcanoes are simply the most spectacular of the destructive natural hazards to life on Earth. To those who are far enough away to view them in safety, volcanoes can offer a truly awe-inspiring pyrotechnic display of the Earth's innate power - a natural, spectacular son et lumiere. For this reason from time immemorial they have exerted a siren-like attraction for geologists, photographers, filmmakers and many others. And, like the sirens of ancient fable, they have lured to their death all too many of those who dared to get too close. Indeed volcanoes inspired such awe in the ancient world that their own mythology sprang up about them. Cyclops, the one-eyed giant who all-unprovoked threw rocks great distances to kill shepherds tending their flocks, we know today as Mount Etna. The giant was also able to cause springs to flow where he struck the ground - it is not uncommon for groundwater flows to be disrupted during volcanic episodes. Like its neighbouring islands in the Caribbean, Montserrat exists solely because of volcanic activity. It is a volcanic island formed by the progressive accumulation of layers of lava and debris erupted on the sea floor. It is a small Caribbean island, which, along with a host of others, is located close to where the slowly moving crust of the Atlantic takes leave of the surface and plunges down into the Earth's interior. All life forms, humans included, are so eager to find new habitats that as soon as a volcano has been inactive for a hundred years or so, and sometimes sooner, it is colonized by a flora and fauna. Montserrat had long been inactive and, besides being well situated for fishing and tourism, and a little agriculture, it supported a resident population of over ten thousand. The art of volcanic prediction is still too poorly developed to be very useful and when in 1995 the volcano showed signs of renewed activity the population simply hoped that it would soon die down. It had been quiescent for about 350 years. Understandably, people who in their own lifetime have known only a gently steaming mountain are not inclined to believe that things are about to change. But change they did. Devastating pyroclastic flows overwhelmed the southern half of the island with its villages and smallholdings. In the north, the infrastructure of life was disrupted and part buried by settling ash. The people had no experience of active volcanoes and could not imagine how rapidly the behaviour of the volcano could change and how unpredictable it was. The inescapable speed and heat of flows of incandescent ash were beyond their comprehension. Although they were warned, many were reluctant to abandon their homes on official advice and chose to take the risk. Sadly some paid the ultimate price. And when this happened in spite of their best efforts, some on the ground had to live with the nagging doubt as to whether, had they tried just one more time, they could have persuaded the farmers to leave. In the world league of volcanic eruptions the ongoing Montserrat eruption does not rate very high. What was unusual, indeed unique, about Montserrat was the combination of two special circumstances. First, because of the risk to life and the presence of an indigenous population with no escape other than to leave the

island, resources were available to monitor the volcano that would simply not have been there for a purely scientific study. Whether those resources were really adequate is another matter. Second, there has been no recent opportunity anywhere in the world to study an oceanic island arc volcano during eruption. Different kinds of recent volcanic activity have been studied elsewhere in Iceland, in Hawaii, in Washington State, in southern Italy and in Japan, but Montserrat offered a unique opportunity to study this kind of oceanic eruption with modern techniques. This combination of circumstances has made it possible to document the behaviour of the volcano in considerable detail and to do so with the collaboration of geologists from a wide range of organizations and from many countries. Intriguingly, there is one important feature of the Montserrat eruption that is little known elsewhere. The fine ash that is common in many eruptions and which buried buildings on Montserrat to depths of three metres or more is very unusual: it contains, and in places largely comprises, very fine particles of silica in an unusual crystalline form - minute particles of the mineral cristobalite. These are uncomfortably similar in their characteristics to other fine particles that damage the respiratory system, and were regarded as potentially a significant health hazard on the island. Occurrences such as this present governments with major moral dilemmas. Montserrat is a British dependency many thousands of kilometres from the UK and therefore difficult, not to say very expensive, to support. In the early years of the renewed eruption, the infrastructure and much of the farming land was destroyed. Resettlement was offered to the inhabitants and eventually the majority of them accepted the offer. But what are our obligations to the others? The economy of the island is now extraordinarily tenuous and life there can continue only if external support is maintained. It looks as if the northern part of the island will be relatively safe from the direct products of eruption for the foreseeable future. But it will be decades before soils develop and before agriculture can be re-established in the south. The eruption is still in progress and has now lasted for more than six years, longer than in virtually all similar lava-dome eruptions around the world. It seems that Soufriere Hill Volcano is evolving into a persistently active state that could continue for decades. And so, alongside the scientific investigations, a complex human drama was playing as well, and geoscientists, for whom volcanology had been their somewhat esoteric and rather academic specialization, suddenly found themselves at the frontline of its practical application where life at times had much in common with a war-zone. There are other lessons to be learned as well. Volcanology has been very much a minority discipline within UK Earth Sciences. It has been kept alive by the efforts of a few outstanding and energetic individuals of real academic distinction. No value for money or relevance criteria would have suggested that volcanology was worthy of more than peripheral support. But without this infrastructure of knowledge and experience the UK would have had no indigenous capability to cope with the events on Montserrat. Many of the results of the scientific studies at Montserrat are presented in this volume. They represent an unparalleled suite of detailed observations that will add significantly to the understanding of volcanic hazards. In due course, they will lead to a better understanding of how volcanoes work and to a better ability to predict their behaviour. I am proud that the Geological Society is the publisher of this volume. Ron Oxburgh President of the Geological Society of London

Preface The andesitic dome-building eruption of Soufriere Hills Volcano has wreaked havoc on the small Caribbean island of Montserrat. About half of this 'Emerald Isle' has been rendered barren and uninhabitable, almost two-thirds of the original population has left, and 19 lives have been lost, all as a direct result of the volcanic activity. Many Montserratians have suffered multiple evacuations and displacements from their homes, and the economy has been severely affected by the loss of infrastructure and farmland, and by the adverse impact on tourism. The centre of the former capital, Plymouth, today lies partially buried under metres of volcanic debris, and several villages have been swept away by catastrophic flows of incandescent ash. As this book goes to press, the eruption continues and as yet shows no signs of abating. The still-populated northern half of the island is to some extent recovering and the people are learning to live with the volcano. Nevertheless, fine airborne ash from intermittent major events continues to penetrate into many homes and to pose a health hazard, while boulder-laden floods following the torrential rains of all-too-frequent hurricanes impede recovery of property. This Memoir presents results of monitoring and associated research over a five-year period, from the onset of the eruption in July 1995 until November 1999. Scientists on active volcanoes need to balance their essential activities in monitoring, assessing hazards, advising local and national authorities, and informing the public, with programmes of basic research into causes, mechanisms and consequences of the volcanic activity under scrutiny. Mitigation of volcanic hazards is most effective when there exists good understanding of the physical and chemical processes controlling the system. Monitoring and research on Montserrat evolved from mainly remote surveillance by the Seismic Research Unit of the University of the West Indies in Trinidad, before the eruption, to establishment of a multinational and multidisciplinary team at the Montserrat Volcano Observatory (MVO). The observatory, which moved its location three times as the eruption slowly escalated between July 1995 and September 1997, developed enhanced capabilities over the period as funding and technical assistance became available. The slow escalation of the crisis gave the scientists time to build effective monitoring infrastructures and team management before the most devastating phases of the eruption in 1997. Monitoring included use of short-period and broadband seismometers to detect and locate all types of earthquakes, Global Positioning Satellite (GPS) and Electronic Distance Measurement networks to detect upheaval of the volcano, measurements by Correlation Spectrometer and Fourier Transform Infra-red spectroscopy to monitor gas exhalations, and photogrammetry and GPS-based methods to determine magma extrusion rates and volumes. Petrological and geochemical studies of magmatic products were carried out as the eruption progressed. Monitoring was initially aided by the close proximity to the volcano of the MVO, which until September 1997 benefited from line-of-sight observations. Analysis and cross-correlation of emerging multi-parameter data sets, together with the development and application of physical models of magma ascent, degassing and extrusion, led to greatly increased understanding of the system dynamics and origin of the various signals being monitored as the eruption unfolded. This in turn influenced subsequent monitoring strategies. The large numbers of visiting scientists, who undertook diverse studies at the MVO, greatly enhanced the effectiveness and innovation of the work of the core teams. Many diverse phenomena related to the ascent and extrusion of andesitic magma have now been studied at close quarters. These include phreatic explosions, dome collapses with formation of pyroclastic flows and surges, dispersal of ash plumes, vertically directed explosions with fountain collapse, sector collapse followed by debris avalanche, lateral blast and violent pyroclastic density current, and a remarkable period of disintegration of a lava dome with little or no accompanying magma extrusion or precursory seismic activity. Cyclic activity, registered in seismic and deformation data, in compositions and fluxes of magmatic gases, and in rates of magma

extrusion, has been attributed to interactive physical processes in the conduit and extruding lava dome. Cycles with timescales ranging from several weeks to several hours are now understood in terms of the ascent, degassing, rheological stiffening and pressurization of crystal-rich andesitic magma, with complex feedback effects and multiple regimes of behaviour. The eruption has permitted detailed documentation and better understanding of processes and associated signals that lead to lava dome instability and to explosive decompression, and of the physical behaviour and characteristics of associated pyroclastic currents. It has also provided the opportunity for development of different methods in hazards assessment and zonation, including formal elicitations of international scientific expertise and statistical treatments of eruption-scenario models and associated uncertainties. The eruption and its consequences constitute an important case of volcanic crisis management and of the interactions between scientists, authorities and populace on a small island, with significant lessons for the future. The decision was taken to publish as much as possible of the material pertaining to the 1995-1999 eruptive period in one book. Previous notable benchmark eruptions, such as of Mount St Helens in 1980 and Mount Pinatubo in 1991, have been similarly followed by scientific monographs that are the main repositories of data and interpretations concerning those eruptions. The books constitute an invaluable resource for those researchers, volcano observatories and government agencies concerned with volcanic risk minimization. Two series of short reports on the Montserrat eruption up to August 1997 were published during 1998 in Geophysical Research Letters. This Memoir contains 30 papers, many of which address the chronology, dynamics, products and associated hazards of the eruption. It also includes papers specifically on the associated geophysics and geochemistry, although geophysical aspects in particular constitute significant parts of many of the other papers. Four introductory papers provide overviews of the eruption chronology and consequences, of the scientific results, of the evolution, organization, role and activities of the Montserrat Volcano Observatory, and of the volcanic development of Montserrat through time. A large photographic record of the 1995-1999 eruptive period is included. Using author consensus, we have adopted certain conventions for clarity and consistency throughout the Memoir. We consider the recent events to represent a single eruption of Soufriere Hills Volcano, which, following six years of increased seismicity, started on 18 July 1995 and continues at the time of publication. We use the convention that there have been two lava domes to date during the eruption. One grew intermittently from mid-November 1995 to midMarch 1998, undergoing multiple collapses as it did so. This was followed by a twenty-month interval during which sectors of the first dome collapsed with little or no associated extrusion and commonly without seismic precursors. The second dome started growing in midNovember 1999. This Memoir concerns only that period of the eruption up to the resumption of magma discharge in mid-November 1999. Individual lava extrusions during dome growth are referred to as shear lobes, or, for brevity, lobes. Lobes are referred to by the date of their first appearance (e.g. 17 July 1996 Lobe). We have also tried, for the benefit of non-specialist readers, to simplify the terminology used in the Memoir, particularly that concerning pyroclastic density currents. In most cases during the 1995-1999 period it was possible to distinguish fairly clearly, using conventional criteria, between pyroclastic density currents predominantly of high-concentration and those predominantly of lowconcentration, and also between their respective deposits. We have retained the terms pyroclastic flow and pyroclastic surge respectively for these phenomena. We distinguish genetically between domecollapse, fountain-collapse and surge-derived pyroclastic flows. The descriptive terms block-and-ash flow and pumice-and-ash flow are also used respectively for the former two phenomena. Only in one case, the catastrophic collapse of 26 December 1997, have we retained the general term pyroclastic density current, because significant vertical and lateral gradients in particle size and

PREFACE concentration are inferred to have been present in the moving current so that neither pyroclastic flow nor pyroclastic surge seems appropriate. Monitoring the eruption from 1995 to 1999 was the work of more than 120 people, including Montserratian scientific, technical and clerical staff along with scientists and technicians from the British Geological Survey, the Seismic Research Unit on Trinidad, the US Geological Survey Volcanic Crisis Assistance Team, and universities in the UK and other countries including the USA, Puerto Rico and France. Many were graduate students in volcanology or young, MVO-trained staff and much of the collection and analysis of data from the eruption was accomplished by these young people, who also participated in the daily jobs of hazards assessment and public education. Although tragic in so many ways, the eruption has served as a training ground for a large cohort of young volcanologists who are now well prepared to deal with future volcanic crises around the globe. It is a tribute to all of the people involved

that, after more than five years of volcanic activity and with constant changing of observatory personnel, the vast amount of information collected is in such good order that it could be compiled and analysed for the papers in this Memoir. A hundred years ago. on 8 May 1902, a cloud of hot gas and pyroclastic debris from a lava dome on Montagne Pelee swept over the town of St Pierre on the French Caribbean island of Martinique and claimed about 28000 lives. Many of the processes that drove that eruption probably resembled those at Soufriere Hills. This Memoir registers the enormous advances made since that time, both in our understanding of lava-dome eruptions and in the sophistication of the tools we use to monitor them. Much, however, remains to be learned. Nature will not reveal many secrets in one go. Tim Druitt & Peter Kokelaar Editors

Acknowledgements The editors and authors of this book wish to pay their respect to the people of Montserrat, who have suffered greatly during the present crisis. They have borne their hardship and, for many, the loss of homes, family and friends, with courage, good nature and dignity. The contributions of all staff members of the Montserrat Volcano Observatory (MVO) over the 1995-1999 period are acknowledged. The local staff in particular formed the backbone of the team, often under conditions that were stressful for them, their families and their communities. They are: Venus Bass, Owen Butler, Levar Cabey, Sharon Charles, Thomas Christopher, Paulette Cooper, Billy Darroux, Deneese Fenton, Franklyn Greenaway, Linda Halloran, David Lea, Sunny Lea, Chelston Lee, Leroy Luke, Pops Morris, Karney Osborne, Joel Osborne, Alwyn Ponteen, Graham Ryan, Patch Silcott, George Skerrit, Tappy Syers, Bill Thorn, Bill Tonge, Jackie Weekes, Daisy Weekes, Dave Williams and Pyiko Williams. We also list the overseas scientists and technicians that rotated through the MVO during the period, while apologizing to any inadvertently forgotten: Stephane Acounis, Godfrey Almorales, Christian Antenor, Sayyadul Arafin, Willy Aspinall, Wilkie Balgobin, Brian Baptie, Jenni Barclay, Peter Baxter, Costanza Bonadonna, Eliza Calder, Robert Carsley, Tom Casadevall, Caroline Choux, Amanda Clarke, Paul Cole, Mark Davies, Peter Day, Pierre Delmelle, Joe Devine, Laurance Donnelly, Sarah Dornan, Tim Druitt, Hayley Duffell, Peter Dunkley, Neil Dyer, Marie Edmonds, John Ewert, Glenn Ford, Peter Francis, Davie Galloway, Cynthia Gardner, Gilbert Hammouya, Chloe Harford, Richard Herd, Rick Hoblitt, Claire Horwell, Paul Jackson, Mike James, Art Jolly, Chris Kilburn, Pete Kokelaar, Jean-Christophe Komorowski, Joan Latchman, Anne-Marie Lejeune, Andy Lockhart, Sue Loughlin, Richard Luckett, Lloyd Lynch, Adam Maciejewski, Maggie Mangan, Glen Mattioli, Gari Mayberry, Bill McGuire, Angus Miller, Dan Miller, Kate Moore, Mick Murphy, Tom Murray, Jurgen Neuberg, Gill Norton, Clive Oppenheimer, Ouchi Osuji, Dave Petrie, Luchman Pollard, John Power, Dave Pyle, Chandrapath Ramsingh, Tony Reedman, Lucy Ritchie, Ritchie Robertson, Geoff Robson, Lizzette Rodriguez, Keith Rowley, Willie Scott, Desmond Seupersad, John Shepherd, Bennett Simpson, Alan Smith, Stefan Soil, Steve Sparks, Mark Stasiuk, Nicki Stevens, Rod Stewart, Dai Stewart, John Stix, Sharon Teebenny, Glenn Thompson, John Tomblin, Jane Toothill,

Patrick Tuchais, Terry Turbitt, Jean-Pierre Viode, Barry Voight, Geoff Wadge, Colin Walker, Matthew Watson, Robert Watts, Randall White, and Simon Young. It has been a privilege for MVO scientists to have worked, often under hazardous conditions, with helicopter pilots of exceptional skill, in particular Jim McMahon, Alex Grouchy, Laurance Linskey and Barry Lashley. The MVO was financed mainly by the UK Government Department for International Development (DFID), formerly the Overseas Development Administration (ODA), and by the Government of Montserrat. Research grants and studentships were provided by the UK Natural Environment Research Council (NERC), the US National Science Foundation (NSF) and the French Centre National de la Recherche Scientifique (CNRS). The work at the MVO during 1995-1999 was supported on Montserrat by two Governors, HE Frank Savage and HE Tony Abbott, by three Chief Ministers, Reuben Meade, Bertrand Osborne and David Brandt, and by Frankie Michael and Horatio Tuitt of the Emergency Department. The observatory also benefited from the help of Frank Hooper and Chris Burgess, the police force, the Royal Montserrat Defence Force, Radio ZJB, the Government Information Service, the Aid Management Office of DFID, the Governor's Office, and ministers and officers of the Government of Montserrat. The production of this Memoir was skilfully overseen by Angharad Hills. Funding for colour printing was generously provided by the Geological Society of London. The following people kindly took the time to provide reviews of the articles, and helped us ensure high standards of science and presentation: Willy Aspinall, Charlie Bacon, Peter Baxter, Mike Branney, Ray Cas, Bernard Chouet, Brian Dade, Mike Dungan, Jon Fink, Armin Freundt, JeanLuc Froger, Jennie Gilbert, Lori Glaze, Richard Hiscott, Rick Hoblitt, Jean-Christophe Komorowski, Tak Koyaguchi, Steve McNutt, Pete Mouginis-Mark, Setsuya Nakada, Augusto Neri, Chris Newhall, Harry Pinkerton, John Power, Richie Robertson, Bill Rose, Mauro Rosi, Dave Rothery, Steve Self, Steve Sparks, John Stix, Brad Sturtevant, Don Swanson, Greg Valentine, Jim Vallance, Ben van Wyk de Vries, Barry Voight, Richard Waitt, Randy White, Stan Williams, Lionel Wilson, Andy Woods and Simon Young.

In Memorium PETER WILLIAM FRANCIS, 1944-1999 Professor of Volcanology, The Open University

Like many people, I first encountered Peter through his book Volcanoes. It was listed amongst about thirty titles that I was advised by my college to digest before going to university (I confess it was the only one of them I read). It accompanied me for several months on a trip to Indonesia - my dishevelled copy is annotated with comments on the characteristics of Javanese volcanoes that resembled those Peter described and illustrated. I had no idea that, five years later, I would begin my doctoral studies at the Open University under Peter's supervision. Numerous colleagues have related very similar stories of how Peter's books inspired them to follow careers in volcanology. Peter was born in 1944 in Mufulira, Zambia, where his father worked as a pharmacist. After his family resettled in England, he attended Bournemouth Grammar School and Reading School, and then studied at Imperial College, London, where he earned a first class honours degree in geology in 1966. While at Imperial, Peter went to a talk at the college's Expeditions Society given by a filmmaker, Tony Morrison, on the subject of a 'ghost' mining town high in the southern Bolivian Andes. The lecture fired the imagination of a small group, which put together an expedition to the region. Peter joined as Equipment Officer, and took responsibility for the medical kit. They set off for South America by sea in July 1966, taking a Landrover with them, and before long Peter was enjoying his first taste of remote volcanoes. They produced a weighty report on return, which included a contribution from Peter that reveals youthful manifestations of his well known sense of humour. In a section captioned 'Unnecessary Items' he admitted that: 'an overestimate on the part of the Equipment Officer led to rather too much toilet paper being carried. In this case, better too much than too little'. Under the heading 'Items we should have had', and following a list that includes pressure lights and a water pump, he reported that: 'certain members of the party acquired an obsession for looking at themselves in a looking-glass. Future expeditions are recommended to carry a full-length mirror to satisfy this desire'. Peter remained at Imperial for his doctoral research, which he completed in 1969. Supervised by Janet Watson, he wrote his thesis on the structural geology of the island of Barra in the Outer Hebrides of Scotland. Peter's adventurous, but sometimes accidentprone, character is typified by a story he used to tell of solo climbing in his field area (presumably in pursuit of some lofty rock samples). Negotiating a tricky cliff section, he lost his footing, but dropped short of the ground thanks to a rope securely attached to an overhang. He then swung helplessly and hopelessly for several hours before being able to reattach himself to the rock face. As an old friend of Peter's recalled, another of his tricks in this period was to coast his Landrover down the 260 metre descent into Castlebay without touching the brakes. He also took the opportunity to dive, retrieving items from Second World War wrecks, and enjoyed taking ammunition apart, so that he could dry the cordite and throw handfuls on to the stove for a bit of light relief while waiting out the autumn storms. Peter's flair for organizing fieldwork in remote and sometimes hostile environments was put to use on a further expedition to the Central Andes in 1970, this time to north Chile. He contributed his expertise in structural geology to a group that included George Walker. He was now truly engrossed by volcanology, and the research led to his early paper on the San Pablo and San Pedro volcanoes and their pyroclastic deposits (Francis et al. 1974). Recruited by Ian Gass to the Department of Earth Sciences at the Open University in 1971, Peter initiated geochemical studies of Andean volcanic rocks (chiefly in collaboration with Richard Thorpe) and made further contributions to the understanding of

ignimbrites. Always excited (if sometimes frustrated) by new technology, he was quick to recognize the potential of the first Landsat multispectral satellite images for mapping such vast and arid terrains as the Central Andes. By this means, he was the first to observe that Cerro Galan, in remote northwest Argentina, is the core of a resurgent caldera (Francis et al. 1983). He followed this discovery up by leading an Anglo-Argentinian expedition to the volcano in 1982, which included Steve Sparks. Steve later recalled driving across the fascinating and breathtaking altiplano landscape on the way up to the base camp, whilst listening to Peter's tape of Bach's St Matthews Passion, as a 'high point' in his life (anyone who telephoned Peter at home regularly will have spotted his enthusiasm for baroque music). The project received military support, and was the last joint operation between the British and Argentine armies before the Falkland Islands/Las Malvinas conflict. His name was subsequently splashed over the Argentine tabloid papers as the man behind 'COMPLOT, a fictional campaign of British espionage. But coming under suspicion for spying was nothing new to Peter. During fieldwork in Peru, he once spent three nights in police custody after being apprehended photographing some intriguing ash dunes of the 1600 Huaynatputina fall deposit, a little too close to a military base. He had with him a set of US aeronautical charts, the only maps available of the region, and for fear of what might happen if these were discovered, tore them to shreds which he slipped through the cracks in the floorboards. He was later moved to another station for detention, where he assisted the police by picking the lock for the main door, which they were unable to open. Flying above the freshly emplaced debris avalanche and blast deposits of the 1980 Mount St Helens eruption, the new landscape around the volcano reminded Peter of the hummocky geomorphology visible on satellite images of many Andean volcanoes. This led him to make important contributions to the understanding of catastrophic sector collapses and debris avalanche deposits. This included the first description (Francis et al. 1987) of what is arguably the best-preserved debris avalanche deposit in the world at Volcan Socompa in northern Chile. At this time, he also engaged in archival research on the tsunami generated by the mighty 1883 eruption of Krakatau (Francis 1985). In the 1980s, taking leave from the Open University, Peter held an appointment as Senior Visiting Scientist at the Lunar and Planetary Institute in Houston, Texas, where he studied Martian volcanism (e.g. Francis & Wood 1982; Francis & Wadge 1983), and was subsequently a Visiting Professor in the Planetary Geosciences Division of the University of Hawaii at Manoa until 1991. This led to several productive collaborations with US-based scientists. Peter became involved in the volcanology team for NASA's current Earth System Enterprise spaceborne mission. It was while in Washington DC for a NASA team meeting that he met Mary George, a senior British civil servant then seconded to the British Embassy, whom he married in London in June 1991 (while Pinatubo erupted!). By the mid-1980s, a new generation of Landsat satellites had arrived, providing higher spatial resolution and extending into the short wavelength infrared region. While surveying Andean volcanic summits for signs of recent activity using these new image data, Peter made the chance discovery of an infrared 'glow' within the summit crater of Lascar volcano in northern Chile in 1985 (Francis & Rothery 1987). Until then, this volcano had not been recognized as active but, one year later, its lava dome exploded, showering the distant town of Salta in Argentina with ash. The modern era of eruption detection and early warning from satellites had arrived.

OBITUARY

Fig. 2. Peter would throw on the brakes while driving around Montserrat if he saw decent-looking mango windfalls on the road. His Montserratian colleagues considered him eccentric for this, but he enjoyed supplementing the limited island diet of roti and goat stew with fresh fruit in this wav.

Fig. 1. Peter in the operations rooms at MVO, Mongo Hill. Peter was usually distinctively dressed for the job, though unusually in this scene he has not put his sandals over dark socks. In the early 1990s, Peter alighted on the volcanological potential of developments in infrared spectroscopy that had led to the manufacture of portable and robust equipment suitable for field use. Peter obtained a commercial Fourier transform spectrometer and initiated the most complete investigations carried out with this technique to date. These included field measurements in Italy, Nicaragua and Montserrat (e.g. Francis et aL 1998). Peter led a major ECfunded project to study gas and aerosol emissions from Mount Etna, which was completed shortly before his death. This brought together colleagues from Italy, France and the UK in a memorably convivial campaign that combined in-plume sampling, ground-based, airborne and satellite remote sensing studies, and outstanding dining on funghi porcini dishes in Nicolosi's fine restaurants. Peter was particularly enchanted by the simple beauty of rural Sicily, and was seldom happier than when telephoning Mary from the hilltop village of Novara di Sicilia, high in the Peloritan mountains, and commanding a magical view of the Eolian islands on a clear day. Peter carried out three tours of duty as Deputy Chief Scientist at Montserrat Volcano Observatory (MVO), and attended several scientific assessment meetings (Figs 1 and 2). He was on-island during the extraordinary period of cyclic activity in late 1997, and the 3 July 1998 dome collapse, and co-wrote the corresponding scientific reports (Young et al. 1997; Antenor et al. 1998). His contributions at MVO spanned research, surveillance and hazard management, focusing on many aspects including the crater wall stability, lahars, gas monitoring, and delineation of hazard zones. Many people came to respect him through the positive, considered and equitable way in which he engaged in scientific debate on the crisis. His decision-making was always informed by the immense breadth and depth of his expertise and reading, his practical knowledge of the vagaries and traits of volcanoes, and his irreproachable humanitarian instincts. He was renowned for his modesty, fairmindedness and level-headed judgement, even during times when egos obstructed the path of science, and when contentious issues raised the hackles of others. His contribution was described by one MVO scientist as the "true and calm voice of reason'.

Perhaps unsurprisingly, he had a love-hate relationship with Montserrat. He found the close proximity to an ongoing eruption utterly absorbing and deeply educational. I remember him remarking how he had never appreciated just how much of a filthy mess a volcano can make when it dumps ash on settled areas. He was by turns fascinated and frustrated by the interwoven scientific, cultural and political complexities to crisis management (Francis 1996; Aspinall et al. 1998). In scientific mode, the torpid pace of island life would get to him; off-duty, he would happily unwind on the verandah at Mongo Hill with a gin and tonic, or go for a world-class sunset swim at Little Bay. Peter w7as instrumental in setting up the long-term gas surveillance programme for MVO. Motivated by the need to complement geophysical and geodetic monitoring data, and his career-long interest in remote sensing techniques, he successfully applied for emergency funds from the UK Natural Environment Research Council in 1996 to purchase a Correlation Spectrometer for use by the observatory. Several scientific papers and PhD thesis chapters arose from the measurements of SO2 flux it obtained (Young et al. 1998; Watson et al. 2000; Oppenheimer et al. 1998c). He also initiated a more experimental programme of remote surveillance of Soufriere Hill's gas composition, using open-path Fourier transform spectroscopy (Fig. 3). The dataset revealed the ratios of sulphur to chlorine in the summit gas plume. It is now the longest running of its kind for any eruption, and several publications have arisen from the work (e.g. Oppenheimer et al. 1998a. b). This work informed Peter's broader perspective on degassing during dome-building eruptions, which he elaborated on at a conference on andesite volcanoes, convened by the Royal Society of London, shortly before his death (Francis et al. 2000). Peter was a co-organizer of the meeting. It was the last time many of us were to enjoy his company. Peter had a tremendous impact on the teaching of Earth sciences at the Open University. Always an innovator, he launched the OU's first Earth system science course. and became Director of Teaching for the Earth Sciences Department in 1996. In 1998. he was awarded the title of Professor of Volcanology in recognition of his research and teaching accomplishments. With a rare gift for writing (inspired by the clarity of Lawrence's prose in Seven Pillars of Wisdom) and a drive to communicate science to the wider audience, Peter was a great popularizer of volcanology and planetary science through books and magazine articles. His early Volcanoes book, published by Penguin in the 1970s (Francis 1976). was the only popular text on volcanoes at the time, and was avidly read by everyone who got involved at Mount St Helens. It turned many people on to a subject that had previously been considered

OBITUARY

Fig. 3. Peter measuring volcanic gases in Plymouth with the infrared spectrometer in 1996. just old rocks. He later revised it for Oxford University Press, whilst on leave in Hawaii (Francis 1993). Peter earned the respect and admiration of colleagues all over the world, and those of us who became his friends deeply miss his intellectual agility, generous spirit, wise counsel, and whimsical, usually provocative, sense of humour. He was an inspirational supervisor who ceaselessly nurtured the intellectual and career development of his research students and postdoctoral assistants. In his spare time, Peter was an amateur pilot, an accomplished sailor, an avid bookworm, and a keen walker and jogger. He loved aphorisms and mottos. One of his favourites was 'carpe diem'. While there was so much he still wanted to do, and so much more he would have contributed to volcanology, no-one could deny that he lived life to the full, along the way inspiring thousands of people with his books, lectures, and Open University courses, and, above all, enriching and changing the lives of Mary, family and friends. Another proverb that Peter used to delight in pointing out was 'man that is born of woman hath but a short time to live'. Perhaps he knew something. Clive Oppenheimer With special thanks and acknowledgements to C. Foster, T. Morrison, S. Self, D. Rothery, S. Sparks, G. Norton, R. Stewart, P. Cole, J. Barclay, B. Voight, W. Aspinall, G. Woo, P. Baxter, K. Rowley, and M. Francis. References ANTENOR, C., BONADONNA, C., FRANCIS, P., LUCKETT, R., ROBERTSON, R., NORTON, G., ROWLEY, K., WALKER, C., WATTS, R. & MVO STAFF. 1998. The events of July 3, 1998. MVO Special Report 7.

ASPINALL, W., FRANCIS, P., LYNCH, L., ROBERTSON, R., ROWLEY, K., SPARKS, S. & YOUNG, S. 1998. Scientists at the sharp end in a disaster zone. Nature, 393, 728. FRANCIS, P. 1976. Volcanoes. Penguin, Middlesex. FRANCIS, P.W. 1985. The origin of the 1883 Krakatau tsunamis. Journal of Volcanology and Geothermal Research, 25, 349-363. FRANCIS, P. 1993. Volcanoes: a Planetary Perspective. Oxford. FRANCIS, P. 1996. Volcanoes - Dangers hang over dependency. Nature, 383, 28. FRANCIS, P. W. & ROTHERY, D. A. 1987. Using the Landsat Thematic Mapper to detect and monitor active volcanos: an example from Lascar volcano, northern Chile. Geology, 15, 614-617. FRANCIS, P. W. & WADGE, G. 1983. The Olympus Mons aureole - formation by gravitational spreading. Journal of Geophysical Research, 88, 8333-8344. FRANCIS, P. W. & WOOD, C. A. 1982. Absence of silicic volcanism on Mars - implications for crustal composition and volatile abundance. Journal of Geophysical Research, 87, 9881-9889. FRANCIS, P. W., ROOBOL, M. J., WALKER, G. P. L., COBBOLD, P. R. & COWARD, M. 1974. The San Pedro and San Pablo volcanoes of northern Chile and their hot avalanche deposits. Geologische Rundschau, 63, 357-388. FRANCIS, P. W., O'CALLAGHAN, L., KRETZSCHMAR, G. A., Thorpe, R. S., SPARKS, R. S. J., PAGE, R. N., DEBARRIO, R. E., GILLOU, G. & GONZALES, O. E. 1983. The Cerro Galan ignimbrite. Nature, 301, 51-53. FRANCIS, P. W., GARDEWEG, M., RAMIREZ, C. F. & ROTHERY, D. A. 1987. Catastrophic debris avalanche deposit of Socompa volcano, northern Chile. Geology, 13, 600-603. FRANCIS, P., BURTON, M. & OPPENHEIMER, C. 1998. Remote measurements of volcanic gas compositions by solar FTIR spectroscopy. Nature, 396, 567-570. FRANCIS, P., HORROCKS, L. & OPPENHEIMER, C. 2000. Monitoring gases from andesite volcanoes, Philosophical Transactions of the Royal Society, 358, 1567-1584. OPPENHEIMER, C., FRANCIS, P. & MACIEJEWSKA, A. 1998#. Volcanic gas measurements by helicopter-borne fourier transform spectroscopy. International Journal of Remote Sensing, 19, 373-379. OPPENHEIMER, C., FRANCIS, P. & MACIEJEWSKI, A. 19986. Spectroscopic observation of HC1 degassing from Soufriere Hills volcano, Montserrat. Geophysical Research Letters, 25, 3689-3692. OPPENHEIMER, C., FRANCIS, P. & STIX, J. 1998c. Depletion rates of SO2 in tropospheric volcanic plumes. Geophsyical Research Letters, 25, 2671-2674. WATSON, I. M., OPPENHEIMER, C., VOIGHT, B, FRANCIS, P. W., CLARKE, A., STIX, J., MILLER, A., PYLE, D. M., BURTON, M. R., YOUNG, S. R., NORTON, G., LOUGHLIN, S., DARROUX, B. & MVO STAFF. 2000. The relationship between degassing and deformation at Soufriere Hills volcano, Montserrat. Journal of Volcanology and Geothermal Research, 98, 117-126. YOUNG, S., FRANCIS, P. W., BARCLAY, J., CASADEVALL, T. J., GARDNER, C. A., DARROUX, B., DAVIES, M. A., DELMELLE, P., NORTON, G. E., MACIEJEWSKI, A. J. H., OPPENHEIMER, C., STIX, J. & WATSON, I. M. 1998. Monitoring SO2 emission at the Soufriere Hills volcano: implications for changes in eruptive conditions. Geophysical Research Letters, 25, 3681-3684. YOUNG, S. R., COLE, P. D., CALDER, E. S., BAPTIE, B. A., BONADONNA, C., FRANCIS, P. W., HERD, R. A., JACKSON, P., LOUGHLIN, S. C., LUCKETT, R., NORTON, G. E., ROWLEY, K., SPARKS, R. S. J., WATTS, R. & MVO STAFF. 1997. Dome collapse and Vulcanian explosive activity, September to October 1997. MVO Special Report 5.

Setting, chronology and consequences of the eruption of Soufriere Hills Volcano, Montserrat (1995-1999) B. P. KOKELAAR Earth Sciences Department, University of Liverpool, Liverpool, L69 3BX, UK (e-mail: [email protected])

Abstract: The eruption on Montserrat during 1995-1999 was the most destructive in the Caribbean volcanic arc since that of Mont Pelee (Martinique) in 1902. It began on 18 July 1995 at the site of the most recent previous activity, on the flank of a c. 350-year-old lava dome within a sector-collapse scar. Phreatic explosivity occurred for 18 weeks before the onset of extrusion of an andesitic lava dome. Dome collapses produced pyroclastic flows that initially were confined by the sector-collapse scar. After 60 weeks of unsteadily accelerating dome growth and one episode of sub-Plinian explosivity, the dome eventually overtopped the confining scar. During 1997 almost two-thirds of the island was devastated following major dome collapses, two episodes of Vulcanian explosivity with fountain-collapse pyroclastic flows, and a flank failure with associated debris avalanche and explosive disruption of the lava dome. Nineteen people were killed directly by the volcanic activity and several were injured. From March 1998 until November 1999 there was a pause in magma ascent accompanied by reduced seismic activity, substantial degradation of the dome, and considerable degassing with venting of ash. The slow progress and long duration of the volcanic escalation, coupled with the small size of the island and the vulnerability of homes, key installations and infrastructure, resulted in a style of emergency management that was dominantly reactive. In order to minimize the disruption to life for those remaining on the island, following large-scale evacuations, scientists at the Montserrat Volcano Observatory had to anticipate hazards and their potential extents of impact with considerable precision. Based on frequent hazards assessments, a series of risk management zone maps was issued by administrative authorities to control access as the eruption escalated. These were used in conjunction with an alert-level system. The unpreparedness of the Montserrat authorities and the responsible UK government departments resulted in hardship, ill feeling and at times acrimony as the situation deteriorated and needs for aid mounted. Losses and stress could have been less if an existing hazards assessment had registered with appropriate authorities before the eruption.

The eruption of Soufriere Hills Volcano during 1995-1999 devastated the small Caribbean island of Montserrat, which is an Overseas Territory of the UK. Approximately 60% of the island, including the most densely populated districts, was designated unsafe for human habitation. Of the original population of c. 10 500, 92% suffered evacuations and many families were relocated two or three times. At the climax of the crisis, in 1997, almost 1600 people were accommodated in basic temporary shelters, and by early 1998 roughly 70% of the population had left the island. Most of the administrative, commercial and industrial facilities were destroyed or rendered inaccessible, as were the airport, harbour and prime agricultural land. Also lost was much of the verdant paradise that attracted tourists and numerous residential migrants from the North American winter, all of whom contributed significantly to Montserrat's economy. More than two-thirds of businesses were closed by October 1998. Insurance companies curtailed or withdrew cover as the eruption escalated in August 1997, which was just before most of the losses were incurred. Consequently, the local financial institution concerned with mortgages and savings collapsed. The Montserrat economy, only recently in budgetary surplus, was plunged back into dependency on UK financial aid. Unofficial insurance industry sources estimate that total losses could be as much as £1 billion if real estate is not recovered. Whereas health problems were less than in many other natural disasters with catastrophic onset (e.g. monsoon floods), the protracted emergency led to considerable psychological distress and related health problems for many Montserratians (Clay et al. 1999). Perhaps more challenging still for the future population will be the linking of disaster recovery with sustainable development (see Possekel 1999). The eruption and associated hazards escalated only slowly, step by step from 1995 through to 1997, and from the outset many inhabitants indicated a strong preference to remain on the island. Understandably, the Government of Montserrat wished to preserve life as near to normal as possible and to avoid jeopardizing the longterm viability of the island community. The UK Government policy was that people would be supported to remain on the island as long as there was a viable safe area. Given this scenario, a reactive strategy for emergency management was inevitable. The management strategy adopted was to react to changing levels of risk as they were identified, rather than immediate and complete evacuation to

an entirely safe area. Consequently, considerable importance was placed on scientific monitoring, hazard anticipation, risk assessment and communication of risks to officials and the public. There were no contingency plans. Many actions taken by both the UK Government and the Government of Montserrat were driven stepwise by events in the volcanic escalation. Initially, because there was no clear understanding of how the eruption might develop, much of the on-island emergency management involved solutions for the short term. Similarly, UK Government departments attempted initially to deal with the crisis using normal institutional arrangements. However, as the eruption escalated it became clear that some aspects of the handling of the emergency were unsatisfactory and that longer-term solutions were required. In 1997 the House of Commons Select Committee on International Development recommended an independent evaluation of the UK Government's response to the Montserrat volcanic emergency (IDC 1997). The terms of reference requested identification of key findings and lessons learnt. The present author was asked to evaluate the scientific monitoring and risk assessment, and to lay out the course of the volcanic developments against which the emergency management response could be charted. Aspects of this paper originate in work for the evaluation study, which was published recently (Clay et al. 1999). This paper is mainly to provide a factual record that sets the scene for the more analytical scientific contributions that follow. Plates 1-20 provide a pictorial narrative, principally concerning the characteristic styles of eruption and their effects. Some aspects of the handling of the emergency are analysed to distinguish problems and their possible solutions. Facets of the history of Montserrat are given (mainly tabulated), because they are relevant to understanding the plight of the people and their perceptions regarding both the handling of the emergency and the aid provided by the UK Government (see also Fergus 1994; Pattullo 2000). The judgements in this paper are those of the author. However, the author has benefited from full co-operation of the scientists involved, with generous provision of information and guidance. Although not explicitly woven into the narrative, it should be noted that, in addition to the core scientific team from the Caribbean and the UK, scientists from France, Puerto Rico and the USA made valuable and considerable contributions in the crisis management.

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 1-43. 0435-4052/02/S15 © The Geological Society of London 2002.

B. P. KOKELAAR Setting of the volcanic crisis Geological setting Montserrat is in the northern part of the Lesser Antilles volcanic island arc (Fig. 1). The arc results from westward subduction of Atlantic oceanic lithosphere beneath the Caribbean plate. From Martinique southwards to Grenada the arc comprises a closely spaced double chain of volcanoes, the eastern elements of which date back to the Eocene. From Martinique northwards an extinct eastern volcanic chain, of Eocene to mid-Oligocene age, diverges via Marie Galante to Sombrero, while Montserrat lies in the western active chain that extends to seamounts up to 100 km NW of Saba. The northern arc volcanoes are founded on a Cretaceous oceanic island arc (Bouysse & Guennoc 1983; Wadge 1986). Earthquake studies indicate that the subducted oceanic slab is segmented into three main parts with differing dips and slip vectors (Wadge & Shepherd 1984). Montserrat is above the northern segment, overlying crust that is no more than 30 km thick, an asthenospheric wedge that extends to 130km depth, and a Benioff zone that dips westwards at 50-60°. Montserrat is the top of a compound volcanic

edifice that extends from 1 km above sea level, at Soufriere Hills Volcano, to 700-900 m below sea level, where the basal diameter is c. 25-30 km. Building of a volcano here probably initiated in the Miocene (c. 9 Ma; Briden et al. 1979). when the axis of the northern part of the arc migrated westwards. However, the oldest exposed rocks are Pliocene (c. 2.6 Ma) and most are Pleistocene or younger in age (see Harford et al. 2002). Recent convergence of the Atlantic and Caribbean plates has been quite slow, at 20-40 mm a !. and magma productivity has consequently been low ( 3-5 km 3 M a - 1 km-1 of arc). especially in northern and southern parts of the arc (see Wadge 1984; M acdonald et al. 2000). Basalts of the northern volcanic chain are predominantly of low-K or medium-K type (low-K tholeiite and low-K calcalkaline; Rea 1982). with compositional trends through to andesite and dacite mainly controlled by polybaric crystal fractionation with limited magma hybridization. Montserrat is predominantly composed of porphyritic andesites. with basaltic rocks represented in volumetrically minor outcrops (South Soufriere Hills; Rea 1974) and common mafic inclusions in the more evolved rocks. The Soufriere Hills andesites erupted during 1995-1998 clearly implicate basalt in their petrogenesis and ascent. Experimental and petrological

Fig. 1. Location of Montserrat in the Lesser Antilles volcanic island arc (modified from Wadge 1986).

SETTING AND CHRONOLOGY OF THE ERUPTION

Fig. 2. Map showing Montserrat as it was before the eruption, which initiated in July 1995 through Castle Peak (lava dome). studies show that basalt at 1050°C invaded and mingled with hydrous andesitic crystal mush (c. 60-65% crystalline with interstitial melt containing c.4-5wt% H2O) at c. 830-860°C, heated it, and then erupted with it as inclusions (Barclay et al. 1998; Devine

et al. 1998; Murphy et al. 1998, 2000). These studies, together with analyses of seismic and deformation signals of conduit processes (Aspinall et al. 1998; Mattioli et al. 1998; Voight et al. 1999), are consistent with tapping of a long-lived reservoir >6 km below the

Fig. 3. Maps of (a) prehistoric fans of pyroclastic and lahar deposits of Soufriere Hills Volcano (modified from Roobol & Smith 1998). showing these as preferred sites for homes, key installations and infrastructure, and (b) the extent of areas devastated by pyroclastic flows during 1995-1999 (after Cole et al. 2002).

SETTING AND CHRONOLOGY OF THE ERUPTION vent. The form of the reservoir, however, is not well defined and the controls and duration of its replenishment are only just beginning to be understood (see Murphy et ai 2000). Cyclicity of seismic and magmatic activity detected on scales of about six to seven weeks can be related to processes in the conduit and magma chamber (Denlinger & Hoblitt 1999; Melnik & Sparks 1999, 2002; Voight et al. 1999; Wylie et al. 1999). The recurrence interval of approximately 30 years for volcanoseismic crises at Montserrat (Table 1) seemingly reflects the frequency of substantial perturbation of the magmatic system by influx of basalt from greater depths. The understanding of this volcanic plumbing system behaviour is an exciting prospect for the future, with particular significance for possible quantitative forecasting of eruptions. Physiography Montserrat is a small island, approximately 16.5km north to south and 10km east to west (c. 100km 2 ; Fig. 2, Plates 1,2). Its topography is dominated by four main volcanic massifs, each with many valleys and ridges radiating towards and truncated at a coastline predominantly of steep cliffs. The massifs, from north to south, Silver Hill (403m), Centre Hills (740m), Soufriere Hills (preemption 914m at Chances Peak) and South Soufriere Hills (756m), each represent composite eruptive centres, mainly of andesitic lavas, although deep erosion has modified most original volcanic landforms. The less substantial St George's Hill, Garibaldi Hill and Roche's Bluff mainly comprise volcaniclastic deposits. New representative 40 Ar/ 39 Ar age determinations are presented by Harford et al. (2002). Soufriere Hills Volcano, which is the youngest centre, retained little-modified primary features in its sector-collapse scar (English's Crater) and Castle Peak lava dome within (see Fig. 2, Plate 2B), although these are now substantially obliterated. Thermal waters found widely on Montserrat, including the numerous hot springs and fumaroles (soufrieres) on the flanks of Soufriere Hills Volcano (main ones labelled 1-4 in Fig. 2), reflect a sustained deep supply of both magmatic heat and volatiles to overlying aquifers (Chiodini et al. 1996; Hammouya et al. 1998). The gentler slopes flanking the volcanic massifs in the southern two-thirds of Montserrat are composed of volcaniclastic deposits from Soufriere Hills Volcano and were the sites favoured for habitation and infrastructure (Figs 2, 3a, Plates 1, 2A). The capital town, Plymouth, and its environs on the west coast, the airport on the east coast, and numerous communities in between on the northern flanks of the volcano, were all built on incised prehistoric fans primarily of pyroclastic flow and lahar deposits from the volcano (Plate 1A, B; see Roobol & Smith 1998). Montserrat's climate is maritime subtropical. In the period from 1992 to 1997, winds towards the west tended to prevail at low levels (1-5 km altitude) and high levels (20-30 km), and towards the east at intermediate levels (8-18 km), with standard deviations for directions over the 30km of altitude of 30-162 (Bonadonna et al. 2002). During 1997, ash plumes from dome-collapse pyroclastic flows and Vulcanian explosions commonly ascended to 15 km, and tephra was dispersed by intermediate-level winds, at different times, towards the north, NW, NE, south, SW and SE. Average rainfall ranges from 1 ma" 1 near sea level to >2.5ma - 1 in the hills (Possekel 1999). Torrential rain is associated with hurricanes that all too frequently track northwestwards through the eastern Caribbean. Before the 1995 eruption the vegetation in the hills was mainly secondary forest (little indigenous forest remaining), whereas on the less steep slopes and volcaniclastic fans there was mainly bush or cultivated land (Plates 1, 2, 3A). It was Montserrat's originally lush and exotic vegetation that earned it the epithet 'Emerald Isle of the Caribbean', recalling the verdant homeland of the early Irish colonists. Deforestation and inappropriate land use, however, coupled with the torrential rain, left many slopes eroded and susceptible to landslides. Montserrat's small size and predominantly rugged terrain severely constrained on-island options for volcanic risk mitigation. The location of most human activity and infrastructure in highly vulnerable areas maximized the impact of the eruption (Fig. 3b)

5

A brief history of Montserrat leading to the eruption during 1995-1999 In 1998 Montserrat became a UK Overseas Territory, having previously been a UK Dependent Territory. Although tragically the eruption had just rendered Montserrat once more dependent on UK financial aid, the change of title constituted one further advance from a history of some 300 years of British colonial status, commercial exploitation and slavery (see Fergus 1994). Basic features of social justice were secured only quite recently. Power invested in a Montserratian Chief Minister with a ministerial government dates from 1961, although issues concerning national security and international relations remain the business of the Governor of Montserrat, who is answerable to the UK Government. The duality of governance of Montserrat caused some problems in the management of the volcanic crisis, as described by Aspinall et al. (2002). Table 1 gives key developments in Montserrat's emergence into a free and economically viable small-island nation. Alongside this are charted the volcanic activity and volcanoseismic surveillance, both on Montserrat and on other Caribbean islands, as they bear on Montserrat's preparedness for the Soufriere Hills eruption. Prior to the devastation inflicted by Hurricane Hugo in 1989, Montserratians had acquired good standards of accommodation, education and health-care services, and, with UK aid following the storm, they were on the verge of almost complete recovery when disaster struck again. Chronology, nature and nomenclature of the volcanic crisis The 1995 eruption of Soufriere Hills Volcano involved a slow, incremental escalation of volcanic activity and associated hazards, after several years of precursory seismic activity. With the small size of the island, and with the population located mainly on the flanks of the active volcano (Fig. 3a, Plates 1, 2A), the slow escalation caused several distinct problems in emergency management. These are outlined both in the following narrative chronology of events and in succeeding sections. Key developments of the eruption, emergency responses and effects on the population of Montserrat are listed in Table 2 (see also Plates). Precursors Increased seismicity in the vicinity of Montserrat was initially detected in April 1989. Eighteen low- to moderate-intensity swarms of volcanotectonic earthquakes close to Soufriere Hills Volcano were registered at intervals from January 1992 and particularly from mid- to late 1994, before the first phreatic explosion on 18 July 1995 (Ambeh & Lynch 1996; Aspinall et al. 1998; Table 1, Fig. 4). Hot springs and fumaroles (soufrieres) on the volcano flanks showed little change prior to the eruption, although in March 1995 Galway's Soufriere showed pronounced magmatic signatures in 3 He/ 4 He and 13C, like those subsequently measured in gas from the andesitic lava dome (Hammouya et al. 1998). The unrest was on schedule according to the previously recognized c. 30 year cyclicity of volcanoseismic crises at Montserrat (Table 1; Wadge & Isaacs 1988), but none of the detected precursors was a clear, unambiguous indicator of an imminent eruption. Styles of eruption and pyroclastic fallout Following an opening phreatic phase, most of the eruption from 1995 to 1998 involved slow, unsteady ascent and extrusion (0.511 m3 S-1) of andesitic magma of high to extremely high viscosity (c. 106 to >10 I4 Pas), which formed a composite lava dome comprising numerous shear lobes (Sparks et al. 1998; Voight et al. 1999; Sparks & Young 2002; Watts et al. 2002). However, magma also erupted explosively from the conduit on several occasions, including two protracted intervals. On the first occasion (17-18 September 1996), during sub-Plinian explosive activity, fragmentation in the

B. P. KOKELAAR

6

Table 1. Historical development of Montserrat and the region leading to the 1995 eruption Occurrence

Volcanic activity

c. 3950 BP

English's Crater forms (Roobol & Smith 1998).

Sector collapse.

c. 3000 BP 1493

South American Amerindians first settle on island. Columbus sails along west coast of island (apparently deserted), and names it Santa Maria de Monserrate after an abbey in mountains near Barcelona (Spain) where a similar rugged profile occurs.

1500s to early 1600s; likely 1620s

Castle Peak dome forms in English's Crater with several pyroclastic layers deposited (Young et al. 1996, 1998; S.R. Young pers. comm. 2000).

1632

Irish Catholics first colonize as religious refugees from nearby St Kitts. shortly afterwards joined by Irish Catholic dissidents from Virginia.

Mid- 1600s

English colonial control established. Arrival of exiles from Ireland (deported by Cromwell) and transportees (criminals). African slaves imported, mainly to work in sugar plantations. Population comprises Anglo-Irish plantation owners, poorer Irish servants and increasing numbers of slaves. Frequent raids by French and Caribs. French capture island: restored in Treaty of Versailles.

Date

Late 1600s- 1700s 1782-1783 Early 1800s 1834

1838-

1866

1897-1898

1902 1933-1937

1936 1951 1959 1961

1966-1967

1967

1971-1972

1976 1979 1981 1987 1988 April 1989

Slave population exceeds 6500. Abolition of slavery by Act of UK Parliament. Full emancipation of slaves. Former slave population continues to struggle to establish independent peasant culture, being compromised by land-lease arrangements (share-cropping) and political control by whites. UK tightens control by establishing Crown Colony rule. Governor (appointed in UK) heads Legislative Council with six members appointed by him. New hot springs and fumaroles (Gages Lower Soufriere) initiated on volcano flank. Eruption at St Vincent kills c. 1500 persons. Eruption at Martinique (Mont Pelee) kills c. 29 500. Perret makes observations 1934-1937 (Perret 1939); Royal Society expedition in 1936 consequent upon petition following destructive earthquakes in 1935 and concern about possible major eruption (MacGregor 1938). Gages Upper Soufriere reactivated.

Andesitic lava and ash eruption(s).

Volcanoseismic crisis.

Volcanoseismic crisis.

New constitution includes four elected members of Legislative Council. Universal adult suffrage introduced. Share-cropping ended. Political power wrested from white merchant-planter class. Ministerial government established and led by first Chief Minister. Governor retains responsibility for national security, civil service and foreign relations. Galway's and Tar River Soufrieres become more active. Movement of magma from > 10km depth inferred (Shepherd et al. 1971). Affiliation with West Indian Federation rejected, effectively reaffirming colonial status; Montserrat becomes a UK Dependent Territory. Sale of 600 acres of prime land to North Americans and Europeans initiates relative economic boom.

Volcanoseismic crisis.

Eruption at St Vincent. On neighbouring Guadeloupe, phreatic explosions at La Soufriere lead to evacuation of 72000 persons (Fiske 1984); estimated cost c. £200 million. Eruption at St Vincent. Montserrat no longer in need of UK budgetary aid. Wadge & Isaacs (1987) report on hazards due to Soufriere Hills Volcano submitted to sponsors, noting Plymouth to be vulnerable. Lesser Antilles Volcanic Assessment Seminar hosted by Seismic Research Unit (Trinidad) and attended by Montserrat government representative(s). Wadge & Isaacs" (1988) findings published in an international journal. Seismic Research Unit prompted to deploy second and third seismic stations at Montserrat.

Seismic activity escalates above background. continued

7

SETTING AND CHRONOLOGY OF THE ERUPTION Table 1. (continued] Date

Occurrence

17 September 1989

Hurricane Hugo totally destroys 20% of homes, with 50% severely damaged; nearly 25% of population homeless. Plymouth devastated; total damage estimated at £150 million. 1 1 persons killed. £3 million from UK as emergency aid. Average wind speeds c. 240 km h^1 (c. 67ms~') with gusts over 300 km h" 1 . All three seismic stations destroyed. £16.8 million capital aid programme approved in UK. New government headquarters, library and hospital, all built in Plymouth. Seismic stations restored by Seismic Research Unit.

1991

January 1992 July 1993

Mid- 1994

End Nov.-Dec. 1994

Volcanic activity

Earthquakes occur in distinct swarms. Hypocentres located up to 1015km depth (Robertson et al 2000). Governor assists Montserrat Government in initiating upgrade of disaster preparedness. Three additional seismic stations established. Direct links of two stations to Seismic Research Unit, via Antigua, restored. Hypocentres up to 10-1 5 km depth. Head of Seismic Research Unit gives public interviews to reassure population concerning the earthquakes.

Early 1995

Data from six seismic stations telemetered to Emergency Operations Centre in Plymouth, with possible events forwarded to Seismic Research Unit. National Disaster Action Plan (manual) delivered, with virtually no reference to volcanoes. Potato crops fail on volcano flank.

18 July 1995

Early mid-afternoon: jet-engine-like roaring noises, sulphurous smell and ash fallout. Population of Montserrat c. 10500.

Start of volcanoseismic crisis.

Increasing seismicity.

Intense swarm of felt earthquakes.

Soufriere Hills Volcano erupts.

Table 2. Progress of volcanic activity with related emergency responses and effects on the population of Montserrat Date

Volcanic activity

Response

18 July 1995

Onset of eruption.

Emergency Operations Centre activated in Plymouth. On-island population c. 10500.

28 July 1995

Military contingency evacuation plans completed both for removal to north and off-island. Long Ground temporarily evacuated. First major evacuation of 6000 from southern and eastern areas, which lasted for 2 weeks.

21-22 August 1995

Major phreatic explosions.

Mid- to late November 1995 1-2 December 1995

Onset of lava-dome growth.

Long Ground and White's Yard evacuated.

Onset of minor pyroclastic flows.

Second major evacuation of 6000, which lasted for 1 month.

January 1996 21 March 1996 31 March 1996 3-4 April 1996

Onset of major pyroclastic flows.

April 1996

Civilian contingency evacuation plans (Operation Exodus) initiated. Government of Montserrat confirms acceptance of budgetary aid conditions. On-island population c. 9000 Governor declares state of public emergency. Plymouth and southern areas evacuated finally. Population in temporary shelters 1366 (gradually declined until August 1997). Voluntary Evacuation Scheme gives evacuees leave to remain in UK for 2 years. Risk management zone map introduced (Fig. 5a). On-island population c. 7500. £25 million aid for 2 years agreed.

May 1996 August 1996

Pyroclastic flows reach the sea.

September 1996

First major magmatic explosion; ballistic Revised risk management zone map issued, dated October (Fig. 5b). blocks >1 m diameter wreck Long Ground. First Galway's Wall crisis. Revised risk management zone maps issued dated November and December (Fig. 5c, d). Red Alert requiring further evacuations (19 December) largely ignored. Dome material overtops Galway's Wall On-island population c. 6000. Revised risk management zone map for first time. issued (Fig. 5e).

End November into December 1996

February 1997

continued

B. P. KOKELAAR

8

Table 2. (continued) Date

Volcanic activity

Response

May 1997

Dome growth switches to north and escalates.

Early to mid-June 1997

Increasing dome-collapse and pyroclastic flow activity in northern drainages.

25 June 1997

Pyroclastic flows kill 19 persons and injure 8. Surge-derived pyroclastic flow unexpectedly reaches vicinity of Cork Hill.

Population in temporary shelters 775: some residents still refuse to evacuate high-risk zones. Revised risk management zone map issued 6 June (Fig. 5f ): increased risk at airport is explicit. Emergency jetty handed over to Government of Montserrat. Cork Hill evacuated; airport and Plymouth port closed. Search and rescue initiated.

27 June 1997 4 July 1997

Revised risk management zone map issued, requiring further evacuation of western areas (Fig. 5g). Large pyroclastic flows frequent and encroaching Plymouth.

Revised risk management zone map issued (Fig. 5h). abandoning microzonation and designating all hazardous areas as Exclusion Zone.

First series of (13) cyclic repetitive Vulcanian explosions and associated radially directed fountain-collapse pyroclastic flows.

1 160 persons in temporary shelters. £6.5 million emergency housing scheme announced to accommodate 1000 in north of island. Evacuation of areas just north of Belham River. 1598 persons in temporary shelters. Formal assessment by MVO presented to UK Government. UK Foreign Secretary establishes inter-departmental Montserrat Action Group with Ministerial and Cabinet Office monitoring. Assisted Passage (to UK) Scheme announced, to aid relocation.

July 1997 August 1997

September 1997

22 September-21 October 1997

Second series of (75) cyclic repetitive Vulcanian explosions

November 1997 2-5 December 1997 26 December 1997

Revised risk management map issued places Salem and Old Towne in Exclusion Zone (Fig. 5i). Chief Minister visits London securing commitments to aid development of northern Montserrat. Frequent ashfall causes extreme nuisance: many remaining islanders decide to leave. On-island population 3338 Scientists meet in Antigua to produce formal assessment for UK Government: validated on 19 December by Chief Scientific Adviser.

Galway's Wall sector collapse and violent pyroclastic density current.

Early 1998 February 1998

Heavy ashfall in Central Zone.

March 1998

Cessation of magma ascent.

20-21 April 1998 21 May 1998

On-island population 3000. Recommended evacuation of Central Zone generally not heeded. UK Foreign Secretary visits Montserrat. Robbery of bank vault in Plymouth constitutes most significant opportunistic crime of the emergency. UK Government spend on aid related to the volcanic emergency totals c, £56 million Scientists meet in UK to produce formal assessment for UK Government. Evacuees allowed to settle indefinitely in UK.

11 June 1998

UK Government commitment of £75 million over 3 years to 2001. and indicative £25 million for 2001-2002.

14-16 July 1998

Scientists meet on Montserrat: formal assessment confirms lower risk levels.

30 September 1998 October 1998

Revised risk management map issued (Fig. 5j). Phased reoccupation of areas north of Belham River allowed. 427 people still housed in shelters. Montserrat Sustainable Development Plan published. Inquest verdict on deaths of June 1997 published (11 January); it criticizes both UK and Montserrat Governments. Montserrat Country Policy Plan agreed. Revised risk map issued allows Daytime Entry north of Plymouth (Fig. 5k). Assisted Return Passage Scheme (from UK) begins. On-island population recovered to c. 4500.

November 1998 January 1999 12 April 1999 1 May 1999 Mid-1999 November 1999

Lava dome substantially degraded by collapses; ash-venting frequent.

Renewed dome growth.

SETTING AND CHRONOLOGY OF THE ERUPTION

Fig. 4. Hypocentres of earthquakes that occurred in early to middle stages of the Montserrat volcanic emergency (after Aspinall et al. 1998). conduit may have descended to depths close to the magma reservoir (Robertson et al. 1998). By March 1998 a total cumulative volume of 0.3 km3 of magma had been erupted (Watts et al. 2002). From March 1998 until November 1999 there was a pause in magma extrusion,

9

during which time the dome became substantially reduced by collapses and ultimately divided by a deep chasm (Norton el al. 2002). The renewed extrusion from November 1999, which is not dealt with in this Memoir, is considered as forming a second dome (Watts et al. 2002; Sparks et al 2002). Thus the adopted convention is that there is one eruption, ongoing at the time of writing (April 2001), and only one dome formed until November 1999. The growth and partial collapse behaviour of this first dome, however, was extremely varied and on several occasions involved rebuilding from the mouth of the volcanic conduit. The eruption during 1995-1999 involved five main styles of volcanic activity. Each characterized a distinctive phase in the eruptive history, but overlapped with other styles. Phreatic explosions. These were produced by sudden and/or sustained jet-like releases primarily of heated groundwater. They blasted out mainly old volcanic rock, forming small craters, and characterized the opening phase of the eruption (see Plates 2B, 3). A powerful explosion on 28 July was associated with the opening of a new vent and another, on 21 August, produced a slow-moving, cold, dilute pyroclastic density current that precipitated the first evacuation of Plymouth (Table 2). Technically, if the explosions produced ash that included fragments of new (juvenile) magma, the explosions would be referred to as phreatomagmatic. Although juvenile fragments may have been included (e.g. Boudon et al. 1998), this was not clearly demonstrated. The term "phreatic phase' applies to this early activity as it registers the significant involvement of groundwater, irrespective of whether explosions ejected any juvenile material. The heat source is inferred to have been newly arisen magma and associated released volatiles (Gardner & White 2002). Lava-dome growth with dome-collapse pyroclastic flows. Extrusion of andesitic lava, at a rate mainly in the range 0.5-11 m3 s"1 (Sparks et al. 1998; Sparks & Young 2002), formed a steep-sided composite dome comprising numerous shear lobes and spines (e.g. Plates 4, 7, 17). Partial dome collapse due to gravitational instability commonly produced, in genetic terms, dome-collapse pyroclastic flows (Plates 5B, 6A, 8). Watts et al. (2002) document the dome growth, while Calder et al. (1999, 2002) tabulate volumes of deposits formed from dome-collapse pyroclastic flows, their runout parameters and the dates of occurrence. Cole et al. (2002) present sedimentological analyses of the pyroclastic deposits and their parent flows. To derive approximate dense-rock equivalent (DRE) volumes of andesite from deposit or collapse-scar volumes, Calder et al. (2002) utilize bulk densities of 2 x 10 3 kgm~ 3 for deposits, 2.2 x 10 3 kg m~3 for the dome and associated carapace breccias and 2.6 x 103 kgm~ 3 for the andesite. In this paper, the volume of material given as having collapsed is that of the partially fragmental dome, with a bulk density of 2.2 x 103 kgm~ 3 . Minor collapses that formed the talus apron around the dome and had runouts of <0.51 km were designated as rockfalls (Plate 7B, C; Calder et al. 2002). In lithological terms, larger collapses produced high-concentration block-and-ash flows. These typically produced an over-riding lowconcentration ash cloud, part of which commonly flowed downslope as a pyroclastic surge while part ascended buoyantly to form a plume (Plates 5B, 8A). These phenomena characterized most of the eruption during 1996, 1997 and early 1998. A spectrum of collapse types and associated flows has been recognized (Cole et al. 1998; Calder et al. 1999, 2002; Sparks & Young 2002; Sparks et al. 2002), in part related to gas content of the collapsing lava. Because deeper parts of the dome tended to have a greater content and pressure of gas, larger collapses that involved such deeper parts yielded lava fragments that decrepitated more and with greater explosivity. Thus increasingly substantial associated pyroclastic surges and plumes were formed, as well as flows with longer runouts. This occurred in the extreme when a large part of the dome collapsed almost instantaneously due to flank collapse on 26 December 1997 (see below). Magmatic explosions, commonly with fountain-collapse pyroclastic flows. These followed substantial dome collapses and consequent depressurization of volatile-rich magma. Explosive disruption of the magma formed an eruptive jet, part of which ascended buoyantly as

10

B. P. KOKELAAR

Fig. 5. Risk management maps based on hazards assessments made by the Montserrat Volcano Observatory (see text and Table 2). Until July 1997. the maps were used in conjunction with an alert-level system that changed the advice and or access status relating to specific zones (see Table 3). Moving from Zones 4 to 1 or G to A (earlier maps) represents increasing risk. The May 1996 map (a) utilized field assessment and simulation (modelling) of pyroclastic flow behaviour on the volcano slopes (e.g. Wadge et al. 1998). The October 1996 map (b) primarily reflected MVO consensus on the hazards of dome-collapse pyroclastic flows, with enhanced risks from major magmatic explosions (as occurred on 17 September). The November and December 1996 maps (c and d) reflected concern for a possible collapse to the SW of Galway's Wall, with a major explosion, and the February 1997 map (e) registered a diminishment of this. The 6 June 1997 map (f) took account of the new threats from pyroclastic flows travelling northwards. The 27 June 1997 map (g) registered the aftermath of the 25 June tragedy and continuing volcanic escalation. The 4 July 1997 map (h) constituted acknowledgement of the continued escalation and need for rationalization of the management system for simplicity: the September 1997 map (i) took account of a possible explosion with ten times the intensity of that of 17 September 1996. The 30 September 1998 map (j) reflects the cessation of emplacement of new magma but persistent threat of pyroclastic flows. The 12 April 1999 map (k) recognized diminished eruptive energy, although substantial ashfall rendered the Daytime Entry Zone far from comfortable. a plume to several kilometres (up to 15km) and part of which commonly (not always) collapsed back to form fountain-collapse pyroclastic flows and associated pyroclastic surges (Plates 12A, 15). The first major explosive activity, during 17-18 September 1996,

produced a short-lived sub-Plinian eruption plume after Vulcanian vent clearance that showered ballistics 1.5m in diameter up to 2.1 km from the vent (Robertson el al. 1998). Two later episodes of cyclic Vulcanian explosive activity (August and September-October

SETTING AND CHRONOLOGY OF THE ERUPTION

11

Fig. 5. (continued) 1997) were associated with fountain-collapse pyroclastic flows that were directed radially down most valleys (Druitt et al. 20026). The relatively gas-rich nature of the magma was reflected in these flows being predominantly pumiceous (lithologically, pumice-and-

ash flows; see deposits in Plates 12B, C, 14B, C), with upper parts forming less dense pyroclastic surges and buoyant ash plumes. Catastrophic sector collapse with associated explosive dome disruption and pyroclastic density current. This was caused by

B. P. KOKELAAR large-scale structural instability of the volcano leading to sudden flank failure and depressurization of gas-rich dome lava. It was anticipated during November-December 1996 (Young el aL 2002), but occurred over a year later, on 26 December 1997, when failure of hydrothermally weakened flank rocks caused a debris avalanche (Voight el al. 2002), closely succeeded by a catastrophic flow from the explosively disintegrating unsupported dome (see Plate 16). The term 'pyroclastic density current' is adopted for the latter phenomenon (Sparks el aL 2002; Woods el al. 2002), recognizing that the catastrophic disintegration of the dome produced an exceptionally energetic particulate flow in which vertical and cross-flow changes in particle size and concentration were marked, but, at least initially, were probably not sharply gradational. The sector collapse is also referred to as the Boxing Day collapse, from the name given in the UK to 26 December, following Christmas Day, when gifts ('boxes') traditionally were given to trades folk. Ash-venting. This occurred periodically from the actively growing lava dome, but also characterized the phase after mid-March 1998 when there was no substantial extrusion of magma. It occurred when copious volatiles (mainly water, with SO2 and HC1) were released, transporting ash derived by magma fragmentation and conduit erosion. Some explosive releases were akin to Vulcanian activity and produced small-volume block-and-ash flow deposits with relatively long runouts (Norton el aL 2002). Pyroclastic fallout was associated with all five styles of activity and generally increased as the eruption escalated. Ash plumes lower than c. 5 km were commonly deflected westwards by the prevailing winds and substantial ashfall was frequent over the most densely inhabited western and northwestern parts of the island (Plates 8A, 14B, 18). Taller plumes, reaching to 15km altitude from explosive activity, produced fallout in many sectors through time owing to the variability of the wind direction between 8 and 18km altitude (Bonadonna el al. 2002). While the ashfall constituted a major nuisance and caused some secondary hazards (e.g. from roofcollapse and reduced visibility; Plate 18A, B), its effects were mainly in the risk domain of long-term health. Respirable cristobalite, abundant in the dome-derived ash, is a matter of concern regarding lung-scarring silicosis (see Baxter el al. 1999). Ballistic projectiles capable of causing serious injury or death (fragments larger than c. 50mm in diameter) fell mainly within zones that had been previously evacuated because of the possibilities of other hazards, e.g. pyroclastic surges. Almost half of the houses in Long Ground (Fig. 2; Plate 1C), c. 2km NE of the vent, were hit by ballistic blocks up to 1.5m in diameter from the initial Vulcanian explosion(s) of 17 September 1996.

Seismicily and ground deformation Monitoring of seismicity and ground deformation throughout the eruption in 1995-1999 was a major part of the scientific effort and was used to anticipate volcanic activity and thus to mitigate risk (Miller el aL 1998; Voight el al. 1998, 1999; Gardner & White 2002). The effects of earthquakes were trivial relative to those of the volcanic activity. Earthquakes centred beneath Montserrat during initial through to advanced stages of the eruption are shown in Figure 4. The dense clusters of hypocentres extending to 5 km below sea level (6 km below the vent of Soufriere Hills Volcano) preclude existence of a large magma body shallower than that. Earthquakes early in the crisis beneath St George's Hill (Fig. 4a), and in a linear array trending SW towards the volcano, are interpreted as due to localized stress relief triggered by changes in the Soufriere Hills magmatic plumbing system, rather than from peripheral movement of magma (Aspinall el al. 1998). Volcanotectonic earthquakes (with a distinct shear-wave component) characterized the opening phase of the activity (see also Gardner & White 2002), but these soon became subordinate to signals due to gas and magma movement (lacking identifiable shear waves), with hypocentres focused in a relatively narrow cylindrical volume mostly <4 km beneath the vent (Fig. 4c). As lava extrusion escalated, swarms of 'hybrid' earth-

quakes (sensu Miller el al. 1998) occurred, sometimes merging into periods of harmonic tremor. From December 1996. and particularly at the height of the emergency in May to August 1997, distinctive inflation-deflation cycles of 6-30 hours' duration, linked with seismicity, were recognized. Inflation at the volcano summit accompanied by hybrid earthquakes or tremor was followed by deflation, hybrid-earthquake quiescence, lava extrusion and rockfall or pyroclastic flow activity. The inflation climax and following deflation were marked by long-period seismic signals, due to resonance from high-pressure gas flowing in cracks, while degassing also was marked. Hybrid earthquakes virtually ceased with the temporary cessation of lava extrusion from March 1998 (see Miller el aL 1998; Voight el al. 1998, 1999; Sparks & Young 2002). Global positioning system (GPS) geodesy and electronic distance measurement (EDM) were employed at various times throughout the crisis, and proved deformation both in the near-field of the conduit and further afield due to fault movements (see Jackson el al. 1998; Mattioli el al. 1998; Shepherd el al. 1998; Sparks & Young 2002).

Emergency management Table 2 outlines the progress of the eruption against the responses in terms of emergency management. By early April 1996 most homes in southern Montserrat were evacuated. However, wholesale withdrawal was neither desirable nor readily achievable on the island. In order to manage access to the evacuated area, the administrative authorities in May 1996 published the first of a series of 11 risk management maps. These changed as the eruption escalated and then paused (Fig. 5; see below). The zones of each map reflected the nature and extents of hazards anticipated by the Montserrat Volcano Observatory (MVO; Aspinall el al. 2002). Up to July 1997, the maps depicted zones with limits placed according to fine judgements of the possible impacts of hazardous phenomena. This microzonation procedure was to minimize the disruption affecting those people living on Montserrat. There was a pressure to keep risk management zones as narrow and precisely located as possible. (1) because of the stresses caused by evacuations, access restrictions and losses of property. (2) because there was a general determination not to leave the island, and (3) because there was a political wish to keep life as near to normal as possible (although it was quite abnormal). The microzonation maps were used in conjunction with an alertlevel system (Table 3; see also Aspinall el al. 2002). which could alter the access regulations of the zones according to changing levels of risk. At a particular alert level, certain restrictions and or actions applied to the various zones. For example, with a change of alert level from Orange to Red, Zones C and D. previously with limited access only, would become exclusion zones and hence rapidly evacuated (see Table 3). To a certain extent the restrictions were treated flexibly by administrative officials and by the general public. If new scientific consensus found it necessary, information was provided by the MVO as a basis for the maps and or alert levels to be changed. In this process, during early stages of the eruption, it was in the main helpful that the Governor and Chief Minister attended many scientific meetings. At times, however, the changing regulations of the complex zone maps and alert system could be difficult to communicate to the public. Significantly prompted by the 25 June 1997 tragedy and the continuing escalation of volcanic activity (Table 2). the management system was modified, in July 1997. to a simpler and more conservative division of the island into an Exclusion Zone and a Northern Zone, with an intervening Central (buffer) Zone. Through the early stages of the eruption, the processes of achieving and communicating scientific consensus evolved towards formal quantitative risk assessment. From December 1997 there were periodic (about twice yearly) formal elicitations of international scientific expertise to assess hazards and report on risks (Table 2; Aspinall el al. 2002; Sparks & Young 2002). These involved Monte Carlo statistical treatments of eruption-scenario models and related uncertainties in expert opinions, and the

SETTING AND CHRONOLOGY OF THE ERUPTION

13

Table 3. Alert system for Montserrat, March 1999 Volcanic activity

Alert stage

Actions by administrators and general public*

Background seismicity with no new surface manifestation of volcanic activity. Low-level local seismic activity, ground deformation and mild phreatic activity.

0 (White) 1 (Yellow)

Dome-building in progress, periodic collapses generating rockfalls and occasional pyroclastic flows. Moderate level of seismic activity with no sudden changes.

2 (Amber)

Change in style of activity anticipated within a few days. Pyroclastic flows common with associated light ashfall. High level of seismic activity.

3 (Orange)

All zones occupied. Review and update emergency plans on an ongoing basis. Maintain readiness of key personnel, systems and procedures. Keep stock of critical supplies. Local evacuations may be necessary in Zone A. Zones B and C on standby. Zone A - No access. Zone B - Access limited to short visits by residents, officials and approved visitors with rapid means of exit. Zone C - Daytime-only visits by residents, for approved commercial activities and agriculture. Zone D - Day and night-time occupation by residents, high level of alert maintained. Zones E, F - Full occupation by residents with national contingency plan for evacuation in readiness. Zone G - Full occupation. Zones A B - No access. Zone C - Access limited to short visits by residents and workers with means of rapid exit. Zone D - Daytime occupation for essential services and agriculture, residents allowed access in daytime. Essential services operate with standby transport and evacuation plans in place. Zone E - Full occupation with high level of alert maintained. Schools operate with standby transport. Zone F - Full occupation by residents with contingency plan for evacuation. Warn of ashfalls in Zones E and F. Zone G - Full occupation.

Major dome collapse under way, with large pyroclastic flows and heavy ashfall. Explosive event possible if the activity continues.

4 (Red)

Montserrat Standing Operation Procedures for Red Alert in place. All schools closed as required. People with special needs removed from Zones E and F. Zones A, B, C, D - No access, rapid evacuation of all remaining persons. Zone E - Rapid evacuation. Warn of potential for gravel, pumice and ashfall. Zone F - Warn of potential for gravel, pumice and ashfall. Zone G - Full occupation.

Ongoing large explosive eruption with heavy ashfall.

5 (Purple)

Zones A, B, C, D, E - No access, rapid evacuation of all remaining persons. Zone F - Initiate evacuation. Warn of potential for gravel, pumice and ashfall. Zone G - Warn of potential for ash hazards.

* Recommended actions to minimize significant casualties. Zones as defined on the Volcanic Risk Management Map. structured approach constituted a significant advance in volcanic emergency management. It improved communication of crucial information to those decision-makers responsible for policies and actions regarding evacuations and/or various types of preparedness. A fully formalized assessment by the MVO in August 1997 served to focus the attention of the UK Government on the seriousness of the situation (Table 2). Nevertheless, quantitative assessments remain difficult for administrators to use in making decisions that involve risk to life, because in fact some uncertainty always remains and the cutoff for what is acceptable risk, decided by political and/or humanitarian considerations, is subjectively set and can change with circumstances. Indeed, the UK Government's Chief Scientific Adviser, in his appraisal and validation of the report of the comprehensive December 1997 scientific elicitation, initially expressed a degree of scepticism centred on the extent of uncertainty regarding the understanding of lava-dome eruptions. Had his view prevailed, it might have critically reduced the scientific contribution to the decision-making process in this particular case, when the safety of the north of the island was a problematic issue. Course of the volcanic crisis in 1995 On 18 July 1995 phreatic explosions and jetting of gases started from a vent that opened on the western flank of Castle Peak dome, in English's Crater (Fig. 2; Plates 2B, 3). The first fallout of ash was

towards the populated western coast of Montserrat. Personnel from the Seismic Research Unit (SRU) in Trinidad, who were primarily responsible for providing advice in such emergencies in the region, arrived on 19 July, and the US Geological Survey's Volcanic Crisis Assistance Team arrived on 25 July (see Aspinall et al. (2002) regarding scientific personnel). By 26 July, UK military personnel, who were in the region coincidentally, had formulated both onisland and off-island evacuation plans. In the first six weeks of the eruption other vents opened around and across the old dome, and the explosions became more energetic. An increasing flux of SC>2 peaked and then diminished (Robertson et al. 2000; Gardner & White 2002). On 21 August at 08:02 local time (LT), on what became known as Ash Monday, a particularly energetic phreatic explosion produced an ash plume that reached c. 3 km altitude, as well as a slow-moving, cold, dilute pyroclastic surge that enveloped Plymouth for approximately 15 minutes. Car headlights were virtually useless in the darkness there. The surge cloud appeared menacing (more threatening than the actuality) and many inhabitants became frightened and spontaneously evacuated the town. As there could be no guarantee that conditions would not rapidly become dangerous, an official evacuation was ordered. Areas south of Richmond Hill, including Plymouth (Fig. 2), were evacuated on 22 August (c. 5000 people), and, with explosions continuing, the entire area south of the Belham River was evacuated on 23 August (a further c. 1000 people). On 27 August a new vent opened on the north flank of Castle Peak dome as

14

B. P. KOKELAAR

explosions continued. There was controversy between the scientific teams regarding the safety of a return of the population to the south, but partial reoccupation north of Plymouth was permitted on 4 September in order to allow evacuees to gain shelter and prepare for Hurricane Luis. Many had been accommodated only in tents. Hurricane Luis struck on 5 September, with storm noise disrupting much of the ongoing seismic recording. Remaining evacuated areas, including Plymouth, were reoccupied on 7 September or shortly afterwards, while there were more phreatic explosions. The US Geological Survey's Volcanic Crisis Assistance Team departed on 10 September. On 25 September, a new, small (c.4 x 10 4 m 3 ) mound-like mass of lava with a central spine was confirmed to be growing on the SW side of Castle Peak dome (see Plate 3C; Robertson et al. 2000, fig. 3). This growth was thought to reflect either upheaval of old rock due to the shallow emplacement of new magma, or possibly the emergence of a cooled part of a new plug. Throughout the period from the initial explosions until 25 September, there was no clear evidence within the seismic activity of ascent of a substantial amount of new magma, and, similarly, ground deformation monitors in position at that time detected none. However, the flux of SO2 and the variety of seismic signals at shallow levels are taken to indicate that magma may already have migrated to a shallower level before and/or soon after the first phreatic explosion (Gardner & White 2002). This being the case, the upheaval of late September represented renewed ascent of magma. Between 25 September and 30 November (65 days), the volcano was mostly obscured by low atmospheric cloud and there was disagreement concerning the risk attributable to the presence of new magma. Scientists from the UK and USA, mindful of the rapid onset of dangerous activity at St Vincent (1979), Mount St Helens (1980) and Pinatubo (1991), were concerned that the risk on the volcano flanks might be considerable and they recommended evacuation of Long Ground, near to the open side of English's Crater (Fig. 2). Concern was also expressed that the probable existence and implications of the occurrence of new magma had not yet been communicated to Montserratians. On the other hand, the leader of the scientific team (from SRU) adopted a less precautionary stance, which was supported by the Chief Minister. This position was probably influenced by the considerable economic impacts of the evacuation of 72000 people from the flanks of La Soufriere on Guadeloupe in 1976, which was deemed, by some, to have been unnecessary (see Fiske 1984). The elderly and infirm were re-evacuated from southern Montserrat on 5 October. The airport was temporarily closed owing to ash fallout on 23 October, Plymouth was inundated with ash on 31 October, 4 November and 9 November, and southern villages were similarly affected in early November, all due to continued phreatic explosivity. The initiation of continuous lava-dome growth, on 15 November, was marked by remotely monitored deformation in the area of English's Crater and by intense hybrid earthquake activity at depths <3km beneath the vent (Aspinall et al. 1998; Kilburn & Voight 1998; White et al. 1998; see Fig. 4b). Eventually the vulnerable eastern communities of Long Ground and White's Yard were evacuated, on 29 November, and a new andesite dome was first observed on 30 November. It was incandescent and lay within the enlarged crater at the site of the initial (18 July) explosions. A minor collapse from the dome and associated minor pyroclastic flow (rockfall) occurred on 1 December, and by 2 December all sectors east, south and west of the volcano, including Plymouth, were again evacuated. Some 6000 people were relocated to the northern part of Montserrat.

Course of the volcanic crisis in 1996 While the new andesitic dome grew slowly (c. 0.5m 3 S-1; Sparks et al. 1998; Plates 4, 5A), the population returned to the southern and western areas of Montserrat on 1-2 January 1996 and to the eastern villages on 16 January. In mid-February the dome-growth rate had increased (c. 2 m 3 S - 1 ) and the stability of the western

volcano flank above Plymouth gave cause for concern (Gages Wall; Fig. 2, Plate 2B). However, in early March the focus of dome growth switched to the NE. where a steep unstable flank developed. On 27 March, dome collapse produced the first really significant block-and-ash flow. It travelled 1 km into the upper reaches of the Tar River valley, where trees were burned, and its associated ash cloud rose to c. 2km. On 27 and 28 March both the Governor and the Chief Minister were advised by volcanologists that Long Ground was at risk if part of a large pyroclastic flow or pyroclastic surge overtopped the northern wall of the Tar River valley (Fig. 2). or if there was a NE-directed explosion. They were advised to consider the immediate evacuation of the village. The leader of the scientific team, however, did not advocate this. On 31 March, at 20:43 LT. a substantial flow travelled some 1.5km and consequently Long Ground was evacuated early on 1 April. (See Cole et al (1998, 2002) and Calder et al. (1999) for details of all substantial dome-collapse pyroclastic flows and their deposits.) Collapses continued into early April (Plate 5B). Plymouth was evacuated for the third and final time on 3 April, and the remainder of the southern part of the island was evacuated on the following day. The plumes above the pyroclastic flows reached 9 km altitude, becoming a significant aviation hazard. On 12 May block-and-ash flows travelled 2.7km east of the growing dome, via the Tar River valley, and reached the sea for the first time. (Tar River valley is the general topographic depression within and east of English's Crater. The main drainage to the sea, initially followed by block-and-ash flows, was via Hot River; see Fig. 2 and Plates 2B & 5C.) Prevailing winds deflected the associated ash plumes westwards and northwestwards, adversely affecting Cork Hill and areas further north. Numerous large spines were extruded and their collapse formed flanking talus slopes. On 19 May, the first map depicting risk management zones was produced, based on hazards anticipated by the MVO (Fig. 5a). Zones were delimited according to judgements regarding the limits of possible domecollapse block-and-ash flow's and the pyroclastic surges that might detach from them. These projections were aided by computer modelling of pyroclastic flow runout (see Wadge & Isaacs 1988; Wadge et al. 1998). although detailed field examination of the topography was the main basis. During the period of late July to early September, several major dome-growth pulses ( 4 m 3 S - 1 ) were marked by increases in shallow seismicity and ground deformation determined by EDM (Jackson et al. 1998; Sparks et al. 1998). These led to protracted collapses that produced multiple or sustained pyroclastic flows down the Tar River valley and extended a pyroclastic fan some 400m into the sea (Cole et al. 1998. 2002). On 17 September 1996, nine hours of dome collapse(s) with pyroclastic flows (Plate 6A). followed by 2.5 hours of quiescence, led to the first major magmatic explosive activity at 23:42 LT. Overall the explosivity lasted c. 40 minutes and 3.2 ) x 106 m3 of new magma (DRE) was erupted, producing a sub-Plinian ash plume that reached between 11.3 and 15km altitude. Magma drawdown. 3-5 km into the conduit, probably approached close to the magma storage depth. In an initial Vulcanian vent-clearing phase, ballistic blocks up to 1.5m in diameter were thrown 2.1 km northeastwards, into Long Ground (Plate 1C). During the precursory collapse(s) approximately one-third of the dome (c. 11 x 106 m3) was removed and a pyroclastic surge destroyed the Tar River Estate House on the north flank of the Tar River valley (Plate 6; Robertson et al. 1998; Calder et al. 2002). Pumice clasts 50 mm in diameter fell 3 km north of the airport on the NE coast and near the NW coast (Olde Towne; Fig. 2). and some roofs collapsed beneath accumulated ash. Cork Hill was partially evacuated on 18 September and dome growth resumed on 1 October. As a result of these developments a second risk management map was issued in October (Fig. 5b). On 30 October, with seismicity intensifying again and a perception of increased risk WNW of the volcano, Richmond Hill was evacuated and Cork Hill schools w;ere closed. Near the end of November and into December, earthquake swarms developed as cracks opened in Galway's Wall, the steep southern flank of the volcano summit above Galway's Soufriere (Fig. 2, Plates 2, 5A). and numerous small cold-rock avalanches

SETTING AND CHRONOLOGY OF THE ERUPTION were shed from it (see Young el al 2002). The 17 September collapse scar had refilled (Plate 7A) and this activity was accompanied by internal dome growth against the wall. It caused concern that the southwestern volcano flank might collapse catastrophically with an associated laterally directed explosion, as had occurred at Mount St Helens (USA) in 1980. Such a collapse had the potential to generate a hazardous tsunami if a large mass of debris rapidly impacted the sea, with consequences for coastal Montserrat and possibly for nearby Guadeloupe (Fig. 1). This new increased risk to the southwestern flank resulted in two revisions of the risk management map (Fig. 5c, d) in November and December (the latter labelled Temporary Revision'). At an advanced stage during this Galway's Wall crisis, on 19 December, a Red Alert warning was issued with the consequent implication that large numbers of inhabitants should retreat further northwards, especially from the vicinity of Cork Hill (see Fig. 5c and Table 3). This warning originated automatically as an administrative stipulation according to terms of the alert scheme established earlier in the autumn. However, few people responded. External dome growth had previously resumed, on 11-12 December, and pyroclastic flows reached the sea again via the Tar River valley on 19 December, following a somewhat explosive collapse (these were the first flows with pumiceous fragments). The lack of any SW-directed threatening activity that was obvious for the public to see, both before and following the Red Alert, did not advance public confidence in the scientific advice that was given then, but it was not a simple matter for the MVO to prevent or rescind the warning. Galway's Wall did not collapse catastrophically until 13 months later (see below). Towards the end of December and into the New Year, rapid dome growth (Plate 7B, C) and increased seismicity were accompanied by 6-8 hourly cyclic inflation and deflation of the dome and nearby flanks (Voight et al. 1998, 1999).

Course of the volcanic crisis in 1997 During January 1997 the stability of Galway's Wall again gave cause for concern and several eastward-directed dome collapses sent more pyroclastic flows to the sea down the Tar River valley, the floor of which by now was mainly buried by pyroclastic flow deposits (Fig. 6a). On 10 February, dome material cascaded southwards over Galway's Wall for the first time and from the end of March through to 11 April a series of collapses produced pyroclastic flows down the White River valley. One on 11 April reached 4.1 km from the dome (Fig. 6b). A new risk management zone map was issued in February (Fig. 5e), reverting to the November 1996 version. However, in mid-May, dome growth switched to the north and collapses in the following month led to numerous minor pyroclastic flows and several large ones. These travelled down Tuitt's Ghaut as far as 2.8km on 5 June, Fort Ghaut as far as 2km on 16 June, and Mosquito Ghaut as far as 4km on 17 June (e.g. Plate 8A, B; data from Cole et al. 2002). Dome growth was sustained at >4m 3 S-1 (Sparks et al. 1998). A revised risk management zone map was issued on 6 June (Fig. 5f), explicitly showing the airport to be at increased risk in changing from Zone E to Zone C. By mid-June, the dome was approximately twice as large as it had been at around the time of the 17 September 1996 magmatic explosion(s). Its growth had increased to moderately high rates in April and May (Sparks et al. 1998), while the shedding of mass by collapse had not been commensurate. Nevertheless, successive pyroclastic flows travelled further and further down valleys in the north and east sectors, gradually filling the upper reaches with deposits and thus increasing the possibility that later flows might surmount barriers and spill into adjacent valleys (Cole et al. 1998, 2002; Calder et al. 1999). The increasing risk of pyroclastic flows impacting the airport was generally acknowledged, and on 16 June the airport was closed for the day owing to a surge in activity on the dome (Plate 8A, B). At this time activity at the volcano was characterized by cyclic episodes of hybrid earthquake swarms that

15

peaked with inflation, followed by deflation and relative seismic quiescence with increased dome growth and collapse-flow activity (Voight et al. 1998, 1999). SO2 emission (measured by COSPEC) increased in concert with intensifying earthquake and inflation activity (Watson et al. 2000). At 12:55 LT on 25 June 1997 a collapse of roughly 6 x 10 6 m 3 of the lava dome started. It produced a pyroclastic flow that travelled initially down Mosquito Ghaut and then mainly via Paradise River and Pea Ghaut (refer to Fig. 2). In three main pulses over a period of about 20 minutes it devastated villages in central and eastern areas, killed 19 people, and reached to within 200m of the airport terminal buildings, which were successfully evacuated and immediately closed (Loughlin et al. 20020, b\ see Plates 8C, 9 10, 11). The upper, dilute parts of the second and third pulses detached at a constriction and bend in Mosquito Ghaut and travelled as pyroclastic surges northwards and westwards across the gentle slopes around Farrell's Yard and towards Streatham, ultimately running onto Windy Hill (Plate 9A, B). Here seven people fleeing the rapid but silent advance of the searing clouds were killed (Loughlin et al. 20026). As the pyroclastic surges lost capacity they rapidly dumped much of their suspended load into a thin, dense, granular flow of ash that drained into Tyre's Ghaut (Plate 9A, B) and Dyer's River valley and then along the Belham River valley. This surge-derived pyroclastic flow (Calder et al. 1999; Druitt et al. 2002a) terminated in the vicinity of Cork Hill (Fig. 2, Plate 9C), close to the school but 50m topographically below it. The flows prompted evacuation of some 1500 persons from western areas, and a new risk management zone map was issued on 27 June (Fig. 5g). On 28 and 30 June, dome material avalanched over Gages Wall and pyroclastic flows encroached the outskirts of Plymouth. For two to threee days, pyroclastic flows swept down Mosquito Ghaut and Fort Ghaut regularly every 8-12 hours (Cole et al. 2002). Figure 6c shows the marked increase in the extent of impact by pyroclastic flows due to the May-June escalation. On 4 July, a new and simpler risk management map designated all western areas along and south of Belham River valley as Exclusion Zone (Fig. 5h) with a Central (buffer) Zone to the north. Through July, increased extrusion rates (5-1 Om3 s"1; Robertson et al. 2000) and collapses caused infilling of the upper reaches of northern and western valleys. On 3 August a major dome collapse (c. 8 x l 0 6 m 3 of dome material) formed pyroclastic flows that reached the harbour and destroyed much of Plymouth (Fig. 6d). This led to the first of two dramatic series of repetitive Vulcanian explosions, from a vent on the NW side of the dome (Druitt et al. 20026; Clarke et al. 2002). From 4 to 12 August, 13 magmatic explosions, mostly on a 10-12 hour cycle, produced eruption plumes up to c. 14km altitude as well as radially directed pumiceous pyroclastic flows and pyroclastic surges that formed by eruptive-fountain collapse. These travelled up to several kilometres down most flanks of the volcano (Plate 12). Scientific concern for further large, northward-directed collapses and increased explosivity resulted in another revision of the risk management map, dated September 1997 (Fig. 5i), and further northward evacuation. This involved Salem and Old Towne, substantial communities north of the Belham River (Fig. 2), and for many evacuees it constituted a third or fourth upheaval and relocation. The MVO itself, then in Old Towne, was moved to Mongo Hill, to the north of Centre Hills (Aspinall et al. 2002). On 21 September another major dome collapse (c. 13 x 10 6 m 3 of dome material) occurred. This had been to an extent anticipated, according to the emerging pattern from May 1997 of a six to seven week cyclicity (B. Voight pers. comm.). It produced pyroclastic flows NE of the dome that wrecked Tuitt's village, destroyed the airport and entered the sea nearby (Plates 13, 14A). The collapse led to the second series of Vulcanian explosions. This started on 22 September, lasted until 21 October, and involved 75 major explosions recurring on an average 9.5 hour cycle with plumes to between 3 and 15km altitude. These were mostly associated with radially directed pumiceous fountain-collapse pyroclastic flows (Plates 14B, C, 15; Druitt et al. 20026). A crater 300m wide was reamed out in a scar on the northern part of the dome at the

16

B. P. KOKELAAR

Fig. 6. Progressive inundation of southern Montserrat by pyroclastic flow deposits during 1995-1999 (modified from Cole et al. 2002). Each map shows the extent of deposits at the end of the time indicated (see text for details). The areas impacted by pyroclastic flows are almost entirely the same as those anticipated in the hazards assessment made by Wadge & Isaacs (1987, 1988) before Hurricane Hugo damaged many key installations in Plymouth. Despite the contrary advice in the assessment, key installations were rebuilt in Plymouth, only to be lost to the volcano.

SETTING AND CHRONOLOGY OF THE ERUPTION

location of the initial (18 July 1995) phreatic vent, ballistics landed as far as 1.6km away, and ash from the explosions was distributed over much of the northeastern Caribbean (Young et al. 1998). Figure 6e shows the increased extent of pyroclastic flows from this dome-collapse and explosion episode. Heightened activity in November was more confidently anticipated according to the six to seven week cyclicity (Sparks & Young 2002). Renewed dome growth rapidly filled the vent of the September-October explosions (at c. 7-8 m3 s- 1 ) and on 4 and 6 November dome collapses (c. 7x 106m3 of dome material) sent pyroclastic flows down the White River valley, building a significant fan at the coast (Fig. 6f). Following continued dome growth and increasing hybrid earthquake activity, the flank sector including Galway's Wall collapsed at 03:01 LT on 26 December (Boxing Day). A portion of the old edifice, with a large overburden of the new dome and its talus, detached at the hydrothermally altered and hence structurally weak level of Galway's Soufriere. The flank rocks and some dome talus formed an extensive debris avalanche that spread deposit along the lower reaches of the White River valley (Voight et al. 2002). The dome, suddenly unsupported, disrupted explosively to produce a violent pyroclastic density current with unconfined upper parts that swept devastatingly across a broad swath (c. 10km2) radially towards the SW (Plate 16A). The villages of St Patrick's and Morris' were all but obliterated (see Fig. 6g, Plate 16B, C; Sparks et al. 2002; Ritchie et al. 2002). The duration of the main collapse and associated pyroclastic density current was about 15 minutes, and the associated ash plume rose to c. 15 km altitude. The volume of explosively disrupted dome lava and talus (35-45 x 106m3; Sparks et al. 2002) was considerably greater than any previous collapse volume, and the amount of collapsed flank material, including a significant volume of dome talus, was also large (c.46 x 106m3). Much of the explosively disrupted dome debris entered the sea and a small tsunami impacted the shore at Old Road Bay to the north, at the mouth of the Belham River (Fig. 2).

Course of the volcanic crisis in 1998 In January the UK Government's Chief Scientific Adviser and Chief Medical Officer recommended in the strongest possible terms that everyone, but especially children and asthmatics, should leave the Central Zone, including Woodlands, near the west coast (Fig. 5i). This recommendation was based on their reading of a scientific assessment that had been validated by the Chief Scientific Adviser on 19 December 1997. It was founded on a perception of significant primary volcanic risk in this zone and the uncertainties concerning health deterioration due to protracted exposure to respirable ash. The recommendation, issued by the administrative authorities, was largely ignored. It was not practical to enforce an evacuation of the Central Zone (using emergency regulations) and still maintain a viable island community. This was because there was insufficient accommodation available further north, and because the area concerned included residences of key personnel and had by this stage of the crisis become the administrative and commercial centre. There really was nowhere left on-island for the personnel or various facilities to move to. The insufficient provision in the north of the island of accommodation for evacuees and storage facilities for businesses was a continuing problem in the emergency management. It was poignantly reflected in the large numbers of evacuees who inhabited basic temporary shelters for many months (Table 2), and in the considerable losses of capital assets and stocks not removed from the evacuated zone into storage. This insufficient provision was also held to be partly responsible for some of the deaths on 25 June (Inquest Report published January 1999). The situation that had evolved during the slow volcanic escalation reflected inadequate medium- to long-term foresight in UK and Montserrat government departments, and tardiness in implementation of emergency administration in the UK (Clay et al. 1999). (Establishment of the Montserrat Action Group by the UK Government's Foreign Secretary in August 1997 (Table 2) effectively altered this for the

17

better.) This was one occasion when the full implementation of emergency actions for risk mitigation according to scientific advice was not feasible, owing both to inadequate provision for relocation and to the likelihood that full implementation would render the continued function of the remaining community non-viable. The scars of the 26 December (1997) collapse were filled by early February, with the dome initially growing at an estimated 10-11 m3 s- 1 , but then more slowly (Sparks & Young 2002). By 10 March 1998, when the total (cumulative) volume of erupted magma was c. 0.3 km3, the ascent of new magma effectively ceased. A prominent summit spine took the final elevation to 1031 m a.s.l. (Plate 17; Norton et al. 2002). Subsequent minor collapses and pyroclastic flows were related mainly to gravitational stabilization of the slowly cooling and degassing dome. On 3 July a protracted collapse and pyroclastic flow down the Tar River valley to the sea was accompanied by an ash plume to 14 km altitude, with fallout of coarse ash and lapilli over Salem to the NW. The collapse involved roughly 22 x 106 m3 of lava and talus, removing about one-fifth of the dome, and was followed by an explosion that hurled ballistic blocks 1 km from the vent (Robertson et al. 2000; Norton et al. 2002). It left a deep elongate scar in the dome and a pyroclastic surge impacted Long Ground for the first time (see Fig. 6h). Two small collapses on 13 August left horseshoe-shaped scars on the dome and sent pyroclastic flows 1.8km down the White River valley, and on 16 August a pyroclastic flow reached the coastal fan below the Tar River valley (see Plate 17A). Emission of SO2 waxed and waned through August and September, with periods of vigorous degassing and venting of ash that correlated with lowamplitude seismic tremor (Robertson et al. 2000). Three small pyroclastic flows occurred in September and torrential rain associated with Hurricane Georges (20-21 September) formed largevolume lahars down the main drainages. The lahars encroached on the (abandoned) airport runway, further buried Plymouth and incised a new channel there, and extended the delta at the mouth of Belham River (Plates 19, 20). Ash-venting episodes recurred frequently on the north side of the dome in the period 26-30 September, along with increased SOa emissions. On 30 September a revision of the Exclusion Zone boundary (Fig. 5j) returned Salem and Old Towne to habitable status. A small lava spine (<10m) that appeared in October 1998 possibly registered a very minor extrusion. Collapses during mid- to late October left minor deposits up to 3km from the dome in several valleys and were accompanied by substantial venting of gas with entrained ash. Significant collapses occurred on 3, 5, 8, 9 and 12 November. That on 12 November sent an ash plume to 8km altitude and pyroclastic flows into the sea via the Tar River and White River valleys, as well as into Plymouth (Norton et al. 2002). The October and November collapses left a WNW-trending chasm, 150m deep, through the dome. Ash-venting occurred throughout December and collapses on 14 and 19 December were accompanied by significant explosivity and large ash clouds that rose rapidly to 6km altitude. Associated pyroclastic flows reached the sea again along the Tar River valley, producing loosely packed, gas-rich deposits (very high voidage). Explosions on 21 December blasted black jets of ash and blocks to heights of 80 m above the vent, after which vigorous ash-venting persisted for at least 30 minutes. The last significant event of 1998 was a minor collapse on 27 December. Ground deformation studies suggested that fault-block movement had occurred east of the volcano after March 1998.

Course of the volcanic crisis to mid-November 1999 Small explosions, vigorous ash-venting and minor dome collapses continued through January 1999, and by the end of the month the chasm had deepened so as to cut some 100m into the pre-1995 edifice (Norton et al. 2002). Explosive ash-venting characterized activity in the next two months, with plumes to 7 km altitude and some fountain-collapse pyroclastic flows. Ground deformation was slight. In April 1999 a revised risk management map (Fig. 5k)

18

B. P. KOKELAAR

reflected the perceived reduced risk of large-scale explosions or flows. The map removed communities just south of Belham River (lies Bay; Fig. 2) from the Exclusion Zone and established a Daytime Entry Zone north of Plymouth. However, substantial ash from explosions and continued collapses westwards into the catchment of Fort Ghaut (Gages valley) restricted access and posed serious problems for reoccupation. In May and June further explosions produced ash plumes to up to 9 km altitude and pyroclastic flows entered several drainages, reaching the sea via the Tar River valley. A moderate-sized collapse on 20 July left the remnant dome, estimated at 63 x 10 6 m 3 , split into three parts. Although six to seven week cyclicity of residual activity was discerned from March through to August, there were few hybrid earthquakes and diminishing exhalation of SO2, so that there appeared to be a general decline in eruptive energy. Hurricanes in September, October and November all produced major lahars. Explosions in the period 23-28 October produced plumes to 7.5km altitude. During 3-8 November, a swarm of hybrid earthquakes heralded renewed ascent of magma. Juvenile vesicular ash was erupted explosively on 8 and 9 November, a plug of old lava was extruded shortly afterwards, and new lava was emergent by 19 November. This was the initiation of the second dome of the eruption (as defined in this Memoir), which started to grow at rates comparable to those measured early in 1996, despite 20 months of magma stagnation and degassing. Scientific appraisal of the situation by December 1999 concluded that further protracted dome growth was likely, and hence that the pre-April 1999 Exclusion Zone-Safe Zone boundary would be valid for the foreseeable future. (At the time of writing (April 2001), with the large size of the second dome, it is conceivable that the boundary might have to be moved further north again if growth occurs in the western sector; G. E. Norton pers. comm.)

Preparedness for the Soufriere Hills eruption Despite considerably increased awareness in the Caribbean region of the hazards of volcanism, largely due to the efforts of the Seismic Research Unit in Trinidad and particularly following the eruptions in the 1970s at St Vincent and nearby Guadeloupe (Table 1), it is clear that in 1995 the governing authorities on Montserrat were not prepared for the Soufriere Hills eruption. Similarly, the branches of the Foreign and Commonwealth Office (UK Government) with responsibility for Dependent Overseas Territories were unprepared (Clay et al. 1999). In its first report on the Montserrat emergency (IDC 1997), the House of Commons Select Committee on International Development concluded (para. 20 iv): 'Many of the imperfections and difficulties in the delivery of aid are simply due to the complicated, ongoing and unpredictable nature of the volcanic activity. ... There are numerous lessons to be learned from events in Montserrat both from the viewpoint of science and of emergency planning'. The sixth report of the Select Committee (IDC 1998, para. 2), in the section The Unanswered Recommendations', noted that the Committee's findings with regard to the Wadge and Isaacs Report were 'ignored by the Government' (of the UK), and went on (para. 6) to urge 'that there are organisational lessons for the future that can be learned'. The Wadge and Isaacs report was an up-to-date hazards assessment for Montserrat. published in 1987, in which it was anticipated that Plymouth might be seriously impacted during an eruption. The tale regarding the lack of notice taken of the assessment does embody organizational lessons for the future, as well as pathos, and it is briefly reviewed here. A particularly poignant facet regarding the state of preparedness in 1995 concerns the fact that Montserrat had only recently recovered from devastation in 1989 by Hurricane Hugo, which post-dated the Wadge and Isaacs hazards assessment. With UK aid, Montserrat had rebuilt its key facilities in Plymouth (see Table. 1). When the volcano erupted, the new hospital was not yet fully commissioned, the new library was still unfinished, and the new government buildings were yet to see a full Legislative Council meeting. All were

lost to the volcano, as had been anticipated as a possibility in the hazards assessment. The Wadge and Isaacs report (1987) The study entitled Volcanic Hazards from Soufriere Hills Volcano, Montserrat, West Indies was commissioned in 1986 under the umbrella of the Pan-Caribbean Disaster Prevention and Preparedness Program and funded by the UK Natural Environment Research Council. It was commissioned on the understanding that Soufriere Hills Volcano was potentially dangerous. According to the report (Wadge & Isaacs 1987). the impetus for the study came from the Government of Montserrat, which wanted a full assessment to be made. Apparently there were plans to combine the hazards analysis with new census data to produce a risk assessment. The latter never materialized. The report (Wadge & Isaacs 1987) discussed a range of eruption scenarios, provided maps that implied devastation on various scales up to and including most of southern Montserrat (as actually occurred), made reference to the likelihood of an eruption, and advocated local emergency planning. The study utilized three main classes of data: (1) the distribution of prehistoric pyroclastic flow (and other) deposits on Montserrat (Fig. 3a); (2) age determinations of the pyroclastic deposits; and (3) computer models of pyroclastic flows down the (digital terrain) slopes of the volcano. The modelling simulated phenomena akin to the fountain-collapse pyroclastic flows of the August and September-October 1997 explosions, and predicted extremes of runout likely to be valid for dome-collapse pyroclastic flows. The existing deposits confirmed the extent of impact inferred by the modelling, and the age data yielded some information on prehistoric frequency. The focus on pyroclastic flow behaviour was highly relevant for Montserrat, because the favoured sites for the towns and villages, as well as the airport and farming sites, were on the gentle slopes of pyroclastic flow deposits derived from Soufriere Hills Volcano (Figs 2, 3. Plates 1. 2A). The report concluded that eruption emergency planning should allow for three types of eruptions: (1) a small eruption that would directly threaten Long Ground, which should be evacuated as soon as the eruption began; (2) a moderate to large eruption for which most of southern Montserrat should be evacuated according to priorities indicated in the sequential hazard zone map; (3) a collapsing dome lateral blast eruption - a very remote but dangerous possibility requiring immediate evacuation of the relevant 180 sector of the volcano. It went on to suggest that some consideration be given to strategies for mitigating the damage done to Montserrat by the loss during an eruption of the centralized facilities at Plymouth. Copies of the report were delivered to the Governor's office and to the Commissioner of Police on Montserrat, but not to the Foreign and Commonwealth Office in the UK. In 1988 a distillation of the report was published and distributed internationally (Wadge & Isaacs 1988). In 1989 one of the hazard maps was reproduced in a children's book in a series on World Disasters (Knapp 1989). The caption (abbreviated) reads: This computer map shows one way the volcano on Montserrat may erupt. It shows areas where people need to be evacuated. The government can also use these maps to show areas where it is safe to build in the future. Hospitals and control centers should all be placed away from danger spots'. The text states: The latest scientific techniques have been used to help the government of the Caribbean island of Montserrat to make plans in case there should be an eruption in the future. ... Scientists have pinpointed the likely danger spots so that evacuation plans can be prepared taking these spots into account'. Evidently, the Wadge and Isaacs report was never received in a way that allowed it to be used as advocated in the children's book. On 25 July 1995, one week into the eruption, the Governor of Montserrat and the leader of the scientific team from the Seismic Research Unit, in a telegram to the UK Foreign and Commonwealth Office, expressed astonishment that recommendations in the

SETTING AND CHRONOLOGY OF THE ERUPTION report had not been noted and also that there were no contingency plans. On Montserrat, losses of documents from the Governor's office and other government offices were attributed to Hurricane Hugo (Pattullo 2000). The regional organization charged with ensuring preparedness, which commissioned the report in 1986 (Pan-Caribbean Disaster Prevention and Preparedness Program), was found wanting after Hurricanes Gilbert and Hugo respectively devastated Jamaica in 1988 and Montserrat in 1989, and it was superseded in 1991 (by the Caribbean Disaster and Emergency Response Agency). Evidently, institutional memory can be as short as the period leading to a change of key personnel, organizational structure or government. It is conceivable that //the report had been read thoroughly by interested parties in Montserrat, its findings on the long-term recurrence probability of pyroclastic-flow inundation of Plymouth, of about 1% per century, might have been interpreted as acceptable odds justifying no action on rethinking the island's infrastructure. Similarly, a sentence near the end of the report, under Long-Term Planning - 'Soufriere Hills Volcano is not a very active volcano and it may be centuries before it erupts again on the scale requiring mass evacuation' - might have been interpreted as a rationale for doing nothing. The estimation of future-event probability was necessarily crude. It is nevertheless remarkable that the report's recommendations were not recollected when, in mid- to late 1994, earthquakes very obviously located beneath Soufriere Hills Volcano were detected and escalating. Although this escalation constituted no strong reason to expect an eruption (at the highest level of probability), evacuation plans, at least, should have been considered then. Former volcanoseismic crises of greater seismic energy had not led to eruption, but the key issue, recognized in the commissioning of the Wadge and Isaacs study in 1986 and so pertinent for the region, is one of disaster preparedness. Just before the eruption in 1995, a National Disaster Action Plan (handbook) was published (Government of Montserrat 1995). It dealt primarily with hurricanes, floods, earthquakes and 'mancaused disasters', and, surprisingly, contained only one sentence referring to the threat of volcanic eruption, on the basis of Montserrat being a volcanic island. This occurred despite the earlier impetus from the government that led to the Wadge and Isaacs study, and despite the presentation of volcanic hazards information to Montserratian officials by the Seismic Research Unit, via both seminars (in Trinidad 1988) and poster displays (in Plymouth 1994). It is regrettable that the Wadge and Isaacs study did not register fully in Montserrat or with the UK Government Department responsible for the administration of its Overseas Territory. Full consideration of it would: (1) have forced authorities in advance to evaluate the priorities and possible consequences of their actions or inaction, particularly with regard to rebuilding key installations in Plymouth; and (2) have provided them with enhanced forwardlooking capability at the onset of the eruption. Montserrat was less prepared than it might have been had the Wadge and Isaacs report found its mark. As regards institutional planning for risk mitigation, it should be acknowledged that political prioritization tends to operate for the short term. Wadge and Isaacs used no language that referred to volcanic risks in the short term. Census data were never utilized to produce risk assessments, which might have transformed the findings into terms more readily understood by planners and policy-makers. The lessons in this for other communities at risk from volcanoes are non-trivial. With hindsight, the tragic losses on Montserrat could have been considerably fewer if the volcanic hazards assessment had triggered institutional planning for risk mitigation. The challenge for the future in vulnerable areas lies in ensuring that full institutional preparedness actually does follow from soundly based scientific appraisal of hazards and robust assessments of consequent risks. Keeping the airport open until 25 June 1997 At times during the crisis, MVO scientists were inexorably drawn towards informal or unstructured interactions with politicians and

19

the public (for an overview of MVO developments, see Aspinall el al. 2002). Although MVO Chief Scientists interpreted their role as advisory, some interactions with Montserrat authorities concerning emergency management went significantly further. It is now well known, acknowledging the peculiarities of the crisis (slow escalation, long duration, small island, problems of evacuation; see above), that the evolved practices of MVO were worthwhile and crucial to the mitigation of risk and saving of many lives (see Voight 1998). However, the circumstances relating to the operation of the airport up to and on 25 June 1997, until its evacuation during the advance of a pyroclastic flow that reached 200m from the terminal building (Plate 11C), invite reflective consideration. The airport is situated NNE of Soufriere Hills Volcano, on a coastal fan of prehistoric pyroclastic and laharic debris (Fig. 3a, Plate IB). The Government of Montserrat wished to keep it open for two linked main reasons. First, the airport was symbolic for Montserratians regarding the future viability of the island, even after almost two years of the crisis and the related depopulation (Table 2). It provided the major connection with the outside world for Montserratians, foreign residents and tourists. Closure of the airport would mark the end of any pretence of a near-future return to 'business as usual'. Second, it provided an important evacuation route (e.g. for medical emergencies). The new small jetty in the north (Little Bay; Fig. 2), intended for evacuation, only became functional in February 1997 (formally handed over on 18 June 1997), but even so it could be rendered inoperable if there was much sea swell or wave activity. In early June 1997 pyroclastic flows on the northern flanks of Soufriere Hills Volcano posed the most serious risks to date, and it was clear that the potential for dangerous activity was escalating (see above). The MVO's worst-case scenario for the northern flanks at this time was a collapse involving 10 x 10 6 m 3 of the dome, directed down Tuitt's or Mosquito Ghauts. It was estimated that 3-4 x 10 6 m 3 discharged in a single collapse into Tuitt's Ghaut could generate a pyroclastic flow capable of reaching the airport in as little as 90 seconds. Moderate flows in Fort Ghaut with some also in Mosquito Ghaut (Plate 8A, B) resulted in temporary closure of the airport on 16 June. A large pyroclastic flow reached the vicinity of Harris via Mosquito Ghaut on 17 June and, on the following day, a passenger aeroplane was turned back by its pilot during a flight to Montserrat because of conditions there. Despite the volcanic escalation, the administrative authorities certainly wanted the airport to be kept open. The MVO made it clear, in a written statement of 9 June, and in discussions the next day (involving two Chief Scientists, the Acting Governor, the Chief Minister, plus other officials), that there was a serious risk involved in continuing to use the airport. Nevertheless, the scientists made a commitment to help, if the authorities decided on a course of action to keep the facility operational for as long as possible. A temporary commitment was made to post a scientist at the airport to provide up-to-the-minute advice to the operating airline company, who required this as a precondition for continuing flights to Montserrat. The scientist was also to assist with the implementation of official evacuation procedures in the event of a volcanic emergency. The MVO had autonomy over its own activities in the current restricted areas on the volcano. For internal operational purposes, the team decided upon a threshold length of pyroclastic flow runout towards the airport, which, if closely approached or passed, would signal that the direct threat to any scientist there (or in the field on eastern sides of the volcano) was then too great. The threshold runout was placed at the bridge at the lower end of the Paradise River valley, near Bethel and the head of the volcaniclastic fan, 1.6km from the airport terminal buildings (Fig. 2; see Plate 10A). Once this threshold was passed and the eastern sector hence closed to MVO personnel on their own initiative, the corollary would be that operation of the airport would be untenable. The temporary arrangements for using the airport, however, lasted far longer than was expected. They became a significant drain on MVO staff in terms of both availability and stress on individuals, with an impact on scientific work, and in the event it might be judged that they became critically dangerous.

20

B. P. KOKELAAR

By 24 June, MVO personnel were seriously concerned that their role at the airport was becoming untenable, and that unreasonable responsibilities were being put on the individuals involved. One of them referred to the role as 'front-line defence'. When there was poor visibility, the MVO observer at the airport would be unable to see any flows initiated and so needed to rely on the seismic interpretation communicated from the operations room at the MVO, or other observers, to trigger the alarm. Even with good visibility, the middle of the course down Mosquito Ghaut to Paradise River valley was hidden from view. Late on 24 June a switch in the direction of collapse activity on the top of the dome, back in the direction of Mosquito Ghaut, was noted, and at dawn on 25 June small pyroclastic flows were observed descending into the top of that drainage. The collapse that initiated at 12:55 LT on 25 June involved c. 6 x 106m3 of the dome, roughly twice the amount of material involved in all but one of the previous collapses (17 September 1996), but only 60% of that of the MVO worst-case scenario. While the collapse volume was well within the range of possible scenarios anticipated by the MVO, the associated pyroclastic flow easily exceeded the (internal MVO) threshold runout distance that had been set (Plate 10A). The flow comprised three main pulses, reflecting unsteady retrogressive failure of the dome rather than an abrupt single collapse, and it was material of the second pulse that extended furthest towards the airport. Mainly because a significant collapse was anticipated and employees were alert, the airport was evacuated very quickly. Phased evacuation was advised by the MVO at 12:58 and immediate evacuation at 13:00. This took about 3 minutes. The evacuees included passengers who had only just arrived, one of whom was the returning Governor of Montserrat. The aeroplane took off as the flow front approached the sea near the end of the runway (see Plate 11C). Another part of the flow reached to within 200m of the terminal building. This was at 13:07, about 4 minutes after the evacuation and 12 minutes after initiation of the collapse. It took c. 1 minutes for the pyroclastic debris to reach here from the dome (the second flow pulse initiated c. 5 minutes into the collapse; data from Loughlin et al. 2002a, b). It is inescapable to conclude that risk had increased and safety margins reduced to extremely dangerous levels. Apparently the situation deteriorated progressively without there being a major stimulus for changing arrangements for keeping the airport open. Seemingly, on 25 June, the only things that could have stopped the continuation of airport operations were the withdrawal of services by individuals or groups (e.g. the airline, airport staffer MVO staff), or an event of dramatic scale and consequence. The latter occurred just as staff withdrawal from the airport was imminent, but the official call should have been sooner. It has been suggested that the Governor's absence from the island created to some extent an official-decision vacuum during this most critical stage in the emergency management. The problem here is epitomized in the adage: Throw a frog into hot water - it jumps out and survives; place it in cold water and slowly heat it - the frog stays put and dies'. Akin to the latter scenario, the slowly progressive escalation of the eruption apparently affected the way in which the emergency was handled by authorities. For example, agencies of the UK Government seemingly procrastinated as if unwilling to accept what really was happening (Clay et al. 1999; Pattullo 2000). Similarly, the maintenance of the airport function until 25 June became increasingly dangerous gradually while the functional role there of the scientists became increasingly inappropriate, On Montserrat the credibility of the MVO was all-important for the successful implementation and periodical adjustment by the authorities of what was a fairly complex microzonation and alertlevel system (see above, and Aspinall et al. 2002). The civil authorities wished to maximize access and preserve as much as possible of a normal life, and the scientists had to be concerned not to give advice that would be interpreted as unduly precautionary. The working margin for maintaining the operation of the airport was very tight. The official status of the airport location (Zone C Alert-stage Orange) was 'access limited to short visits by residents and workers

with means of rapid exit\ and the scientists were concerned that the civil authorities who set the status (with MVO information) would not appreciate the increases in risk immediately at the times when they developed. Also, while the public watched and was influenced by the scientists' activities in the field, there was a danger that people would be tempted to guide their own risk-taking practices accordingly, failing to recognize that the scientists were better equipped to receive warnings and evacuate quickly. As conditions worsened there was a distinct possibility that the civil authorities' emergency arrangements and response capabilities could be outstripped by events. Additionally, there were some people, on the volcano flanks, who relied entirely on their own judgement, rather than on the official advice being given. All this contributed to the increasing concern at the MVO that the progressively worsening situation was rapidly becoming untenable. Such problems deriving from progressive escalation of risks (as opposed to rapid onset) might be reduced if civil plans for disaster preparedness necessarily embed formal requirements for any emergency management team, comprising both scientists and officials, (1) to check regularly and deliberately, perhaps daily or even more frequently, for actual or potential progressive development towards dangerous or inappropriate activity, and (2) to moderate any such activity by default unless actions can be proved safe and/or appropriate to a high degree of probability. The slowly heated frog should be obliged to ask itself Ts this comfortable?' and, if it is not quite certain, must jump out. The scientists, however, should always be careful not to force issues by withdrawing their involvement before their own objectively determined conservative safety margins are reached. Discussion The eruption on Montserrat during 1995-1999 was the most destructive in the West Indies since 1902 (Mont Pelee, Martinique). By 1998, when magma ascent paused and it was hoped that the eruption had ended, the resources deployed by agencies of the UK Government for scientific monitoring of the volcano, risk assessment and dissemination of advice, amounted to more than £3.8 million, including provision of a helicopter (c. £1.2 million). This appears as quite a small proportion of the total cost of the emergency, especially when compared with the expenditure on. for example, other advice and technical assistance (Fig. 7). The slow progress of the volcanic escalation coupled with the small size of the island seems to have had several key effects. While initially it was considered that evacuation of the entire island might

Fig. 7. UK Government expenditure in Montserrat owing to the volcanic crisis, by functional use over the three fiscal years corresponding to the escalating crisis, 1995/1996-1997/1998 (after Clay et al. 1999). External transport costs include c.£1.2 million for the helicopter used by the MVO. The total expenditure represented is almost £56 million, which was disbursed by the Department for International Development (DFID).

SETTING AND CHRONOLOGY OF THE ERUPTION be needed, and early contingency plans were formulated (Table 2), that prospect receded as the eruption became established. Had the island been much smaller, total evacuation would have been inevitable. The island was just sufficiently large for it to have a 'safe zone' throughout the eruption, with the Centre Hills providing an important protective barrier, but the majority of the inhabited areas and associated infrastructure were not in the safe zone and had progressively to be abandoned. While conditions slowly worsened, the MVO was under pressure to provide robust advice for risk mitigation without excessive precaution, and, with the advent of the risk management maps, to keep risk zones as narrow and precisely located as possible. This pressure arose from the political desire to keep life as near to normal as possible (albeit quite abnormal) and the general public determination not to leave the island. The 'microzonation system' developed by the MVO for Montserrat (see Fig. 5) was at first appropriate, although it was at times difficult to communicate efficiently and to enforce. The 25 June 1997 tragedy and subsequent volcanic developments that placed numerous sectors simultaneously at risk inevitably led to abandonment of the complex zone maps in favour of the simpler tripartite division of the island into an Exclusion Zone and a (safe) Northern Zone, with an intervening Central Zone. Had the eruption escalated rapidly the microzonation system probably would never have been developed. The slow volcanic escalation also seems partly to underlie the tardiness with which departments of the UK Government responded to the emergency. The various shortcomings in the provisions for adequate shelter in the north of the island at the time of, and long after, the evacuations, relate to the departments attempting to manage the emergency within normal institutional arrangements. Early establishment of an inter-departmental crisis management team with executive authority to fast-track procedures and with direct access to responsible government ministers would have been more effective (see Clay et al. 1999), but the slow volcanic escalation gave little impetus for this. Even in 1997, after the tragic deaths of 25 June and while the Vulcanian explosions were gaining international media attention, it was believed by the MVO scientists and the Governor of Montserrat that politicians and civil servants in London still did not appreciate the seriousness of the situation (e.g. see Pattullo 2000). This state of affairs can be contrasted with other natural disasters that involved a more catastrophic onset and

21

were immediately followed by visits by high-ranking government officials. Shortly following the 18 May 1980 debris avalanche and blast at Mount St Helens, the US President (Carter) visited the site and the scale of the disaster was immediately and effectively communicated to the US Government. In terms of emergency management at Montserrat, the slow escalation also caused some problems with enforcement of evacuations. Whereas risk could be assessed and evacuations advised according to robust criteria, long periods with little change perceptible to the population contributed to several instances of hazardous disregard of advice or instructions. One such instance contributed to some of the tragic deaths on 25 June 1997. Some survivors commented that had they really known what the volcano could do, they might not have exposed themselves to the hazard (Loughlin et al. 20026). A more catastrophic onset to the eruption might have engendered more cautious attention to given advice and to issued risk notices. The slow eruption escalation did, however, allow the development of scientific understanding, monitoring techniques, reporting procedures and public awareness campaigns that, together with the training of many tens of personnel, did succeed in protecting many lives. This paper is dedicated to all of the staff and other scientists who have served the Montserrat Volcano Observatory and hence given vital assistance to the beleaguered people of Montserrat. The author is greatly indebted to W. Aspinall, S. Loughlin, S. Sparks, B. Voight and S. Young for their patient guidance and substantial contributions to the paper. W. Aspinall, P. Cole, T. Druitt, C. Gardner, R. Herd, S. Loughlin, G. Norton and G. Wadge assisted with material for the figures and plates, K. Lancaster drafted the figures, and T. Druitt and M. Howells improved the text. Helen Kokelaar patiently assisted in many ways, for which I am grateful. Plates Refer to Figure 2 for locations. Plates captioned © NERC are published under Permit IPR/20-10C British Geological Survey. All rights are reserved. These materials are a product of the DFID programme of work carried out in Montserrat by the BGS.

Plate 1. (A) View ENE over Plymouth and St George's Hill (centre), with Soufriere Hills Volcano shrouded in cloud to the right, although showing a plume of condensed steam from the 27 August 1995 phreatic vent. Plymouth harbour is at bottom right, with Fort Ghaut extending inland behind. Communities on the left are Richmond Hill (foreground) and Cork Hill (middle ground), with Centre Hills in the distance. The gentle slopes favoured for habitation in the foreground and distance to the right are of pyroclastic flow and lahar deposits from previous eruptions of Soufriere Hills Volcano (see Fig. 3a) (photo by C. A. Gardner. 28 August 1995).

Plate 1. (B) View north along the centraleast coast. Tuitt's village is middle left, with Bethel beyond and Spanish Point centre, and with the airport on the coast in the distance. All were built on pyroclastic flow and lahar deposits from Soufriere Hills Volcano (see Fig. 3 A). The volcaniclastic fan is cut by Irish Ghaut (foreground leading to precipitous cliffs). White's Ghaut (forested valley centre) and Paradise River-Pea Ghaut (along distant fan margin) (photo by T. Casadavall. 6 August 1995).

Plate 1. (C) Village of Long Ground, highest on the volcano (c. 300m) and nearest to the open, eastern side of English's Crater. Roughly half of the houses here were damaged by ballistic blocks up to 1.5m in diameter during the first magmatic explosion(s) of 17-18 September 1996; many consequently caught fire. Although Long Ground was the first of the eastern villages to be evacuated, it was the last to be impacted by a pyroclastic flow, mainly because of its elevated location between major drainages (photo by T. Casadavall. 6 Aueust 1995).

Plate 2. (A) Southwestern coastal villages of St Patrick's (centre) and Morris1 (right), built on a relatively steep fan of volcaniclastic deposits. The valley high on the right is the upper part of White River valley, leading to the steep, non-vegetated Galway's Wall, with part of Galway's Soufriere visible on the extreme right (bare patch in forest). The highest summit (centre) is Chances Peak andesitic lava dome, with Gages Mountain and Galway's Mountain andesitic domes to the left and right respectively. The prominent drainage to the sea on the left, from Chances Peak, is Gingoes Ghaut. Almost all of this area was devastated by the flank collapse and ensuing pyroclastic density current on 26 December 1997 (see Plate 16) (photo by S. R. Young, December 1995; © NERC).

Plate 2. (B) English's Crater (sector-collapse scar) and Castle Peak andesitic lava dome within, viewed towards the NE on 18 October 1995. Explosion craters, formed during the phreatic phase of the eruption, are evident on the NW flank of the lava dome, as is a small protrusion of andesite (near site of fumarolic activity) that appeared in September 1995 (see Plate 3C). The various named walls of English's Crater are: Galway's Wall, at bottom right; Gages Wall, at bottom left; and Farrell's Wall, at mid-left. The three prominent valleys at mid-left are Mosquito Ghaut (extreme left, mostly out of view), Tuitt's Ghaut (middle) and White's Ghaut (right), all of which drain towards the volcaniclastic fan in the distance where eastern coastal villages and the airport are located. The broad valley extending to the sea eastwards from English's Crater is Tar River valley (photo by G. Wadge).

Plate 3. (A) Andesitic lava spines on the c. 350-year-old Castle Peak dome, viewed towards the WSW. Phreatic explosivity began on 18 July 1995, low on the dome to the far right. Much of the originally verdant forest in the foreground is damaged by almost five weeks of ashfall and acid emissions (photo by C. A. Gardner. 20 August 1995).

Plate 3. (B) View NNE across Galway's Wall (foreground), showing phreatic explosion craters along a fissure extending NW towards the 18 July vent (far left). During a pause in magma extrusion in 1998-1999, explosion craters and vents again developed along a NW-SE fissure in this vicinity. The summit of Castle Peak dome is on the right. Defoliation is advanced in the vent area after nearly seven weeks of phreatic explosivity. The eastern coastal fan is visible in the distance (photo by C. A. Gardner. 3 September 1995).

Plate 3. (C) Altered andesite (oxidized; nearest the viewer) was first extruded on the SW flank of Castle Peak dome in late September 1995. Although it was unclear whether this was old andesite or a cooled part of a new plug, it was certainly taken to register shallow magmatic upheaval. It would be some nine weeks from the first extrusion before incandescent magma was observed, partially filling the enlarged 18 July phreatic vent visible just behind this first extrusion. Moat ponds of discoloured water are evident in the middle foreground and to the left (photo by K. West, late September 1995).

Plate 4. (A) The new andesitic dome in January 1996. Rockfall debris has started to infill the moat beneath Farrell's Wall. Destruction of the original dense vegetation revealed for the first time the internal structure of the domes and flanking deposits that were cut by the c. 4000-year-old sector-collapse scar. The hillside village of Harris is visible on the left, with Bethel, Spanish Point and the airport in the distance. All of these places would be impacted by pyroclastic flows when the dome overtopped Farrell's Wall (photo by S. R. Young; © NERC).

Plate 4. (B) View westwards on 25 January 1996 of the new dome growing on the NW flank of Castle Peak lava dome. The steep confines of English's Crater and the Tar River valley directed early dome-collapse pyroclastic flows towards the east. Tar River Soufriere is in the non-vegetated ground low on the right, above the Hot River drainage, which initially channelled pyroclastic flows to the sea. Plymouth is visible on the west coast, to the right of Chances Peak (in cloud) (photo by R. P. Hoblitt).

Plate 5. (A) View towards NW. The fumarolic activity is from the new dome growing on the NW flank of Castle Peak dome (note old lava spines stripped of vegetation in foreground; see Plate 3A). As the new dome grew above the old one. the prominent gully on the latter would act as a significant conduit for dome-collapse debris (see Plate 7C). Streatham and Dyer's villages are visible at top right. The narrow and precipitous Galway's Wall is seen in profile on the left (photo R. P. Hoblitt. 2 February 1996).

Plate 5. (B) Dome-collapse pyroclastic flow, on 3 April 1996. exiting the moat on the northern side of the dome above Tar River Soufriere (right foreground). This was the first major pyroclastic flow of the eruption; it came within an hour or tw?o of the decision to evacuate southern Montserrat. Note the production of copious ash from the rapid fragmentation of lava blocks, which is producing an over-riding pyroclastic surge as well as a coeval ash plume blown NW by prevailing winds (photo by S. R. Young; f NERC).

Plate 5. (C) Lower reach of the Hot River drainage, on the east coast, after the first pyroclastic flows reached the sea on 12 May 1996. Note the narrow flanking singe zones (vegetation turned brown) and trees remaining standing not far from the valley axis (photo by S. R. Young, late May 1996; < NERC).

Plate 6. (A) View south of Tar River Estate House on 17 September 1996, with a dome-collapse pyroclastic flow passing behind in the Tar River valley. The dome collapse continued for nine hours, leading ultimately to magmatic explosivity shortly before midnight (photo by S. R. Young; u) NERC).

Plate 6. (B) Tar River Estate House at 06:00 LT on 18 September 1996, following the major dome-collapse pyroclastic flows of 17 September. A pyroclastic surge detached from the main flow and devastated this flanking ridge and building. This was the first building wrecked by the volcano. Wind-blown condensed steam and ash drift over the block-and-ash flow deposits that line the floor of the Tar River valley (photo by S. R. Young; © NERC).

Plate 6. (C) Tar River Estate House following ruin by the pyroclastic surge(s) of 17 September 1996. Impact from the WSW (top right) is evident in the tree blow-down, damage to the house and scatter of debris (photo by S. R. Young; 18 September 1996; (D NERC).

Plate 7. (A) By late November 1996, dome growth had almost completely filled the scar from the 17-18 September collapse and explosive activity. The rim of the scar is plainly visible truncating outer slopes of the dome and is open to the east, where it connects to a prominent sinuous channel formed by pyroclastic flows. Dome talus is close to overtopping the confinement of English's Crater. The sites and times of eventual overtopping would depend on the switching locus of asymmetric dome growth. It was at this time that Galway's Wall (lower right foreground) started to crack and cause concern for collapse of this flank (photo by S. R. Young; c NERC).

Plate 7. (B) The andesitic dome at night in late December 1996, viewed towards SW from White's Yard. Sites of small avalanches are marked by bright incandescent patches, with rockfalls beneath appearing like rivulets of light (photo by S. R. Young: f NERC).

Plate 7. (C) The dome in January 1997, viewed towards SW. Collapsing debris has eroded deeply into remnants of the old Castle Peak dome (left-hand side), following a pre-existing gully there (photo by S. R. Young: ( NERC).

Plate 8. (A) Consequence of the first major dome collapse over Gages Wall, on 16 June 1997, viewed from Windy Hill. The pyroclastic flow travelled c. 2 km westwards, as far as Gages Lower Soufriere (upper reaches of Fort Ghaut). The ash plume moved west as a result of prevailing winds and heavy ashfall affected the Plymouth area. Much of Plymouth was destroyed by block-and-ash flows and associated pyroclastic surges that descended the same drainage on 3 August 1997 (photo by S. C. Loughlin; © NERC).

Plate 8. (B) Ten minutes after the collapse over Gages Wall (A), a substantial domecollapse pyroclastic flow also travelled down Mosquito Ghaut, prompting evacuation of the airport. This view, SSE from Windy Hill across the pyroclastic fan of Farrell's plain, illustrates the vulnerability of the workers who continued to tend crops in this area (see Plate 9A) (photo by S. C. Loughlin, 16 June 1997; © NERC).

Plate 8. (C) The dome-collapse pyroclastic flow of 25 June 1997, near to its termination in Spanish Point on the eastern coastal fan (see Plate 11B). The flow involved three main pulses, registering unsteady retrogressive collapse of the dome. The first and second pulses substantially infilled the Mosquito Ghaut to Pea Ghaut drainage, so that the third spilled eastwards through Bethel and into Spanish Point. Two flow lobes are seen following shallow gullies, while others have recently halted or are about to cease advance. Evidently the flow lobes were thin near the terminations, moving largely as dense but mobile granular flows that were readily impeded by low barriers (Loughlin et aL 2002#). The onshore wind, from the east (left), prevented the clouds of elutriated ash from moving ahead of the denser flows (photo by P. Cole).

Plate 9. (A) Aftermath of 25 June 1997. Upper reaches of Mosquito Ghaut (centre) are substantially filled with block-and-ash flow deposits. Pyroclastic surge deposits blanket Farrell's plain west of Mosquito Ghaut, northwards as far as lower parts of Windy Hill and Streatham village (middle and right). Houses of Dyer's village are visible on the extreme right, across Tyre's Ghaut. Note the broad singe zones (vegetation turned brown) and tendency for the deposit extremities to extend into topographic depressions (photo by G. E. Norton. 28 June 1997; < NERC).

Plate 9.(B) Aftermath of 25 June 1997. View northeastwards over Streatham and Windy Hill. Much of the surge-derived pyroclastic flow that travelled a further 1.5 km beyond the limit of the parent surge drained into Tyre's Ghaut and then Dyer's River valley via the gullies in the foreground. Eyewitnesses reported seeing ash flowing down the road (middle of view), 'boiling' and 'moving round bends like a vehicle" (Loughlin et til. 2002<:/. h}. The parent block-and-ash flow followed the Paradise River northeastwards, to the right of the prominent small hill in the distance, virtually in the opposite direction to the surge-derived flow (photo by G. E. Norton. 28 June 1997: < NERC).

Plate 9. (C) Aftermath of 25 June 1997. View showing the trace of the surge-derived pyroclastic flow that penetrated the Belham River valley as far as Cork Hill, marked by singed vegetation along the drainage. Cork Hill school, which was not evacuated, is the large (pale-roofed) building towards the top left. This hazard was not anticipated (photo by G. E. Norton. 5 August 1997: ( NERC).

Plate 10. (A) Aftermath of 25 June 1997. View southwestwards. Upper reaches of Tuitt's Ghaut are clearly visible (top centre) with Mosquito Ghaut on the far right. The first pulse of the 25 June block-and-ash flow took c. 5 minutes from inception at the head of Mosquito Ghaut to emergence at the bend near the confluence with Tuitt's Ghaut (left of centre), averaging some 15m s~[ (Loughlin et al. 2002(7). The second pulse, travelling somewhat faster to the same bend (c. 20ms~ 1 ), substantially filled the drainage channel and reached to within 50m of the sea. The third pulse, travelling at a similar velocity to the second pulse, partially escaped the channel at the bend below the confluence with Tuitt's Ghaut, inundating parts of Bethel (middle left), and it also partially inundated the village of Farm (lower right). The bridge at the lower end of the Paradise River valley, which marked the internally set MVO threshold for runout distance (see text), was located some 200-300 m upstream from the houses that remained at Farm village at this time. Note the large andesitic boulders stranded on the block-and-ash flow deposit surface; some were almost the size of the houses. The village of Harris (Plate 10B) is out of sight behind the hill on the right (photo by S. C. Loughlin, June 1997; © NERC).

Plate 10. (B) Aftermath of 25 June 1997. The sinuous course of Paradise Ghaut channelled most of the block-and-ash flow past the village of Harris, although the sharp turn from NE to SE (upper-middle right) caused it to surmount the steep banks and inundate lower parts of the community (see Plate 11 A). Note that defoliated trees remain standing close to the edges of the main flow path. In the distance, the deposit extends past Farm village (Plate 10A) towards the airport (photo by S. C. Loughlin, June 1997; © NERC).

Plate 11. (A) Aftermath of 25 June 1997. The picture shows sharp bends in the river valley (viewed eastwards and foreshortened) at Harris where the flow's momentum caused it to surmount steep banks and spill destructively into a low part of the village, where all vegetation and several houses were obliterated. A coeval pyroclastic surge spread more widely into the village, where the church (middle right) and other buildings w;ere burned, although trees remained standing (photo by G. E. Norton. 1 Julv 1997: ( NERC).

Plate 11. (B) Aftermath of 25 June 1997. Block-and-ash flow deposit terminations across the volcaniclastic fan from Bethel (top right) through Spanish Point (middle) to Farm village (largely buried in middle-right foreground). Note the narrowness of the singe zones, partly due to a strong prevailing wind from the east (left to right) on 25 June. Compare with Plate 8C (photo by G. E. Norton. 1 July 1997: ( NERC).

Plate 11. (C) Aftermath of 25 June 1997. Former site of Trant's (centre) buried beneath distal block-and-ash flow deposits. c. 200m from the airport terminal buildings (extreme left). Note the apparent tendency for the flow near its terminations to have been restricted by low obstacles while extending further where unimpeded along roads and channels (photo by G. E. Norton. 1 July 1997: r NERC).

Plate 12. (A) Vulcanian explosion of 12:05 LT on 7 August 1997. This photograph, taken some 36 seconds after the seismically denned initiation, shows three vigorously ascending ash plumes (two on the right are one in front of the other), which soon merged into a single plume that reached an altitude of about 13 km (Druitt et al 20026). A pyroclastic surge formed by fountain collapse has surmounted Gages Mountain (middle), while other pyroclastic surges form on northern flanks (beyond on left) (photo by R. Herd; © NERC).

Plate 12. (B) During the Vulcanian explosivity of 4-12 August 1997, fountain-collapse pyroclastic flows produced pumiceous deposits in several drainages, including Tuitt's (centre) and Mosquito (right) Ghauts. The andesitic dome has overtopped the northern crater wall (photo by G. E. Norton, 7 August 1997; © NERC).

Plate 12. (C) White pumiceous fountain-collapse pyroclastic flow deposits of the August 1997 series of Vulcanian explosions overlie the grey deposits of the 25 June dome-collapse block-and-ash flow. Sinuous levees of pumice fragments, with lobate deposit terminations and lateral break-outs, characterize these distal parts. Note the contrasting 25 June block-and-ash flow deposits, which thin progressively and become less blocky towards their terminations (photo by S. R Young, September 1997; © NERC).

Plate 13. (A) On 21 September 1997, a major collapse towards the north shed a block-and-ash flow down Tuitf s Ghaut (right). Part of the flow spilled across a low point on the valley side (far centre right) and into White's Ghaut (centre). Tuitt's village, in between the two drainages, was impacted by pyroclastic surges that ignited the buildings (compare with Plate 12B). Note the ponding of ash against low topographic barriers on the interfluves. Until July 1998. the village of Long Ground (upper left) remained mainly unaffected by pyroclastic flows and surges, owing largely to its elevated position and deep flanking drainages (Tar River valley top left) (photo by S. C. Loughlin. 21 September 1997; r NERC).

Plate 13. (B) The 21 September 1997 block-and-ash flow reached the sea immediately south of the airport on the east coast, where boiling seawater formed numerous fumaroles and occasional small secondary explosions (ash cloud just right of centre) (photo by S. C. Loughlin. 21 September 1997: r NERC).

Plate 13. (C) Also destroyed on 21 September 1997 were the airport terminal buildings. Here the weak flow was evidently deflected mainly around or between the buildings, although a significant part evidently entered the doors on one side of the terminal concourse and exited mainly via the doors on the other side (photo by S. C. Loughlin. 21 September 1997: r NERC).

Plate 14. (A) Simon Young of the Montserrat Volcano Observatory measuring temperatures at the margin of the 21 September 1997 block-and-ash now deposits soon after emplacement near the former site of Trant's. The maximum recorded temperature was 365°C, although values up to 640°C were found here for the 25 June block-and-ash flow deposit (Cole et al. 2002). Flow-deposit margins were rich in charcoaled wood and rounded pumice clasts. The pumice clasts could be handled within hours of the emplacement, suggesting that they were incorporated from the (cold) August 1997 fountain-collapse pumiceous flow deposits. In contrast, dense andesite blocks were too hot to handle even with thick gloves (photo by S. C. Loughlin, 21 September 1997; © NERC).

Plate 14. (B) The 21 September collapse led to a protracted series of Vulcanian explosions with associated fountain-collapse pyroclastic flows, during 22 September to 21 October. From left to right, White's, Tuitt's and Mosquito-Paradise Ghauts are seen to have been the main channels for flows to the NE. Explosive activity formed a plume that was strongly deflected by winds towards the west and NW, over inhabited parts of the island. The vigorous ash-venting seen here was transitional from the initial explosion and in general it diminished over tens of minutes to hours. It was often pulsatory. Ash fallout during the September-October 1997 Vulcanian explosivity was a significant factor in causing the departure of many who had remained on-island until then (photo by S. R. Young, September 1997; © NERC).

Plate 14. (C) Lobate terminations of pumiceous fountain-collapse pyroclastic flow deposits, with characteristic finger-like forms, overlie the darker (grey) deposits of the 21 September block-and-ash flow. The villages of Trant's, Farm and Bethel are buried and most buildings at Spanish Point are badly damaged (photo by S. R. Young, September 1997; © NERC).

Plate 15. (A) Buoyant ash plume from a Vulcanian explosion, viewed from the north (Jack Boy Hill). Plumes like this rose to 15 km altitude and ash was dispersed widely in the NE Caribbean region. This plume was relatively steam-rich and ash-poor, according to its pale colour (photo by S. C. Loughlin. late September 1997; f NERC).

Plate 15. (B) Vulcanian explosion of 15:13 LT on 20 October 1997, with associated radially directed fountain-collapse pyroclastic flows and surges travelling northwards in Mosquito Ghaut (left) and westwards towards Plymouth in Gages valley (right). The view is from the NW (from the MVO at Old Towne) (photo by P. Cole).

Plate 16. (A) Aftermath of the flank collapse, debris avalanche and powerful pyroclastic density current of 26 December (Boxing Day) 1997. Southwestern villages are obliterated and the White River valley largely filled with debris (compare with Plate 2A). The debris avalanche stopped within tens of metres of the edge of the coastal fan and the following pyroclastic current entered the sea across a broad front causing a small tsunami (photo by S. R. Young, early January 1998; © NERC).

Plate 16. (B) The village of Morris' before the catastrophic flows of 26 December 1997. The pale brown and green hues register the occasional exposure to fallout of ash (compare with Plate 2A) (photo by G. E. Norton, April 1997; © NERC).

Plate 16. (C) The site of Morris' showing effects of the passage of the strongly erosive pyroclastic density current of 26 December 1997. The area in view is that in the centre of Plate 16B. Most houses have been swept away, with only a few lower storeys or foundations remaining. The numerous downslope-orientated ridges and grooves were formed by deposition in the lee of obstacles, such as tree-stumps or large rocks, with intense scouring or striation in between (Ritchie et al. 2002). In this view some vegetation has re-established (photo by S. R. Young, 2 November 1998; © NERC).

Plate 17. (A) View to SW of Soufriere Hills Volcano on 10 March 1998, when extrusion of magma stopped. The pyroclastic fan at the foot of the Tar River valley extended some 600 m from the original coastline. The upper reach of Dry Ghaut is visible south (far left) of the summit, draining towards the viewer from the flank of South Soufriere Hills. Here the upper parts of the explosively generated pyroclastic current of 26 December 1997 surmounted the drainage divide and flowed eastwards. This formed a surge-derived pyroclastic flow that nearly reached the sea in a direction almost opposite to that of the primary current (photo by R. P. Hoblitt).

Plate 17. (B) The lava dome as it appeared on 23 March 1998. The mass that includes the prominent spine (altitude 1031 m) grew in a scar left by the major collapse and explosion of 26 December 1997. Parts of the scar are visible below and to the right of the summit mass (photo by R. P. Hoblitt).

Plate 18. (A) Fallout of ash, locally referred to as 'ashing' (like raining), associated with Vulcanian explosivity during September 1997. The intense nuisance of the repeated ash fallout due to the numerous (75) explosions and related pyroclastic flows in the September-October period proved stressful for remaining Montserratians, many of whom at last decided to leave the island. This photograph was taken from the MVO at Eiffel House, Old Towne, before it was evacuated later in September 1997 (photo by S. R. Young; © NERC).

Plate 18. (B) Roof collapse due to loading by fallout ash was widespread. The medical school buildings, in the outskirts of Plymouth (Amersham), show the particular susceptibility to collapse of some roofs with relatively broad span and shallow pitch (photo by S. R. Young, 2 November 1998; © NERC).

Plate 18. (C) View westwards from Gages Mountain towards Plymouth with the deep Aymer's Ghaut on the left. This hilltop was subjected to almost continuous ash fallout, because of its location WNW of the lava dome and the prevailing west to northwesterly wind direction. It was impacted by a pyroclastic surge associated with a Vulcanian explosion on 7 August 1997 (see Plate 12A). The flanks below were repeatedly blanketed by fallout of ash and lapilli throughout the phases of the initial phreatic explosions, the many subsequent dome collapses and associated flows, the more violent explosive episodes, and the activity during dome degradation from March 1998 until November 1999 (photo by S. C. Loughlin, May 1998; © NERC).

Plate 19. (A) Major lahars resulted from torrential rains associated with storms, both during and following the period of active dome growth. This view, taken in November 1998 following Hurricane Georges (20-21 September), shows dendritic drainage systems feeding into Plymouth. Fort Ghaut, on the right, has become strongly aggradational and has shifted its main channel in its lower reach near the coast (photo by S. C. Loughlin; NERC).

Plate 19. (B) Mouth of the Belham River valley after the major lahars of September 1998. The braided channels and deposits bury the former golf course, while a new volcaniclastic fan extends into the sea in the distance to the left (photo by G. E. Norton. 1998: NERC).

Plate 19. (C) Access into the areas south of Belham River was considerably impeded by lahar activity, which eventually buried the main road bridge and repeatedly modified the valley floor (the vehicle was previously abandoned) (photo by G. E. Norton. 1998; ( NERC).

Plate 20. (A) The War Memorial clocktower, a telephone box and palm trees near the water-front of Plymouth in January 1996, view towards the volcano. The town had been evacuated twice by this time, most recently in the preceding December for one month (photo by S. R. Young; © NERC).

Plate 20. (B) By February 1998 Plymouth was extensively devastated by pyroclastic flows and surges, and was being progressively buried by successive lahars (photo by S. C. Loughlin; © NERC).

Plate 20. (C) One year and many lahars later, in February 1999, many buildings had been completely buried or swept away, with landmarks lost and entombment of the former capital town advanced, but not finished (photo by S. C. Loughlin; © NERC).

42

B. P. KOKELAAR

References AMBEH, W. B. & LYNCH, L. L. 1996. Seismicity preceding the current eruption of the Soufriere Hills Volcano, Montserrat, West Indies (abstract). In: AHMAD, R. (ed.) Science, Hazards and Hazard Management. The Second Caribbean Conference on Natural Hazards and Disasters, 9-12 October 1996, Kingston, Jamaica. Unit for Disaster Studies, Department of Geography and Geology, The University of West Indies, Jamaica, vol. 1, 30. ASPINALL, W. P., MILLER, A. D., LYNCH, L. L., LATCHMAN, J. L., STEWART, R. C, WHITE, R. A. & POWER, J. A. 1998. Soufriere Hills eruption, Montserrat 1995-1997: volcanic earthquake locations and fault plane solutions. Geophysical Research Letters, 25, 3397-3400. ASPINALL, W. P., LOUGHLIN, S. C., MICHAEL, F. V. ET AL. 2002. The Montserrat Volcano Observatory: its evolution, organization, role and activities. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 71-91. BARCLAY, J., RUTHERFORD, M. J., CARROLL, M. R., MURPHY, M. D., DEVINE, J. D., GARDNER, J. & SPARKS, R. S. J. 1998. Experimental phase equilibria constraints on preemptive storage conditions of the Soufriere Hills magma. Geophysical Research Letters, 25, 3437-3440. BAXTER, P. J., BONADONNA, C., DUPREE, R. ET AL. 1999. Cristobalite in volcanic ash of the Soufriere Hills Volcano, Montserrat, British West Indies. Science, 283, 1142-1145. BONADONNA, C., MAYBERRY, G. C., CALDER, E. S. ET AL. 2002. Tephra fallout in the eruption of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 483-516. BOUDON, G., VlLLEMANT, B., KOMOROWSKI, J.-C., ILDEFONSE, P. & SEMET,

M. P. 1998. The hydrothermal system at Soufriere Hills volcano, Montserrat (West Indies): Characterization and role in the on-going eruption. Geophysical Research Letters, 25, 3693-3696. BOUYSSE, P. & GUENNOC, P. 1983. Donnees sur la structure de l'arc insulaire des Petites Antilles, entre Sainte Lucie et Anguilla. Marine Geology, 53, 131-166. BRIDEN, J. C., REX, D. C., FOLLER, A. M. & TOMBLIN, J. F. 1979. K-Ar geochronology and paleomagnetism of volcanic rocks in the Lesser Antilles island arc. Philosophical Transactions Royal Society London, 291, 485-528. CALDER, E. S., COLE, P. D., DADE, W. B. ET AL. 1999. Mobility of pyroclastic flows and surges at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 26, 537-540. CALDER, E. S., LUCKETT, R., SPARKS, R. S. J. & VOIGHT, B. 2002. Mechanisms of lava dome instability and generation of rockfalls and pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 173-190. CHIODINI, G., CIONI, R., FRULLANI, A., GUIDI, M., MARINI, L., PRATI, F. & RACO, B. 1996. Fluid geochemistry of Montserrat Island, West Indies. Bulletin of Volcanology, 58, 380-392. CLARKE, A. B., NERI, A., VOIGHT, B., MACEDONIO, G. & DRUITT, T. H. 2002. Computational modelling of the transient dynamics of August 1997 Vulcanian explosions at Soufriere Hills Volcano, Montserrat: influence of initial conduit conditions on near-vent pyroclastic dispersal. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 319-348. CLAY, E., BARROW, C., BENSON, C., DEMPSTER, J., KOKELAAR, P., PILLAI, N. & SEAMAN, J. 1999. An evaluation of HMG's response to the Montserrat volcanic emergency. Department for International Development Evaluation Report EV635. COLE, P. D., CALDER, E. S., DRUITT, T. H., HOBLITT, R., ROBERTSON, R., SPARKS, R. S. J. & YOUNG, S. R. 1998. Pyroclastic flows generated by gravitational instability of the 1996-97 lava dome of Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3425-3428. COLE, P. D., CALDER, E. S., SPARKS, R. S. J. ET AL. 2002. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 231-262.

DENLINGER, R. P. & HOBLITT, R. P. 1999. Cyclic eruptive behavior of silicic volcanoes. Geology, 27, 459-462. DEVINE, J. D., MURPHY, M. D., RUTHERFORD. J. J. ET AL. 1998. Petrologic evidence for pre-eruptive pressure-temperature conditions, and recent reheating, of andesitic magma erupting at the Soufriere Hills Volcano, MONTSERRAT, W. I. Geophysical Research Letters, 25. 3669-3672. DRUITT, T. H., CALDER, E. S., COLE. P. D. ET AL. 2002a. Small-volume, highly mobile pyroclastic flows formed by rapid sedimentation from pyroclastic surges at Soufriere Hills Volcano. Montserrat: an important volcanic hazard. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21. 263-279. DRUITT, T. H., YOUNG, S. R., BAPTIE. B. ET AL. 2002b. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306. FERGUS, H. A. 1994. Montserrat: History of a Caribbean Colony. Macmillan, London. FISKE, R. S. 1984. Volcanologists, journalists and the concerned local public: a tale of two crises in the Eastern Caribbean. In: Studies in Geophysics. Explosive Volcanism: Inception, Evolution and Hazards. National Academy Press, Washington, D.C. 170-176. GARDNER, C. A. & WHITE, R. A. 2002. Seismicity, gas emission and deformtion from 18 July to 25 September 1995 during the initial phreatic phase of the eruption of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs. 21, 567-581 GOVERNMENT OF MONTSERRAT 1995. National Disaster Action Plan. Plymouth, Montserrat. HAMMOUYA, G., ALLARD, P., JEAN-BAPTISTE, P., PARELLO, F., SEMET, M. P. & YOUNG, S. R. 1998. Pre- and syn-eruptive geochemistry of volcanic gases from Soufriere Hills of Montserrat, West Indies. Geophysical Research Letters, 25, 3685-3688. HARFORD, C. L., PRINGLE, M. S., SPARKS, R. S. J. & YOUNG, S. R. 2002. The volcanic evolution of Montserrat using 40Ar 39 Ar geochronology. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 93-113. IDC 1997. Montserrat. House of Commons, Session 1997-1998, International Development Committee, First Report, together with the Proceedings of the Committee, Minister of Evidence and Appendices. Stationery Office, London IDC 1998. Montserrat - Further Developments. House of Commons, Session 1997-1998, International Development Committee, Sixth Report, together with the Proceedings of the Committee, Minister of Evidence and Appendices. Stationery Office, London. JACKSON, P., SHEPHERD, J. B., ROBERTSON, R. E. A. & SKERRITT, G. 1998. Ground deformation studies at Soufriere Hills Volcano. Montserrat I: Electronic distance meter studies. Geophvsical Research Letters, 25, 3409-3412. KILBURN, C. R. J. & VOIGHT, B. 1998. Slow rock fracture as eruption precursor at Soufriere Hills volcano, Montserrat. Geophysical Research Letters, 25, 3665-3668. KNAPP, B. 1989. Volcano. Macmillan. LOUGHLIN, S. C., BAXTER. P. J., ASPINALL, W. P.. DARROUX. B., HARFORD, C. L. & MILLER, A. D. 2002a. Eyewitness accounts of the 25 June 1997 pyroclastic flows and surges at Soufriere Hills Volcano, Montserrat, and implications for disaster mitigation. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs. 21,211-230! LOUGHLIN, S. C., CALDER, E. S., CLARKE. A. B. ET AL. 2002b. Pyroclastic flows and surges generated by the 25 June 1997 dome collapse, Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 191-209. MACDONALD, R., HAWKESWORTH, C. J. & HEATH, E. 2000. The Lesser Antilles volcanic chain: a study in arc magmatism. Earth-Science Reviews, 49, 1-76. MACGREGOR, A. G. 1938. The Royal Society Expedition to Montserrat, BWI: the volcanic history and petrology of Montserrat, with observations on Mt. Pele, in Martinique. Philosophical Transactions of the Roval Society, B229. 1-90.

SETTING AND CHRONOLOGY OF THE ERUPTION MATTIOLI, G. S., DIXON, T. H., FARINA, F., HOWELL, E. S., JANSMA, P. E. & SMITH, A. L. 1998. GPS measurement of surface deformation around Soufriere Hills volcano, Montserrat from October 1995 to July 1996. Geophysical Research Letters, 25, 3417-3420. MELNIK, O. & SPARKS, R. S. J. 1999. Nonlinear dynamics of lava dome extrusion. Nature, 402, 37-41. MELNIK, O. & SPARKS, R. S. J. 2002. Dynamics of magma ascent and lava extrusion at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 153-171. MILLER, A. D., STEWART, R. C, WHITE, R. A. ET AL. 1998. Seismicity associated with dome growth and collapse at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3401-3404. MURPHY, M. D., SPARKS, R. S. J., BARCLAY, J. ET AL. 1998. The role of magma mixing in triggering the current eruption at the Soufriere Hills volcano, Montserrat, West Indies. Geophysical Research Letters, 25, 3433-3436. MURPHY, M. D., SPARKS, R. S. J., BARCLAY, J., CARROLL, M. R. & BREWER, T. S. 2000. Remobilization of andesite magma by intrusion of mafic magma at the Soufriere Hills Volcano, Montserrat, West Indies. Journal of Petrology, 41, 21-42. NORTON, G. E., WATTS, R. B., VOIGHT, B. ETAL. 2002. Pyroclastic flow and explosive activity at Soufriere Hills Volcano, Montserrat, during a period of virtually no magma extrusion (March 1998 to November 1999). In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 467-481. PATTULLO, P. 2000. Fire from the Mountain. The Tragedy of Montserrat and the Betrayal of its People. Constable, London. PERRET, F. A. 1939. The volcano-seismic crisis at Montserrat 1933-37. Publications of the Carnegie Institute, Washington, 512. POSSEKEL, A. K. 1999. Living with the Unexpected. Linking Disaster Recovery to Sustainable Development in Montserrat. Springer-Verlag, Berlin. REA, W. J. 1974, The volcanic geology and petrology of Montserrat, West Indies. Geological Society, London, 130, 341-366. REA, W. J. 1982. The Lesser Antilles. In: THORPE, R. S. (ed.) Andesites. John Wiley & Sons, Chichester, 167-185. RITCHIE, L. J., COLE, P. D. & SPARKS, R. S. J. 2002. Sedimentology of deposits from the pyroclastic density current of 26 December 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 435-456 ROBERTSON, R., COLE, P., SPARKS, R. S. J. ETAL. 1998. The explosive eruption of Soufriere Hills Volcano, Montserrat, West Indies, September 17 1996. Geophysical Research Letters, 25, 3429-3432. ROBERTSON, R. E. A., ASPINALL, W. P., HERD, R. A., NORTON, G. E., SPARKS, R. S. J. & YOUNG, S. R. 2000. The 1995-1998 eruption of the Soufriere Hills volcano, Montserrat, WI. Philosophical Transactions Royal Society London, 358, 1619-1637. ROOBOL, M. J. & SMITH, A. L. 1998. Pyroclastic stratigraphy of the Soufriere Hills volcano, Montserat - Impications for the present eruption. Geophysical Research Letters, 25, 3393-3396. SHEPHERD, J. B., TOMBLIN, J. F. & Woo, D. A. 1971. Volcano-seismic crisis in Montserrat, West Indies 1966-67. Bulletin of Volcanology, 35, 143-163. SHEPHERD, J. B., HERD, R. A., JACKSON, P. & WATTS, R. 1998. Ground deformation measurements at the Soufriere Hills Volcano, Montserrat: II: Rapid static GPS measurements June 1996-June 1997. Geophysical Research Letters, 25, 3413-3416. SPARKS, R. S. J. & YOUNG, S. R. 2002. The eruption of Soufriere Hills Volcano, Montserrat (1995-1999): overview of scientific results. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 45-69. SPARKS, R. S. J., YOUNG, S. R., BARCLAY, J. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills Volcano, Montserrat, West Indies: November 1995 to December 1997. Geophysical Research Letters, 25, 3421-3424. SPARKS, R. S. J., BARCLAY, J., CALDER, E. S. ET AL. 2002. Generation of a debris avalanche and violent pyroclastic density current on

43

26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 409-434. VOIGHT, B. 1998. Volcanologists' efforts on Montserrat praiseworthy. Bulletin of Volcanology, 60, 318-319. VOIGHT, B., HOBLITT, R. P., CLARKE, A. B., LOCKHART, A. B., MILLER, A. D., LYNCH, L. & McMAHON, J. 1998. Remarkable cyclic ground deformation monitored in real-time on Montserrat, and its use in eruption forecasting. Geophysical Research Letters, 25, 3405-3408. VOIGHT, B., SPARKS, R. S. J., MILLER, A. D. ET AL.. 1999. Magma flow instability and cyclic activity at Soufriere Hills Volcano, Montserrat, British West Indies. Science, 283, 1138-1142. VOIGHT, B., KOMOROWSKI, J.-C, NORTON, G. E. ET AL.. 2002. The 26 December (Boxing Day) 1997 sector collapse and debris avalanche at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 363-407. WADGE, G. 1984. Comparison of volcanic production rates and subduction rates in the Lesser Antilles and Central America. Geology, 12, 555-558. WADGE, G. 1986. The dykes and structural setting of the volcanic front in the Lesser Antilles island arc. Bulletin of Volcanology, 48, 349-372. WADGE, G. & ISAACS, M. C. 1987. Volcanic Hazards from Soufriere Hills Volcano, Montserrat, West Indies. A report to the Government of Montserrat and the Pan Caribbean Disaster Preparedness and Prevention Project. Department of Geography, University of Reading, UK. WADGE, G. & ISAACS, M. C. 1988. Mapping the volcanic hazards from Soufriere Hills Volcano, Montserrat, West Indies using an image processor. Journal of the Geological Society, London, 145, 541-555. WADGE, G. & SHEPHERD, J. B. 1984. Segmentation of the Lesser Antilles subduction zone. Earth and Planetary Science Letters, 71, 297-304. WADGE, G., JACKSON, P., BOWER, S. M., WOODS, A. W. & CALDER, E. 1998. Computer simulations of pyroclastic flows from dome collapse. Geophysical Research Letters, 25, 3677-3680. WATSON, I. M., OPPENHEIMER, C., VOIGHT, B. ET AL.. 2000. The relationship between degassing and ground deformation at Soufriere Hills Volcano, Montserrat. Journal of Volcanology and Geothermal Research, 98, 117-126. WATTS, R. B., HERD, R. A., SPARKS, R. S. J. & YOUNG, S. R. 2002. Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 115-152. WHITE, R. A., MILLER, A. D., LYNCH, L. & POWER, J. 1998. Observations of hybrid seismic events at Soufriere Hills Volcano, Montserrat: July 1995 to September 1996. Geophysical Research Letters, 25, 3657-3660. WOODS, A. W., SPARKS, R. S. J., RITCHIE, L. J., BATEY, J., GLADSTONE, C. & BURSIK, M. I. 2002. The explosive decompression of a pressurized volcanic dome: the 26 December 1997 collapse and explosion of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 457-466. WYLIE, J. J., VOIGHT, B. & WHITEHEAD, J. A. 1999. Instabiity of magma flow from volatile-dependent viscosity. Science, 285, 1883-1885. YOUNG, S. R., HOBLITT, R. P., SMITH, A. L., DEVINE, J. D., WADGE, G. & SHEPHERD, J. B. 1996. Dating of explosive volcanic eruptions associated with dome growth at the Soufriere Hills volcano, Montserrat, West Indies. In: Second Caribbean Conference on Natural Hazards and Hazard Management. MVO Open File Report 96/22, Kingston, Jamaica. YOUNG, S. R., SPARKS, R. S. J., ASPINALL, W. P., LYNCH, L. L., MILLER, A. D. & ROBERTSON, R. E. A. 1998. Overview of the eruption of Soufriere Hills volcano, Montserrat, 18 July 1995 to December 1997. Geophysical Research Letters, 25, 3389-3392. YOUNG, S. R., VOIGHT, B., BARCLAY, J. ET AL. 2002. Hazard implications of small-scale edifice instability and sector collapse: a case history from Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 349-361.

The eruption of Soufriere Hills Volcano, Montserrat (1995-1999): overview of scientific results R. S. J. SPARKS1 & S. R. YOUNG2 1 Department of Earth Sciences, Bristol University, Bristol, BS8 1RJ, UK (e-mail: [email protected]) Montserrat Volcano Observatory, Mongo Hill, Montserrat, West Indies

Abstract: The eruption of Soufriere Hills Volcano, Montserrat (1995-1999) has displayed a wide range of volcanic phenomena: growth of an andesitic lava dome, generation of pyroclastic flows by lava dome collapse and by fountain collapse in explosive eruptions, Vulcanian and sub-Plinian explosivity with accompanying tephra fall, entrance of pyroclastic flows into the sea, sector collapse with formation of a debris avalanche and a high-velocity pyroclastic density current, and generation of lahars. New phenomena include: cyclic patterns of ground deformation linked with shallow seismicity and eruptive activity; pyroclastic flows formed by rapid sedimentation from pyroclastic surges; and an unprecedented slow escalation of eruption intensity. Magma pulsations with timescales of hours to years have been recognized. Transitions from extrusive to explosive activity were triggered by major dome collapses. Relationships between magma ascent dynamics and geophysical signals have been elucidated. Ascending water-rich andesitic magma becomes Theologically stiffened by degassing and groundmass crystallization. Large magma overpressures are consequently developed in the upper conduit, causing shallow seismicity, radial patterns of ground deformation, and cyclic pulsations of eruptive activity. Lava dome growth involved heterogeneous deformation with formation of spines and lobes along shear zones. Collapse of the pressurized dome resulted in substantial pyroclastic surges forming above pyroclastic flows. Influx of hydrous mafic magma remobilized hot crystalline igneous rocks at depths of 5 to 6 km to form crystal-rich andesitic magma and triggered the eruption.

At the beginning of the twentieth century the eruptions of two Caribbean volcanoes, Mont Pelee, Martinique, and the Soufriere of St Vincent, were the focus for major advances in volcanology (Anderson & Flett 1903; Lacroix 1904). In particular the lethal phenomenon of pyroclastic flows or nuees ardentes was recognized. The death of 28 000 people on 8 May 1902, when a nuee ardente from Mont Pelee swept through the town of St Pierre on Martinique, was the worst volcanic disaster of the twentieth century. At the end of the century once again the Caribbean has become the centre of scientific interest and there is international concern for the plight of the island society of Montserrat. The eruption of Soufriere Hills Volcano, Montserrat (1995 to present) has provided another example of a major andesitic eruption, with and opportunities to advance understanding of volcanic processes and develop innovative mitigation strategies. Montserrat (lat. 16°45'N, long. 62°10/W) is one of the smaller islands in the Lesser Antilles Island Arc. The island consists of four volcanic centres (Silver Hill, Centre Hills, the Soufriere Hills and South Soufriere Hills), which range in age from about 2.5 Ma to the present (MacGregor 1938; Rea 1974; Harford et al 2002). Soufriere Hills Volcano is the youngest of these centres, extending back at least 170ka, and overlaps in age with the basaltic South Soufriere Hills complex which was active at c. 130ka BP (Harford et al. 2002). Soufriere Hills Volcano consists of five andesitic lava domes: Gages Mountain, Chances Peak, Galway's Mountain, Perches Mountain and Castle Peak. The youngest 'Castle Peak dome' occupied the horseshoe-shaped English's Crater, which is 1 km in diameter with walls 100-150m high and is open to the east (Figs 1 and 2). Soufriere Hills Volcano is a Pelean type volcano with andesitic magmas of relatively narrow compositional range (58-64 wt% SiO2) and lava domes and pyroclastic flow deposits as the typical eruption products. The new lava is a porphyritic andesite with a narrow compositional range (57-61.5% SiO2) and contains phenocrysts of plagioclase, hornblende, orthopyroxene and oxide (Devine et al. 1998; Murphy et al. 2000). Recent stratigraphic and geochronological work (Roobol & Smith 1998; Harford et al. 2002) indicates a history extending back at least 170ka. The volcano was very active between 31 and 16kaBP, when voluminous block-andash flows and surges accompanied dome growth. Resumption of activity in c. 4000 years BP is thought by Roobol & Smith (1998) to have resulted in formation of English's Crater by sector collapse. The last eruption before the present one occurred just prior to settlement in AD 1632 and involved small eruptions of andesite to form the Castle Peak dome, nestled in the centre of English's Crater (Fig. 2; Young et al. 1998b).

Despite the lack of eruptive activity during the past 350 years, Soufriere Hills Volcano has not been quiet. Three volcanoseismic crises have occurred during the past 100 years: 18971898, 1933-1937 and 1966-1967 (Powell 1938; MacGregor 1938; Perret 1939; Wadge & Isaacs 1988; Ambeh et al. 1998). All were accompanied by enhanced fumarolic activity and are regarded to have resulted from ascent of magma that failed to reach the surface. The most recent crisis (1966-1967) was interpreted as injection of a relatively small volume of magma from >10km depth, into a complex system of fissures located below the volcano (Shepherd et al. 1971). Eruption of Soufriere Hills Volcano began with phreatic explosions on 18 July 1995 (Young et al. 1997, 1998b; Robertson et al. 2000). The eruption was preceded by a period of elevated seismicity (Aspinall et al. 1998) beginning in 1992, which prompted the installation of additional monitoring equipment. After the initial phreatic phase, most of the eruption has consisted of the emplacement and disintegration of an andesitic lava dome. There have also been significant periods of explosive activity associated with rapid extrusion rates and major dome collapses. The eruption has caused tremendous disruption to the lives of the local population, who had established a viable community on the volcano's fertile pyroclastic apron (Fig. 2). Volcano monitoring and crisis management have had to adapt to a challenging environment in which the resolve of scientists and population have been both tested and strengthened by the evolving eruption. The purpose of this paper is to provide a synthesis of the scientific advances that have been made during monitoring of the eruption. As is usually the case, scientific advances raise more questions and so we also draw attention to issues that will require further investigation and thought. The scientific work on Montserrat contrasts with the studies at Mont Pelee and La Soufriere of St Vincent at the beginning of the century, when a small number of pioneering scientists, with little technical instrumentation or facilities, made critical observations. The Soufriere Hills eruption has been monitored and studied by a multinational and multiinstitutional team using an armoury of sophisticated techniques (Fig. 3; Aspinall et al. 2002). Well over 100 scientists have been involved to different degrees since 1995, and the involvement of large teams is now becoming characteristic of responses to major volcanic eruptions. We provide a brief synopsis of the eruption to provide a context for the discussion of the scientific results. We then consider some of the main geophysical, petrological, geochemical and phenomenological manifestations of the eruption. Our aim is to draw

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 45-69. 0435-4052/02/$15 © The Geological Society of London 2002.

45

46

R. S. J. SPARKS & S. R. YOUNG

Fig. 1. Map of Montserrat showing topography and main geographical locations.

ERUPTION OF SOUFRIERE HILLS VOLCANO: OVERVIEW

47

Fig. 2. View of Soufriere Hills Volcano looking from NE in August 1995, prior to onset of dome growth. The multiple peaks of older domes can be seen from left to right: Roches Mountain, Galway's Mountain, Chances Peak and Gages Mountain. The steaming area is above the Castle Peak dome within English's Crater. The eastern and northeastern flanks in the foreground are composed predominantly of block-and-ash flow deposits related to previous dome eruptions in the 31-16ka BP period.

attention to the new or surprising scientific observations, results and ideas that have emerged from the eruption, which have either advanced the science or at least have posed new questions that science must answer. The tasks of assessment and mitigation of volcanic hazards, analysis of risk, forecasting of hazardous activity and provision of advice to authorities in a volcanic crisis, are underpinned by basic understanding of how volcanoes work. The results of the scientific work on the Soufriere Hills emphasize the benefits from monitoring as many parameters as possible using the most modern techniques. The eruption has been documented by the Monterrat Volcano Observatory (MVO) in great detail. General outlines of the eruption can be found in Young et al (1998b) and Robertson et al. (2000). Many of the preliminary results have now been published in Volume 25 of Geophysical Research Letters (1998) and also in

Fig. 3. (a) Monitoring team at Whites, about 2.6 km to the east of the active dome in English's Crater. The photograph shows instruments used in ground deformation studies: electronic distance measuring (EDM) and differential GPS surveys.

several miscellaneous publications. Other papers in this Memoir develop some of the results and ideas in more detail. Synopsis of the eruption The eruption of the Soufriere Hills Volcano was the first historic extrusion of magma on Montserrat since its settlement in 1632. When seismic activity rose above background levels in 1992 there was no immediate cause for concern, because previous seismic crises had not led to eruption. Between 1992 and July 1995 there were 18 seismic swarms (Ambeh et al. 1998; Young et al. 1998b; Aspinall et al. 1998; Robertson et al. 2000). On 18 July 1995 the eruption began with steam jets accompanied by numerous felt earthquakes. Steam explosions in the 1 km diameter English's

48

R. S. J. SPARKS & S. R. YOUNG

Fig. 3. (b) Use of a theodolite to make topographic surveys of the evolving dome. Photograph taken from Galway's Mountain in January 1999 after major collapses in July and November 1998 had created a large chasm in the dome, effectively splitting it into two parts. Table 1. Chronology of Soufriere Hills eruption Time

Activity

1992-July 1995

Precursor earthquakes in 18 distinct swarms.

18 July 1995

Start of eruption with phreatic explosions, fumarolic jets and condensed steam plumes.

18 July-15 November 1995

Phreatic explosions with cold pyroclastic surges, steam jets and felt earthquake swarms. Notable events: 21 August explosion which resulted in a cold pyroclastic surge moving through Plymouth and in late September appearance of oxidized spine in English's Crater.

15 November 1995

First unambiguous appearance of magma at the surface and start of continuous dome growth in English's Crater.

15 November 1995-January 1996

Slow (0.2-0.5m 3 s - 1 ) dome growth characterized by development of prominent spines.

February-March 1996

Period of increased dome growth (c. 2m 3 s - 1 ) starting in first week of February with prominent swarm of volcanotectonic earthquakes. First rockfalls and minor block-and-ash flows observed.

April 20-July 1996

Period of continued dome growth with increasingly large pyroclastic flow's down the Tar River valley. Prominent flows on 10 April and 12 May 1996. reaching the coast in the later event.

20 July-16 September 1996

Following a prominent volcanotectonic earthquake swarm on 20 July 1996 and acceleration in ground deformation, dome growth increased to 3-5 m 3 s - 1 , leading to a major dome collapse on 28 July 1996. Growth at 2-2.5m 3 s -1 continued in August and September, with large block-and-ash flows from dome collapses on 12-14 August and 2-3 September.

17 September 1996

A period of 9 hours of continuous dome collapse and block-and-ash flow generation began at 11:30 LT. Activity quietened about 20:30 LT and substantial sub-Plinian magmatic explosive activity began at 23:50 LT. lasting about 40 minutes. Ballistic blocks of 1.5 m reached Long Ground (2.1 km from the dome) and eruption plume reached over 14km high. A large scar indicated that 40% of the dome collapsed and the coastal fan of block-and-ash flow deposits was substantially enlarged.

18 September-1 October 1996

Period of quiesence with volcanotectonic earthquake swarm on 25 September.

1 October-12 December 1996

A new dome started to grow in the collapse scar on 1 October and had largely filled the scar by end of October. Dome growth rates were low (0.5-2m 3 s - 1 ) . In late October the Galway's Wall area of English's Crater showed signs of instability. November and early December were characterized by intense hybrid earthquake swarms alternating with aseismic periods of dome growth.

12 December 1996-February 1997

Uplift of the SE part of the dome on 12 December 1996 heralding an acceleration of dome growth with a major new flow emerging on 25 December 1996. Extrusion rates reached 3-4 m3 s-l in January. Major block-and-ash flows occurred on 19 December and in mid-January.

March-April 1997

The focus of dome growth moved to the SW, with a new lobe appearing above Galway's Wall. Dome growth was slower (l-2m 3 s - 1 ) . First major block-and-ash flows moved down the White River valley on 31 March to 3 April.

May 25-June 1997

Dome growth substantially increased to 7-8m 3 s - 1 during May 1997 with a switch of growth to the north above Farrell's Wall (the northern rim of English's Crater) between 14 and 17 May. The dome, with a volume of over 60 x 10 6 m 3 , completely infilled English's Crater. Rockfalls and block-and-ash flows for the first time moved down Mosquito and Tuitts Ghauts on the northern flanks of the volcano during June.

ERUPTION OF SOUFRIERE HILLS VOLCANO: OVERVIEW

49

Table 1. (continued) Time

Activity

25 June 1997

Major collapse of the dome involving 6.4 x 106 m3 occurs at 12:55 LT with three pulses over 20 minutes. Block-andash flows nearly reached Bramble Airport and pyroclastic surges swept over Farrell's Plain. 19 people lost their lives and several were injured. A surge-derived pyroclastic flow moved down the Belham River valley reaching, close to the village of Cork Hill.

25 June-3 August 1997

Continued fast extrusion with dome growth over the Gage's Wall at the head of the valley leading to Plymouth. Major dome collapse on 3 August caused block-and-ash flows to move through central Plymouth with much destruction.

3-12 August 1997

First period of repetitive Vulcanian explosions, one approximately every 12 hours. Each explosion lasted about 1-2 minutes, followed by waning gas venting. Fountain collapse generated pumice-and-ash flows extending up to 5km down several drainages on flanks of volcano. Ballistic blocks of 1-2 m diameter thrown to 1.7km and pumice lapilli and ash dispersed over island. Extrusion rate in this period averaged about 10 m3 s- 1 . Explosion crater 300m in diameter formed.

12 August-21 September 1997

Rapid (7-8 m3 s - 1 ) dome growth resumed filling explosion crater. Instability caused block-and-ash flows in September, culminating in major (14.3 x 10 6 m 3 ) collapse on 21 September. The block-and-ash flow destroyed buildings in the Spanish Point area and the terminal building at Bramble Airport.

22 September-21 October 1997

Second period of repetitive Vulcanian explosions. There were 75 explosions with an average interval of 9.5 hours. Explosions similar to early August with many exhibiting fountain collapse and pumice flow generation. Eruption plumes were up to 15 km high. Average extrusion rate was about 10m 3 s - 1 . A 300m diameter explosion crater formed.

22 October-2 November 1997

Dome growth resumed filling the explosion crater, with a new lobe extending out towards the northern flanks.

2-6 November 1997

Dome growth switched to SW above Galway's Wall. Two major dome collapses occurred on 4 and 6 November with block-and-ash flows (c. 8 x 10 6 m 3 ) being emplaced down the White River valley and forming a coastal fan. Prominent earthquake swarms accompanied this activity.

6 November-25 December 1997

Dome growth at 7-8 m3 S-1 resumed in collapse scar of 6 November and continued throughout period with relatively minor seismic activity. Dome reached largest volume (113 x 106 m3) and greatest height (1030 m) by 25 December.

25-26 December 1997

A major hybrid earthquake swarm started on Christmas Day and built up through the day. Hybrid earthquakes merged into continuous tremor by end of day. At 03:01 LT on 26 December (Boxing Day) a major sector collapse occurred of the SW flank and the dome in the Galway's Wall and Galways Soufriere area. About 40-45 x 106 m3 of hydrothermally altered rocks failed to generate a debris avalanche, which was emplaced down the White River valley. The dome was immediately undermined and about 35-45 x 106m3 disintegrated to generate a high-energy pyroclastic density current, which devastated 10km 2 of southern Montserrat and swept out to sea. Two collapse scars formed, in the Galway's Soufriere area and in the dome.

26 December 1997-10 March 1998

Dome growth resumed in the collapse scar at 10-11 m3 s-1 and slowed somewhat to 7-8 m3 s-1 building up the dome by early March 1998 to almost the same size as on 25 December 1997. There was a period of vigorous ash-venting in early February. Dome growth ceased on about 10 March 1998 with extrusion of a prominent summit spine.

March 1998-November 1999

Residual volcanic activity continued after the dome stopped growing. Activity included substantial dome collapses with block-and-ash flows (3 July 1997, 12 November 1998, April 1998 and 20 July 1999), small-scale Vulcanian explosions and periods of ash-venting.

November 1999-December 2001

Dome growth resumes with large collapse on 20 March 2000 and 29 July 2001.

LT, local time. Dome collapse volumes are given as the volumes (uncompacted) of the resulting pyroclastic deposits (flow plus surge), as given in Calder et al. (2002). Dome volumes are not corrected to DRE. Time (days)

300

15-Nov-95

200

3-Jul-96

400

600

19-Feb-97

800

9-Oct-97

1000

28-May-98

27-Nov-9

Date

Fig. 4. Variations with time of total magma production, dome volume and pyroclastic flow deposit volume in the 1995-1998 period. Magma production and pyroclastic flow deposit volumes are expressed as dense rock equivalents (DRE: density 2600 kg m - 3 ). Dome volume is not corrected to DRE. The time starts on 15 November 1995. Note that the extrusion of magma stopped in March 1998; subsequently the dome volume decreased and pyroclastic deposit volume increased due to several collapses. Dome growth resumed in November 1999.

Crater became increasingly vigorous during August and September 1995. A small oxidized spine of andesite emerged on the crater floor on 25 September. New andesitic magma reached the surface on 15 November 1995. Table 1 summarizes a chronology of the eruption in the 19951999 period, which escalated in vigour over the first three years (Sparks et al. 1998; Fig. 4). After initial slow growth of the dome in the first few months (0.2-0.5m3 s - 1 ) , dome growth escalated in February 1996 to about 2m3 s - 1 . By March 1996 the first block-andash flows, generated by collapse of the dome, moved down the valleys to the east and flows reached the sea in May 1996, building out new land (Fig. 5). The dome and size of block-and-ash flows continued to increase in 1996, culminating in a major 9-hour-duration collapse on 17 September 1996 followed by the first major magmatic explosive activity (Robertson et al. 1998). At this stage the new dome dominated English's Crater. The Tar River valley had been devastated, and the new coastal fan was over 1 km wide and extended the coast by 600m (Fig. 5c). Although dome growth was sluggish for a while thereafter, activity picked up on 12 December 1996, with the extrusion rate reaching 3-4m 3 s - 1 in January 1997 following appearance of a vigorous new lava lobe on 25 December 1996. By March 1997 the dome had filled much of English's Crater (Fig. 6a) and block-and-ash flows started to inundate the other flanks of the volcano.

50 (a)

(b)

R. S. J. SPARKS & S. R. YOUNG

ERUPTION OF SOUFRIERE HILLS VOLCANO: OVERVIEW

51

(c)

Fig. 5. (a). Typical small block-and-ash flows from dome collapses moving down the Tar River valley in July 1996 showing tapering lofting ash plumes. (b). Block-and-ash flow entering the sea from the Tar River valley on the east coast, (c) View of the Soufriere Hills Volcano in October 1996 from the east showing the new dome within the crater formed by the 17 September 1996 dome collapse and sub-Plinian explosive eruption, devastation of the Tar River valley, and the new coastal fan of block-and-ash flow deposits.

A major collapse on 25 June 1997 resulted in the deaths of 19 people on the northern flanks, which were overwhelmed by a pyroclastic surge that detached from a large block-and-ash flow. Large block-and-ash flows destroyed the centre of the capital, Plymouth, on 3 August 1997 (Fig. 6b), an event which was followed by a series of 13 Vulcanian explosions. Magma extrusion rates were now reaching 7-8m 3 s - 1 (Fig. 4) with all flanks of the volcano threatened. Another major collapse on 21 September 1997 destroyed the airport terminal building and was followed by a series of 75 Vulcanian explosions over the following month, on average recurring every 9.5 hours (Fig. 7a). Dome growth resumed in late October and by late December 1997 the dome had a volume of over 110 x 106 m3. On 25 December 1997 (Christmas Day) a 24-hour intense earthquake swarm was followed on 26 December 1997 by a major failure of old altered rocks and 50% of the dome on the SW flank (Sparks et al. 2002). A large volcanic landslide (debris avalanche) almost reached the sea and was accompanied by explosive disintegration of the dome and a volcanic blast, which devastated 10 km2 of southern Montserrat (Fig. 7b). The dome then regrew to its pre-collapse size by March 1998 (Fig. 8), when growth and magma ascent abruptly stopped. For the next 20 months residual volcanic activity consisted of collapses of the unstable dome, occasional moderate Vulcanian explosions with columns up to 6km a.s.l. and vigorous venting of ash and gas (Norton et al. 2002). Some of the collapses were substantial, creating a deep chasm, effectively slicing the dome into two parts (Fig. 3b). Dome growth resumed in November 1999 and has continued into 2002. The new andesite is porphyritic with 55-65% of phenocrysts and microphenocrysts (Fig. 9a; Devine et al. 1998; Murphy et al. 2000).

The phenocrysts include zoned plagioclase, sometimes with calcic rims surrounding more sodic cores, hornblende, orthopyroxene, titanomagnetite and minor quartz (Fig. 10). Clinopyroxene occurs instead of hornblende in the microphenocryst population. The groundmass consists of variable amounts of high-silica (76-80%) rhyolitic glass and microlites. The magma composition has fluctuated slightly during the eruption with SiO2 in the range 57 to 61.5%, but these variations are not systematic with time. Magmatic mafic inclusions are also ubiquitous within the andesite (Figs 9b and lOd; Murphy et al. 1998, 2000). Petrological and experimental studies (Barclay et al. 1998; Devine et al. 1998; Murphy et al. 1998, 2000) indicate that the magma had a temperature of about 850°C, and that about 4-5% water was dissolved in a rhyolitic melt phase at pressures of 120-140 MPa. These conditions are consistent with a chamber depth of 5-6 km. Petrologically the lava is dominated by disequilibrium features, such as reversely zoned phenocrysts and reaction textures (Fig. 10). Studies of the argon isotope compositions of plagioclase phenocrysts and the hydrogen isotope compositions of amphiboles using the ion probe indicate disequilibria, with old xenocrystal igneous material forming an important component of the magma (Harford et al. 2002; Harford & Sparks 2001). The eruption on Montserrat has had a major impact on the island, with the original resident population of 10 500 being reduced at one time to less than 3500 (Kokelaar 2002). As the eruption escalated, increasing areas of Montserrat were threatened and then abandoned. The escalation was insidious, with activity alternating between periods of apparent calm with low rates of dome growth and periods of heightened activity punctuated by episodes of major dome collapse and episodes of explosive activity. This style of activity over a

52

R. S. J. SPARKS & S. R. YOUNG

Fig. 6. (a) The lava dome in July 1997 towering above the rim of English's Crater. View from the east, (b) Plymouth in August 1996. showing the centre buried by block-and-ash flow deposits (right-hand side) and substantial damage to buildings by pyroclastic surges (left-hand side).

Fig. 7. (a) Typical Vulcanian explosion with pumiceous pyroclastic flows moving down the flanks of the volcano in the main ghauts (valleys), (b) Southern Montserrat shortly after the sector collapse of 26 December (Boxing Day) 1997 looking towards the SE. The White River valley has been infilled with a debris avalanche deposit and the deposits of a high-energy pyroclastic density current. The coastline has been extended by over 400 m along a length of 2 km. The large mountain in the background is the South Soufriere Hills whose north-facing side has been affected by the pyroclastic density current, which reached the summit. In the foreground (left-hand side) is the area of St Patrick's and Morris's villages, which have been completely destroyed, with only the foundations of the buildings remaining.

54

R. S. J. SPARKS & S. R. YOUNG

(a)

(b)

Fig. 8. (a) The lava dome in March 1998 after magma ascent had stopped. There is a prominent late spine on the summit (about 40m high). View from the south. Chances Peak is on the left-hand side of the dome and Galway's Mountain is in the foreground on the right, with the remnants of the connecting Galway's Wall in between buried underneath the new dome. The steaming ridge on the right is the wall of the collapse scar formed in the new dome on 26 December 1997. (b) Close-up of the dome in March 1998 with an excellent example of a spine and the back wall of the collapse scar formed on 26 December 1997.

ERUPTION OF SOUFRIERE HILLS VOLCANO: OVERVIEW

55

Fig. 9. (a) Andesite lava of the Soufriere Hills Volcano showing porphyritic texture with prominent black hornblende and white plagioclase. Field of view about 6cm across, (b) Several mafic inclusions in the andesite host. Camera lens 5cm diameter.

Fig. 10. Key petrological features of the Soufriere Hills andesite and associated mafic inclusions, (a) Electron micrograph of reversely zoned plagioclase phenocryst showing sodic core, sieve textured zones and calcic overgrowth rim with 10um scale bar at the top left, (b) Electron micrograph of reversely zoned orthopyroxene phenocryst showing more Mg-rich rim and homogeneous core with 100um scale bar at bottom right, (c) Photomicrograph (plane-polarized light) with field of view of 3 cm of hornblende phenocryst showing abundant microphenocryst inclusions and thin decompression rim of pyroxenes, plagioclase and oxide (marked by arrow), (d) Photomicrograph (cross-polarized light) with field of view of 3 cm of typical dixytaxitic texture of a mafic inclusion with fine-grained chilled margin and acicular amphiboles and plagioclases.

56

R. S. J. SPARKS & S. R. YOUNG

four-year period has presented challenges to the management of the volcanic crisis. Eventually about two-thirds of the island was abandoned and major assets of Montserrat as an island nation destroyed, including the capital Plymouth, the port, rice mill and airport. Rebuilding the island community constitutes a major challenge.

Slow escalation of the eruption Perhaps the most surprising aspect of the eruption of Soufriere Hills Volcano has been its slow escalation. Data on the volume of erupted magma against time (Fig. 4) show an increase, albeit rather irregular, in magma extrusion rate with time between mid-November 1995, when magma started to discharge continuously at the surface, and March 1998, when dome growth ceased. Also the intensity and magnitude of pyroclastic flows and explosive activity tended to increase as the dome grew larger and the extrusion rate increased (Table 1). Most of the large pyroclastic flows that formed by dome collapse occurred in the second half of 1997. There were two series of repetitive Vulcanian explosions (4 to 12 August 1997 and 22 September to 21 October 1997). The largest and most destructive activity of all occurred on 26 December 1997, only two months before dome growth ceased in March 1998 (Fig. 8). The escalation of activity is also mirrored in the increasing height of the summit of the dome (Watts et al 2002). A pattern of the eruptive behaviour that follows long periods of dormancy has been widely recognized at volcanoes of silicic to intermediate composition. Typically the most intense and usually highly explosive activity occurs in the early stages of the eruption, generally in the first few days or weeks, and then the activity declines, often over periods of months or years. The eruptions of Mount St Helens in 1980-1986 (Christiansen & Peterson 1981), Mount Pinatubo in 1991 (Wolfe & Hoblitt 1996), La Soufriere of St. Vincent in 1979 (Shepherd et al. 1979), Mont Pelee on Martinique in 1902-1906 (Lacroix 1904; Anderson 1908) and Mount Lamington, Papua New Guinea, in 1951-1955 (Taylor 1958) all conform to this pattern, although there are major differences between each of these eruptions in terms of eruptive style, intensity and magnitude. Mount Unzen in Japan, which has often been seen as closely analoguous to Soufriere Hills Volcano, also reached its highest magma extrusion rate, and had its only two explosive phases in, the first two months of a fouryear eruption (Nakada el al. 1999). Persistently active and more steady-state volcanoes include Santiaguito in Guatemala (Rose 1973), Galeras in Colombia (Cortes & Raigosa 1997), Lascar in Chile (Matthews et al. 1997) and Popocatepetl in Mexico. Thus the behaviour of Soufriere Hills Volcano, in terms of a major eruption of a long-dormant volcano, appears to be atypical. The slow escalation of the eruption posed problems in forecasting and interpretation in the context of the provision of scientific advice. Early on, the approach of high-viscosity andesitic magma to the surface, after a period of at least 350 years of dormancy, was considered by many scientists as ominous. The worst volcanic disaster of the twentieth century occurred on Mont Pelee, Martinique, in 1902 when a Caribbean volcano similar to the Soufriere Hills Volcano erupted catastrophically (Lacroix 1904) after only a brief period of precursory activity. Similar situations at Mount St Helens in 1980 (Christiansen & Peterson 1981) and Mount Pinatubo in 1991 (Wolfe & Hoblitt 1996) led quickly to major and catastrophic explosive eruptions within only a few weeks. The concern that emerged from comparison of the situation at Soufriere Hills Volcano with these sinister case histories was partly ameliorated by geological studies on Montserrat (Wadge & Isaacs 1988), which found no evidence for large-magnitude (Plinian style) explosive eruptions in the recent history of the volcano. As the eruption evolved there was a temptation (largely resisted) to draw exponential curves through the volume data (Fig. 4) and see the escalation as a build-up to an eruption of a scale that threatened the whole island. The atypical slow escalation and the lack of understanding of the volume data made confident forecasting diffi-

cult and empirical extrapolation questionable. There was always a background concern, however, that the slow escalation might evolve into a runaway situation. Almost as surprising as the slow escalation of the eruption was the abrupt cessation of dome growth in early March 1998 (Figs 4 and 8). In January and February 1998 regrowth of the dome after the major dome collapse of 26 December 1997 began at an extrusion rate of about 8-1 Om 3 s - ! , making this period, in terms of extrusion rate, one of the highest of the eruption. When the dome reached about 1030m a.s.l. in early March 1998. growth slowed and then ceased after final extrusion of a spine (Fig. 8). Any curvefitting through the volume versus time data would have failed to forecast what eventually happened. During the period of residual volcanic activity (March 1998 to November 1999) there was some evidence that magma ascent continued at a much reduced rate. Activity was characterized by occasional swarms of volcanotectonic earthquakes, several substantial collapses of the lava dome to generate pyroclastic flows, moderate Vulcanian explosions and periods of sustained ash venting (Norton et al. 2002). Dome growth resumed in November 1999 and has continued into 2002.

Pulsatory magma extrusion Although extrusion of magma was nearly continuous in the domebuilding phase, it has not been steady. There have been marked pulsations on various timescales from hours to years during the eruption. Historical data indicate timescales of 30 years for episodic magma ascent into the upper crust. Further geochronological data (Harford et al. 2002) indicate that there have been alternations of periods of elevated eruptive activity and dormancy over timescales of the order of 103-104 years. The detailed and various data collected by the MVO during the eruption show that changes of extrusion rate are almost always associated with other changes, such as nature and intensity of seismicity, rate and patterns of ground deformation, gas fluxes as monitored by SO2 observations using correlation spectroscopy and other volatiles using Fourier transform infrared spectroscopy (Young et al. 1998A; Edmonds et al. 2001), occurrences of major episodes of dome collapse with generation of pyroclastic currents, and explosive activity. The pulsations and their relation to geophysical and volcanic phenomena do not follow a single simple pattern. The shortest timescale and most carefully documented cyclic pulsations were documented by the tiltmeters on Chances Peak (Fig. 11). These remarkable cyclic tilt patterns have been described and discussed in detail by Voight et al. (1998, 1999). We therefore provide only a brief summary here. The tiltmeter data show cyclic patterns of inflation and deflation with time periods from 4 hours to about 36 hours. The tilt data show strong correlation with hybrid earthquake swarms, which build in the inflation part of a cycle and decline rapidly as the tilt cycle reaches a peak. The cycles also show a strong correlation with eruptive activity, with the onset of dome growth accompanied by rockfalls and block-and-ash flows or the occurrence of Vulcanian explosions or the onset of vigorous ash and gas venting at the peak of a cycle. The regularity of the cycles over periods of days to a few weeks gave the MVO the possibility of short-term forecasting as discussed by Voight et al. (1998). Tiltmeter data were collected only in the December 1996 to February 1997 and May 1997 to early August 1997 periods, but the occurrence of regular quasi-periodic hybrid earthquake swarms throughout much of the eruption indicates that this kind of periodic behaviour was a ubiquitous dynamical feature. A series of 75 Vulcanian explosions in the period 22 September to 21 October 1997 also provides evidence of a basic mechanism for these pulsations. These observations have led to the idea that the regular short-term pulsations result from the interplay between degassing, pressurization and rheological stiffening in the conduit flow feeding the dome extrusion (Voight et al. 1999; Wylie et al. 1999; Melnik & Sparks 1999, 2002a,b), with likely stick-slip flow in the conduit (Voight et al. 1999; Denlinger & Hoblitt 1999: Wylie et al. 1999).

57

ERUPTION OF SOUFRIERE HILLS VOLCANO: OVERVIEW

CHANCES PEAK TILTMETER 2

19

20

21

22

23

24

25

26

START DAY IS MAY 18 1997 GMT - JULIAN DAY 138 Fig. 11. Example of cyclic tilt pattern over period 18 to 27 May 1997, showing data from the Chances Peak tiltmeter (after Voight et al. 1998). The tiltmeter was approximately 400 m from the centre of the dome with the tilt axis for data shown being approximately radial to the dome centre. The earthquake event frequency in events per hour (right-hand vertical axis) at the Gage's seismometer is shown as histograms. The tilt variation in microradians (left-hand vertical axis) is shown as the continuous curves, (a) Tilt data along the horizontal axis radial to the dome; (b) tilt data along the horizontal axis tangential to the dome. All the instrument output displays the cyclic pattern of deformation and seismicity, with hybrid earthquakes occurring in the inflation periods and rockfall signals occurring during the deflation periods.

There were also pronounced fluctuations of extrusion rate on longer timescales (Sparks et al. 1998). Changes in extrusion rate were commonly related to major switches in the direction of dome growth, with development of lobes in which the lava expanded asymmetrically from the vent along shear zones (Watts et al. 2002). A particular shear lobe often remained active for several weeks and would then stagnate. A new shear lobe would then develop with extrusion in another direction and often an increase in extrusion rate. In many instances, enhanced extrusion rates appeared to precede major dome collapses. In other instances enhanced extrusion rates may have followed major dome collapse. In yet other cases there was no obvious change in extrusion rate associated with a dome collapse. The pyroclastic flows of 29-30 July 1996 are an example in which enhanced extrusion rates preceded a major dome collapse episode. A major earthquake swarm on 20 July 1996, combined with a marked increase in ground deformation (Jackson et al. 1998), was associated with an increase of extrusion rate from about 2-3 m3 s-1 to about 4-5 m3 s - ! . Approximate estimates indicated that extrusion rates exceeded 10m 3 s - 1 in the two days before collapse (Sparks et al. 1998). After dome collapse, growth continued at an elevated

rate until the collapses of 12-14 August 1996. In contrast, the major dome collapse and sub-Plinian explosive activity of 17 September 1996 were preceded by extrusion rates of about 2 m 3 s - 1 in September, followed by a two-week hiatus before dome growth resumed on 1 October 1997 at an extrusion rate of about 2m3 s - ! . The explosive dome collapses of 19 December 1996 and major collapses of mid-January 1997 occurred in a period of generally accelerating dome growth. An important cause of dome instability was the relationship between the active dome lobes and topography. The new dome developed over the Castle Peak dome, which formed an irregular platform, with a steep slope on the eastern margins in the early stages and on all margins once the new dome had breached the low points on the rim of English's Crater. When dome growth occurred on the top of Castle Peak or within the scar of a previous collapse it was stable. Once the dome had reached a size where the flow front reached the steep front of Castle Peak or the outer walls of English's Crater, large collapses inevitably followed (Watts et al. 2002). In principle, major collapses should be followed by enhanced extrusion rate due to the sudden reduction in load pressure of the

58

R. S. J. SPARKS & S. R. YOUNG

dome on the conduit. There is some evidence that this happened. As dome height increased, the time required for growth back to the height before collapse decreased (Watts et al. 2002). This observation, however, merely reflects the general increase of extrusion rate and dome size with time. One clear case of accelerated dome growth occurred after the largest collapse on 26 December 1997, when the dome regrew on the floor of the collapse scar from 650m a.s.l. to over 950m a.s.l. in about a week. The clearest correlation of accelerated extrusion rate and dome collapse is with the onset of explosive activity. All the major episodes of explosive activity almost immediately followed a major dome collapse. Druitt et al. (2002b) estimate that magma extrusion rates reached values of about l O m 3 s - 1 during the two series of Vulcanian explosions in 1997. These high extrusion rates can be linked to the absence of resistance by a dome overlying the conduit in these periods. In fact most of the major increases of extrusion rate occurred quite abruptly, were not associated with any major dome collapse episode, and were often linked to changes in seismicity or ground deformation. This can be seen in Figure 4 and height versus time data (Watts et al. 2002). The first marked increase in extrusion rate occurred in early February 1996, and coincided with a swarm of volcanotectonic earthquakes with depths up to 6km, but no obvious change in ground deformation rates. The large increase in extrusion rate in late July 1996 seems to have begun fairly abruptly, with a volcanotectonic earthquake swarm on 20 July (Miller et al. 1998) and a marked increase in the outward movement of radial lines in the deformation network (Jackson et al. 1998). The marked pulsation of December 1996 to February 1997 followed a period of endogenous growth and strong cyclic ground deformation, and was initiated by a sudden uplift of the SE sector of the dome with movement along a well-defined shear zone on 12 December 1996 (Watts et al. 2002). This uplift preceded extrusion of a major new lobe on'25 December 1996 and acceleration of extrusion rate. The most significant increase in extrusion rate occurred in May 1997, when the long-term extrusion rate increased by over a factor of two from about 2.5m 3 s -1 to between 4 and 5m 3 s - 1 . This change was linked to onset of repetitive hybrid swarms (although not to any discrete volcanotectonic earthquake swarm), to increased outward movement rates on the northern flanks, and to a general reorganization of ground deformation patterns away from simple radial movements to more complex movements focused on the northern flanks. These major shifts in extrusion rate were accompanied by shifts in the direction of dome growth. These observations indicate that the major changes in extrusion rate had a deep origin unrelated to shallow-level processes such as dome collapses. However, they also indicate some complexity. The most compelling evidence for pulsatory cyclic behaviour on a timescale of several weeks comes from tilt and seismic data (Voight et al. 1998, 1999). Data from mid-May to early August 1997 (Fig. 12) shows a gradual deflation overprinted by the shortterm cycles discussed earlier. The onset of three cycles can be identified on 17 May 1997, 22 June 1997 and 31 July 1997. Each cycle began remarkably abruptly; indeed the onset can be identified within less than an hour in the case of the one beginning 22 June 1997 (Voight et al. 1999). The beginning of a cycle consists of a sharp increase in the amplitude and decrease in the period of inflation/deflation tilt cycles, and onset of prominent hybrid earthquake swarms (Fig. 12). In these three examples the intensity of the hybrid earthquake swarms gradually decreases and they gradually disappear. At the same time the amplitude of tilt cycles decreases and their period increases. Significant volcanic activity characterizes the onset of a cycle. For example the major dome collapse of 25 June 1997 followed the onset of the cycle on 22 June, and the major collapse of 3 August 1997 followed the start of the new cycle on 31 July 1997. Unfortunately the tiltmeters were destroyed in the Vulcanian explosions of early August 1997, but seismic data and eruptive events indicated that the several-week cycles continued. The MVO inferred that the subsequent cycle started on 21 September 1997, with a major dome collapse followed by a four-week series of

31 July

22 June

17 May

urad

20

l() JUN JUL AUG START DAY IS MAY J7 IW GMT - JULIAN DAY 137

Fig. 12. Tilt and seismic patterns at Montserrat in the period May to August 1997. The vertical lines represent a hybrid earthquake swarm of the kind shown in detail in Figure 11. The tilt meter data are for the radial component of the Chances Peak tiltmeter 2 (see Voight et al. 1998). The dates of the start of each major cycle are indicated at the top of the diagram.

repetitive Vulcanian explosions. Another cycle began in early November with the onset of intense hybrid swarms, a major switch in dome growth direction (Watts et al. 2002), and major dome collapses on 4 and 6 November 1997. The start of the next cycle was on 24 December 1997 with a major hybrid swarm leading to the major flank failure and dome collapse of 26 December 1997. Strong seismicity and ash-venting in early February 1998 may represent the start of a final cycle. These timings give 36, 52, 43, 52 and 43 days for the recurrence intervals of the cycles. The possibility of cycles with major events every several weeks was recognized by the MVO and used in communications to the public on Montserrat to warn of the possibility of escalations in activity. The several-week cycles were identified from May 1997 onwards, but it may in fact be misleading to consider them simply as time cycles. One alternative interpretation is to consider them as volume cycles in which an approximately equal volume is erupted in each cycle (Table 2). In this context the five cycles each have a volume of about 30 x 106 m3 (Table 2). The origin of these cycles is unclear and will be a focus for future research. The suddenness of their onset suggests a deep origin (Voight et al. 1999). The hypothesis that they are volume rather than time cycles implies that they could be related to the size of the magma chamber, which can store an excess volume that is then released in pulses. Another possibility is that convective processes in the chamber are responsible, in which overturn of gasrich magma triggers onset of a cycle (Couch et al. 2001). Finally, the onset of a new period of dome growth in November 1999 indicates a possible cyclicity with a timescale of several years.

Table 2. Estimates of erupted volumes (DREj and durations of larger scale eruptive cycles on Montserrat Time interval

17 May-22 June 1997 31 July-21 Sep. 1997 21 Sep.-4 Nov. 1997 4 Nov.-26 Dec. 1997 27 Dec.-7 Feb. 1997

Volume (x 10 6 m 3 )

Length of cycle (days)

26.1 29.7 34.6 30.8 28.0

36 52 43 52 43

ERUPTION OF SOUFRIERE HILLS VOLCANO: OVERVIEW

Understanding volcano geophysics Significant advances have been made at Soufriere Hills Volcano in understanding the relationship between dynamics of the eruptive processes and the principal geophysical manifestations that are routinely monitored, namely seismicity and ground deformation. Here we summarize the pertinent geophysical observations and then discuss ideas that have emerged during the eruption on how these observations might be explained. Five main kinds of seismic signal have been recorded in the eruption (Miller et al. 1998; White et al 1998): (1) volcanotectonic earthquakes; (2) a shallow type of earthquake locally termed hybrid; (3) long-period earthquakes; (4) rock fall and pyroclastic flow signals; and (5) explosion earthquakes. Here discussion is focused on the hybrid and long-period earthquakes that characterize dome growth. Hybrid earthquakes on Montserrat contain a mixture of high- and low-frequency components. S waves are not recognizable at seismic stations on the island. Hybrid earthquakes are located at shallow depths of less than 2 km, although accurate determination of depth is made difficult by the lack of detection of S waves. The hybrid earthquakes typically occur in discrete swarms of a few hours' to a few days' duration. During the more intense swarms, tens to hundreds of earthquakes can occur per day (Fig. 11), although the energy of individual earthquakes is low, with magnitudes estimated to be no more than about 3 on the Richter scale. In some of the earlier hybrid swarms of the eruption the earthquakes were repetitive, occurring with similar waveforms and at regular intervals of a few seconds to a few minutes. However, later in the eruption the hybrid swarms were non-repetitive. Occasionally the hybrid swarms merge into continuous harmonic tremor (Neuberg et al. 1998; Neuberg 2000). Long-period earthquakes are an endmember of the hybrid type and consist dominantly of low-frequency signals (1-2.5 Hz). During the eruption some long-period earthquakes occurred simultaneously with vigorous ash and gas venting from fractures at the dome surface, a phenomenon also recorded on the lava dome at Galeras Volcano, Colombia (Gil Cruz & Chouet 1997). Luckett et al. (2002) have shown that rockfall and block-andash flow signals also contain long-period signals. Long-period and hybrid shallow seismicity can be related to movement of high-pressure gases along fractures with low-frequency components being related to resonation on fractures due to flow of high-pressure gas (Chouet 1996) or to the resonation of a bubbly magma conduit (Neuberg 2000). In the case of the Soufriere Hills eruption the source of the gases is attributed to degassing of the magma. Clear links have been established between magma ascent and occurrence of hybrid earthquakes. Hybrid seismicity was dominant while the dome was actively growing. Volcanotectonic earthquakes were the dominant type prior to magma reaching the Earth's surface, but then became subordinate (Aspinall et al. 1998). Hybrid earthquakes also virtually ceased when dome growth abruptly stopped in early March 1998, and this observation was used by the MVO as part of the evidence to infer that magma ascent had indeed ceased (albeit temporarily). There has also been a clear linkage of hybrid earthquake swarms with the short-term cyclic patterns of magma pressurization (Voight et al. 1999). The observations of the cyclic patterns suggest that extrusion rate decreases or is even zero during the hybrid swarms, which occur in periods of inflation (Fig. 11) and internal pressurization in the upper parts of the conduit. The peaks of these cycles are sometimes associated with strong degassing and even Vulcanian explosions, and the deformation is almost entirely recovered in the deflation phase of a cycle. These observations are consistent with the hybrid earthquakes being related to build-up of gas pressure and the development of shallow fracture networks in the magma and in the surrounding wallrocks to the conduit, which allow high-pressure gases to escape. The direct venting of high-pressure gas during some long-period events and long-period components in the rockfall signals (Luckett et al. 2002) also link the seismicity strongly with degassing processes. The development of high overpressures and conditions to develop fracturing can be related to the coupled processes of degassing, microlite crystallization and marked rheological stiffen-

59

ing with development of strength in the ascending magma (Sparks 1997; Voight et al. 1999; Wylie et al. 1999; Melnik & Sparks 1999, 20020; Sparks et al. 2000). The observations at Soufriere Hills Volcano have elucidated the strong link between magma flow, degassing and seismicity. Ground deformation on Montserrat has been monitored by global positioning system (GPS) data, electronic distance measuring (EDM) and tiltmeters (Jackson et al. 1998; Mattioli et al. 1998, 2000; Shepherd et al. 1998; Voight et al. 1998). Conventionally, ground deformation at volcanoes has been attributed largely to pressure changes in a magma chamber. Near-field ground deformation at Soufriere Hills Volcano may also be interpreted as a shallow phenomenon involving pressurization within the upper parts of the conduit due to magma flow (Melnik & Sparks 1999). Pressurization could also cause movement along reactivated faults in the shallow crust. Shepherd et al. (1998) showed that ground movements in 1996 were radial from the dome and that there was a systematic decrease of outward velocity with distance. They were able to model these movements well by invoking a shallow pressure source at less than 750m below the dome with an overpressure of about 10 MPa. Voight et al. (1999) modelled radial tilt data recorded on instruments close to the dome as a shallow pressure source at about 400m depth. In contrast Mattioli et al. (1998) interpret their high-precision GPS data on vertical displacements in terms of a deflating magma chamber at 6 km depth with horizontal displacements related to deformations associated with dykes for the period of major surface magma flux from 1995 to 1997. More recent ground deformation data from early 1998 to late 1999 may involve movement of fault blocks, particularly on the eastern flanks of the volcano, in response to magma pressure and loading by the dome (Norton et al. 2002). From February 1998 until November 1999, continuous GPS sites showed inflation or uplift relative to the Earth's centre of mass (Mattioli et al. 2000) during the period of no surface magma flux. GPS data from the period 1998 to late 1999 are best fitted by an inflating point-source at approximately 6 km depth, embedded in a simple elastic half-space (Mogi 1958). The derived depth is similar to that obtained by Mattioli et al. (1998) for the GPS data from 1995 to mid-1997, although the dilation is opposite in sign and about one-half the magnitude. As with seismicity, there is a strong link between ground deformation and the dynamics of magma ascent and shallow conduit pressurization (Voight et al. 1999). These observations have spawned new explanations for the shallow pressurization based on models of magma flow in conduits, which take into account degassing and groundmass crystallization (Fig. 13). Degassing induces very large changes of the rheological properties in silicic magmas as they ascend, decompress and exsolve gas (mainly water). Much of the frictional resistance to flow from the magma chamber occurs in the last several hundred metres of flow where very high viscosities are developed (Sparks 1997; Melnik & Sparks 1999). Mathematical models of the pressurization indicate that an overpressure maximum consequently developed in the uppermost few hundred metres of the conduit (Melnik & Sparks 1999, 20020). The value of the pressure maximum is calculated to be several megapascals and is therefore comparable to the strength of the volcanic edifice and to estimates of the overpressures needed to cause the ground deformation (Shepherd et al. 1998). Another factor in pressurization is the development of overpressures caused by microlite crystallization (Stix et al. 1997; Sparks 1997). Large pressure increases are expected as crystals grow from undercooled and partially degassed melt. Sparks (1997) and Melnik & Sparks (1999) showed that the maximum in overpressure is expected at depths of several hundred metres in volcanic conduits. Role of degassing in dome emplacement The dome has displayed a wide range of behaviour, patterns of growth and surface morphologies, as described by Sparks et al. (1998, 2000) and Watts et al. (2002). An important notion emerging from the documentation and study of the dome growth is that

60

R. S. J. SPARKS & S. R. YOUNG

Gas Plume

Fig. 13. Schematic illustration of the key features of the variation of porosity and overpressure in the dome and conduit. Overpressure is defined as the difference between the magma pressure and lithostatic pressure. The overpressure reaches a maximum at several hundred metres below the dome and causes a focus for deformation of the wallrocks and edifice. The porosity reaches a maximum just below the exit of the conduit. Within the dome, pressurized gases escape particularly along fractures and shear zones to form gas plumes emerging from the dome surface.

Porosity Maximum

Overpressure Maximum

Deformation

emplacement is dominated by solidification caused by degassing rather than by cooling (Sparks et al. 2000). In models and studies of lava dome emplacement, emphasis has been given to the role of cooling in forming a resistant carapace to the lava, which increases in viscosity, thickness and strength with time. The carapace slows the advance of the lava and eventually brings it to a halt (e.g. Fink & Griffiths 1990; Griffiths 2000). Less attention has been given to the role of degassing, in which the exsolution of gas causes a marked increase in the liquidus of the melt phase. In the Soufriere Hills case the crystal-rich magma had an estimated temperature of about 850CC with a rhyolitic melt phase containing 4-5% water in the magma chamber (Barclay et al. 1998; Murphy el al. 2000). On decompression and degassing to the Earth's surface pressure, the liquidus of the melt phase increases by at least 200CC and there is pervasive groundmass crystallization as a consequence. The magma undergoes profound rheological stiffening, being transformed from a crystal-rich Newtonian magma to a hot crystalline solid with small amounts (<10%) of residual rhyolitic melt. The initial viscosity of the Soufriere Hills magma is estimated as around 7 x 10 6 Pas and the final degassed properties indicate a viscosity exceeding 1014 Pa s and strength of at least 1 MPa (Voight el al. 1999; Sparks el al. 2000). Essentially, solidification induced by degassing results in an immobile dome, which is capable of only limited lateral spreading and whose emplacement can involve heterogeneous deformation along ductile shear zones (Watts el al. 2002). We now outline the observations that support such a view. The importance of groundmass crystallization can be assessed by comparing samples that have erupted at different rates. Pumice samples from the explosive eruptions were erupted rapidly and contain up to 35% rhyolitic glass and only sparse microlites (Murphy el al. 2000). However, the groundmass of most samples from the dome is highly crystalline with typically only 5-10% residual rhyolitic glass and sometimes abundant cristobalite (Baxter el al. 1999). The dome samples were collected from block-and-ash flow deposits from major collapses where the large collapse scars show the interior of the dome at 100 to 200m depth. Most of the material in the block-and-ash flow deposits must be derived from the interior of the dome, where cooling cannot have been significant. External cooling by conduction over a few weeks or months can only affect a few metres of the near-surface dome material, which is a small proportion of the collapse material. Good observations of the dome interior were made sometimes within a day of a major collapse (Figs 3b and 14a). The interior of the dome is pervasively fractured with some large faults. For much

of the eruption much of the extrusion of the dome has been along ductile shear zones as spines and asymmetric lobes (Watts el al. 2002). Spines (Fig. 8b) typically extrude at rates of a few metres per day and lobes at a few tens of metres per day. Both spines and lobes show discrete striated surfaces. Lava extrusion is often confined to the summit area and only in a few cases was significant lateral spreading of lava observed to form pancake morphology (Fig. 14b). These observations demonstrate that the dome typically does not deform like a Newtonian fluid, but deforms in highly localized zones. The curved character of the striated surfaces (Watts et al. 2002) suggests that the conduit wall provides the main detachment surface. The pervasive internal fracturing shows that the dome is capable of brittle fracture in its deep interior. Hydrous porphyritic andesite like the Soufriere Hills magma is the most viscous kind of magma on Earth. The combination of high original crystal content, rhyolitic melt composition, low eruption temperature and extensive groundmass crystallization on loss of a few per cent water during ascent gives exceptionally high viscosities in the magma, which is on the limits of being capable of erupting. Pyroclastic processes The eruption of Soufriere Hills Volcano has been an outstanding opportunity to investigate pyroclastic processes. During the eruption there have been two main kinds of pyroclastic density current: those formed by dome collapse and those formed from fountain collapse during Vulcanian explosions (Cole et al. 1998, 2002). There was also a high-energy and very destructive pyroclastic density current, or volcanic blast, formed during the sector collapse of 26 December 1997 (Sparks et al. 2002) and surge-derived pyroclastic flows formed by rapid sedimentation from pyroclastic surges (Druitt et al. 2002a). There have been many tephra falls from the lofting plumes above the pyroclastic flows and from the eruption columns associated with the magmatic explosive activity (Bonadonna et al. 2002). These processes have been directly recorded by video, including the unique opportunity to observe block-and-ash flows moving into the sea (Fig. 5b). Here we draw attention briefly to some of the important findings that are discussed in detail in many of the papers in this volume. One interesting issue is the role of pore pressure in the generation of dome-collapse block-and-ash flows (Sato et al. 1992; Fink & Kieffer 1993). High pore pressures are indicated by a number of

ERUPTION OF SOUFRIERE HILLS VOLCANO: OVERVIEW

61

(a)

(b)

Fig. 14. (a) The interior of the dome exposed after a major collapse of the dome on 17 September 1996, revealing that the interior is heavily fractured. The photograph was taken of the headwall of the collapse scar eight days after formation and shows about 70m of exposed dome interior. Large fractures of at least 50m in length can be seen, (b) Pancake flow morphology, where the dome has spread laterally. The picture was taken on 5 October 1996. This dome lobe appeared on 1 October 1996 in the collapse scar of 17 September 1996.

observations. Many collapses that generate block-and-ash flows have an impulsive element recorded in the seismic signals, with the majority of signals containing low-frequency components consistent with gas expansion (Luckett et al. 2002). Occasionally, clasts were observed to be ejected from flows, consistent with impulsive gas expansion. Also there were often jets of high-pressure gas and condensed steam emerging from fractures in the dome surface, implying high internal pressures. All the block-and-ash flows developed pyroclastic surges and lofting buoyant ash plumes. Data presented by Cole et al. (2002), however, show that the development of the pyroclastic surge is not related simply to block-and-ash flow volume. At Soufriere Hills Volcano the block-and-ash flows range from small rockfalls to major collapses with volumes of several million cubic metres (Cole et al. 1998, 2002). Cole et al. (2002) document that, for flows of similar volume, some have very weak pyroclastic surge development and others have very widespread and violent pyroclastic surges. The most compelling case for high internal pore pressures comes from studies of the deposits of the

dome collapse of 26 December 1997, when a high-energy pyroclastic density current formed as a consequence of the largest dome collapse of the eruption. Sparks et al. (2002) and Ritchie et al. (2002) describe observations and evidence from this event and its products, which imply that the pyroclastic density current was initiated from a dispersion formed by explosive expansion of pressurized pore gases as the dome rock disintegrated during collapse. Woods et al. (2002) have modelled these processes by considering the expansion of high-pressure gas within the dome as collapse takes place. They estimate that the expansion phase occurs on the order of 10 seconds and creates marked gradients in particle size and concentration during radial expansion. In this process fine particles are transported with the expanding gas while large particles are left behind due to their inertia. As a consequence the current becomes stratified in both grain size and density. The rapid expansion of the pore gases created a thick (hundreds of metres), dilute but density-stratified mixture, which then spread across the landscape under gravity.

62

R. S. J. SPARKS & S. R. YOUNG

Observations of surge and blast development can be explained by variable pore pressures within the dome. As water-rich andesite magma ascends, gas exsolution occurs as pressure reduces. Near to the surface, gas can start to escape when the magma develops permeability (Eichelberger et ai 1986; Jaupart 1998) and strong gradients in pore pressure can be expected as the gas changes from magmatic pressures to atmospheric pressure at the dome surface. That this happened at Soufriere Hills Volcano is certain. A persistent gas plume was always observed emerging from the dome (Young et al. 1998a), the dome rocks had porosities and residual water and chlorine contents (Edmonds et al. 2001) implying that most gas had been lost during ascent and extrusion, and dome rocks show extensive groundmass crystallization attributed to gas loss. The block-and-ash flow deposits acted as significant sources of gas for the few days following emplacement, indicating that residual gas was still being released. The gas pressure within the dome will depend on a number of factors, of which permeability and extrusion rate are the most important. Models estimating gas pressures in the upper conduit (Melnik & Sparks 2002a) and in the dome (Woods et al. 2002) indicate typical values of several megapascals gas pressure at the base of the dome and top of the conduit. The effectiveness of pressurized gas expansion can be comprehended with the knowledge that dome rock with 2 MPa internal pressure and 10% porosity (a gas mass fraction of 5 x 10 - 4 ) will expand by a factor of 50 when equilibrated to one atmosphere. High internal gas pressure can help explain the variability in pyroclastic surge development. Small collapses sample the exterior of the dome where gas pressures are relatively low so that sub-

stantial surge clouds are not expected. Larger collapses, sampling deeper into the dome, tend to sample regions with higher pore pressures, which results in more substantial pyroclastic surges as the rock mass disintegrates into a block-and-ash flow. However, a simple relationship between pyroclastic surge development and dome volume is not expected, because pore pressure also depends on the age of the mass that collapses. Collapses that take place from active regions of dome growth are expected to have higher pressures than collapses of older parts of the dome, where time has elapsed to allow dissipation of pore pressure. As discussed by Cole et al. (2002). there are several factors in controlling pyroclastic surge development, including topography. An important discovery from the Soufriere Hills eruption, with particular relevance to volcanic hazards, is the generation of surgederived pyroclastic flows, and the ability of these flows to move down drainages in unexpected directions well outside the area inundated by the surge cloud (Druitt et al. 2002a). On 25 June 1997 pyroclastic surges detached from major block-and-ash flows at a prominent bend in the valley, in which the dense concentrated parts of the flows were confined (Loughlin et al. 2002). The block-and-ash flow travelled down drainages to the NE. while detached surges moved northwards over Farrell's Plain and were then blocked by the opposing hillside (Fig. 15). The deceleration of the surge cloud generated a dense pyroclastic flow by rapid sedimentation, and this drained into the Belham River valley where it moved about 3 km to reach the outskirts of the still-populated village of Cork Hill. A similar process occurred on 26 December 1997. when the upper parts of the pyroclastic density current spilled over the saddle between

Fig. 15. Map showing the distribution of 25 June 1997 block-and-ash flow deposits (after Cole et ai. 1998. 2002). The surge-derived pyroclastic flow deposits in the Belham River valley formed from the pyroclastic surge that detached from the main flow in Mosquito Ghaut and swept across Farrell's Plain and up Windy Hill. The surge-derived pyroclastic flow was formed by rapid sedimentation from the surge cloud (Druitt et al. 2002a) and reached near the village of Cork Hill, which was occupied.

Galway's Mountain and the South Soufriere Hills. Here another dense surge-derived pyroclastic flow formed and travelled 2km down Dry Ghaut. Similar phenomena have also been called secondary pyroclastic flows, and have been recognized at Mount St Helens, where they formed simultaneously by drainage of surge deposits off steep slopes (Hoblitt et al. 1981). The new observations at Montserrat show that such flows can be formed by rapid sedimentation from dilute surge clouds as they decelerate on quite modest slopes. The observations also establish that such flows are capable of flowing well outside of the area inundated by the parent surge cloud. The Soufriere Hills eruption has also given the opportunity to compare the mobilities of different kinds of pyroclastic flow. Calder et al. (1999) compiled data on the runout distances, volumes, areas inundated and height changes of the pyroclastic flows from the eruption. They analysed the data in terms of a new model for rock avalanches (Dade & Huppert 1999). The pyroclastic flows from

Fig. 16. (a) Slickensided friction mark with polished slickensided surface on a block from the 21 September 1997 block-and-ash flow deposit. Camera lens is 5 cm diameter, (b) Backscattered electron image of a polished thin-section of a friction mark showing pseudotachylite overlying cataclasite with a transitional zone of pseudotachylite and crystal fragments. The shear surface is at the bottom of the photograph.

dome collapses had comparable mobilities to cold rock avalanches, as previously recognized (Sparks 1976; Hayashi & Self 1992). The pumice flows and surge-derived pyroclastic flows in contrast were much more mobile, inundating much larger areas than block-andash flows of comparable volume. It is not yet clear what are the critical factors in explaining the marked differences in mobilities. Calder et al. (1999) related mobility to the observation that the more mobile flows were generated by rapid sedimentation from dilute suspensions, whereas the block-and-ash flows were generated by disintegration of dense rock masses. Another new observation on Montserrat is the occurrence of friction marks on the large blocks (> 1 m) in pyroclastic flow deposits (Grunewald et ai 2000). The friction marks occur as polished and striated slickensided surfaces on a thin layer (5-10 mm) of cataclastic rock (Fig. 16a). Friction marks occur on all sides of blocks, commonly in several directions. Petrographic studies show

64

R. S. J. SPARKS & S. R. YOUNG

that the friction marks are similar to tectonic mylonites and that in the largest marks frictional melting took place to form pseudotachylites (Fig. 16b). Grunewald et al (2000) calculated that the imposition of the weight of a large block (1-10 m) on a small area of a few square centimetres with relative sliding velocities of a few metres per second is sufficient to cause confining stress conditions that are comparable to those experienced on large tectonic faults and to cause frictional melting at over 1400CC. A major implication of the formation of friction marks is that Bingham or sliding-block models of pyroclastic flows cannot be realistic. The flows are best perceived as rapid agitated granular flows in which blocks jostle, rotate and occasionally slide past one another with strong frictional contacts (Drake 1990; Druitt 1998). The magmatic explosive activity during the eruption has provided a superb opportunity to study Vulcanian explosions, which were recorded in detail by video, photography and seismicity (Druitt el al. 2002b). The explosive activity always followed a major collapse of the dome and thus was triggered by rapid unloading of the conduit. As discussed previously and in Voight el al. (1999), the two the series of Vulcanian explosions in 1997 involved repetitive and cyclic explosions over extended periods that can be linked to ground deformation and seismic cycles. Typical Vulcanian explosions lasted a few tens of seconds and were followed by vigorous ash-venting lasting from 30 minutes to an hour. Eruption columns ranged from 3 to 15km high and all but one of the Vulcanian explosions involved fountain collapse, generating pumice-and-ash flows, which were emplaced along all the main drainages (Fig. 7a). The explosions were pulse-like and involved ejection of ballistics up to 1.7km from the dome. The origin of the cyclic and pulse-like character of the explosions has been considered by Melnik & Sparks (2002b), who found that pulse-like behaviour is predicted if the magma fragments at some critical overpressure. Pumice clasts from the explosions are typically angular and platy (see photographs in Druitt et al. 2002b) and are similar to the shapes of clasts produced in laboratory experiments by Adilbirov & Dingwell (1996), in which natural dome rocks and pumices, containing a high-pressure gas, were abruptly decompressed. This similarity supports the idea of brittle fragmentation of a vesicular magma as the cause of the explosions. Melnik & Sparks (20026) found that the characteristics of the Vulcanian explosions can be described by models of the explosive expansion of already vesiculated magma in the upper conduit with no further mass transfer during the event. The eruption has also provided the opportunity to compare models of interaction with the atmosphere and fountain collapse during explosive eruptions, as discussed further by Clarke et al. (2002). The eruption has given the opportunity to investigate the contrasted features of tephra fall deposits generated from plumes above block-and-ash flows related to dome collapses and from magmatic explosive eruptions. The dispersal of tephra is dominated by the local wind conditions in the Caribbean, with low-level (up to 4-6 km) winds blowing to the west and high-level winds to the east. As expected, the tephra deposits from the lofting ash plumes above the block-and-ash flows are very fine-grained and aggregation processes, often with formation of accretionary lapilli, are dominant (Bonadonna et al. 2002). In contrast, the Vulcanian tephra fall deposits are usually strongly bimodal as a consequence of having two sources of tephra, namely relatively coarse-grained lapilli and ash derived from the high vertical columns above the vent, and finegrained plumes of low height generated above the pumice flow (Bonadonna et al. 2002). The abundant cristobalite discovered in the finest respirable components of the ash fall deposits associated with block-and-ash flows has posed a possible health threat to the islanders (Baxter et al. 1999). The cristobalite forms by vapourphase crystallization and devitrification in the dome and is concentrated in the fine elutriated ash by selective crushing of the groundmass in the pyroclastic flows. Detailed geochemical studies (Horwell et al. 2001) show that the physical fractionation is a consequence of the fine-grained groundmass being significantly weaker than the phenocrysts. Baxter et al. (1999) found that the dome rock contained 3-6% cristobalite, but the respirable ash (<10/um) in the tephra fall deposits contained 10% to over 25% cristobalite. The

occurrence of abundant fine cristobalite in fine-grained tephras and the longevity of the eruption with persistent ash production have raised concerns about the long-term health effects on the population. Cristobalite is twice as toxic as quartz and with long-term exposure can lead to illnesses such as silicosis. An interesting feature of the studies of Baxter et al. (1999) is the observation that concentrations of suspended fine respirable ash in the air can be significantly greater days or weeks after an eruption in dry conditions. When ash first falls, much of the fine ash is sequestered in aggregates and accretionary lapilli. When the ash dries the aggregates are broken up by human activities such as sweeping, driving and walking, releasing the fine particles into suspension.

Role of basaltic magma The importance of basaltic magma in silicic volcanism in arcs and continents has been widely recognized (e.g. Sparks et al. 1977; Heiken & Eichelberger 1980; Hildreth 1981; Clynne 1999; Miller et al. 1999). The emerging results on the petrology, geochemistry and volcanology of the Soufriere Hills eruption (Devine et al. 1998; Murphy et al. 1998, 2000) all point to the fundamental role of hydrous basalt magma invading an upper crustal, highly crystalline magma body of andesitic composition to trigger and drive the eruption. Similar petrological relationships and interpretations to those described from Montserrat have been documented for the lava domes of Mount Unzen (Nakada & Motomura 1999), Mount Dutton, Alaska (Miller et al. 1999). and Lascar Volcano. Chile (Matthews et al. 1999). Two key observations of the new lava indicate remobilization of a body of highly crystalline magma or partially molten rock by invasion of hydrous basaltic magmas. First, the andesite contains ubiquitous mafic inclusions of 1-50cm diameter (Figs 9b and lOd). which make up about 1% of the rock. The mafic inclusions show the characteristic features of quenching of magma, with acicular amphiboles and plagioclase in a glassy matrix (Murphy et al. 1998, 2000). Second, the andesite shows strong evidence of reheating, as detailed by Devine et al. (1998) and Murphy et al. (2000). The andesite was generated by remobilization of hot solid igneous rock and partially melted rock rather than a mobile magma. The injection of mafic magma added heat and probably volatiles to the andesite source, remelting it to sufficient mobility that it became a crystalrich magma, which was able to erupt. The mafic magma may also have added chemical components although there is no unequivocal evidence for this beyond the existence of the mafic inclusions. An important aspect of the petrological features of the andesite is the highly heterogeneous populations of crystals. The phenocrysts show evidence of disequilibrium, such as zoning patterns and reaction textures (Fig. 10). which can be related to reheating and remobilization (Couch et al. 2001). A population of a given phenocryst phase is typically highly variable in texture within a single thin-section, implying that different individual crystals have had contrasted histories of growth and resorption. The microscopic heterogeneity is in contrast to the macroscopic homogeneity of the whole-rock chemistry. Murphy et al. (2000) have interpreted these observations to imply that different parts of the system had been heated and remelted by different amounts and that convective stirring had thoroughly mixed up parcels of magma with contrasted thermal histories to produce a rather homogeneous magma. Some andesite blocks also show strong layering (Fig. 17). which relates to very minor compositional and textural variations. The layering might be explained by stirring motions in a convecting chamber. Heterogeneous layering can be expected in the remobilization of crustal rocks or magma chambers by influx of mafic magma on the basis of fluid dynamic principles (Huppert & Sparks 1988). Couch et al. (2001) have proposed that the textures and mineral compositions in the andesite can be explained by a process of convective self-mixing, in which the andesite was heated from below by basalt. The andesite is heated to form a hot boundary layer adjacent to the basalt and rises up into the cooler andesite within the interior of

ERUPTION OF SOUFRIERE HILLS VOLCANO: OVERVIEW

65

Fig. 17. Pronounced flow layering in the andesite, which is interpreted as indicative of strong convective stirring of magmas with slightly different physical properties prior to extrusion. Note that chemical analysis of the contrasted layers shows that they have the same composition within analytical uncertainties.

the chamber. Mixing results in quenching of the heated andesite and formation of a disequilibrium mineral assemblage. There is evidence for the assimilation of older igneous material into the magma. 40 Ar/ 39 Ar studies of feldspar phenocryst separates (Harford et al. 2002) show that there is a significant component of older feldspar xenocrysts with model ages of at least 0.5 Ma. Harford & Sparks (2001) have found from ion probe analyses that amphibole phenocrysts erupted prior to about April 1996 have cores enriched in deuterium. In contrast, the rims of these same phenocrysts and the amphiboles erupted after April 1996 have smaller D/H ratios, which have typical magmatic values. These observations can be explained if the early amphiboles represent preexisting andesitic intrusions that had exchanged hydrogen with hydrothermal fluids. Harford & Sparks (2001) propose that early andesitic intrusions have been heated and assimilated by the new andesite. The preservation of these isotopic heterogeneities also provides time constraints. The 40 Ar/ 39 Ar heterogeneities could not be preserved if the feldspars had been at magmatic temperatures (c. 850°C) for more than a few hundred years. The hydrogen isotope heterogeneities would be eliminated within a few years at these temperatures, indicating that assimilation has been taking place during the eruption. Some of the key features of the eruption could be related to the influx of hydrous mafic magma. The influx of new magma will cause magma chamber pressurization by a variety of mechanisms (e.g. Sparks et al. 1977; Folch & Marti 1998). The escalation of the eruption during 1995-1998 could be attributed to an increase in chamber pressure as more mafic magma arrived from depth and to the accompanying remobilization of the andesite. The main consequences of the transfers of heat and volatiles from the mafic magma to the silicic magma are predominantly in the direction of pressure increase so the escalation could also indicate that the volume of remobilized magma increased with time. Heating of the andesite

and transfer of water from the basalt to the andesite can reduce the viscosity of the andesitic magma, thereby increasing discharge rate with time even at fixed pressure. Another possible cause of the volume or time pulses discussed before (Table 2) is convective overturn in the chamber, periodically transferring volatile-rich magma to the top of the chamber and causing a new spurt of dome growth. Involvement of basalt can also help explain the anomalous SO2 fluxes observed at Soufriere Hills (Young et al. 19980; Edmonds et al. 2001). As is the case at many other volcanoes (Wallace 2001), the SO2 fluxes measured during the eruption are significantly higher than can be reconciled with the andesitic magma. Fluxes of SO2 throughout the eruption have been in the range 100 to over 2000 tonnes per day, with values of several hundred tonnes per day being typical. However, for andesitic magma with a rhyolitic melt phase constituting no more than one-third of the erupted mass, the sulphur concentration at saturation should be only about lOOppm. At the observed range of magma extrusion rates, the values of SO2 flux should be only tens rather than hundreds and thousands of tonnes per day. The most compelling explanation for the excess sulphur flux is that most volatiles come from the mafic magma, as has also been proposed for Mount Pinatubo (Matthews et al. 1992; Hattori 1993; Wallace & Gerlach 1994). This concept makes sense in relation to the role of remobilization of the andesite source. Remelting of hydrous andesite rocks, which in this case was largely under conditions where amphibole remained stable, implies that the magma should have been undersaturated in volatiles as melting progressed. Hydrous mafic magma cools and crystallizes during invasion and released volatiles can be stirred into the andesite. There is great scatter in the SO2 data (Young et al. 1998a; Edmonds et al. 2001), but an increase in gas flux can be broadly correlated with the increase in magma extrusion rate while the dome was growing. This correlation broke down after magma ascent and dome growth ceased. The SO2 flux remained substantial

66

R. S. J. SPARKS & S. R. YOUNG

in the period of no dome growth, sometimes reaching a few thousand tonnes per day (Norton et al. 2002). This continued degassing after dome growth ceased can be interpreted in terms of the influx of large amounts of hydrous basalt at the base of the magma chamber. After the eruption has ceased the body of basalt intruded into the chamber should be comparable to the volume of erupted andesite (>0.4km 3 ). Thus the basalt magma can be expected to continue to cool and crystallize and supply further gases during residual volcanic activities. The interpretations on the role of the mafic magma in some of the eruptive phenomena discussed above must remain speculative. Clearly there is a major challenge for volcanology: to monitor and understand these deeper processes, which are quite likely to be fundamental to the eruptive process.

Concluding remarks At the close of the twentieth century substantial progress has been achieved in understanding volcanoes. Significant advances have been made as a consequence of technical developments, enhanced funding of science in general and an increasingly quantitative approach. An important factor in progress has been the detailed study of particular eruptions by large multidisciplinary teams using sophisticated techniques. Such concentrations of effort and expertise on single eruptions have been the hallmark of volcanology in the last decades of the twentieth century with investigations of Kilauea, Etna, Piton de la Fournaise on Reunion, and Krafla being examples for basaltic volcanoes and Mount St Helens, Mount Pinatubo, Mount Redoubt and Mount Unzen being examples for more silicic volcanoes. Each investigation has contributed to identifying new phenomena, compiling substantial databases of geophysical, geological and geochemical information, and stimulating development of theoretical work and new ideas. Such case studies have added to the fundamental understanding that ultimately underpins the practical objective of mitigating the effects of volcanic eruptions. The investigations at Soufriere Hills Volcano have benefited from these previous studies. This paper has outlined some of the particular contributions of the Soufriere Hills eruption to the reservoir of knowledge and scientific understanding. Despite the advances in volcanology, eruptions like that at Soufriere Hills Volcano highlight that there is much to do. In assessing the scientific work on Montserrat, Sir Robert May (Chief Scientific Adviser of the UK Government in 1995-2000) was generally complimentary about the work of the MVO, but remarked that the understanding of volcanoes was still largely based on natural history rather than on quantitative science. While we do not completely accept Sir Robert's analysis, his remarks highlight a number of issues that will occupy volcanologists in the next century. Despite the importance of natural history, it is remarkable how few really well-documented examples of eruptions actually exist to develop robust empirical or natural laws of how volcanoes behave. During the Soufriere Hills eruption, the MVO team found that they had remarkably little past experience and information to draw on. It soon became apparent that the only compelling analogue for the eruption, for which there was a significant amount of high-quality documentation and modern data, was the dacitic dome eruption of Mount Unzen (1991-1995). We also found that much of the information that we needed to know about dome eruptions was not in a readily accessible form, with only the Smithsonian database and the pioneering paper of Newhall & Melson (1983) providing some systematic information on dome-forming eruptions. There were several stratigraphic studies of dome complexes and their products, including Wadge & Isaacs (1988) for Montserrat. The information in such studies is necessarily partial and incomplete due to the vagaries of geological preservation. This information needed to construct a natural history of dome-forming eruptions is actually very limited. Our science might be compared with meteorology, where scientists are faced with phenomena of equal complexity, such as hurricanes.

Meteorologists, however, have a good understanding of the natural laws of hurricanes, because there are several hurricanes per year to study. At the moment there are very few cases of well-documented dome eruptions. Sir Robert was perhaps a little unfair in his implication that volcanology has not advanced into a quantitative science. Certainly there have been substantial developments in volcano geophysics and modelling of volcanic processes, which have been aided by the huge increases in computer power. Mathematicians, engineers and physicists have involved themselves in the science and geologists have tended to embrace these more quantitative approaches. Nevertheless Sir Robert's perspective has some truth to it. To a large extent we do not yet understand the data that come from monitoring active volcanoes very well; at least not well enough to be able to make confident forecasts. A few examples are worth giving. The SO? flux has varied greatly during the Soufriere Hills eruption (Young et al. 1998a), but was a decrease in flux good or bad news? A flux decline might mean gas was being stored and pressure was building prior to explosive activity (bad news) or it might mean that magma ascent was declining (good news). In the period of residual volcanic activity from March 1998 to November 1999 there were sporadic volcanotectonic earthquake swarms and small-scale but still impressive explosions and ash-venting (Norton et al. 2002). Prior to the resumption of dome growth in November 1999 the scientists involved in Montserrat struggled with interpreting this residual volcanic activity. Was this activity the sign of new attempts of the magma to rise again or the sign of settling of a disturbed system? The volcano rather than the scientists answered the question. The advances made at Soufriere Hills suggest that the challenges are even more formidable than we had thought. For Soufriere Hills one major advance has been to develop a strong case that the main parameters that are monitored (seismicity, ground deformation and gas fluxes) are governed largely by shallow-level processes in the conduit. The signals seem largely to be telling us about the processes of magma ascent, degassing and rheological stiffening with consequent pressure fluctuations in the upper parts of the conduit (Voight et al. 1999; Denlinger & Hoblitt 1999; Wylie et al. 1999; Melnik & Sparks 1999, 2002a). While this may be a significant advance, the basic driving forces for the eruption are much deeper. It is disconcerting to be faced with the notion that monitoring data may contain little information on what is going on in the magma chamber and deeper crust. We have little to go on with regard to deeper processes. Petrology and geochemistry can provide important constraints and help unravel complex processes, and they have provided important inputs to the long-term assessments of the volcano during the Soufriere Hills crisis. However, such studies are too long and complex to make substantial contributions to real-time monitoring, forecasting, and on-the-spot judgements. Seismic tomography is both expensive and timeconsuming and its resolution insufficient to indicate any more than that there are anomalous volumes of melted rock beneath volcanoes, a fact that most petrologists would no doubt argue is selfevident! There are therefore formidable challenges in development of instruments, techniques and understanding that will allow us to see through the dominant shallow processes to decipher the magma chamber and source processes. Advances in modelling work are improving understanding, but also emphasizing just how far we have still to go in prediction and forecasting. Published models of volcanic flows are predominantly steady-state and deterministic. Only recently have models been developed that start to explore strong non-linear behaviour and unsteady flows. The implications of these new models are not entirely encouraging. Models that explore gas loss and transitions between extrusive and explosive activity show that systems can be extremely sensitive to very small changes in conditions, and can show multiple steady solutions for exactly the same conditions (Jaupart & Allegre 1991; Woods & Koyaguchi 1994; Woods 1995; Melnik & Sparks 2002a). Thus some forms of volcanism may prove at a certain level of detail inherently unpredictable. Switches can occur between one eruptive state and another with little warning.

ERUPTION OF SOUFRIERE HILLS VOLCANO: OVERVIEW and the issue of whether some changes in eruptive state can be forecast or predicted may become a major focus in research. Unsteadiness is a major feature of much volcanic activity and models will need to develop an understanding of unsteadiness. One example of such a study concerns the explosive activity on Montserrat (Melnik & Sparks 2002b). It seems probable that volcanic systems, as anticipated by Shaw (1988), can exhibit chaotic behaviour. Inspection of the mathematical description of many of the processes that control volcanic flows shows that they are often highly non-linear with mathematical structures that make chaotic behaviour likely. This topic of mathematical modelling applied to volcanology has yet to be developed in earnest. Thus advances in modelling concepts are casting some doubt on determinacy as a valid concept. Chaos, however, does not necessarily mean that accurate forecasting is an unattainable objective. Hurricanes are part of a chaotic system (the atmosphere), but the combination of theoretical advances and observations on many examples allow hurricanes to be tracked, courses anticipated in probabilistic terms, and energetics and dynamics understood. The word 'prediction' may become outmoded in the new century, with the concepts of forecasting and anticipation becoming more appropriate to the nature of volcanoes. Another issue concerns the funding of volcano science. To a considerable extent funding in volcanology revolves around crises. Rather large amounts of money become available once a crisis starts, particularly if a wealthy developed country is involved in some way. A conservative estimate of the cost of the scientific work so far on the Soufriere Hills Volcano is US$ 9 million, far more than the UK has spent on academic volcanological research over the past decade. Funding then declines if the activity declines and there is a struggle to support even the most basic monitoring capability on dormant but dangerous volcanoes. There are still remarkably few volcanoes with good baseline monitoring and detailed stratigraphic studies combined with precision geochronology and quantitative hazards investigations. With the exception of funding quanta acquired during a crisis, funding levels make volcanology a poor relation of other areas of the science. It is not quite clear why this should be. Volcanoes excite the public's interest just as much as black holes and exploration of the planets and arguably more than particle physics. Yet the Earth's volcanoes largely remain poorly monitored and few of the 40 to 50 eruptions per year are documented in much detail. We seem unable to make our presence felt as a big science. However, with an expanding world population and some megacities adjacent to dangerous volcanoes it is likely that this coming century will have eruptions that will seriously threaten large numbers of people. Montserrat is a microcosm of the problems that will arise when volcanoes like Vesuvius and Popocatepetl have major eruptions. The authors acknowledge the work of our many colleagues at the Montserrat Volcano Observatory, some of which is summarized in this overview. R.S.J.S. acknowledges a NERC Research Professorship. Thanks to G. Norton, R. Stebbles, P. Cole and M. Murphy for help with some of the figures. P. Kokelaar and R. Hoblitt provided helpful reviews. References ADILBIROV, M. & DINGWELL, D. B. 1996. Magma fragmentation by rapid decompression. Nature, 380, 146-148. AMBEH, W. B., LYNCH, L. L., CHEN, J. & ROBERTSON, R. E. A. 1998. Contemporary seismicity of Montserrat, West Indies (abstract). In: An, W., PAUL, A. & YOUNG ON, V. (eds) Transactions of the 3rd Geological Conference of the Geological Society of Trinidad and Tobago and the 14th Caribbean Geological Conference, Vol. 1. ANDERSON, T. 1908. Report on the eruptions of the Soufriere in St. Vincent in 1902, and on a visit to Montagne Pelee in Martinique. The changes in the district and the subsequent history of the volcanoes. Philosophical Transactions of the Royal Society, Series A 208(ii), 275-332. ANDERSON, T. & FLETT, J. S. 1903. Report on the eruptions of the Soufriere St Vincent, and a visit to Montagne Pelee in Martinique. Part 1. Philosophical Transactions of the Royal Society, Series A 200, 353-553.

67

ASPINALL, W. P., MILLER, A. D., LYNCH, L. L. ETAL. 1998. Soufriere Hills eruption, Montserrat 1995-1997: volcanic earthquake locations and fault plane solutions. Geophysical Research Letters, 25, 3397-3400. ASPINALL, W. P., LOUGHLIN, S. C., MICHAEL, F. V. ET AL. 2002. The Montserrat Volcano Observatory: its evolution, organization, role and activities. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 71-91. BARCLAY, J., CARROLL, M. R., RUTHERFORD, M. J., MURPHY, M. D., DEVINE, J. D., GARDNER, J. & SPARKS, R. S. J. 1998. Experimental phase equilibria: constraints on pre-eruptive storage conditions of the Soufriere Hills magma. Geophysical Research Letters, 25, 3437-3440. BAXTER, P. J., BONADONNA, C., DUPREE, R. ET AL. 1999. Cristobalite in volcanic ash of the Soufriere Hills Volcano, Montserrat: Hazards implications. Science, 283, 1142-1145. BONADONNA, C., MAYBERRY, G. C., CALDER, E. S. ET AL. 2002. Tephra fallout in the 1995-1998 eruption of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 483-516. CALDER, E. S., COLE, P. D., DADE, W. B. ET AL. 1999. Mobility of pyroclastic flows and surges at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 26, 537-540. CALDER, E. S., LUCKETT, R., SPARKS, R. S. J. & VOIGHT, B. 2002. Mechanisms of lava dome instability and generation of rockfalls and pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 713-190. CHOUET, B. A. 1996. Long period volcano seismicity: its source and use in eruption forecasting. Nature, 380, 309-316. CHRISTIANSEN, R. L. & PETERSON, R. L. 1981. The 1980 Mount St. Helens eruption: Chronology of the 1980 eruptive activity. In: LIPMAN, P. W. & MULLINEAUX, D. R. (eds) The 1980 Eruptions of Mount St. Helens, Washington. US Geological Survey, Professional Papers, 1250, 17-30. CLARKE, A. B., NERI, A., VOIGHT, B., MACEDONIO, G. & DRUITT, T. H. 2002. Computational modelling of the transient dynamics of August 1997 Vulcanian explosions at Soufriere Hills Volcano, Montserrat: influence of initial conduit conditions on near-vent pyroclastic dispersal. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 319-348. CLYNNE, M. A. 1999. A complex magma mixing origin for rocks erupted in 1915, Lassen Peak, California. Journal of Petrology, 40, 105-132. COLE, P. C., CALDER, E. S., DRUITT, T. H., HOBLITT, R., ROBERTSON, R. E. A., SPARKS, R. S. J. & YOUNG, S. R. 1998. Pyroclastic flows generated by the eruption of the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3425-3428. COLE, P. D., CALDER, E. S., SPARKS, R. S. J. ETAL. 2002. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 231-262. CORTES, G. P. & RAIGOSA, A. 1997. A synthesis of the recent activity of Galeras volcano, Colombia:seven years of continuous surveillance 1989-1995. Journal of Volcanology and Geothermal Research, 77, 101-114. COUCH, S., SPARKS, R. S. J. & CARROLL, M. R. 2001. Mineral disequilibria in lavas explained by convective self-mixing in open magma chambers. Nature, 411, 1037-1039. DADE, W. B. & HUPPERT, H. E. 1999. Long run-out rockfalls. Geology, 26, 803-806. DENLINGER, R. P. & HOBLITT, R. P. 1999. Cyclic behavior of silicic volcanoes. Geology, 27, 459-462. DEVINE, J. D., MURPHY, M. D., RUTHERFORD, M. R. ETAL. 1998. Petrologic evidence for pressure-temperature conditions and recent reheating of andesite magma at Soufriere Hills Volcano, Montserrat, West Indies. Geophysical Research Letters, 25, 3669-3672. DRAKE, T. G. 1990. Structural feature in granular flows. Journal of Geophysical Research, 95, 8681-8696. DRUITT, T. H. 1998. Pyroclastic density currents. In: GILBERT, J. S. & SPARKS, R. S. J. (eds) The Physics of Explosive Volcanic Eruptions. Geological Society, London, Special Publication, 145, 145-182.

68

R. S. J. SPARKS & S. R. YOUNG

DRUITT, T. H., CALDER, E. S., COLE, P. D. ET AL. 20020. Small-volume, highly mobile pyroclastic flows formed by rapid sedimentation from pyroclastic surges at Soufriere Hills Volcano, Montserrat: an important volcanic hazard. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 263-279. DRUITT, T. H., YOUNG, S. R., BAPTIE, B. ET AL. 20026. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306. EDMONDS, M., PYLE, D. M. & OPPENHEIMER, C. 2001. A model for degassing at the Soufriere Hills Volcano, Montserrat, West Indies, based on geochemical data. Earth and Planetary Science Letters, 186, 159-173. ElCHELBERGER, J. C., CARRIGAN, C. R., WESTRICH, H. R. & PRICE, R. H. 1986. Non-explosive silicic volcanism. Nature, 323, 598-602.

FINK, J. H. & GRIFFITHS, R. W. 1990. Radial spreading of viscous gravity currents with solidifying crust. Journal of Fluid Mechanics, 221, 485-510. FINK, J. H. & KIEFFER, S. W. 1993. Estimates of pyroclastic flow velocities from explosive decompression of lava domes. Nature, 363, 612-615. FOLCH, A. & MARTI, J. 1998. The generation of overpressure in felsic magma chambers by replenishment. Earth and Planetary Science Letters, 163, 301-314. GIL CRUZ, F. & CHOUET, B. A. 1997. Long period events, the most characteristic seismicity accompanying the emplacement and extrusion of a lava dome in Galeras volcano, Colombia in 1991. Journal of Volcanology and Geothermal Research, 77, 121-158. GRIFFITHS, R. W. 2000. The dynamics of lava flows. Annual Review of Fluid Mechanics, 32, 477-518. GRUNEWALD, U., SPARKS, R. S. J., KEARNS, S. & KOMOROWSKI, J. C. 2000. Friction marks on blocks from pyroclastic flows at the Soufriere Hills Volcano, Montserrat: implications for flow mechanisms. Geology, 28, 827-830. HARFORD, C. L. & SPARKS, R. S. J. 2001. Recent remobilisation of shallowlevel igneous rock on Montserrat revealed by hydrogen isotope composition of amphiboles. Earth and Planetary Science Letters, 185, 285-297. HARFORD, C., PRINGLE, M. S., SPARKS, R. S. J. & YOUNG, S. R. 2002. The volcanic evolution of Montserrat using 40 Ar/ 39 Ar geochronology. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 93-113. HATTORI, K. 1993. High-sulfur magma. A product of fluid discharge from underlying mafic magma: evidence from Mount Pinatubo, Phillipines. Geology, 21, 1083-1086. HAYASHI, J. N. & SELF, S. 1992. A comparison of pyroclastic flow and debris avalanche mobility. Journal of Geophysical Research, 97, 9063-9071. HEIKEN, G., & EICHELBERGER, J. C. 1980. Eruptions at Chaos Crags, Lassen Volcanic National Park, California. Journal of Volcanology and Geothermal Research, 7, 443-481. HILDRETH, W. 1981. Gradients in silicic magma chambers: implications for lithospheric magmatism. Journal of Geophysical Research, 86, 10153-10192. HOBLITT, R. P., MILLER, C. D. & VALLANCE, J. W. 1981. Origin and stratigraphy of the deposit produced by the May 18 directed blast. In: LIPMAN, P. W. & MULLINEAUX, D. R. (eds) The 1980 Eruptions of Mount St Helens, Washington. US Geological Survey, Professional Papers, 1250, 401-419. HORWELL, C. J., BRANA, L. P., SPARKS, R. S. J., MURPHY, M. D. & HARDS, V. 2001. A geochemical investigation of fragmentation and physical fractionation in pyroclastic flows from the Soufriere Hills Volcano. Montserrat. Journal of Volcanology and Geothermal Research, 109, 247-262. HUPPERT, H. E. & SPARKS, R. S. J. 1988. The generation of granite by intrusion of basalt into the continental crust. Journal of Petrology, 29, 599-624. JACKSON, P., SHEPHERD, J. S., ROBERTSON, R. E. A. & SKERRIT, G. 1998. Ground deformation studies at the Soufriere Hills Volcano, Montserrat, 1 Electronic distance meter studies. Geophysical Research Letters, 25, 3409-3413. JAUPART, C. 1998. Gas loss from magmas through conduit walls during eruption. In: GILBERT, J. S. & SPARKS, R. S. J. (eds) The Physics of Explosive Volcanic Eruptions. Geological Society, London, Special Publication, 145, 73-90.

JAUPART, C. & ALLEGRE, C. 1991. Gas content, eruption rate and instabilities of eruption in silicic volcanoes. Earth and Planetary Science Letters, 102, 413-429. KOKELAAR, B. P. 2002. Setting, chronology and consequences of the eruption of Soufriere Hills Volcano, Montserrat (1995-1999). In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs, 21, 1-43. LACROIX, A. 1904. La Montague Pelee and ses eruptions. Masson, Paris. LOUGHLIN, S. C., CALDER, E. S., CLARKE. A. B. ET AL. 2002. Pyroclastic flows and surges generated by the 25 June 1997 dome collapse, Soufriere Hills Volcano, Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21, 191-209! LUCKETT, R., BAPTIE, B. & NEUBERG. J. 2002. The relationship between degassing and rockfall signals at Soufriere Hills Volcano. Montserrat. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 595-602. MACGREGOR, A. G. 1938. The Royal Society expedition to MONTSERRAT. B.W.I. The volcanic history and petrology of Montserrat with observations on Mt Pele in Martinique. Philosophical Transactions of the Royal Society, B229, 1-90. MATTHEWS, S. J., JONES, A. P. & BRISTOW. C. S. 1992. A simple magmamixing model for sulfur behaviour in calcalkaline volcanic rocks: mineralogical evidence from Mount Pinatubo 1991 eruption. Journal of the Geological Society of London, 149, 863-866. MATTHEWS. S. J, GARDEWEG. M. C. & SPARKS. R. S. J. 1997. The 1984 to 1996 cyclic activity of Lascar Volcano. Northern Chile: cycles of dome growth, dome subsidence, degassing and explosive eruption. Bulletin of Volcanology, 59, 72-82. MATTHEWS, S. J., SPARKS. R. S. J. & GARDEWEG. M. C. 1999. The Piedras Grandes-Soncor eruptions. Lascar Volcano. Chile; Evolution of a zoned magma chamber in the Central Andean upper crust. Journal of Petrology, 40. 1891-1919. MATTIOLI, G. S., DIXON, T. H., FARINA. F.. HOWELL. E. S., JANSMA, P. E. & SMITH, A. L. 1998. GPS meausrements of surface deformation around Soufriere Hills Volcano. Montserrat from October 1995 to July 1996. Geophysical Research Letters, 25. 3417-3420. MATTIOLI, G. S., RODRIGUEZ, L. SMITH. A. L.. JANSMA. P. E. & HERD. R. A. 2000. Changes in magmatic flux recorded by campaign and continuous GPS geodesy of Soufriere Hills. Montserrat. BWI (abstract). Geosciences 2000. Geological Society of London. MELNIK, O. & SPARKS, R. S. J. 1999. Nonlinear dynamics of lava dome extrusion. Nature, 402, 37-41. MELNIK, O. & SPARKS, R. S. J. 2002#. Dynamics of magma ascent and lava extrusion at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London. Memoirs, 21, 153-171. MELNIK, O. & SPARKS, R. S. J. 20026. Modelling of conduit flow dynamics during explosive activity at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London, Memoirs, 21. 307-317. MILLER, A. D., STEWART, R. C., WHITE, R. A. ET AL. 1998. Seismicity associated with dome growth and collapse at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25. 3401-3405. MILLER, T. P., CHERTKOFF, D. G., EICHELBERGER. J. C. & COOMBS. M. L. 1999. Mount Dutton Volcano, Alsaka: Aleutian Arc analogue to Unzen Volcano, Japan. Journal of Volcanology and Geothermal Research, 89. 255-302. MOGI, K. 1958. Relations between the eruptions of various volcanoes and the deformations of the ground surfaces around them. Bulletin of Earthquake Research Institute, 36, 99-134. MURPHY, M. D., SPARKS, R. S. J., BARCLAY, J. ET AL. 1998. The role of magma mixing in triggering the current eruption of the Soufriere Hills Volcano. Montserrat. Geophysical Research Letters, 25. 3433-3436. MURPHY, M. D., SPARKS, R. S. J., BARCLAY. J.. CARROLL, M. R. & BREWER, T. S. 2000. Remobilisation of andesite magma by intrusion of mafic magma at the Soufriere Hills Volcano. Montserrat. West Indies. Journal of Petrology, 41. 21-42.

ERUPTION OF SOUFRIERE HILLS VOLCANO: OVERVIEW NAKADA, S. & MOTOMURA, Y. 1999. Petrology of the 1991-1995 eruption at Unzen: effusion pulsation and groundmass crystallization. Journal of Volcanology and Geothermal Research, 89, 173-196. NAKADA, S., SHIMIZU, H. & OHTA, K. 1999. Overview of the 1990-1995 eruption at Unzen Volcano. Journal of Volcanology and Geothermal Research, 89, 1-22. NEUBERG, J. 2000. Characteristics and causes of shallow seismicity in andesite volcanoes. Philosophical Transactions of the Royal Society Series A 358, 1533-1546. NEUBERG, J., BAPTIE, B., LUCKETT, R. & STEWART, R. C. 1998. Results from the broadband seismic network on Montserrat. Geophysical Research Letters, 25, 3661-3665. NEWHALL, C. G. & MELSON, W. G. 1983. Explosive activity associated with the growth of volcanic domes. Journal of Volcanology and Geothermal Research, 17, 111-131. NORTON, G. E., WATTS, R. B., VOIGT, B ETAL. 2002. Pyroclastic flow and explosive activity at Soufriere Hills Volcano, Montserrat, during a period of virtually no magma extrusion (March 1998 to November 1999). In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 467-481. PERRET, F. A. 1939. The volcano-seismic crisis at Montserrat 1933-37. Publications of the Carnegie Institute, Washington, 512, 1-76. POWELL, C. F. 1938. The Royal Society expedition to Montserrat, B.W.I. Final Report. Philosophical Transactions of the Royal Society, Series A 237, 1-34. REA, W. J. 1974. The volcanic geology of Montserrat. Journal of the Geological Society, 130, 341-366. RITCHIE, L. J., COLE, P. D, & SPARKS, R. S. J. 2002. Sedimentology of deposits from the pyroclastic density current of 26 December 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 435-456. ROBERTSON, R., COLE, P., SPARKS, R. S. J. ET AL. 1998. The explosive eruption of Soufriere Hills Volcano, Montserrat, West Indies, 17 September, 1996. Geophysical Research Letters, 25, 3429-3432. ROBERTSON, R. E. A., ASPINALL, W. P., HERD, R. A., NORTON, G. E., SPARKS, R. S. J. & YOUNG, S. R. 2000. The 1995-98 eruption of the Soufriere Hills volcano, Montserrat. Philosophical Transactions of the Royal Society, 358, 1619-1637. ROOBOL, M. J. & SMITH, A. L. 1998. Pyroclastic stratigraphy of the Soufriere Hills Volcano, Montserrat: implications for the present eruption. Geophysical Research Letters, 25, 3393-3396. ROSE, W. I. 1973. Pattern and mechanism of volcanic activity in Santiaguito volcanic dome. Bulletin Volcanogique, 37, 73-94. SATO, H., FUJII, T. & NAKADA, S. 1992. Crumbling dacite dome lava and generation of pyroclastic flows at Unzen Volcano. Nature, 360, 664-666. SHAW, H. R. 1988. Mathematical attractor theory and plutonic-volcanic episodicity. In: Cni-Yu KING & SCARPA, R. (eds) Modelling of Volcanic Processes. Vieweg, Berlin, 161-206. SHEPHERD, J. B., TOMBLIN, J. F. & Woo, D. 1971. Volcano-seismic crisis in Montserrat, West Indies. Bulletin Volcanologique, 35, 143-163. SHEPHERD, J. B., ASPINALL, W. P., ROWLEY, K. C., PEREIRA, J., SIGURDSSON, H., FISKE, R. S. & TOMBLIN, J. F. 1979. The eruption of the Soufriere Volcano, St. Vincent, April-June 1979. Nature, 282, 24-28. SHEPHERD, J. B., HERD, R. A., JACKSON, P. & WATTS, R. B. 1998. Ground deformation studies at the Soufriere Hills Volcano, Montserrat, 2, Rapid static GPS measurements, June 1996 to June 1997. Geophysical Research Letters, 25, 3413-3416. SPARKS, R. S. J. 1976. Grain size variations in ignimbrites and implications for the transport of pyroclastic flows. Sedimentology, 23, 147-188. SPARKS, R. S. J. 1997. Causes and consequences of pressurization in lava dome eruptions. Earth and Planetary Science Letters, 150, 177-189. SPARKS, R. S. J., SIGURDSSON, H. & WILSON, L. 1977. Magma mixing: mechanism of triggering explosive acid eruptions. Nature, 267, 315-318. SPARKS, R. S. J., YOUNG, S. R., BARCLAY, J. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills volcano, Montserrat: November 1995 to December 1997. Geophysical Research Letters, 25, 3421-3424.

69

SPARKS, R. S. J., MURPHY, M. D., LEJEUNE, A. M., WATTS, R. B., BARCLAY, J. & YOUNG, S. R. 2000. Control on the emplacement of the andesite lava dome of the Soufriere Hills volcano, Montserrat, by degassing-induced crystallization. Terra Nova, 12, 14-20. SPARKS, R. S. J., BARCLAY, J., CALDER, E. S. ETAL. (2002) Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 409-434. STIX, J., TORRES, C. R., NARVAEZ, M. L., CORTES, J. G. P., RAIGOSA, J. A., GOMEZ, M. D. & CASTONGUEY, R. 1997. A model of vulcanian eruptions at Galeras volcano, Columbia. Journal of Volcanology and Geothermal Research, 77, 285-304. TAYLOR, G. A. M. 1958. The 1951 eruption of Mount Lamington, Papua. Bureau of Mineral Resources of Australia: Geology & Geophysical Bulletin, 38, 1-117. VOIGHT, B., HOBLITT, R. P., CLARKE, A. B., LOCKHART, A. B., MlLLER, A. D.,

LYNCH, L. L. & MCMAHON, J. 1998. Remarkable cyclic ground deformation monitored in real time on Montserrat and its use in eruption forecasting. Geophysical Research Letters, 25, 3405-3408. VOIGHT, B., SPARKS, R. S. J., MILLER, A. D. ET AL. 1999. Magma flow instability and cyclic activity at Soufriere Hills Voclano, Montserrat, British West Indies. Science, 283, 1138-1142. WADGE, G. & ISAACS, M. C. 1988. Mapping the volcanic hazards from Soufriere Hills Volcano, Montserrat, West Indies using an image processor. Journal of the Geological Society, 145, 541-551. WALLACE, P. J. 2001. Volcanic SO2 emissions and the abundance and distribution of exsolved gas in magma bodies. Journal of Volcanology and Geothermal Research, 108, 85-106. WALLACE, P. J. & GERLACH, T. M. 1994. Magmatic vapour source for sulfur dioxide released during volcanic eruptions: evidence from Mount Pinatubo. Science, 265, 497-499. WATTS, R. B., HERD, R. A., SPARKS, R. S. J. & YOUNG, S. R. 2002. Growth patterns and emplacement of the andesite lava dome at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 115-152. WHITE, R. A., MILLER, A. D., LYNCH, L. & POWER, J. 1998. Observations of seismic hybrid events at Soufriere Hills Volcano, Montserrat: July 1995 to September 1996. Geophysical Research Letters, 25, 3657-3660. WOLFE, E. W. & HOBLITT, R. P. 1996. Overview of the eruptions. In: NEWHALL, C. G. & PUNONGBAYAN, R. S. (eds) Fire and Mud: Eruptions and lahars of Mount Pinatubo, Philippines. University of Washington Press, 3-20. WOODS, A. W. 1995. The dynamics of explosive volcanic eruptions. Reviews of Geophysics, 33, 495-530. WOODS, A. W. & KOYAGUCHI, T. 1994. Transitions between explosive and effusive volcanic eruptions. Nature, 370, 641-644. WOODS, A. W., SPARKS, R. S. J., RITCHIE, L. J., BATEY, J., GLADSTONE, C. & BURSIK, M. I. 2002. The explosive decompression of a pressurized volcanic dome: the 26 December 1997 collapse and explosion of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoir, 21, 457-466. WYLIE, J. J., VOIGHT, B. & WHITEHEAD, J. A. 1999. Instability of magma flow from volatile-dependent viscosity. Science, 285, 1883-1885. YOUNG, S. R., SPARKS, R. S. J., ROBERTSON, R. E. A., LYNCH, L. L. & ASPINALL, W. P. 1997. Eruption of Soufriere Hills in Montserrat continues. Transactions of the American Geophysical Union, 78, 401-405. YOUNG, S. R., FRANCIS, P. W., BARCLAY, J. ETAL. 19980. Monitoring SO2 emission at the Soufriere Hills volcano: implications for changes in eruptive conditions. Geophysical Research Letters, 25, 3681-3684. YOUNG, S. R., SPARKS, R. S. J., ROBERTSON, R., LYNCH, L., MILLER, A. D., SHEPHERD, J. B. & ASPINALL, W. A. 19986. Overview of the Soufriere Hills Volcano and the eruption. Geophysical Research Letters, 25, 3389-3392.

The Montserrat Volcano Observatory: its evolution, organization, role and activities W. P. ASPINALL1 2, S. C. LOUGHLIN1'3, F. V. MICHAEL4, A. D. MILLER1'5, G. E. NORTON1 6, K. C. ROWLEY1'7, R. S. J. SPARKS 18 , S. R. YOUNG1 1 Montserrat Volcano Observatory, Mongo Hill, Montserrat, West Indies 2 Aspinall & Associates, 5 Woodside Close, Beaconsfield, Bucks, HP9 1JQ, UK (e-mail: [email protected]) 3 British Geological Survey, Murchison House, West Mains Road, Edinburgh, EH9 3LA, UK 4 Emergency Department, Government of Montserrat, St John's, Montserrat, West Indies 5 GEOWALKS, 24 Argyle Place, Edinburgh, EH9 1JJ, UK 6 British Geological Survey, Keyworth, Nottingham, NG12 5GG, UK 1 LAN DAT A Ltd, Trinidad & Tobago 8 Department of Earth Sciences, University of Bristol, Bristol, BS8 1RJ, UK

Abstract: The Montserrat Volcano Observatory (MVO) is a statutory body of the Government of Montserrat and is the organization responsible for volcano monitoring operations on the island. It was formed shortly after the first phreatic explosions from Soufriere Hills Volcano occurred on 18 July 1995, and evolved from a hastily created, interim entity to a fully established volcano monitoring operation. Participating scientific teams have been drawn mainly from the Seismic Research Unit of the University of the West Indies, the US Geological Survey, the British Geological Survey and universities from various countries including the USA, UK, France and Puerto Rico. Despite its hurried inception, the MVO has been able to provide timely, high quality hazard advice to the civil authorities and has maintained an exceptional documentary record of all scientific aspects of the eruption. Its public education and information efforts have been extensive and there have been unusually high levels of interaction between scientists and the civil authorities, and between scientists and the public, both within Montserrat and outside in the wider world. The experience of setting up and running the MVO, under difficult and stressful conditions, has exemplified the advantages of teamwork and flexibility within monitoring operations and the benefits of openness and clarity in public interactions. Novel techniques have been applied to the appraisal of hazards and advances in scientific understanding have proved invaluable for risk assessment and management.

The first historic eruption of Soufriere Hills Volcano, on the island of Montserrat (Fig. 1), began on 18 July 1995 with phreatic explosions from within the central prehistoric collapse scar (English's Crater) of this dome-complex volcano. This activity lasted until November 1995, and was followed by an extended period of dome growth that ended abruptly in early March 1998 (Young et al. 1998&). There then followed a 20-month interval in which no new magma was extruded and there was only mild activity involving ashventing, explosions, generally low levels of seismicity, and several collapses of the dome (Norton et al. 2002). In mid-November 1999, however, a new lava dome began to grow in the big chasm that had been cut into the remnants of the 1995-1998 dome by collapses in 1998 and 1999. This new dome increased steadily in size until it collapsed almost entirely on 20 March 2000, in a sequence of flows to the east, down the Tar River valley. After this, dome growth recommenced with increased vigour and has continued into 2002. Magma production rates have largely determined eruptive style, with high rates being associated with major dome collapses and explosions. This eruption, protracted over six years, has caused tremendous disruption to the lives of all residents on the island and has necessitated the continued presence on the island of a full-time scientific team and a dedicated monitoring facility. The initial slow rate of dome growth and gradual escalation of the crisis were major factors in the development of the monitoring effort, bearing in particular on early prognostications on the course of the volcanic activity. The setting, major event chronology and most important consequences of the eruption of Soufriere Hills Volcano from 1995 to 1999 are described by Kokelaar (2002). For the past six years, the Montserrat Volcano Observatory (MVO) has been the organization responsible for continuous monitoring of the volcano. The formation of a smoothly functioning scientific team during a crisis, with staff drawn from various nationalities and organizations, is inherently difficult. For the establishment of a full-scale observatory operation, the stressful situation in Montserrat, with a complex and unsatisfactory overall management structure and funding deficiencies, combined to create some significant problems. Notwithstanding these difficulties, however, a great deal of valuable scientific work has been done by the MVO, particularly in the documentation

of the eruption. There has been rapid publication of accounts of activity and scientific results in various media (e.g. Ahmad 1996; Montserrat Volcano Observatory Team 1997; Young et al. 1997; Ambeh et al. 1998). A collection of papers focusing on the first two years of the eruption was published in a two-part special section of Geophysical Research Letters (Aspinall et al. 1998a; Young et al. 1998c), and overviews of the scientific advances appear in Robertson et al. (2000) and in Sparks & Young (2002). There have been extensive efforts in public education and the MVO has become a primary source of information for the local administrative authorities and the public, as well as a major attraction for visitors to the island. More importantly, the continued operation of a volcano-monitoring facility on Montserrat is essential to the present and future occupation of the island, and recent administrative changes to the status of the MVO have placed its managerial and financial structure on a stable basis. There are many valuable lessons to be learned from the experience of setting up and running the MVO; important insights and issues of a detailed scientific nature are treated in the companion papers in this volume and elsewhere (see references). This paper summarizes some of the unique features of the new institution and outlines essential elements in its evolution, structure and organization. However, myriad other factors and influences shaped events and outcomes in Montserrat during the eruption from 1995 onwards, and these cannot all be covered in one paper. A book by Pattullo (2000) and a comprehensive report by Clay et al. (1999) each document many other aspects of the broader picture.

Administrative setting The inception and evolution of the MVO took place against a background of complex administrative arrangements, which characterized Montserrat as a British Dependent Territory (now a United Kingdom Overseas Territory). The administration of the island involves both a Governor, who represents the Queen and the UK government, and a locally elected Government, led by a Chief

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 71-91. 0435-4052/02/S15 © The Geological Society of London 2002.

71

72

W. P. ASPINALL ETAL.

Fig. 1. Map of Montserrat showing the location of the volcano, main place names used in the text and the migration of the MVO to the north of the island during the period 1995-1999.

THE MONTSERRAT VOLCANO OBSERVATORY

Minister. There is limited self-government, with a legislature and civil service, but the UK has responsibility for defence, internal security and foreign relations. For self-governance, there is an Executive Council consisting of the Governor, the Chief Minister, the Attorney-General, the Financial Secretary and three more ministers, and an eleven-member Legislative Council, of which seven are elected by universal adult suffrage (from age 18), two are nominated and two are appointed ex-officio. The UK government department with administrative responsibility for Montserrat is the Foreign and Commonwealth Office (FCO). However, this department has virtually no financial resources with which to respond to a natural disaster and the actual financial and logistic responsibility for Montserrat in an emergency devolved to the Overseas Development Administration (ODA, now called the Department for International Development - DFID). Another complication in relation to the Foreign Office is that Montserrat affairs have been dealt with internally both by the West Indies and Atlantic Department in London and by the Dependent Territories Regional Secretariat based in Barbados. The intricate relationships between all these ministries, departments and organizations, and their respective roles in the crisis, were not always clear and, as a result, were confusing to many of the scientists responding to the emergency. When activity escalated dramatically on Montserrat in July 1995, an operational base was immediately established on the island by the Seismic Research Unit (SRU) of the University of the West Indies (UWI) and, from this foundation, the Montserrat Volcano Observatory evolved into a fully functional local organization staffed by a multinational, multi-institutional team, utilizing a variety of geophysical, geological and geochemical techniques to monitor the volcano. Funding for these activities was provided mainly by the Government of Montserrat and by the UK government. The initial absence of suitably trained local staff and the longevity of the eruption necessitated a major turnover of staff at the MVO: over 100 scientists and technical staff have worked there during the past six years. At the height of the crisis, a staff complement of up to eight UK scientists (generally appointed through the British Geological Survey), three SRU scientists and technical staff could be mustered, eventually supplemented by seven Montserratian technical and administrative staff. In addition, staff and scientists from the US Geological Survey (USGS) and from universities in the UK, the USA, France and Puerto Rico have been involved in MVO operations from time to time. Generous support has been received also from the Institut de Physique du Globe de Paris, through staff from its Volcano Observatories in Guadeloupe and Martinique. From the resident Montserrat population itself, many volunteers have provided invaluable assistance with observations, routine support work and night-duty shifts at the Observatory. Evolution of the Montserrat Volcano Observatory The genesis of a permanent volcano observatory on Montserrat can be traced back to the periods of increased seismic activity that began in Montserrat in the early 1990s, first recognized by the SRU from their regional monitoring activities. This activity prompted SRU to strengthen the seismic network on the island. Within the eastern Caribbean, SRU is the regional agency for earthquake and volcano surveillance and this group monitored activity on Montserrat for several decades with permanently installed seismographs and regular field visits. Indeed, the history of the Unit's active involvement in Montserrat goes back to, and beyond, the last significant volcanoseismic crisis there, in 1966 (Shepherd et al. 1971). Their experience of handling volcanic crises in the region includes the eruptions of the Soufriere of St Vincent, in 1971-1972 (Aspinall et al. 1972) and 1979 (Shepherd et al. 1979), as well as numerous other volcanic-seismic swarms and related studies. Between January 1992 and July 1995, the seismic stations on Montserrat and the surrounding islands recorded more than 15 episodic swarms of small earthquakes. These originated at depths

73

of 10-15 km and had epicentres defining a NE trend off the east coast of Montserrat (Ambeh et al. 1998). As the swarms recurred, the SRU upgraded and augmented its seismic network on Montserrat, made field measurements of dry tilt, and observed the volcano to watch for changes in fumarolic activity. They were assisted by scientists from the Guadeloupe Volcano Observatory, who undertook annual sampling visits to the main fumaroles and thermal spring areas for physicochemical monitoring (Hammouya et al. 1998). These latter efforts, however, did not reveal any significant changes in fumarole activity or chemistry indicative of an impending eruption.

Establishing monitoring operations on Montserrat Immediately following the first phreatic explosions on 18 July 1995, the SRU established an operational base on Montserrat to provide scientific advice on the state of the volcano directly to local authorities. Acting on the advice of SRU, the Government of Montserrat then invited scientists from the USGS Volcano Disaster Assistance Program (VDAP) and the Guadeloupe Volcano Observatory to join the SRU team on-island to assist with their monitoring activities. The SRU and USGS scientists, together with one UK scientist and two student volunteers recruited in early 1995 from the local secondary school, formed the embryonic observatory team. A temporary facility was established near the government headquarters in Plymouth in July 1995 and was first called the Soufriere Hills Volcano Observatory. The physical location of the Observatory moved (Fig. 1) from Plymouth to the Vue Pointe Hotel, on the northern shore of Old Road Bay in August 1995, and then to a rented villa in Old Towne in October 1995 (Fig. 2), at which time the name changed to the Montserrat Volcano Observatory. The USGS VDAP team spent six weeks on Montserrat, from late July 1995, and reinforced the pre-existing seismic network with the establishment of additional short-period seismograph stations. In addition, three electronic tiltmeters were installed and a programme of SO2 monitoring was started, using a correlation spectrometer (COSPEC). An automated seismic data acquisition and processing system was also installed, which eventually replaced the pre-existing Soufriere system software that had been the basis of the data acquisition for the SRU network. At these early stages of the crisis (i.e. from August to November 1995), the Foreign and Commonwealth Office commissioned short visits to Montserrat by individual UK consultants with experience of volcanic activity in the Caribbean. These scientists functioned essentially as advisors to the Governor and to the UK government, but also assisted the MVO with routine monitoring work. In October 1995, the first of a series of short-term contracts from the ODA supported the direct involvement of staff from the British Geological Survey (BGS), to work with the incumbent monitoring team. Later fixed-term contracts with the BGS supported staff from that organization, UK university scientists and students, and other specialists from time to time (the number of UK personnel at the MVO increased from two in January 1996 to seven in July 1996). The local complement of staff at the observatory also increased in October 1995 with the secondment of several Montserratian civil servants from other government departments. The day-to-day running of the MVO was managed by a Chief Scientist, who was responsible for co-ordinating the scientific work and for reporting to the Government of Montserrat and to the Governor. During the first year of its operation, the Head of SRU fulfilled this role, with various other members of SRU acting on his behalf when he was off-island. By this time, the need was being recognized for a sizeable scientific, technical and administrative staff to monitor the volcano 24 hours a day. In an extended undertaking, personnel were drawn from various institutions: most came from the SRU and from the UK. UK scientists were assembled from the British Geological Survey and from the universities with strong research expertise in volcanology (principally, the universities of Bristol, Lancaster, the Open University and Cambridge). The USGS, the Institut de Physique du Globe de Paris, the

74

W. P. ASPINALL ETAL.

Fig. 2. Team members observing a small Vulcanian eruption cloud on 3 July 1997, from the pool-deck of Eiffel House, Old Towne, the location of MVO (South) from October 1995 to September 1997 (photo t W.P. Aspinall).

Universite Blaise Pascal, Penn State University, Brown University and the University of Puerto Rico provided significant assistance. However, it was initially far from clear who was responsible for what amongst all these different institutions and individuals. At times, relationships became very strained over priorities for scientific work and responsibilities for giving advice.

Early challenges and difficulties In part at least, the initial tensions over scientific obligations and priorities can be traced to the very difficult question of what level of response was appropriate at the time, given the uncertainty surrounding the future course of the crisis. Only two decades earlier, in 1976, a volcanic crisis on the neighbouring island of Guadeloupe, with many volcanological similarities to conditions in Montserrat in 1995, had failed to develop into a significant, lifethreatening eruption after a major evacuation of the population had been ordered. The circumstances surrounding the decision to instigate the evacuation became the focus of intense debate among volcanologists (see Tazieff 1980 (and preceding contributions in the same journal referenced therein); Fiske 1984), and the management of the emergency was the subject of a wide-ranging international commission of inquiry, sponsored by the French government. One major factor that was considered in analyses of the Guadeloupe episode was the enormous modern economic cost of an unfulfilled

warning of a dangerous eruption, notwithstanding the haunting political spectre of an earlier Caribbean volcanic catastrophe, the 1902 eruption of Mount Pelee in Martinique that had claimed nearly 30000 lives. Against this backdrop, it is understandable that in 1995 some in the eastern Caribbean, particularly in Montserrat itself, were reluctant to initiate urgent precautionary measures until the extent of the threat could be better ascertained. They would have been very conscious of the potential for a devastating impact on the economy of their small island, which had only just been reestablished after the ravages of Hurricane Hugo, in 1989. On the other hand, the USGS VDAP scientists, some of whom had witnessed the catastrophic consequences of the eruptions of Mount St. Helens and Pinatubo, took a more pessimistic view of the outlook and advised accordingly. In these circumstances, the UK government had to be mindful of its ultimate responsibility for the welfare of the people of Montserrat. As the higher political directorate, and one steeped in recent public disasters and health scares, UK politicians and civil servants were probably more 'risk averse' than their counterparts in the Montserrat government. They were certainly in the position to assert their judgement and authority more forcefully. This dichotomy in political attitudes generated dissonances within the scientific group. Contention developed as differences emerged as to the perceived level of risk and, with it, other difficulties arose, relating to attitude, personality and management styles. These were exacerbated in the first weeks of the crisis by the disparities in funding and resourcing that were initially available to. or deployed by, the various groups making up the team. It was not conducive to team spirit and collaboration that the regional scientists were obliged to make use of separate, cheaper accommodation, away from others, and had to endure restrictive transport arrangements and so on; this provoked a 'them-and-us' atmosphere and reduced opportunities for the informal scientific dialogue (and banter) that is so invaluable in crisis circumstances. A simple instance of a contentious issue was the accumulating cost of frequent and extended international phone calls from the nascent observatory facility. This was taken for granted as a necessity by many of the visiting scientists, but represented a major item of expense for the host government. It is unfortunate that, at this early stage, it was not possible to forge a unified approach to these issues within the whole MVO team. A large measure of concurrence existed, and the UK involvement by now included several former staff members of SRU, all familiar with the constraints faced by that organization. Thus, while the regional team from SRU were initially disinclined to widen the scope of monitoring (and scientific advice) much beyond that which could be sustained by or through the local government, some other scientists were less restrained about pressing forward with recommendations for strengthening the scientific efforts. This, in effect, implied seeking direct financial support from the UK government. As things stood then, it is perhaps not surprising that differing perceptions of the balance to be struck between economic imperative and acceptable risk overlapped into another area where substantial difficulties always arise in a volcanic crisis: the articulation of scientific uncertainty in the forecasting of future activity. Where the public is exposed to a given hazard, natural or otherwise, the lack of sufficient knowledge to quantify the attendant risks with useful accuracy impinges upon a government's ability to comprehend, and then justify to itself and the public, the extent to which steps should be taken to eliminate or reduce those risks. In the first few weeks of the Montserrat crisis there was perhaps, at times, some unwarranted scientific dogmatism about what might or might not happen at the volcano, especially in terms of the eruption turning magmatic and explosive. The confounding effect of these diverging, categorical scientific stances was then compounded for a short while by an overall diminution in communication between scientists and the civil authorities. The result was a dip in confidence in the MVO team and. with it, some loss of public credibility; this was not fully restored until later, when a consensual approach was achieved.

THE MONTSERRAT VOLCANO OBSERVATORY

Widening the scientific involvement Thus a situation developed, as the crisis started to drag on and deepen slowly, in which there was increasing pressure from the UK government in particular for the wider involvement in their deliberations of more scientists and specialists. This reflected both the natural instinct of politicians to canvass as many opinions as possible, and the anticipated strictures of impending new UK government internal policy guidelines. The latter enjoined politicians and civil servants to organize and engage wide-ranging scientific consultation in any science-related decision-making process, the relevant precepts emerging from high-profile health risk crises, such as BSE (as formulated in documents such as Office of Science and Technology 1997, 1998, for example; see also The Royal Society 1999; Curnow 1999). This pressure, to extend the spectrum of scientific feedback in respect of Montserrat, and the need to develop a team capable of functioning 24 hours per day, resulted in an increasing presence of UK personnel at the MVO and this, in turn, put additional strain on SRU's efforts to manage all the scientific elements of the crisis. In short, SRU was under-resourced for sustaining what was turning out to be a protracted and growing full-scale crisis management role on one island. As an institution it still had many other regional commitments to fulfil, but initially showed some reluctance to seek additional technical and scientific assistance once the VDAP team had left. Moreover, during the first two years of the eruption, ODA/ DFID, as the main funding agency, had virtually no direct contact with the SRU concerning either monitoring in Montserrat or the needs of the MVO. There was only very limited involvement of SRU/UWI in this process at any level, although they were the regional organization which, up to that time, had held full responsibility for volcano monitoring in Montserrat. The UK government (ODA/DFID) placed all contracts for scientific monitoring work in Montserrat through BGS as their principal source of geological advice. This one-sided arrangement served to sour inter-institutional relationships, at both administrative and scientific levels, and matters were not helped by changes in SRU's status within the University of the West Indies. In 1996, the SRU, having been a selfadministered regional research unit since its inception, was subsumed into the bigger physics department, where teaching and theoretical science are the mainstays. Thus, at an administrative level within the university, there was a shift to a directorate with other concerns, and with little experience of the demands and ramifications of mounting a major crisis response. Apparently, there was limited appreciation of the opportunity that such a prominent eruption afforded for reinforcing the SRU group's work by international exposure and external funding. The changes impinged on the deployment in the field of individual SRU scientists and technical staff, and, together with some residual internal difficulties that were legacies from the start of the eruption, the group's former responsiveness, resilience and flexibility were significantly curtailed. With all these internal and external factors taken together, it is not surprising that SRU, a small team labouring under severe constraints in difficult conditions, started to lose some momentum in its lead role at the MVO. Fundamentally, however, the difficulties faced by all the MVO scientific personnel, jointly and as members of separate institutions, stemmed mainly from the political duality that existed in a territory with the simultaneous involvement of two sets of governments.

Changes in the MVO management Recognizing the unsatisfactory nature of the arrangements at the MVO, an 'audit' of the situation was commissioned by the Governor in January 1996. Its main purpose was to consider medium-term needs (i.e. the following three years) and the longer-term development of the MVO and associated scientific programmes. The Audit Report, published immediately following the audit on-island, concluded that the eruption could be long-lived and that there was a

75

need to establish and staff a well equipped and effective monitoring facility. The report recommended the installation of new seismological equipment, the training of local staff and the construction of a purpose-built permanent observatory building. It also recommended the establishment of a management board and greater coordination in handling the UK contribution to the monitoring effort. While the Audit Report was duly completed in January 1996 and its recommendations apparently accepted, most of these (save for the procurement and installation of a new state-of-the-art seismic network) proved to have an inordinately long gestation period, or were not implemented at all. During the course of 1996, there was a shift in the role and responsibilities of the main agencies involved in, and responsible for, the management of the MVO. At the start of the eruption, the MVO was administered locally through the Development Unit within the Chief Minister's Office. In August 1996, responsibility for the MVO along with all other aspects of emergency operations was transferred to the Governor's Office. An administrative manager was employed to assist with the co-ordination of non-scientific aspects of the MVO work and to manage local staff. In September 1996, a decision was taken that the MVO should be jointly managed by the BGS and SRU, with the position of Chief Scientist alternating between staff drawn from the SRU and the UK. Each Chief Scientist took on a tour of duty that lasted between four and six weeks, typically requiring a commitment of 14 or more hours per day, seven days a week. Added to this were frequent disturbed nights when the volcano was very active. The rotation arrangement was intended to reduce the stress associated with taking responsibility for all scientific endeavours at an erupting volcano and to ensure that the individuals concerned functioned at optimal efficiency in the circumstances, without suffering undue fatigue that might affect judgement. The changes of Chief Scientist caused some inconsistency in management, unavoidable shifts in emphasis and difficulties in developing long-term strategic policies for the MVO. However, these shortcomings were ameliorated by posting to the team selected scientists who undertook longer stints of duty as Deputy Chief Scientist. These individuals took on a significant proportion of the routine administrative work of the observatory and provided much-needed continuity in management, while still acting fully as members of the scientific team. As the crisis developed, the direct load of responsibility on the Chief Scientist was also relieved by posting nominated Senior Scientists on missions to the Observatory. These were all experienced volcanologists who could assist with the tasks of interpreting observations and data for day-to-day hazard assessment and risk mitigation. With the establishment of a Chief Scientists' Committee in early 1997, an attempt was made to develop policies that were more consistent. This committee met three times during 1997 and 1998 and, in between, there were frequent contacts with the Chief Scientist in post via the Internet, so that assessments and views coming from the MVO had the understanding and support of all the scientists who filled the Chief Scientist role at other times. Before this happened, however, the BGS had submitted in November 1996 a proposal to DFID for rationalizing the management of all scientific input to the MVO that was funded directly by the UK government. The project was to cover a two-year period beginning April 1997 and provided for the overseas input to the MVO to be managed by the BGS, through whom all UK funding to the MVO would be channelled. For a variety of administrative reasons, the project was not fully approved until mid-1997, and then there were further delays in the signing of a contract between UWI and the UK Natural Environment Research Council (NERC) for the involvement of the SRU. This was not finalized until early 1998. The very slow speed of completion led to further difficulties between the two principal institutions involved in monitoring the volcano. With the eventual inception of the two-year contract, the formal reporting line to the UK government then became one that went from BGS to DFID. This involved meetings with DFID staff and weekly reports from the Observatory to DFID on the state of the volcano. DFID also received scientific feedback from the Foreign Office through the informal input of scientific advice or indirectly

76

W. P. ASPINALL ET AL.

via the Governor's Office. These multiple reporting channels and complex inter-relationships continued to lead to occasional misunderstandings and confusion during the crisis. The on-island reporting mechanism - bi-weekly briefings by the Chief Scientist to local administrators (see below) - remained the most important direct interface to the civil authorities and was largely unaffected by the overarching administrative changes.

Funding arrangements By late 1997, funding for the MVO was thus derived jointly from the Government of Montserrat (local and recurrent expenses) and from the UK government through DFID (foreign scientific support and major expenditure on monitoring equipment). The Government of Montserrat supported the local staff, MVO's local budget and the direct participation of the SRU during the period July 1995 to April 1997. Various short-term contracts between DFID, the BGS and individual consultants supported the UK and other nonSRU overseas scientists until April 1997. The new two-year DFID contract, which began (back-dated) in April 1997, provided the means for continued involvement of all overseas scientists at the MVO, including those by now being seconded from the SRU. Initially, the authorities had been slow to appreciate the significance of the longevity of the eruption and the crisis, and to understand the implications of this for scientific monitoring and its funding at an active volcano. This had led to a number of administrative responses of a short-term nature and the early development of the MVO as if it were in a conventional civil service setting, which in turn allowed little flexibility in spending. The commitment by DFID to a two-year period of stable funding for the monitoring activities improved matters, but the scientific team was still unable to respond as rapidly to breakdown of equipment or a need to acquire new equipment as they would have liked. This was because the BGS, as manager of a publicly funded project, was also bound by restrictive procedures that allowed little room for local flexibility (Clay el al 1999).

MVO management board Formation of a Management Board for the MVO, which had been recommended by the January 1996 Audit Report, was finally completed in October 1996. However, the first meeting of the Board did not occur until 7 March 1997. At that meeting, the MVO was

declared an entity of the Government of Montserrat; a mission statement was approved and decisions taken on the management structure, staffing policy, funding and long-term development of the MVO. As was the case with the Audit Report, few, if any, of the Board's recommendations were fully implemented. The Board itself had no direct role in the management of the MVO and did not devise any clear mechanisms for ensuring that decisions were implemented. There was no clear line of authority to link the Board to the MVO operations. Furthermore, the changes in administrative and funding arrangements that were put in place during late 1996 to mid-1997 were paralleled by the shift, noted above, in the relative roles of the SRU and BGS within the MVO: BGS now co-ordinated and directed the involvement of all nonSRU overseas scientists (who were in the majority), and took over effective management of nearly every aspect of scientific work at the MVO.

Relocation, another review of MVO needs, and further management changes Outside the Observatory, the escalating volcanic activity in the middle and later parts of 1997 caused another relocation of the MVO base after nearly two years in Old Towne (Fig. 1). to its present location at Mongo Hill (Fig. 3). The need for a permanent building for MVO had by then been well established. However, recurring problems regarding management and financing, and the longevity of the eruption itself, pointed to the need for a wide-ranging governmental review to ensure the continuance of the MVO beyond the eruption. This led to a Joint Comprehensive Review (JCR) of the MVO, which was commissioned by DFID in February 1998. The main purpose of this was to define the objectives for, and outputs required from, the MVO and to formulate plans on how this could be achieved cost-effectively in the short, medium and longer term. The review was intended to formulate the basis for future funding and management of the MVO. initially for the next threeyear period. The final report was delivered in mid-1998 and contained recommendations on all aspects of the MVO's operations. The JCR recommended that the MVO should become a statutory body of the Government of Montserrat, managed by a twotiered board system. A Management Board would meet annually to determine policy while an Operational Board (with wider representation) would meet quarterly and be responsible for implementation of policy and accountability. A Director with an appropriate

Fig. 3. The Mongo Hill building that became MVO (North) in September 1997. The site is on the northern side of the Centre Hills, near St John's, looking north with no view of the volcano. Windows are shuttered in anticipation of an imminent hurricane (photo G. E. Norton r BGS NERC).

THE MONTSERRAT VOLCANO OBSERVATORY

scientific background was to be recruited as soon as possible and arrangements made for continued provision of specialist advice, whenever needed. It was recommended that all funding to the MVO should be arranged through a single channel. Provision of a forward observation post, which would give the MVO team a clear view of the volcano, was seen as a critical operational issue. On 15 July 1998, the second meeting of the Management Board of the MVO was convened. This meeting had become somewhat overdue and attempted to deal with several important issues. However, most of the discussion centred on the findings of the JCR, while several on-going matters were left outstanding. Despite this, a number of key decisions were taken at this meeting. The recommendation of the JCR to establish the MVO as a statutory body was endorsed; the composition of both boards was discussed and agreed upon. It was decided that an Interim Director, with responsibility for overall management of the MVO, would be appointed as soon as possible. In addition, separate subcommittees were established to make specific recommendations on the future rationalization of the two seismological networks operated by the MVO and on the recruitment of a permanent director, respectively. Unlike previous occasions, follow-up action was taken on all the main decisions of the Board. An Interim Director of the MVO was appointed (who held post from October 1998 to June 1999), members of the Operational Board were selected and a draft bill was formulated for the incorporation of the MVO as a statutory body of the Government of Montserrat. This bill subsequently passed into law in August 1999. The appointment of the Interim Director allowed administrative action to be taken forward in a number of areas. The Operational Board held several meetings and made recommendations for consideration and endorsement by the Main Board on the establishment of a permanent observatory building, monitoring strategy, funding and a number of policy issues. As a result, the MVO now has a full-time Director who currently supervises seven local technical and administrative personnel and one contracted seismologist. They are assisted as needed in the short-term by one or more professional geoscientists, called down through the BGS/DFID arrangements. Funding is provided by the UK government through budgetary aid to the Government of Montserrat and includes provision for the construction of a permanent observatory. The wider social, political and economic contexts of the Montserrat eruption crisis, within which the evolution, vicissitudes and frequent restructuring of the MVO and its management are but a small part, have been exhaustively evaluated by Clay et al. (1999), on behalf of DFID.

77

Fig. 4. Map showing the positions of key stations of the MVO analogue short-period network and the original installed locations of the five broadband and three single-component seismometers that comprise the MVO digital broadband network. (Some instruments of both networks were co-located at the same sites.) The kernel of the short-period network was in place before the eruption started in July 1995, and the combined networks operated together from October 1996. There have been subsequent modifications due to damage from volcanic activity. Also shown are the locations of some temporary short-period analogue stations that were deployed during the crisis.

Monitoring Dedicated monitoring of Soufriere Hills Volcano remains the main objective of the MVO. This has involved routine application of various established techniques for acquiring sufficient data to provide reasoned scientific advice to civilian authorities, but with a strategy allowing flexibility so that additional techniques can be introduced if different hazards emerge as specific threats to the population. Routine monitoring entails three main activities: visual observation of the volcano, operation of the seismological network and ground deformation measurements (Figs 4 and 5). Studies of dome morphology and volume, and deposit mapping are on-going. Gas monitoring is done through direct and remote sampling. Environmental monitoring, in the form of groundwater, rainwater and ash analyses and atmospheric geochemistry measurements, has been undertaken since 1995. Other techniques used include: petrological and geochemical analysis of the erupted products, gravity surveys and rock-strength measurements. Video and still photographs are taken whenever activity dictates and conditions permit. A summary of the main areas of activity at the MVO is given in Table 1; in respect of scientific results, an overview can be found in Sparks & Young (2002). Visual observations have been made from a helicopter and from various observation points around the volcano since eruptive activ-

ity began in July 1995. These observations have been vital to assessing the state of the volcano and have been supplemented with basic photography using small- and medium-format still cameras and video photography. The data collected have allowed the detailed and accurate documentation of the evolution of volcanic activity, and of changes in dome morphology and dome volume. Two seismic monitoring networks have been used by the MVO. A short-period network using analogue telemetry has been in place since July 1995, including some stations that had been installed by the SRU before the crisis. The network consists mainly of vertical 1 Hz instruments, although two three-component instruments have been deployed. Throughout the eruption at least five and usually eight or more stations have been continuously transmitting data by VHP radio telemetry or telephone lines to a PC-SEIS dataacquisition system (Lee 1989) installed at the MVO by the USGS VDAP team. Events are inspected routinely, classified by type (Miller et al. 1998) and locations computed using the HYPO71PC program (Lee & Valdes 1989). The seismic signals are monitored 24 hours a day at the Observatory. Signals from four of the stations are written to paper drum recorders to give a real-time view of the seismic activity. A network of five three-component Guralp CMG40T broadband seismometers was installed in October 1996 and operated in conjunction with the short-period network (Fig. 4).

Fig. 5. (a) Network of stations used for precise levelling (dry tilt) monuments and electronic tiltmeter installations, (b) Electronic distance meter (EDM) networks and measurement sites on and around Soufriere Hills Volcano. Occupation of dry tilt stations was discontinued in 1996. and eruption damage to electronic tiltmeters and EDM targets reduced these networks to a single EDM line by 1998.

THE MONTSERRAT VOLCANO OBSERVATORY Table 1. Summary of main areas ofMVO activity, October 1998 to April 2000 Seismology Processing of earthquake and seismic signals from SP and BB networks Analysis and archiving of seismic data Manning of the operations room Ground deformation GPS measurements (once every one to two months) Electronic distance measurements (depending on visibility) Electronic tilt (continuous, telemetered) Theodolite measurements and photographs for dome volume calculations (depending on visibility) Geology Deposits sampling (as activity permits) Field mapping (as activity permits) Genera! Visual observations (daily) Airfall ash and water sampling (as activity dictates) Gas monitoring with SO2 tubes (every two weeks) and COSPEC (approx. three times per week) Mudflow monitoring (as required) Temperature measurements at pyroclastic flow deposits (every two months) Fumarole and hot springs sampling (two or three times/year, as conditions permit) Video/photos of volcano (as conditions permit) Special projects Hazard mapping Risk assessments and updates Dedicated monitoring for authorized activities in Exclusion Zone Preparation of public education/outreach material Electronics Maintenance and repair of electronic and field equipment Maintenance and operation of computer network Video archive Outreach Briefing of government officials (twice weekly and as activity dictates) MVO reports (daily, weekly, monthly, special) Press interviews Weekly radio interviews Public lectures; guided tours of the MVO (as necessary; officially once to three times per month on weekdays) Posters at the MVO and elsewhere Videos, and television and radio programmes on the volcano and volcanism Web pages Collaboration with other scientists and groups

79

stations located on the western flank of the volcano (Fig. 5a, b). Concern over the stability of Gages wall, overlooking Plymouth, led to the introduction of a Leica TCI 100 Total Station. This was used with single or triple prism targets to monitor possible movement on Gages Wall and Castle Peak dome. Six target sites were initially established in late September 1995. The network was radically modified between November 1995 and January 1996 when further instrument and target sites were added (Jackson et al. 1998). Destruction of reflectors during the progressive encroachments of the volcanic activity onto the surrounding flanks undermined the effectiveness of this network, which, by mid-1998, was abandoned when the single remaining reflector on the northern flank was lost. A new network was established in and around Long Ground in February 1999 to monitor possible movement on the eastern side of the volcano and the line from the side of Chances Peak to Amersham was reoccupied in June 1999 after a two-year period of inaccessibility (Fig. 5b). The ground deformation programme was reinforced from April 1996 by measurements using the portable Leica System 300, dualfrequency, differential global positioning system (GPS; Fig. 6). This allowed resolution of changes in line lengths between points on the volcano into vertical and horizontal components. The equipment was used in the rapid static mode and data processed using the Leica SKI software system (Shepherd et al. 1998). In addition to the MVO GPS system, the University of Puerto Rico (UPR) has maintained a GPS network at Soufriere Hills Volcano since July 1995 (Mattioli et al. 1998). At present, this latter network consists of three permanent GPS stations, which continuously transmit data to the MVO and to the UPR via the Internet. The MVO established a permanent GPS station at Harris in April 1998. Closer collaboration with the UPR GPS programme from February 1998 allowed the use of data collected by the UPR network in processing data by MVO GPS stations. During early 1998 and again in March 1999, the MVO GPS network was upgraded with new receivers and another permanent station established at South Soufriere Hills. A further MVO permanent GPS station was installed at Spring Estate in June 1999. Details on the early crisis period results, obtained from the GPS programme at Soufriere Hills Volcano, can be found in Shepherd et al. (1998) and Mattioli et al. (1998). Electronic tiltmeters have been used intermittently since July 1995 when three stations were initially deployed at Spring Estate, Amersham and Long Ground (Fig. 5a). The stations used high-gain tilt sensors with bubble-type biaxial platform tiltmeters and digital telemetry designed and built by the USGS Cascades Volcano Observatory (Murray et al. 1996). Two instruments were moved to Chances Peak in December 1996 because of increased concern over

Management Work programme and staffing etc. Helicopter operations Acquisition, repair and maintenance of monitoring equipment Scientific and other professional visitors Housing/accommodation Collaboration with National Trust and Tourist Board on development of visitors centre Specific operational tasks can change from time to time, but it is anticipated that most of these MVO responsibilities will remain for the foreseeable future.

Three single-component Integra LA100/F seismometers were also included in the network. These data are transmitted as 24-bit digital signals by UHF radio telemetry to the Observatory where it is acquired by a SEISAN data-acquisition system. Further details of these networks along with the data collected during the first two years of operation can be found in Aspinall et al. (1998b), Miller et al. (1998) and Neuberg et al. (1998). Ground deformation monitoring began with the use of a Wild NA2 precision level and 3-metre invar levelling rods at three dry tilt

Fig. 6. MVO field team, Rick Herd and Tappy Syers, having uncovered a marker pin beneath thick new surge deposits, are setting up the Tar River GPS station for overnight measurement, January 1999. The ruins of the Tar River Estate House are in the background. The view is looking NE, out to sea from the volcano (photo © R. E. A. Robertson).

80

W. P. ASPINALL ET AL.

the possibility of sector collapse, and others were placed at Hermitage and Gages Mountain in 1998 (Fig. 5a). However, apart from the stations on Chances Peak, there have been no perceptible variations or trends in the data collected from these tiltmeters (Voight et al. 1998). The Chances Peak stations were important for discerning cyclic patterns of conduit and dome pressurization and these variations were used, together with seismicity changes, to provide accurate short-term forecasts of activity during June to August 1997 (Voight et al. 1998, 1999). Estimates of dome volume have been made since January 1996 using compass and Abney level surveys, supplemented by photographs and theodolite measurements (Watts et al. 2002). Repeated observations and photos from fixed positions allowed estimates of dome volume to be made from trigonometry and triangulation, and from simple geometric measurements between successive photographic images (Sparks et al. 1998). In August 1996, a GPS kinematic survey method using mobile equipment and laser rangefinding binoculars was developed and used to determine dome volume. During late 1998, a computer program for such GPS data, based on an approach similar to the photographic method, was developed and used for estimating dome volumes (Herd et al. 1998). Measurements of the volume, distribution and character of pyroclastic deposits have also been vital for quantifying and characterizing the evolution of the eruption. Regular calculations of deposit volumes using the laser range-finding binoculars and kinematic GPS method, together with isopach mapping of ash-fall thickness (Bonadonna et al. 2002), have provided information on the total volume of magma extruded (Sparks et al. 1998). Changes in topography of pathways for pyroclastic flows were monitored giving direct input to hazard assessments, such as the numerical modelling of runouts (Wadge et al. 1998). Detailed mapping, sampling and temperature measurements of the new deposits (Cole et aL 1998, 2002; Calder et al. 1999; Komorowski et al. 2000; Druitt et al. 2002; Loughlin et al. 2002b) provided critical data for assessing the likely runouts and areal extents of impact of different phenomena. Coupled with studies of the geology of the Soufriere Hills complex (Roobol & Smith 1998; Harford et al. 2002), this resulted in more accurate assessments of the likely range of future behaviour of the volcano. Petrological work (Devine et al. 1998(7,6; Barclay et al. 1998; Murphy et al. 1998, 2000) has given indications as to the state of the magma chamber and the ascent rate of the magma. Gas monitoring has involved various techniques throughout the course of the eruption. The environmental effects of sulphurous gases have been monitored by filter packs (Allen 1996), SO2 diffusion tubes and measurements of acidity levels in rainfall and surface water (Norton 1997). Remote sampling of gas concentrations has been undertaken using a Mini-COSPEC correlation spectrometer (Young et al. 1998a) and open-path Fourier transform infra-red spectroscopy (Oppenheimer et al. 1998; Edmonds et al. 2001); these measurements helped elucidate cyclic emission patterns and deeper processes. Direct sampling of dome gases was done on one occasion and sampling of the associated hydrothermal system (Hammouya et al. 1998; Boudon et al. 1998) was carried out routinely during the early stages of the eruption until sample sites became too dangerous or covered by volcanic deposits. All these diverse scientific monitoring activities have been directly geared to providing information for hazard assessments and crisis management.

Crisis management The crisis management of the Montserrat volcanic eruption has been characterized by extremely close and frequent interactions of the scientific team with the civilian authorities and the public. In addition, a custom Montserrat Alert System was drafted that was specific to the conditions on the island. This has been supplemented by volcanic hazard mapping, vulnerability assessment, risk maps, formal and informal reports, scientific briefings, public meetings, news-media interviews and articles.

During the first few months of the eruption, a framework was set up to facilitate interaction among scientists and also between the scientists and the administrative authorities. An existing volcanic hazard and risk assessment, prepared originally by Wadge & Isaacs (1987, 1988), was initially adopted and then adjusted, as the nature of the eruption became clearer. (Although this work had been in existence for several years, its findings were not utilised in good time for mitigation prior to the start of the eruption; see Kokelaar 2002). The MVO subsequently produced a series of hazard maps for the authorities and worked closely with civilian authorities to produce risk maps. These showed geographical zones indicating the level of danger locally and also whether occupation was advised or allowed. The maps were published in the local newspaper and widely circulated as hand-outs. The conclusions of specially convened scientific meetings of the MVO team provided the basis for an accompanying consensus opinion on the appropriate state of activity to apply. At government briefings, this advice was normally presented by the Chief Scientist on behalf of the team.

Government briefings The MVO has provided frequent, regular briefings to the local authorities on the state of the volcano ever since the onset of eruptive activity at Soufriere Hills Volcano in July 1995. As a result, key officials have become thoroughly familiar with both the extent of scientific understanding of the volcano and the implications of any scientific uncertainty. These briefings were, and remain, integral to decision making for official response and mitigation measures. From July to September 1995, the Chief Scientist, often accompanied by another senior team member, provided daily briefings to the crisis management teams. As the pattern of events became more certain and dome growth began, the briefings became more structured and were given at least weekly to the crisis management team. This consisted of the Governor and the Chief Minister, senior civil servants and aid-agency personnel. In addition, the Governor and Chief Minister attended at least one of two internal scientific meetings held weekly at the MVO to discuss the state of the volcano. This continued until the middle of 1997. when a system of bi-weekly briefings was initiated, which remains in effect to the present. The first of these is usually given on a Monday to members of the Volcanic Executive Group, the highest decision-making body involved in disaster management on the island, comprising the Governor, the Chief Minister, senior ministers of government, the Chief of Police, the Chief Scientist and the Head of the Emergency Table 2. The Montserrat Alert Scheme, December 1995 Volcanic activity

Alert stage

Background seismicity with no new surface manifestation of volcanic activity.

0 (White)

Low-level local seismic activity, ground deformation and mild phreatic activity.

1 (Yellow)

Dome building in progress, periodic collapses generating rockfalls and occasional pyroclastic flows. Moderate level of seismic activity with no sudden changes.

2

(Amber)

Change in style of activity anticipated within a few days. Pyroclastic flows common with associated light ash fall. High level of seismic activity.

3

(Orange)

Major dome collapse under way, with large pyroclastic flows and heavy ash fall. Explosive event possible if the activity continues.

4

(Red)

Ongoing large explosive eruption with heavy ash fall.

5

(Purple)

The Alert Scheme was used as a guide when providing advice to the authorities about activity at the volcano. Although no specific actions are listed for the administrators, contingency plans draw ? n up by the Government's Emergency Department were based on the identified alert levels.

THE MONTSERRAT VOLCANO

Department. The second briefing is given on a Friday to a larger group called the Volcanic Management Support Group, which consists of senior civil servants and key sector agencies. The briefings are usually given by the Chief Scientist (now the Director) or Deputy Chief Scientist and consist of a synopsis of activity for the week along with an up-to-date assessment of what may be expected from activity during the ensuing week. Advice is also given on various aspects of crisis management, from the scientific perspective. As noted earlier, one of the peculiarities of Montserrat is the existence of a complex civil administration, with two major ministries of the UK government, the Governor and the local government all involved in the response to the crisis. Inevitably these arrangements at times led to a lack of clarity about where, and in what form, essential scientific information should be provided. The drawbacks of such a situation were exemplified early in the eruption by the appointment of independent consultants to the FCO and DFID at the same time as there was an on-going reporting line in Montserrat, originating from the embryonic MVO. This caused the SRU scientists in Montserrat to feel that their role was being undermined and their advice countered by separate, independent opinion being expressed in the UK. While their concern was soon acknowledged and resolved within the scientific team, the issue of the garbling or misinterpretation of the content of urgent scientific advice by uninitiated intermediaries remained a source of problems, even when only one channel of reporting was meant to be active. Thus, the problem of how to communicate uniform and coherent scientific advice to several interested arms of two governments, in parallel, in two separate countries in different time zones, has

OBSERVATORY

81

been a significant challenge throughout the eruption. It is a complication that may arise in other countries where there are both federal and local strands of government, for instance. The introduction of six-monthly risk assessments from August 1997 onwards, formally prepared by a group of senior scientists, improved the situation. These reports are delivered to all interested parties simultaneously and the principle has been established that the Chief Scientist (more recently, the Director) should be the channel for all immediate scientific advice. The problems that were encountered earlier on in Montserrat serve to emphasize how important it is for scientists to establish clear lines of communication in a volcanic crisis, although this can prove far from straightforward if duality of political accountability exists.

Alert Scheme From the onset of the eruption, an Alert Scheme has served as a useful tool for guiding civil response actions to varying levels of volcanic activity. The scheme has undergone several iterations. The first system used was a generic one patterned on that from the Office of the United Nations Disaster Relief Co-ordinator (UNDRO) handbook on Volcanic Emergency Management (Table 2). It was not specifically tailored to the peculiarities of the environment on Montserrat and proved inadequate to guide all necessary mitigation efforts as the eruption slowly evolved. The explosive activity that occurred in mid-September 1996 was a significant departure from the extrusive style of eruption that had

Table 3. The Montserrat Alert Scheme, March 1997 Volcanic activity

Alert stage

Background seismicity with no new surface manifestation of volcanic activity.

0 (White)

All zones occupied. Review and update emergency plans on an ongoing basis.

Low-level local seismic activity, ground deformation and mild phreatic activity.

1 (Yellow)

Maintain readiness of key personnel, systems and procedures. Keep stock of critical supplies Local evacuations may be necessary in Zone A. Zones B and C on standby.

Dome building in progress, periodic collapses generating rockfalls and occasional pyroclastic flows. Moderate level of seismic activity with no sudden changes.

2 (Amber)

Zone A: No access Zone B: Access limited to short visits by residents, officials and approved visitors with rapid means of exit Zone C: Daytime-only visits by residents, approved commercial activities and agriculture Zone D: Day- and night-time occupation by residents; high level of alert maintained Zones E, F: Full occupation by residents with national contingency plan for evacuation in readiness Zone G: Full occupation

Change in style of activity anticipated within a few days. Pyroclastic flows common with associated light ashfall. High level of seismic activity.

Major dome collapse under way, with large pyroclastic flows and heavy ash fall. Explosive event possible if the activity continues.

Ongoing large explosive eruption with heavy ashfalls.

(Orange)

4 (Red)

(Purple)

Actions by administrators and general public*

Zones A, B: No access Zone C: Access limited to short visits by residents and workers with means of rapid exit Zone D: Daytime occupation for essential services and agriculture, residents allowed access in daytime. Essential services operate with standby transport and evacuation plans in place Zone E: Full occupation with high level of alert maintained. Schools operate with standby transport Zone F: Full occupation by residents with contingency plan for evacuation. Warn of ashfalls in Zones E and F Zone G: Full occupation Montserrat Standing Operating Procedures for Red Alert in place All schools closed as required. People with special needs removed from Zones E and F Zones A, B, C, D: No access; rapid evacuation of all remaining persons Zone E: Rapid evacuation; warn of potential for gravel, pumice and ashfall Zone F: Warn of potential for gravel, pumice and ashfall Zone G: Full occupation Zones A, B, C, D, E: No access, rapid evacuation of all remaining persons Zone F: Initiate evacuation; warn of potential for gravel, pumice and ashfall Zone G: Warn of potential for ash hazards

* Recommended actions to minimize significant casualties. Zones as defined on Volcanic Risk Management Map. This Alert Scheme was drafted with closer collaboration between the disaster managers and the MVO and, as such, incorporated specific actions to be taken by administrators and the general public. The Scheme had accompanying risk maps (e.g. Fig. 7) that illustrated the zones mentioned. These maps, and a detailed listing of the specific villages within each zone, were distributed to the public by the civil authorities.

W. P. ASPINALL ET AL.

occurred up to that date. Consequently, it was necessary to revise the Alert Scheme to incorporate new conditions for potential explosion hazards, with corresponding changes concerning vulnerability and response. This resulted in the first draft of an Alert Scheme specifically for Montserrat (Table 3), which provided a detailed guide for actions to be taken by civil authorities within defined risk zones around the volcano at specified alert levels determined by the MVO team (Fig. 7). This, in turn, was revised a number of times as activity escalated. During early December 1996, when the Galway's segment of English's Crater appeared to be under severe stress, it emerged that differences in interpretation of the emergency procedures existed between scientists at the MVO, the civil administrators and the public. These procedures came under closer scrutiny again in mid-December 1996 when, upon raising the alert level in response to increased activity at the volcano, only a small percentage of the residents in a particular community that was deemed vulnerable heeded the official notice to evacuate. The increased activity that triggered the raising of the level on this occasion had all but died down by the time the public announcement for evacuation was made. Once volcanic activity had encroached on most of the built-up areas and vital services around the volcano in the tragic events of 25 June 1997, the Alert Scheme could no longer be retained as a microzonation management tool. The approach was then simplified to a system whereby the entire island was divided into a Northern Zone, a Central Zone and an Unsafe or Exclusion Zone, with severe restrictions being placed on entry into the Exclusion

Moving from Zones G to A represents an increasing risk, based on an evaluation of the volcanic hazard. The status of each zone is dependent on the Alert Level. Fig. 7. Volcanic Risk Management Map from 6 June 1997, drafted by the MVO in consultation with Montserrat government officials and disaster managers. This and the maps in Figures 8 and 9 were based on hazard maps produced for the main types of volcanic events that were expected from Soufriere Hills Volcano and were used along with population distribution and infrastructure maps to delineate risk boundaries (map redrawn here for clarity).

Zone (Fig. 8). This system was considered applicable for as long as the dome-building eruption continued and, with minor administrative changes, has remained in effect ever since. Small modifications to the boundary line of the Exclusion Zone were made as activity at the volcano decreased during 1998 (Fig. 9). However, scientific consensus on what activity level was appropriate at any particular time was seldom easy to achieve and this stimulated the development by the MVO team of structured decision procedures for the purpose.

Expert elicit at ion As an aid to discharging the responsibility to provide good and timely advice to the civilian decision makers in the form of an agreed scientific position, a formalized procedure for eliciting expert judgement was adopted by the MVO in August 1995. The need became clear at this early stage in the crisis when the nature and magnitude of future volcanic activity was most uncertain and time for repeated and protracted scientific debates amongst a large group was just not available. Arriving, by committee, at an agreed position on what advice to give the authorities against a deadline each day, took an increasing amount of the scientists' time, just when they urgently needed to press ahead with observations and measurements. This became frustrating both for them and for the authorities, who preferred to have a rapid and definitive answer provided as punctually as possible. The formalized procedure that was introduced is based on the 'classical model' for structured expert judgement (Cooke 1991), following the suggestion first made at the International Symposium on Large Explosive Eruptions in Rome in 1993 to consider using this approach for producing a collective scientific opinion in a volcanic crisis (Aspinall & Woo 1994). The method performs weighted combinations of expert judgements, where weights are determined by 'calibration* and 'information' performance on questions for which the true values are known, or become known a posteriori. However, in application, the procedure had to be adapted to the needs of real-time crisis management. In open meetings, where some preferred to hold their counsel, others were more forceful in giving their opinions so. from the outset, it was considered better to focus the procedure on quantifying the range of conservatism ("informativeness') of each individual's views, rather than rely too heavily on a hurried and questionable calibration score. (The problems of calibrating the expertise of a group of volcanologists are non-trivial at the best of times, let alone with an eruption going on outside the window!). By emphasizing individual informativeness in this way, there was an implicit assumption that all members of the team had roughly equal expertise, a reasonable supposition for the initial scientific group assembled in Montserrat. In fact, when the concepts of this approach were being introduced to the administrators and scientists they were novel to most, and several pressed for the scheme to be administered in such a way that no single participant was ever zero-weighted: all views would be used with some weight in the decision process. However, in order to bring some element of calibration into play, a preliminary set of five suitable seed questions was hastily prepared. While this set was aimed primarily at measuring the individual's informativeness factor, it was also used to provide a basic measure of how well each person might make quick judgements on issues related to safety and hazard mitigation in an emergency. The latter element of the exercise was undertaken, with general agreement, as an exploratory trial of the method in application in a live crisis. As the emergency progressed, three limitations to the elicitation procedure emerged. First, having once used the initial set of 'seed questions' for calibration, for uniformity these had to be repeated for scientists subsequently joining or replacing others in the team. Over the course of three years, a total of more than 60 individual scientists and technical specialists participated in the different elicitations (although there were generally only between five and 20 present at any one time). To everyone's credit, the answers to the

83

THE MONTSERRAT VOLCANO OBSERVATORY

N

Fig. 8. Volcanic Risk Management Map for Montserrat as it stood in September 1997. Microzonation was discontinued after 4 July 1997, when volcanic activity became more severe.

Exclusion Zone Central Zone

No admittance except for scientific monitoring and National Security Matters Residential area only, all resident on heightened state of alert. All resident to have rapid means of exit 24 hours per day Hard hat area all residents to have hard hats and dust masks.

Northern Zone

Area with significantly lower risk, suitable for residential and commercial occupation Limit of high direct volcanic hazard

seed questions were not leaked to new arrivals at the Observatory. Secondly, an experienced technical facilitator was not always available in Montserrat to supervise the elicitations and thirdly, there was a fluctuating mandate to undertake formal elicitations, as Chief Scientists rotated, or as conditions and levels of anxiety varied. There were divided opinions amongst the scientific group as regards the utility and soundness of the method, based as it is on concepts of subjective probability. These were new to many, and, with an experienced technical facilitator not always available at the MVO, the technique was not used continuously. However, the majority of scientists supported the approach, particularly for issues requiring frequent but tricky, marginal decisions, for which it proved a very successful aid. After being discontinued in late 1995, the methodology was reintroduced following the 17 September 1996 explosion. It was then used more widely in the decision-making process, being integrated into the Alert Scheme and used on a weekly basis to assist the Chief Scientist to determine the appropriate level of alert. The method continued in operation in this context until the Alert Scheme was simplified in July 1997. Its use for assessing the appropriate alert level gave continuity to the decision-making

process and provided a traceable record through time of the views of the scientific team (see, for example, Fig. 10). Further details on the theory and potential application of this technique to volcanic crises are given in Aspinall & Woo (1994) and an account of its introduction and use at the MVO is contained in Aspinall & Cooke (1998). The same structured elicitation procedure was also employed by the scientific team to progress the strategic hazard and risk assessment reviews that were commissioned later on in the crisis.

Risk and hazard assessments With the change in eruptive style that occurred in 1997, a more formalized approach to hazard and risk assessment was invoked. The first of what became regular reviews of the state of the volcano and its activity was undertaken in December 1997 and involved most of the senior scientists who had worked at the MVO. These assessments provided a considered and comprehensive review of the most recent activity and, to meet the requirements of the civil

84

W. P. ASPINALL ETAL.

proper consistency to the outcomes. The structured elicitation approach to this problem is one more example of how creative hazard management techniques have been developed within the MVO during the crisis. All the risk assessments have been updated at intervals of approximately six months (or as circumstances demanded). They have depicted most likely outlooks for future activity which consistently, as time proceeded, were not overturned by more extreme events or major surprises in the behaviour of the volcano. These assessments thus form a valid basis for addressing the question of 'acceptable risk'.

Acceptable risk

Bottom of Belham River valley and all areas south, across Centre Hills to Pelican Ghaut Limit of high direct volcanic hazard

Fig. 9. Volcanic Risk Management Map of Montserrat, revised in September 1998. The boundaries of this map remained unchanged until April 1999 when some minor administrative changes were implemented by the civil authorities in consultation with the MVO.

authorities, arrived at a prognosis for the immediate following period (six months) and the more distant future (five years or more). Various possible eruptive scenarios for this volcano, which these reviews developed on the basis of available evidence, were usually summarized in the form of event-probability trees. This provides a compact format for presentation of such information to decision makers. The formalized elicitation methodology, used with the Alert Scheme, has also been used extensively to provide consensus probability value distributions for input onto the branches of the event trees (see Fig. 11), in conjunction with traditional decisionconferencing and numerical modelling. In turn, these probability distributions serve as inputs to Monte Carlo models for simulating population casualty numbers, which form an integral part of the volcanic risk assessments that the MVO undertakes for guiding civil mitigation decisions. As the eruption progressed and its behavioural patterns changed, the event trees became increasingly more specific in terms of threats to particular areas and, as population zone occupancies changed, the casualty impact simulations also grew in complexity and detail. Figure 12 shows examples of successive F-n exposure curves for Montserrat (where the F-n curve is a risk assessment concept representing the exceedance probability of frequency of a cumulative number of casualties). In constructing and running these models from one assessment to the next, it was essential that the varied opinions from a diverse and changing group of experts could be incorporated in a coherent manner in order to maintain a rational balance to the input assumptions and a

In Montserrat, it was acknowledged by all parties that the administrative authorities held responsibility for specifying the level of acceptable risk throughout the volcanic crisis. From the onset of the eruption, their stated policy was to set the tolerability threshold at a low level, and to maintain it there. Whenever the risk to an area near the volcano was deemed to be approaching life-threatening levels, evacuation of such areas was undertaken. Having accepted accountability for determining the threshold of risk acceptability, the government also assumed the responsibility for providing relief assistance to the population that was evacuated from threatened areas. Bans were placed on re-entry until the danger was judged to have subsided and. in order to apply them, police checkpoints were established at key entry points. The purpose of these checkpoints was to limit access to authorized essential personnel only, thereby restricting the number of entries to a manageable level, in accordance with emergency and riskmanagement plans and capability. However, because of a number of social and political factors, these exclusions were not totally enforceable. Furthermore, for some people living and working near the boundary of an exclusion zone, it was inevitable that the small, residual risk involved there (about which they were advised) was not an effective restraint. By making unauthorized entry to the Exclusion Zone, many individuals in Montserrat have been able to set their own level of acceptable risk. The actual threshold that such individuals assigned to themselves varied throughout the crisis. At the start of the crisis, public knowledge of the volcano was not as informed as later, and uncertainty about the future was greatest. Many persons, having heard accounts of the 1902 Martinique disaster, accepted only a low risk and heeded evacuation recommentions. As the eruption progressed, and it was felt that their knowledge of the hazards and experience of this particular volcano increased, personal risk thresholds among some of the population appeared to change and many individuals seemed willing to accept higher levels than previously. The majority, however, were prepared to be guided by the advice and information provided by the MVO's outreach efforts.

Outreach Throughout the full duration of this persisting crisis, the MVO has taken many positive steps to help inform all the public of Montserrat on the hazards posed by their volcano (see Table 4 for a summary). They have undertaken a series of educational activities, guided by principles of openness about what is known and frankness about what is uncertain or unpredictable.

Policy A commitment to an open policy as regards information has been maintained by the MVO throughout the crisis and, as far as is practicable, a public open-door policy within the Observatory itself has been the norm (at whichever location it happened to be).

THE MONTSERRAT VOLCANO OBSERVATORY

85

Fig. 10. Time-line chart of repeated appraisals of the volcanic Alert Level using expert judgement elicitations in a weighted voting scheme, together with a record of the Alert Level in force. This representation of scientific opinion, in which the current level setting is assessed against thresholds to move up or down, provided decision makers with both the central (optimal) result of the poll and a measure of the spread of views, thus allowing them to select what level of conservatism to accept. More often than not, the distributions of scientific opinion were skewed, with longer tails to the more cautious end of the range.

This policy attitude evolved directly from public concerns about the openness of the MVO during the early stages of the crisis, when, crucially, there were differences of scientific opinion about whether the volcano would erupt. Then, it was perceived by some that the scientists and civil authorities were jointly withholding information and bad news from the public; later, others felt it was possible that some scientists were working to a hidden political agenda. (As a UK territory, there was no freedom of information act enshrined in law in Montserrat.) To minimize these concerns, the civil authorities endorsed the MVO open information policy and actively supported the scientists in their public outreach.

Outreach activities For five years, the MVO team has engaged in an extensive programme of public information and communication, in partnership with civil officials, welfare and health professionals, schools, churches and the media. Many public education exercises were co-ordinated by the Government Information Service and the Government of Montserrat Emergency Department in response to specific, identified requirements. All these efforts have resulted in a highly informed population. However, it would be unrealistic to pretend that even this intensive programme has resolved all related problems: the longlived character of the eruption and the protracted nature of the crisis on a small island have caused particular difficulties.

During July-September 1995, daily reports and ad hoc scientific briefings were given by the Chief Scientist on local radio and television. From early 1996 to mid-1998, morning and evening reports (which were circulated to key government agencies and other interested organizations by fax) were read out on the local radio station. During periods of heightened activity, additional lunch-time radio updates were, and still are, presented live by MVO staff. These enable topical issues to be explained and ensure that the latest volcanic situation is promptly understood by the public. From early in 1997, one or more of the MVO senior scientists (and usually the Chief Scientist) has been interviewed on the radio, usually once a week on a Friday, for half an hour or more. Special radio call-in programmes, allowing the public to interact directly with the scientists and local technical staff, have been aired frequently. These are effective in clarifying issues of concern to the individual caller and important topics of consequence for the wider public. From time to time, individual MVO staff have given public lectures on more specialized aspects of volcanology. These are always broadcast live on national radio and are invariably followed by intense, but enlightened, questioning from the audience. Articles about the volcano have been contributed regularly to the main local newspaper. In addition, the Montserrat National Trust and the MVO published jointly a monthly magazine called SeismiCity News, which provided further explanation of the science. This publication contained sections aimed specifically at children, because the latter were believed to be influential opinion-formers

W. P. ASPINALL ETAL.

ACTUAL COURSE OF ERUPTION

Fig. 11. A summary event probability tree for early stages of the Montserrat eruption, showing elicited probability values attached to branches depicting different potential eruptive scenarios. Such trees prompt the scientists to identify all possible scenarios and help communicate their forecasts to public officials, summarizing on various timescales the hierarchy of hazards against which mitigation measures can be planned. At the MVO. the structured expert elicitation methodology was used to update probabilities in a consistent manner; associated uncertainty distributions were, in turn, used as inputs to Monte Carlo risk simulations of population impact (see Fig. 12).

within families. The magazine was started in early 1997, following a period in December 1996 when escalating volcanic activity, rapid and frequent changes in alert level, and general public unease raised concerns about the MVO's credibility (see below). SeismiCity News was sponsored by DFID and circulated to the public free of cost. Other public information bulletins and information sheets, circulated freely to the public, have been prepared in collaboration with the Government of Montserrat's Emergency Department. Any organization or group on Montserrat that wished for further information or advice from the MVO scientists has been able to obtain it. Informal talks and tours of the Observatory are given to groups and individuals on request. Scientists make visits to schools, workers' groups, religious groups and commercial enterprises whenever possible or appropriate. For instance, from April 1996 to June 1997, when Plymouth was evacuated but the essential services still maintained their operations there, a great deal of effort was invested in making certain that the workers involved were fully aware of the risks that they faced. During visits made to these workplaces, talks were given and lengthy question periods allowed. When some residents of Spanish Point on the northeastern flank of the volcano refused to move away from danger, MVO scientists visited them in their homes with social workers and presented information about the risks that they were running. Bulletin boards were placed in central locations in the main occupied town of Salem and at the airport, with weekly updates to the displays being prepared by two school students, helped by Observatory staff.

Negative feedback All these public education and information efforts have been fully supported by the Government of Montserrat, the UK government and other agencies, and there have been strong, positive comments on their benefits from all sections of the Montserratian community (including those abroad). Of course, there have also been times

when negative comments were received from the public. Some were made on occasions when the public became aware of internal difficulties within the MVO and these usually concerned the independence, impartiality, cultural sensitivity, or otherwise, of the contributions being made by some members of the team (because of weight of numbers and the political backdrop, a lot of these criticisms tended to be aimed at UK scientists). More generally, however, adverse comments were forthcoming during periods when there appeared to be differences in perception between the MVO and the public regarding the level of hazards posed by the volcano and what mitigation steps were most appropriate. At times when strong eruptive activity waned and visible activity was low for weeks on end. many of the public quickly felt that the volcano was doing nothing at all and posed little or no threat, despite the fact they were informed that the monitoring instruments continued to indicate that the eruption had not ended. To make matters worse, these periods were commonly followed by episodes when the volcano suddenly escalated to highly dangerous activity, often in only a matter of hours. Keeping the population sufficiently alert to the dangers during such lulls in activity was a continuing, major difficulty for the scientists and, on some occasions, led to a few vocal people expressing public disagreement with MVO advice as to the level of risk involved. Examples included the scientists' recognition in early July 1997 that the style of eruption had changed to a more explosive form that might threaten areas further afield, and later, in 1998. that the key bridge over the Belham River was likely to be submerged by lahars. Both warnings were greeted by disbelief and scorn by a few small sections of the community. The general situation was not helped, however, by the fact that, in some instances, although the scientific team had correctly identified the incipient dangers, these did not always materialize promptly. In the case of the major 26 December (Boxing Day) 1997 collapse and pyroclastic surge (Sparks ei al. 2002) that devastated the southwest part of the island, scientific anxiety had been elevated for over a vear before the event.

87

THE MONTSERRAT VOLCANO OBSERVATORY

0.0001

1

10

100

1000

No. casualties N Fig. 12. Typical cumulative F-n risk curves from three risk assessments for Montserrat. The curves show estimated exceedance probabilities for a given number of casualties occurring in the next six months, based on Monte Carlo risk simulations combining identified volcanic hazard scenarios with population numbers in given areas. Such estimates are useful to civil authorities concerned with the tolerability of risk and may be compared with societal risks from other natural disasters, such as earthquakes or hurricanes. The examples shown here indicate how the perceived risk levels dropped between April 1998 and February 1999 while the volcano was in residual eruptive activity mode.

At other times, the credibility of the scientists seemed to be questioned in the public mind because specific instances of heightened activity were not 'predicted'. One case was the magmatic explosion of 17 September 1996. Even though the general likelihood of such an event taking place at some time had been repeatedly expressed, and despite the fact that the scientists had invariably indicated that specific temporal predictions were not possible, when the explosion did occur there was fairly widespread censure of the MVO for a perceived 'failure to perform'. Nevertheless, throughout the crisis, the majority of the general public were very supportive of the work of the Observatory scientists and appreciative of their efforts. Overall, confidence in the scientists and their assessments generally improved as events that had been forecast did, indeed, come to pass and as the team got better at both anticipating the behaviour of the volcano and communicating their insights and understanding. Perhaps one shortcoming of the public education programme that has been pursued by the MVO has been the absence of a mechanism for gauging the effectiveness of the outreach. This has been due mainly to the lack, until September 1999, of a suitably trained permanent member of staff to co-ordinate this particular activity. Apart from the daily scientific reports, many of the methods used to disseminate information to the public have not been sustained throughout the eruption. There have been one-off responses to particular concerns within the scientific team regarding public awareness and safety, and there have been instances that were the result of the (short-term) presence at the MVO of personnel who had specific interests in public education. Greater support for a wider range of outreach activities became possible with the recruitment to the MVO of a full-time Information Officer. However, there was one occasion when comments were deliberately solicited by the MVO from the public. This was organized immediately following the tragic events of 25 June 1997 and consisted of a series of interviews with people who were in the Exclusion Zone at the time, and with those involved in rescuing survivors (Loughlin et al. 2002a). While the information gathered was intended to provide documentary evidence on details of the volcanic episode itself, the responses also allowed the MVO to investigate how well people felt they had been informed overall on the dangers associated with the eruption. The results suggest that, predominantly, MVO's efforts at public education had been

Table 4. Summary of main methods used by the MVO to disseminate information to the public, August 1995 to April 2000 Method

Start date

End date

MVO reports (daily - morning) MVO reports (daily - evening) MVO reports (weekly) Scientific reports (weekly) Scientific reports (monthly) Scientific reports (quarterly) Special volcanic activity reports Lunch-time updates Friday evening radio interview Radio updates at times of high activity SeismiCity News Government Information Service 'Scientific Explanation' Montserrat Reporter special articles Public posters in Salem and at airport Targeted meetings (e.g. to present risk map revisions) Public lectures (occasional) Visits to schools (occasional) Visits by schools to MVO (occasional) Visits to MVO by various sectors of the population (occasional) Video nights (occasional) Local access TV (occasional) Antigua Broadcasting Service broadcasts (occasional) Off-island visits/talks to Montserratian communities abroad

August 1995 August 1995 April 1999 November 1995 June 1998 May 1999 September 1996 February 1997 February 1997 December 1996 January 1997 1997 1996 December 1996 July 1995 August 1995 November 1995 November 1995 July 1995 November 1995 November 1995 November 1995 November 1995

April 1998 March 1999 Continuing April 1998 April 1999 Continuing Continuing November 1997 Continuing Continuing March 1999 Continuing Continuing June 1997 Continuing Continuing Continuing Continuing Continuing Continuing Continuing Continuing Continuing

W. P. ASPINALL ET AL.

effective in informing the public of the risks. They also showed that individuals' decisions regarding entry into high-risk areas on the volcano were very complex and could not be attributed to any single causative factor. Some people in Montserrat appeared willing to take high risks if the perceived return or benefit to them or the community was considered of sufficient importance (e.g. to continue farming), or if the alternatives were repugnant (e.g. living in shelters). These interpretations, and other contributory factors, were to a large measure confirmed by the findings of the jury in the Coroner's Inquest into the fatalities of 25 June 1997, in the verdict handed down in January 1999.

The Internet One important and, for the region, novel aspect of the MVO's outreach has been the use of the Internet to communicate with the wider community of scientists, with non-resident Montserratians who were otherwise unable to obtain up-to-date information on the state of the volcano and with volcano enthusiasts generally. This was achieved through co-operation with the Michigan Technological University, where a Web page containing MVO reports and other information was maintained for most of the eruption. However, this new medium also showed how relationships between scientists, and between scientists and politicians, can be easily jeopardized by such communications if care is not taken. One group unwisely put more information and interpretation into a report on their homepage than the MVO wished to issue at the time. In addition, insensitive phrasing was used to express some of the group's views. This infelicitous 'report' was openly available to the surfing public and, of course, provoked further claims of information suppression on the island. At the same time, the content of the report also succeeded in irritating or offending some scientific colleagues. In September 1999, the MVO was able to establish its own Web page (). MVO scientists also used the Internet to solicit advice and maintain contact with professional colleagues throughout the world. This enabled them to access a wider variety and greater depth of scientific thought than would have been available otherwise. As one example, in October 1996 there were fears that a sector collapse could generate a large volcanic landslide and associated tsunami. Contact with numerical modelling specialists in France and the UK via the Internet enabled sophisticated calculations to be done, the results of which (Heinrich et ai 1999) allayed concerns as to the potential size of any tsunami generated in this way. Thus, in principle and in practice, the Internet allows an observatory to consult very rapidly with the world's leading experts. For the group of Chief Scientists who were responsible for the MVO during the first three years of the crisis, communication by e-mail was also used effectively to keep each of them informed of developments on Montserrat between individual rotations in post. In the wider context of scientific advancement, the MVO has encouraged collaboration with external scientists who wish to undertake research at Soufriere Hills Volcano. Attempts have been made to ensure that such collaboration is done within a defined framework and following established protocols. A set of guidelines was drafted in 1997 to serve as operational procedures for researchers working at the MVO and for those wishing to collaborate with the MVO, Due to a number of problems, mainly relating to the evolving management structure at the MVO, described above, these protocols have not always been strictly adhered to or consistently implemented. However, most scientists who have worked on Montserrat have respected customary standards of scientific etiquette while on the island. Summing up A new institution, the Montserrat Volcano Observatory, has been established on Montserrat to monitor the Soufriere Hills volcano.

While the MVO owes its existence to the eruption that started in July 1995, it was not created from a standing start: it is founded on the Seismic Research Unit's long experience and knowledge of earthquakes and volcanic crises in the islands of the eastern Caribbean, and on the strength of UK volcanology, supported by the scientific project management experience of BGS. The nature of the present crisis in Montserrat, which changed from one that was initially expected to be short-lived, with high levels of public sensitization-to-threat, to one that became unusually protracted, has presented difficulties in sustaining both public anxiety and official caution. As a result, fresh new challenges were continually arising for the scientific team, both in respect of the behaviour of the volcano itself and in terms of necessary interactions with administrators and the public. Developing new capabilities for monitoring and instituting innovative mitigation measures within a besieged community, with a frequently changing cast of participants, has been at the heart of the effort. Many valuable lessons have been learned about the processes of interfacing with decision makers, key actors and stakeholders, and the public, some of which are being transferred through regional disaster preparedness organizations to other islands. For long periods in the Montserrat crisis, the micromanagement of hazards on a small island with an erupting volcano and a population intent on remaining, pushed the society and its politicians to use to the limit such guidance as the scientists could provide: the implications of this experience, and how it might redefine the scientific role in future volcanic crises, are worthy of more detailed appraisal. It is to the credit of all involved that the performance of the MVO in terms of the quality and timeliness of its scientific advice was as good as it turned out to be. The establishment of a local entity with responsibility for monitoring a single volcano, operating at arm's-length from the regional agency, represents a departure from the conventional way of tackling the problem. Individual observatory buildings and dedicated local teams have been set up before in the region, e.g. on St Vincent and on the French islands of Martinique and Guadeloupe, but these all operate under the aegis of bigger organizations. The exceptional circumstances in Montserrat (dual political hierarchy; protracted, slowly escalating eruption) were the major factors that contributed to the setting up of the MVO as a separate body. Responding to this particular crisis required an extended range of skills and staffing capacity, and thus demanded more resources than the SRU could readily provide. This raises the fundamental question (which will probably recur in future at other small volcanic island states in the region) as to whether an alternative scheme for resourcing the monitoring effort through the SRU could have been implemented. It is too facile to argue that there were traditional colonial political pressures making sure that management control remained in the hands of the home donor country, and that this is the basic explanation for the way things developed as they did. In the earliest stages of the crisis, attempts were made to configure a viable way forward for SRU as lead team in Montserrat, but a combination of singular personalities, unique institutional conditions and characteristics, and many other internal and external factors, were sufficient to divert the evolution of the MVO into other channels, at least for the time being. The SRU experience in Montserrat also underlines yet again the cliched but difficult challenge that volcanologists in such regions need to be able to convince their governments to make long-term investments in hazard mitigation measures and related resources so that good quality, cost-effective solutions are available when disaster strikes. In the present case, the solution to the issue of staffing a monitoring team for dealing with an ongoing crisis was found by combining, on a rotating basis, a corps of young scientists and PhD students with a kernel of experienced volcanologists from various institutions. While this has probably been beneficial overall and has been invaluable as a training regime for the young scientists, it does depart from the way volcanic eruptions have been handled in the region in the past, and at many volcanoes elsewhere in the world. This course of action has not been without drawbacks: for the Chief

THE MONTSERRAT VOLCANO OBSERVATORY Scientist or Director, additional people also mean the managing of extra overheads and resources (vehicles, radios, hot-suits, etc.), and increased anxiety about safety of personnel. The ambitions, goals and experience of some of those involved may be more oriented to research and data collection for that purpose, than to contributing to hazard assessment and mitigation. For students, the circumstances of a major volcanic crisis are not always ideal for fostering their best interests, either academically or in personal terms. Relationships can become very intense and strained working at close quarters in such conditions, and there is seldom the opportunity to relax and regroup that exists in the usual university environment. Furthermore, it is debatable to what extent any student should be immersed in the decision-making process of a live crisis that might result in a bad catastrophe, such as the fatalities of 25 June 1997: some individuals in those circumstances may be inclined to feel, quite unduly, that they share a significant responsibility for the outcome, as a member of the team. That having been said, the Soufriere Hills eruption crisis has clearly been a major stimulus for many young scientists, including, most importantly, several young Montserratians. In recent discussions, the two governments and the various agencies involved in Montserrat have been seeking to map out the long-term direction of the MVO and its eventual integration into regional monitoring. However, the resurgence of dome growth in November 1999 may delay these initiatives and immediate attention is being focused on drawing up a monitoring strategy for coping with yet another major episode of persistent eruptive activity. While this is the current preoccupation, it is foreseen that, in due course, the legacy of resources, knowledge and experience gained at the MVO during this eruption will pass locally into the custodianship of trained Montserratians and will enhance the future management of volcanic crises through a strengthened regional agency. In the meantime, with the eruption ongoing, the new institution has the opportunity to play an important role in advancing volcano monitoring and volcanology, while continuing to fulfil its responsibilities to the people and authorities on Montserrat. The people of Montserrat, so badly affected by the disaster in their island, have given wholehearted support to the MVO and their encouragement and expressions of confidence have been greatly appreciated by all the scientists involved. The MVO has benefited from the hard work and commitment of local staff and individuals who selflessly devoted their time and energies when families were under threat and homes were being lost to the volcano, and from the contributions of scientists and technical staff from the Seismic Research Unit of the University of the West Indies, in particular L. Lynch and R. Robertson. Funding for the MVO has been provided by the Government of Montserrat and by the UK government through the Department for International Development. For relevant authors, this paper is published by permission of the Director, British Geological Survey (NERC) and some images are reproduced by permission of the British Geological Survey NERC - all rights reserved [IPR/7-60]. References AHMAD, R. (ed.) 1996. Abstracts from The Second Caribbean Conference on Natural Hazards and Disasters: Science, Hazards And Hazard Management, 9-12 October 1996, Kingston, Jamaica. Publication No. 1 Unit for Disaster Studies, Department of Geography and Geology, University of the West Indies, Jamaica, 26-41. ALLEN, A. G. 1996. The influence of local volcanic emissions on the atmospheric environment of Montserrat, West Indies (Abstract). Applied Geoscience 96, Warwick. AMBEH, W. B., LYNCH, L. L., CHEN, J. & ROBERTSON, R. E. A. 1998. Contemporary seismicity of Montserrat, West Indies (abstract). In: ALI, W., A. PAUL, A. & YOUNG ON, V. (eds) Transactions of the 3rd Geological Conference of the Geological Society of Trinidad and Tobago and the 14th Caribbean Geological Conference, Port of Spain, Trinidad, Vol. 1, 1-2. ASPINALL, W. P. & COOKE, R. M. 1998. Expert judgement and the Montserrat Volcano eruption. In: MOSLEH, A. BARI, R. A. (eds) Proceedings of the 4th International Conference on Probabilistic Safety Assessment and Management PSAM4, 13-18 September 1998, New York City, USA, Vol. 3, 2113-2118.

89

ASPINALL, W. P. & Woo, G. 1994. An impartial decision-making procedure using expert judgement to assess volcanic hazards. Atti dei Convegni Lincei, Rome, 112, 211-220. ASPINALL, W. P., SIGURDSSON, H. & SHEPHERD, J. B. 1972. Eruption of Soufriere Volcano on St. Vincent Island 1971-1972. Science, 181, 117-124. ASPINALL, W. P., LYNCH, L. L., ROBERTSON, R. E. A., ROWLEY, K. C, SPARKS, R. S. J., VOIGHT, B. & YOUNG, S. R. 1998a. The Soufriere Hills Eruption, Montserrat, British West Indies: Introduction to Special Section, Part 1. Geophysical Research Letters, 25, 3387. ASPINALL, W. P., MILLER, A. D., LYNCH, L. L., LATCHMAN, J. L., STEWART, R. C., WHITE, R. A. & POWER, J. A. 1998b. Soufriere Hills eruption, Montserrat 1995-1997: volcanic earthquake locations and fault plane solutions. Geophysical Research Letters, 25, 3397-3400. BARCLAY, J., RUTHERFORD, M. J., CARROLL, M. R., MURPHY, M. D., DEVINE, J. D., GARDNER, J. & SPARKS, R. S. J. 1998. Experimental phase equilibria constraints on preemptive storage conditions of the Soufriere Hills magma. Geophysical Research Letters, 25, 3437-3650. BONADONNA, C., MAYBERRY, G. C., CALDER, E. S. ET AL. 2002. Tephra

fallout in the eruption of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 483-516. BOUDON, G., VlLLEMANT, B., KOMOROWSKI, J-C, ILDEFONSE, P. & SEMET,

M. P. 1998. The hydrothermal system at Soufriere Hills Volcano, Montserrat (West Indies): characterisation and role in the ongoing eruption. Geophysical Research Letters, 25, 3693-3696. CALDER, E. S., COLE, P. D., DADE, W. B. ET AL. 1999. Mobility of pyroclastic flows and surges at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 26, 537-540. CLAY, E., BARROW, C., BENSON, C., DEMPSTER, J., KOKELAAR, P., PILLAI, N. & SEAMAN, J. 1999. An Evaluation of HMG's Response to the Montserrat Volcanic Emergency. Department for International Development Evaluation, London, Report EV635. COLE, P. D., CALDER, E. S., DRUITT, T. H., HOBLITT, R., ROBERTSON, R., SPARKS, R. S. J. & YOUNG, S. R. 1998. Pyroclastic flows generated by gravitational instability of the 1996-1997 lava dome of Soufriere Hills. Geophysical Research Letters, 25, 3425-3428. COLE, P. D., CALDER, E. S., SPARKS, R. S. J. ET AL. 2002. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoir, 21, 231-262. COOKE, R. M. 1991. Experts in Uncertainty. Oxford University Press, Oxford. CURNOW, R. N. 1999. Unfathomable nature and Government policy (the address of the President of the Royal Statistical Society). The Statistician, Journal of the Royal Statistical Society, Series D 48(4), 463-476. DEVINE, J. D., MURPHY, M. D., RUTHERFORD, M. J. ET AL. 1998a. Petrologic evidence for pre-eruptive pressure-temperature conditions, and recent reheating, of andesitic magma erupting at the Soufriere Hills Volcano, Montserrat, West Indies. Geophysical Research Letters, 25, 3669-3672. DEVINE, J. D., RUTHERFORD, M. J. & GARDNER, J. E. 1998b. Petrologic determination of ascent rates for the 1995-1997 Soufriere Hills Volcano andesitic magma. Geophysical Research Letters, 25, 3673-3676. DRUITT, T. H., CALDER, E. S., COLE, P. D. ET AL. 2002. Small-volume, highly mobile pyroclastic flows formed by rapid sedimentation from pyroclastic surges at Soufriere Hills Volcano, Montserrat: an important volcanic hazard. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 263-279. EDMONDS, M., PYLE, D. M. & OPPEHEIMER, C. 2001. A model for degassing at the Soufriere Hills volcano, Montserrat, West Indies, based on geochemical data. Earth and Planetary Science Letters, 186, 159-173. FISKE, R. S. 1984. Volcanologists, journalists, and the public: a tale of two crises in the Eastern Caribbean. In: Explosive Volcanism: Inception, Evolution and Hazards. National Research Council (US) Geophysics Study, National Academy Press, Washington DC, 170-176. HAMMOUYA, G., ALLARD, P., JEAN-BAPTISTE, P., PARELLO, F., SEMET, M. P. & YOUNG, S. R. 1998. Pre- and syn-eruptive geochemistry of volcanic gases from Soufriere Hills of Montserrat, West Indies. Geophysical Research Letters, 25, 3685-3688. HARFORD, C. L., PRINGLE, M. S., SPARKS, R. S. J. & YOUNG, S. R. 2002. The volcanic evolution of Montserrat using 40 Ar/ 39 Ar geochronology.

90

W. P. ASPINALL ET AL.

In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 93-113. HEINRICH, P., GUIBORG, S., MANGENEY, A. & ROCHE, R. 1999. Numerical modeling of a landslide-generated tsunami following a potential explosion of the Montserrat volcano. Physics and Chemistry of the Earth (A), 24(2), 163-168. HERD, R., CALDER, E., BARCLAY, J. ET AL. 1998. Methods used to survey the lava dome and associated products at the Soufriere Hills Volcano, Montserrat. Montserrat Volcano Observatory Open File Report 98/16. JACKSON, P., SHEPHERD, J. B., ROBERTSON, R. E. A. & SKERRITT, G. 1998. Ground deformation studies at Soufriere Hills volcano, Montserrat I: Electronic distance meter studies. Geophysical Research Letters, 25, 3409-3412. KOKELAAR, B. P. 2002. Setting, chronology and consequences of the eruption of Soufriere Hills Volcano, Montserrat (1995-1999). In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 1-43. KOMOROWSKI, J-C., NORTON, G., BOUDON, G. ET AL. 2000. Field-relationships and texture of the striking debris-avalanche deposit from the 1997 flank-collapse of Soufriere Hills Volcano, Montserrat (abstract). Geoscience 2000. LEE, W. H. K. 1989. PC-SEIS: a toolbox for seismic data acquisition, processing, and analysis. In: LEE, W. H. K. (ed.) IASPEI Software Library, Vol. I. LEE, W. H. K. & VALDES, C. M. 1989. HYPO71PC. In: LEE, W. H. K. (ed.) IASPEI Software Library, Vol. I, 203-236. LOUGHLIN, S. C., BAXTER, P. J., ASPINALL, W. P., DARROUX, B., HARFORD, C. L. & MILLER, A. D. 2002«. Eyewitness accounts of the 25 June 1997 pyroclastic flows and surges at Soufriere Hills Volcano, Montserrat, and implications for disaster mitigation. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 211-230. LOUGHLIN, S. C., CALDER, E. S., CLARKE, A. B. ET AL. 2002/7. Pyroclastic flows and surges generated by the 25 June 1997 dome collapse, Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 191-209. MATTIOLI, G. S., DIXON, T. H., FARINA, F., HOWELL, E. S., JANSMA, P. E. & SMITH, A. L. 1998. GPS measurement of surface deformation around Soufriere Hills Volcano, Montserrat from October 1995 to July 1996. Geophysical Research Letters, 25, 3417-3420. MILLER, A. D., STEWART, R. C., WHITE, R. A. ET AL. 1998. Seismicity associated with dome growth and collapse at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3401-3404. MONTSERRAT VOLCANO OBSERVATORY TEAM 1997. The Ongoing Eruption in Montserrat. Science, 276, 371-372. MURPHY, M. D., SPARKS, R. S. J., BARCLAY, J. ET AL. 1998. The role of magma mixing in triggering the current eruption at the Soufriere Hills Volcano, Montserrat, West Indies. Geophysical Research Letters, 25, 3433-3436. MURPHY, M. D., SPARKS, R. S. J., BARCLAY, J., CARROLL, M. J. & BREWER, T. S. 2000. Remobilization of andesite magma by intrusion of mafic magma at the Soufriere Hills volcano, Montserrat, West Indies. Journal of Petrology, 41, 21-42. MURRAY, T. L., EWERT, J. W., LOCKHART, A. B. & LAHUSEN, R. G. 1996. The integrated mobile volcano-monitoring system used by the Volcano Disaster Assistance Program (VDAP). In: SCARPA, R. & TILLING, R. I. (eds) Monitoring and Mitigation of Volcano Hazards, Springer, Berlin, 315-362. NEUBERG, J., BAPTIE, B., LUCKETT, R. & STEWART, R. 1998. Results from the broadband seismic network on Montserrat. Geophysical Research Letters, 25, 3661-3664. NORTON, G. E. 1997. Summary of the environmental effects of volcano emissions, Soufriere Hills volcano, Montserrat. Montserrat Volcano Obervatory Open File Report 97/15. NORTON, G. E., WATTS, R. B., VOIGHT, B. ET AL. 2002. Pyroclastic flow and explosive activity at Soufriere Hills Volcano, Montserrat, during a period of virtually no magma extrusion (March 1998 to November 1999). In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 467-481.

OFFICE OF SCIENCE AND TECHNOLOGY 1997. The Use of Scientific Advice in Policy Making. Department of Trade and Industry, London. OFFICE OF SCIENCE AND TECHNOLOGY 1998. The Use of Scientific Advice in Policy Making: Implementation of the Guidelines. Department of Trade and Industry, London. OPPENHEIMER, C., FRANCIS, P. & MACIEJEWSKI, A. J. H. 1998. Spectroscopic observation of HC1 degassing from Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3681-3684. PATTULLO, P. 2000. Fire from the Mountain. Constable, London. ROBERTSON, R. E. A., ASPINALL, W. P., HERD, R. A.. NORTON. G. E., SPARKS, R. S. J. & YOUNG, S. R. 2000. The 1995-1998 eruption of the Soufriere Hills volcano, Montserrat, WI. Philosophical Transactions of the Royal Society, London, Series A 358(1770), 1619-1637. ROOBOL, M. J. & SMITH, A. L. 1998. Pyroclastic stratigraphy of the Soufriere Hills Volcano, Montserrat: Implications for the present eruption. Geophysical Research Letters. 25. 3393-3396. SHEPHERD, J. B., TOMBLIN, J. F. Woo, D. A. 1971. Volcano-seismic crisis in Montserrat. West Indies 1966-67. Bulletin Volcanologique. 35. 143-163. SHEPHERD, J. B., ASPINALL, W. P.. ROWLEY. K. C., PEREIRA, J.. SIGURDSSON, H., FISKE, R. S. & TOMBLIN, J. F. 1979. The eruption of Soufriere Volcano, St. Vincent April-June 1979. Nature, 282, 24-28. SHEPHERD, J. B., HERD, R. A.. JACKSON.. P. & WATTS, R. 1998. Ground deformation measurements at the Soufriere Hills Volcano, Montserrat II: Rapid static GPS measurements. June 1996-June 1997. Geophysical Research Letters, 25. 3413-3416. SPARKS, R. S. J. & YOUNG. S. R. 2002. The eruption of Soufriere Hills Volcano, Montserrat (1995-1999): overview of scientific results. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs, 21, 45-69. SPARKS. R. S. J., YOUNG. S. R., BARCLAY. J. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills Volcano. Montserrat, West Indies: November 1995 to December 1997. Geophysical Research Letters, 25, 3421-3424. SPARKS, R. S. J., BARCLAY, J., CALDER. E. S. ET AL. 2002 Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano. Montserrat. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21. 409-434. TAZIEFF, H. 1980. Letter to the Editor. Journal Volcanology and Geothermal Research, 8, 3-6. THE ROYAL SOCIETY 1999. Science and Society: a response to the inquiry by the House of Lords Science and Technology Select Committee. Statement 12 99, London. VOIGHT, B., HOBLITT, R. P., CLARKE. A. B., LOCKHART, A. B.. MILLER. A. D.. LYNCH, L. & McMAHON. J. 1998. Remarkable cyclic ground deformation monitored in real-time on Montserrat, and its use in eruption forecasting. Geophysical Research Letters, 25. 3405-3408. VOIGHT, B.. SPARKS, R. S. J., MILLER. A. D. ET AL. 1999. Magma flowinstability and cyclic activity at Soufriere Hills Volcano. Montserrat. B.W.I. Science. 283. 1138-1142. WADGE, G. & ISAACS, M. C. 1987. Volcanic Hazards from Soufriere Hills Volcano, Montserrat, West Indies. Report to the Government of Montserrat and the Pan Caribbean Disaster Preparedness and Prevention Project. WADGE, G. & ISAACS. M. C. 1988. Mapping the volcanic hazards from the Soufriere Hills Volcano, Montserrat. West Indies using an image processor. Journal of the Geological Society, London, 145. 541-551. WADGE, G., JACKSON, P., BOWER, S. M., WOODS, A. W. & CALDER. E. 1998. Computer simulations of pyroclastic flows from dome collapse. Geophysical Research Letters. 25. 3677-3680. WATTS, R. B., HERD, R. A., SPARKS. R. S. J. & YOUNG, S. R. 2002. Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs. 21. 115-121. YOUNG, S., SPARKS, S., ROBERTSON. R., LYNCH, L. & ASPINALL, W. 1997. Eruption of Soufriere Hills Volcano in Montserrat Continues. EOS, Transactions, American Geophysical Union, 78(38), 404. 408-409. YOUNG, S. R., FRANCIS, P. W., BARCLAY, J. ET AL. 19980. Monitoring SO2 emission at the Soufriere Hills volcano: Implications for changes in eruptive conditions. Geophysical Research Letters. 25. 3681-3684.

THE MONTSERRAT VOLCANO OBSERVATORY YOUNG, S. R., SPARKS, R. S., ASPINALL, W. P., LYNCH, L., MILLER, A. D., ROBERTSON, R. E. A. & SHEPHERD, J. B. 1998b. Overview of the eruption of Soufriere Hills volcano, Montserrat, 18 July 1995 to December 1997. Geophysical Research Letters, 25, 3389-3392.

91

YOUNG, S. R., VOIGHT, B., ROWLEY, K., SPARKS, R. S. J., ROBERTSON, R. E. A., LYNCH, L. L. & ASPINALL, W. P. 1998c. The Soufriere Hills Eruption, Montserrat, British West Indies: Introduction to Special Section, Part 2. Geophysical Research Letters, 25, 3387.

The volcanic evolution of Montserrat using 40Ar/39Ar geochronology C. L. HARFORD1, M. S. PRINGLE2, R. S. J. SPARKS1 & S. R. YOUNG3 1

Department of Earth Sciences, Bristol University, Bristol BS8 1RJ, UK (e-mail: [email protected]) 2 Scottish Universities Research and Reactor Centre, East Kilbride G75 OQF, UK Montserrat Volcano Observatory, Mongo Hill, Montserrat, West Indies

Abstract: 40 Ar/ 39 Ar dating has facilitated a substantial reinterpretation of the volcanic evolution of Montserrat. Three volcanic centres with non-overlapping volcanic activity are identified: Silver Hills (c. 2600 to c. 1200ka); Centre Hills (at least c. 950 to c. 550 ka); South Soufriere Hills-Soufriere Hills (at least c. 170ka to present). The geochronological data show that old xenocrysts are common in the porphyritic andesite, implying that reliable ages are best obtained by dating the groundmass. Soufriere Hills evolved from early eruptions dominated by two-pyroxene andesite to eruptions of hypersthene-hornblende andesite at c. 1 lOka. Between the two varieties of andesite there was an interlude of mafic volcanism at c. 130ka to form South Soufriere Hills. There is evidence of tectonic uplift of early products of the complex along with older submarine volcanic rocks. Consideration of stratigraphy and age data indicates that only a proportion of the dome-forming eruptions are recorded as domes in the geological record. Older products are removed from the subaerial edifice by sector-collapse events. The timeaveraged eruption rate of the South Soufriere Hills-Soufriere Hills centre is estimated at 0.005 m3 s"1 (c. 0.15 km3 ka" 1 ) (dense rock equivalent). The ongoing eruption is very similar in style to previous activity at Soufriere Hills, and future activity is likely to pose similar hazards. Soufriere Hills have been characterized by alternations of periods of enhanced activity and periods of dormancy, both lasting of the order of 104 years. During periods of elevated activity several major dome-forming eruptions are separated by quiescent interludes lasting less than c. 103 years. The ongoing eruption may mark the onset of a fourth period of enhanced volcanic activity at Soufriere Hills.

Geochronology is important for understanding the evolution of young volcanic systems. High-quality age data can help constrain trends of spatial, geochemical and petrological evolution of volcanism. Such data also allow estimation of rates of volcanic processes and magmatic evolution, including eruption rates and fluctuations thereof. Work at Crater Lake, Oregon (Bacon & Lanphere 1990), Mount Adams, Washington (Hildreth & Lanphere 1994), Tatara-San Pedro complex, Chile (Singer et ai 1997) and Santorini Volcano, Greece (Druitt et al 1999) illustrates how a detailed and high-precision geochronological framework can enhance studies of volcanic systems. This paper presents a study of the evolution of volcanism on the island of Montserrat, West Indies, using 40Ar/ 39 Ar geochronology. Methods have been developed only relatively recently to provide precise age information over the age range typical of many volcanoes (tens to hundreds of thousands of years). Radiocarbon dating is only applicable to deposits younger than c. 45 ka, due to the short half-life of 14C. In addition, the radiocarbon timescale is precisely calibrated only back to 21 ka, and is susceptible to contamination problems, particularly in older samples. Furthermore, only deposits containing carbon may be dated, ruling out lava flows, domes and some pyroclastic deposits. K-Ar geochronology, in particular the 40 Ar/ 39 Ar variant, has been the most successfully applied radiometric technique on volcanic rocks less than several hundred thousand years old. Here 40 Ar/ 39 Ar geochronology is applied to a relatively difficult problem, a comparatively low-K volcanic system that is susceptible to alteration in a tropical environment. The island of Montserrat is located in the northern part of the Lesser Antilles island arc, and is over 16km long (north-south) and 10km wide (east-west). Montserrat is built almost exclusively of volcanic rock. The latest manifestation of the island's extrusive history has been the growth of an andesitic lava dome from 1995 to present at Soufriere Hills Volcano, with associated dome-collapse pyroclastic flows and explosions (Robertson et al 2000; Young et al 1998). The island's past volcanic activity is of importance for interpreting this ongoing and any future activity. Previous investigations on Montserrat have suggested that the island has evolved over the last 4 million years with at least six volcanic centres (Fig. 1). MacGregor (1938) carried out the first geological and petrological work on Montserrat. Rea (1970, 1974) developed MacGregor's work and produced a geological map. Two radiocarbon dates and four conventional K-Ar dates (three of the latter were later published in Briden et al. 1979) facilitated Rea's

(1970, 1974) reinterpretation of the geological evolution of the island (Fig. 1). However, the results in Briden et al. (1979) on other islands of the Lesser Antilles have since been found to be inaccurate (Carlut et al. 2000). A further eight conventional K-Ar dates were published by Le Gall et al. (1983), although problems were acknowledged for some of the dates obtained. The precision of all the existing K-Ar ages is poor (Icr = 7 to 41%). The work presented here confirms that significant uncertainties, often even greater than the reported analytical uncertainties, are associated with conventional K-Ar geochronological studies. In order to aid volcanic hazard investigations Baker (1985) and Wadge & Isaacs (1988) studied, and obtained radiocarbon dates for, the youngest deposits of Soufriere Hills. Roobol & Smith (1998) carried out detailed work on the pyroclastic stratigraphy of Soufriere Hills and obtained further radiocarbon ages. The radiocarbon ages cover only the uppermost of the three major units of the pyroclastic stratigraphy and extend back only as far as 31 560 0 years BP. Roobol & Smith (1998) found no carbon to date in the lower stratigraphic units of Soufriere Hills. The geology of Montserrat Montserrat comprises three major massifs: Silver Hills in the north, Centre Hills in the centre, and South Soufriere Hills-Soufriere Hills in the south (Figs 1 and 2). In addition, Garibaldi Hill and St George's Hill form two smaller, isolated topographic highs. The three major massifs are geomorphologically distinct in terms of maturity of erosion (Fig. 2). This prompted Davis (1926) to deduce correctly their relative ages from oldest to youngest as Silver Hills, Centre Hills, South Soufriere Hills-Soufriere Hills. The island's interior, with the exception of areas affected by the ongoing eruption, is densely vegetated. Exposures are thus largely limited to coastal cliffs, roadcuts, and occasional steep interior cliffs. Andesitic deposits produced by dome-forming eruptions dominate the island, although South Soufriere Hills are predominantly basaltic to basaltic-andesite in composition. Major products on Montserrat are remnants of andesitic domes, andesitic breccias representing dome talus, block-and-ash-flow deposits originating from dome collapse, lahar deposits, debris-avalanche deposits, and usually thin and subordinate tephra-fall deposits. There are areas of hydrothermally altered rocks, including active fumarole fields (soufrieres) related to Soufriere Hills Volcano (Fig. 2).

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 93-113. 0435-4052/02/S15 © The Geological Society of London 2002.

93

94

C. L. HARFORD ET AL.

Fig. 1. Shaded relief maps, derived from a digital elevation model (DEM), showing evolution of volcanism according to different authors. This work shows that volcanism has migrated from north to south, in contrast to previous interpretations involving oscillation of volcanism. Montserraf s three major massifs show a marked decrease in erosional maturity from Silver Hills to Centre Hills to South Soufriere Hills-Soufriere Hills.

Silver Hills Silver Hills are deeply eroded. They comprise massive, predominantly andesitic lava, interpreted as parts of eroded lava domes, and sequences of volcaniclastic beds, which once would have formed fan deposits around the lava domes. Extensive areas are hydrothermally altered, for example at Yellow Hole. Debrisavalanche deposits are prominent, for example at Little Bay.

Centre Hills Centre Hills are also significantly eroded. High coastal cliffs surround the hills. The cliffs are significantly higher on the west coast (<140 m) than on the east coast (<75 m), probably reflecting faster erosion in the west due to the prevailing easterly winds and ocean currents. Deep valleys, locally known as "ghauts', incise Centre Hills and have modified the original volcanic morphology. Outcrops are largely limited to the coastal cliffs and small inland exposures formed by roadcuts and steep valley walls. The shallowly sloping flanks (<10 o ) of Centre Hills consist mainly of andesitic volcaniclastic deposits. These deposits comprise predominantly block-and-ash-flow deposits with subordinate pumice-and-ash-flow deposits, pumice-fall deposits, lahar deposits, river deposits and debris-avalanche deposits. The interior mountains consist mainly of massive andesitic lava, which represents the remnants of lava domes that fed the flank pyroclastic deposits. Erosion has degraded the original form of the domes and the stratigraphic relationship of the interior dome-dominated complex to the exterior flanks is unknown. Correlation of units between the west and east coasts proved impossible. A 5-m-thick pumice-fall deposit with no obvious depositional gaps, exposed well in Woodland's Bay and just south of Runaway Ghaut (GR 375575 1852375), indicates that significant magmatic explosive activity occurred at Centre Hills. The hills of Harris and above Trant's are composed of massive lava. These lavas are detached from and at a lower altitude than the

main central massive lava complex of Centre Hills. Thin volcaniclastic deposits overlie the lava in the area of Harris village. These observations suggest that these massive lava deposits may predate other units of Centre Hills.

South Soufriere Hills-Soufriere

Hills

The Soufriere Hills and South Soufriere Hills are little modified by erosion, and so original volcanic morphologies are well preserved. Deposits of basaltic to basaltic andesite composition, comprising lava flows intercalated with volcaniclastic beds, dominate South Soufriere Hills. The volcaniclastic beds consist mainly of breccias, interpreted as related to lava flow collapse, with some scoria-fall deposits and some reworked deposits. The c. 200-m-thick White River Pyroclastic Fall Series of Rea (1974). exposed on the west flank of Fergus Mountain, consists of basaltic to andesitic scoriafall deposits, bedded on a scale of a few centimetres to a few metres (Fig. 3a). These deposits are interpreted as originating from transient explosive (Strombolian to Vulcanian) eruptions, although the ongoing eruption of Soufriere Hills prevented detailed studies of this unit. The stratigraphic relationship between the outcrops on the west flank of Fergus Mountain and the lava flow sequence in the eastern and southern part of the South Soufriere Hills massif is not clear. The summit area of South Soufriere Hills consists of at least four horseshoe-shaped structures, which are open to the ENE and are interpreted as sector-collapse scars (Figs 1 and 3b). The preserved subaerial volume of South Soufriere Hills is only about one-third that of Soufriere Hills. Soufriere Hills are dominantly andesitic in composition and consist of a central mountainous dome complex flanked by an apron of shallow-dipping volcaniclastic deposits (Fig. 4). The central nucleus comprises four large steep-sided domes (Gages Mountain, Chances Peak. Galway's Mountain. Perches Mountain) with associated volcaniclastic aprons. This nucleus is truncated by the 900-m-wide sector-collapse scar of English's Crater, which is open to the east and contains both the 1995-1999 dome and the

EVOLUTION OF MONTSERRAT

Fig. 2. Geological map of Montserrat, including new 40Ar/39Ar data. A and B indicate the position of the profile in Figure 5.

95

96

C. L. HARFORD ET AL.

Fig. 3. South Soufriere Hills: (a) scoria-fall deposits of unit SSH-F on the western flank of Fergus Mountain (GR 379750 1844325): (b) aerial photo of South Soufriere Hills showing horseshoe-shaped collapse scars, taken from north.

smaller, now-buried Castle Peak Dome (Figs 2 and 4). English's Crater formed as a result of sector collapse thought to have occurred at 3950 70 years BP, based on a radiocarbon age obtained by Roobol & Smith (1998). Block-and-ash-flow deposits dominate the flank volcaniclastics, with some outcrops of debris avalanche and lahar deposits. The volcaniclastic sequence is best exposed in the eastern coastal cliffs. The western sea cliffs are significantly lower and only expose the uppermost levels of the stratigraphy. The volcanic stratigraphy of Soufriere Hills presented here has been adapted from that of Roobol & Smith (1998), using our observations and data (Fig. 5). The stratigraphy is divided into three major units, which are separated by erosional unconformities. The basal unit, SH-I (subunit I of Roobol & Smith 1998), exposed at Irish Ghaut, comprises a >l-m-thick block-and-ash-flow deposit, overlain by a 1-m-thick pumice-fall deposit, plus a few metres of thin pumice-and-ash-flow deposits with associated lithic lag breccias. The middle unit, SH-II (subunit II of Roobol & Smith 1998), unconformably overlies SH-I and consists almost entirely of block-andash-flow deposits. We found no evidence for any notable pyroclastic

surge deposits as described by Roobol & Smith (1998). The upper unit, SH-III (subunit III of Roobol & Smith 1998), comprises extensive block-and-ash-flow deposits with minor lenses of pumiceand-ash-flow deposits, which are well exposed at Spanish Point (Fig. 6a). These deposits resemble the deposits of the ongoing eruption of Soufriere Hills Volcano, in which block-and-ash-flow deposits are intercalated with thin anastamosing fingers of pumiceand-ash-flow deposits formed by partial collapses of Vulcanian explosion columns (Cole el al. 2002; Druitt el al. 2002). Thin pumiceand-ash-flow deposits are also observed on the southwestern flanks of Soufriere Hills Volcano, near the top of the section, at the same stratigraphic level as pumice-and-ash-flow deposits on the eastern coast (Wadge & Isaacs 1988). Radiocarbon ages for SH-III indicate that this unit dates from at least 31 560 230 years BP to 16 980 60 years BP (Baker 1985; Rea 1974; Roobol & Smith 1998; Wadge & Isaacs 1988). Overlying the three major Soufriere Hills (SH) units are two minor units, SH-G dated at 3950 0 years BP (Marker G of Roobol & Smith 1998), and unit SH-LPD confined to the Tar River valley and dated at 770 to 200 years BP (Late Prehistoric Deposits

EVOLUTION OF MONTSERRAT

97

Fig. 4. Soufriere Hills: (a) from east (January 1999), showing the central, dome-dominated complex, surrounded by shallow-dipping pyroclastic deposits; (b) from west (28 October 1995).

of Roobol & Smith 1998). These deposits are interpreted as associated with formation of English's Crater and Castle Peak Dome, respectively. Castle Peak Dome and all associated deposits have been buried or removed by the ongoing activity. Unpublished analyses of 14C dates suggest the early seventeenth century as the most likely time for the Castle Peak Dome eruption(s). At the southern end of the eastern coastal cliffs of Soufriere Hills, around Mefraimie Ghaut, units SH-II and SH-III unconformably overlie volcaniclastic deposits of basaltic to basaltic andesite composition (Fig. 5). These volcaniclastic deposits comprise poorly sorted beds, interpreted as mainly block-and-ash-flow deposits related to lava flow collapse, and subordinate well-sorted beds interpreted as scoria-fall deposits. We infer that these deposits were sourced from South Soufriere Hills (SSH) due to compositional and lithological similarities to SSH deposits. We found no evidence for major faulted contacts in this area, although faults are indicated by Roobol & Smith (1998). There is, however, minor faulting along the

eastern cliff section, in particular between Spring Ghaut and Irish Ghaut (Fig. 5). Andesitic block-and-ash-flow deposits representing upper levels of the Soufriere Hills stratigraphy (SH-III) (Rea 1974) overlie pyroclastic-fall deposits of the western flanks of the South Soufriere Hills in the ghauts of SW Soufriere Hills. A 2-m-thick deposit, comprising andesitic pumice-fall and basaltic andesite scoria-fall beds that alternate on a centimetre to decimetre scale, is accessible in Irish Ghaut, and seems to be within unit SH-III (Figs 5 and 6b). Geochemical trends suggest that the basaltic andesite scoria is more similar to Soufriere Hills magmas than to South Soufriere Hills magmas (Fig. 7) (G. Zellmer, pers. comm.). For example, the vanadium content of the basaltic andesite scoria is similar to that of Soufriere Hills mafic inclusions, whereas mafic South Soufriere Hills samples have distinctly higher values (Fig. 7b). The composition of these Soufriere Hills tephrafall deposits may be the result of magma hybridization between andesitic magma and mafic inclusions. Such hybridization has not

98

C. L. HARFORD ET AL.

Fig. 5. Simplified Soufriere Hills stratigraphy, adapted after Roobol & Smith (1998), with new 4() Ar 39 Ar sample locations and ages marked. Location of profile A-B marked on Figure 2. The stratigraphy south of Spring Ghaut has been adapted from that of Roobol & Smith (1998) to fit in with our observations and data. SH-G, SH-III, SH-II and SH-I represent units of Soufriere Hills pyroclastic stratigraphy from youngest to oldest: SSH represents volcaniclastic deposits of South Soufriere Hills; RBSV represents Roche's Bluff submarine volcaniclastics. Radiocarbon sample locations (Roobol & Smith 1998) are shown in italics. The upper surface of the Tar River pyroclastic fan in 1998 is shown. This fan prevented us making observations at lower, previously exposed areas of the stratigraphy.

occurred in the ongoing eruption, and does not seem to have been common in the history of Soufriere Hills. Instead mafic inclusions have been incorporated as discrete inclusions by quenching (Murphy el al. 2000). Roche's Bluff Roche's Bluff is a low hill on the SE coast of the island between Soufriere Hills and South Soufriere Hills (Fig. 2). Sea cliffs extend as high as c.300m. A reddish basaltic andesite lava caps Roche's Bluff and extends down the cliff to sea level at the hill's eastern tip. This lava unconformably overlies a sequence of indurated palecoloured and calcareous volcaniclastic deposits, which are exposed both to the north and south of Roche's Bluff and are pervasively disrupted by faulting (called here Roche's Bluff Submarine Volcaniclastics, unit RBSV) (Figs 8 and 9). Typically faults occur at spacing of the order c. 10m. Faults are dominantly normal and dip steeply (c. 50 to 80°), generally to the SE. Beds dip broadly to the SE at angles generally steeper than the angle of repose (up to c. 40°). A few minor faults extend into the overlying lava through the unconformity. Calcareous components consist of marine macrofossils and fine-grained calcareous material comprising fossil fragments, microfossils (foraminifera and algae) and micrite matrix. MacGregor (1938) identified corals, spines of echinoderms, serpulae, lamellibranches and ostracods. This calcareous material forms thin limestone beds up to 2m in thickness. Other beds comprise dominantly volcanic material, with minor marine fossil fragments. Pumiceous deposits display a strongly bimodal grainsize distribution, consisting of rounded pumice blocks and fine ash particles, consistent with interpretation as subaqueous pumice-andash-flow deposits (Manville et al. 1998). Fine-grained deposits with cross-bedding and volcaniclastic deposits with well rounded clasts, in association with deposits containing marine fossils, suggest deposition in a beach environment. These observations indicate that unit RBSV formed in a shallow-marine depositional environment, where periodic influx of volcanic material mixed with calcareous sediment from the near-shore environment. The steeply dipping beds and strongly faulted character of unit RBSV indicate tectonic disruption since deposition. Evidence for a submarine depositional environment, and the inference that the

current sea level is only a few metres below the glacio-eustatic maximum sea level (Shackleton 1987). indicate that this unit has been uplifted since deposition. Rea's (1974) proposal that the intrusion of Roche's Bluff lava into overlying submarine deposits was responsible for their uplift is inconsistent with the observation that Roche's Bluff lava unconformably overlies the submarine deposits (Fig. 8). Instead. Roche's Bluff lava was deposited on top of an already partly uplifted submarine sequence. The foraminifera species Globorotalia truncatulinoides (d'Orbi ny g ) identified by Paul Pearson (pers. comm.) in unit RBSV, appeared in the Atlantic only around 1.8 Ma (Spencer-Cervato & Thierstein 1997). thus giving an age constraint for the unit of < 1.8 Ma - Nannofossils, which offer more precise age constraints, are absent, probably due to diagenetic micritization. Garibaldi Hill, St George's Hill and Richmond Hill Rea (1974) proposed that Garibaldi Hill and St George's Hill are parasitic centres, the former related to Centre Hills, and the latter as a late-stage vent of Soufriere Hills, which also acted as a source for the pyroclastic deposits of Richmond Hill. Several lines of evidence indicate that this interpretation is incorrect, and instead these volcaniclastic sequences were sourced from one of the major volcanic massifs and were subsequently technically uplifted, St George's Hill consists mainly of andesitic block-and-ash-flow deposits, pumice-and-ash-flow deposits and epiclastic deposits, Pumice-fall deposits are subordinate, in contrast to Rea's (1974) interpretation that the hill mainly comprises such deposits. Garibaldi Hill and Richmond Hill are composed of sequences of similar andesitic pyroclastic and epiclastic deposits. These volcaniclastic deposits are typical of extensive flank deposits derived from a lava dome such as at the currently active Soufriere Hills Volcano, rather than deposits that would form around a small vent. The lithofacies is similar to that of modern flank environments, where original bedding typically displays dips of less than 8 . Good exposures in sea cliffs on the west coast of Garibaldi Hill show that bedding is generally inclined (approximately to the NW) at angles greater than the angle of repose and up to c. 50 . implying tectonic tilting. These cliff exposures also indicate that the volcaniclastic deposits are disrupted by many, dominantly normal faults. Exposure on St George's Hill is

EVOLUTION OF MONTSERRAT

99

Fig. 6. Soufriere Hills deposits: (a) SH-III at Spanish Point, showing block-and-ash-flow deposits with lenses of pumice-and-ash-flow deposits; (b) alternating andesitic pumice-fall and basaltic andesite scoria-fall beds exposed up Irish Ghaut (SH-III).

poor, but some faults are exposed in roadcuts. In addition there is no geological evidence for a vent system centred on these hills. Petrology and geochemistry Montserrat lavas range continuously in composition from 48 to 64 wt% SiO2 (Fig. 7). The dominant composition in Montserrat is andesite, but products of South Soufriere Hills are more mafic. All deposits are low-K to medium-K (Fig. 7) according to the classification of Gill (1981). The andesites are typically porphyritic with phenocrysts (45-55 wt%) set in a microlite-rich groundmass. The andesite and dacite phenocryst phases are plagioclase, oxides, orthopyroxene, iclinopyroxene, ihornblende, iquartz, iolivine (rare). The basaltic andesites lack hornblende and quartz, but contain olivine and commonly clinopyroxene as the dominant pyroxene. The petrological and geochemical features of Silver Hills, Centre Hills and Soufriere Hills are broadly similar. The andesites and dacites commonly contain abundant mafic inclusions, thought to represent hot material injected into the magma chamber and

rapidly quenched in response to the temperature contrast (Murphy el al. 2000). Mafic inclusions contain abundant hornblende (Murphy et al. 2000), whereas South Soufriere Hills mafic lavas contain anhydrous phenocryst assemblages (Baker 1985). Soufriere Hills comprise mainly hornblende-hypersthene lavas. The rocks at the lowest levels of the Soufriere Hills stratigraphy seem dominated by two-pyroxene andesite, whereas the upper levels consist of products of hornblende-hypersthene andesite, as recognized by Rea (1974). All samples studied of SH-II and SHIII comprise hornblende-hypersthene lava. An olivine-bearing twopyroxene lava containing minor hornblende forms Gages Dome, whereas the four remaining domes (Chances, Galway's, Perches, Castle Peak, Fig. 2) are made up of hornblende-hypersthene andesite. Rea (1974) interpreted Perches Dome as an old part of Soufriere Hills, based on the assertion that it comprises two-pyroxene lava. However, our samples of Perches Dome are hornblendehypersthene andesites. Our geochronological data confirm that Gages is the oldest dome and that Perches Dome is quite young; these results are consistent with the observations of the relationship between stratigraphy and petrology.

100

C. L. HARFORD ET AL.

Fig. 7. Variation in SiO2 with (a) K2O and (b) vanadium in Montserrat samples. SvH. Silver Hills: CH. Centre Hills; SSH. South Soufriere Hills: SH. Soufriere Hills. 95-00 represents andesitic samples from the ongoing eruption, and Mafinc represents mafic inclusion samples from this eruption. Analyses by X-ray fluorescence spectrometry from Harford (2000). Baker (1984). Devine et al (1998). Devine (2000). Murphy et al. (2000. 1998). Murphy & Sparks (1999). G. Zellmer (pers. comm.). and major elements of some mafic inclusion samples analysed by J. Devine for the British Geological Survey. Recalculated to 100% and volatile-free, all Fe as Fe2O3. 1 and 2 mark Soufriere Hills samples from Irish Ghaut scoria-fall and pumice-fall beds respectively. Diagonal lines represent boundaries between low-K. medium-K and high-K lavas according to the classification of Gill (1981).

Sample selection A representative suite of samples was selected for 40 Ar/ 39 Ar geochronology following the petrographic criteria of Maniken & Dalrymple (1972), principally by choosing holocrystalline, relatively unaltered samples. Field characteristics used to identify primary pyroclastic-flow deposits, and hence minimize the risk of sampling reworked blocks, included monolithology and absence of rounded clasts. Analytical details are provided in the Appendix. 40

Arj39Ar results

The 40 Ar/ 39 Ar results presented here reveal that Montserrat has evolved as three volcanic centres, distinct in time: Silver Hills (c.2600 to c. 1200ka); Centre Hills (at least c. 950 to c.550ka); South Soufriere, Hills Soufriere Hills (at least c. 170ka to present) (Table 1, Figs 2-10, 1 1 , 1 2 and 13). The new data and sample locations are displayed in Figure 2. These results also show that xeno-

crysts are common in Montserrat lavas, and provide information about the content and distribution of trapped atmospheric argon within the samples. Siiver Hills, Two ages were obtained from massive lava dome rocks. A plagioclase separate from lava near the west coast (MVO144) gives an age of 2580 0 ka. This should be considered as a maximum age due to the presence of a xenocrystic component (Fig. 11), An age of 1160 6 ka was obtained for the groundmass of a massive dacitic lava (MVO755) located to the SW. The high MSWD (mean square weighted deviation) and pattern of decreasing apparent age with increasing temperature in the plateau diagram (Fig. 10), suggests irradiation-induced argon redistribution (presumably 39 Ar recoil, e.g. Pringle 1993). This phenomenon affects fine-grained phases such as clays, and probably indicates that this sample contains fine-grained alteration phases. This is not unexpected owing to the glassy nature of this sample. If all the argon is internally redistributed in the plateau steps, then the average plateau age represents the true sample age. This age is viewed with caution.

EVOLUTION OF MONTSERRAT

101

Fig. 8. Roche's Bluff from helicopter, showing Roche's Bluff lava unconformably overlying older faulted submarine volcaniclastic sequence (unit RBSV).

Centre Hills. Five reliable ages were obtained for the Centre Hills. Massive andesitic lava, which forms the hill on which Harris village lies (MVO148) and is thought to represent part of an old lava dome complex, gave an age of 954 12ka. The pattern of decreasing age with increasing temperature suggests that 39Ar recoil may be the source of the slight excess scatter. All of the 39 Ar released is used in the plateau, so if this minor recoil redistributed argon within the sample, this age would reflect the true eruption age. Any recoil of 39 Ar out of the sample would result in an age that was older than the true crystallization age. The age obtained may therefore be considered a maximum age. A similar andesitic lava from the quarry above Trant's village (MVO131) gave an age of 871 lOka. Blockand-ash-flow deposits at the base and top of the eastern sea cliffs gave ages of a (MVO831) and a (MVO809) respectively. The MSWD for the former age is only very marginally greater than the critical value. This is probably due to a slight underestimation of experimental errors, and can be neglected. In the NW part of Centre Hills, an age of 663 49 ka was obtained for a pumiceous deposit (interpreted as a pumice-and-ash-flow deposit) (MVO147) at the top of the stratigraphic section. This is considered a maximum eruption age because of the possibility of xenocrysts in the plagioclase separate analysed.

South Soufriere Hills-Soufriere Hills. Ages were obtained for five samples from South Soufriere Hills and eight samples from Soufriere Hills. A basaltic lava at Landing Bay (MVO139) gave an age of a for a whole-rock sample, and a for a groundmass separate. The latter age shows slight scatter above the experimental uncertainty. The plateau diagram indicates minor 39 Ar recoil, but by using 87.1% of the sample 39Ar any effect this may have on the age is minimized. The age data from the two experiments are not significantly different. The preferred age for the sample is the weighted mean age of 131 7 ka, due to the high MSWD for the groundmass age. A basaltic lava from near Shoe Rock (MVO791) gave an age of 129 14ka. The lava of basaltic andesite composition at Roche's Bluff (MVO136) shows a saddleshaped spectrum typical of lavas with excess argon (Lanphere & Dalrymple 1971; see discussion below). If the majority of steps are included the isochron 40 Ar/ 36 Ar intercept of 299.7 0.9 also indicates excess argon. When fewer steps are included, no significant

Fig. 9. Photo of calcareous volcaniclastic deposits of Roche's Bluff, showing steeply dipping beds and intense faulting. These deposits are interpreted as having been technically uplifted following deposition in a shallow marine environment.

Table 1. Summary of4(>Ar/J9Ar

incremental heating experiments on Montserrat samples Age spectrum analysis 6

Sample

MVO144 MVO755 MVO135 MVO148 MVO131 MVO831 MVO147 MVO809 MVO785 MVO819 MVO819

Whole-rock K 2 O (wt%)

Total K/Ca2

Total Ara3 Total fusion (molg 1 ) age (ka)

Plagioclase Groundmass Whole- rock Whole-rock Whole-rock Groundmass Plagioclase Groundmass Groundmass Groundmass Groundmass

0.52 0.83 0.87 0.85 0.69 0.76 0.77 0.41 1.02 0.88 0.88

0.010 0.408 0.527 0.162 0.168 0.408 0.009 0.287 0.463 0.321 0.337

2.59E-12 1.49E-11 8.51E-12 5.09E-12 5.30E-12 1.25E-11 6.88E-13 5.25E-12 1.77E-11 2.79E-12 4.07E-12

379275 1847175 379275 1847175 379275 1847175

Whole-rock Groundmass Groundmass

0.85 0.85 0.85

0.293 0.320 0.300

2.21E-12 3.03E-12 3.03E-12

Geological unit 1

Grid reference mEmN

Material

Silver Hills lava Silver Hills lava Roche's submarine bafd Harris lava Trant's lava Lower Centre Hills bafd Upper Centre Hills pfd Upper Centre Hills bafd Garibaldi Hill bafd SH-I SH-I

378025 378650 383975 382150 382050 381763 377600 381475 374990 383238 383238

1857050 1856400 1846150 1850450 1851975 1853525 1855050 1853750 1850025 1849275 1849275

MVO152 Gages Dome MVO152 Gages Dome MVO152 Gages Dome

4 5051 405 422 1009 5 980 32 863 6 836 7 1083 7 538 8 3 8 6

163

N

4

Increments 39 Ar used ( a C) (%) 5

7 of 11 5 of 16 7 of 14 10 of 12 12 of 12 9 of 14 5 of 12 14 of 14 6 of 15 5 of 10 5 of 9

500 1000 825 1025 550-825

550 450 450 900 450 675 725 625

1300 1300 875 1200 1300 950 975 825

65.3 64.6 62.1 93.0 100.0 73.7 76.4 100.0 66.7 91.5 66.7

2 5 5

5 of 14 600 925 2 of 10 775-825 5 of 9 675-950

53.6 53.7 64.2

3

8 of 16 725 1125 5 of 11 675 1150 14 of 17 450 1050

76.0 61.1 87.1

1300 1000 1300 925 850

97.4 79.8 70.7 55.5 63.0

4

5 of 10 700 1025 4 of 10 725 825

58.2 59.4

MVO830 MVO139 MVO139

Roche's Bluff breccia Landing Bay lava Landing Bay lava

383975 1845050 383950 1845000 383950 1845000

Groundmass Whole-rock Groundmass

0.61 0.60 0.60

0.118 0.084 0.110

6.06E-12 5.54E-12 7.02E-12

MVO136 MVO136 a MVO791 MVO1099 MVO1099

Roche's Bluff lava Roche's Bluff lava Shoe Rock lava SSH-F SSH-F

384250 384250 381600 379750 379750

Whole-rock Whole-rock Groundmass Groundmass Groundmass

0.79 0.79 0.58 0.62 0.62

0.180 0.180 0.078 0.137 0.057

3.52E-12 3.52E-12 2.22E-12 3.02E-12 2.I3E-12

18654 18654

Gal way's Dome Galway's Dome

h b

Groundmass Groundmass

1.07 1.07

0.530 0.410

6.54E-12 1.54E-11

MVO777 MVO777

SH-II SH-II

383675 1847250 383675 1847250

Groundmass Groundmass

0.93 0.93

0.288 0.329

1.55E-11 3.96E-11

48 58

18 23

3 of 8 3 of 9

700 825 775 875

63.2 60.4

MVO149 Chances Dome MVO127 S H - I I I MVO154 Perches Dome MVO154 Perches Dome MVO154 Perches Dome

379900 383550 382100 382100 382100

1846825 1849975 1846750 1846750 1846750

Whole-rock Groundmass Whole-rock Groundmass Groundmass

0.76 0.87 0.84 0.84 0.84

0.309 0.258 0.260 0.352 0.379

8.50E-I2 2.86E-12 3.34E-12 2.58E-12 2.86E-12

85 7 64 43 33

22 11 6 3 6

4 3 5 5 4

625 725 725 700 750

825 800 1000 950 975

54.2 72.5 65.3 65.9 68.3

MVO775 S H - I I I MVO104 1996 Dome MVO104 1996 Dome

383425 1847675

Groundmass Groundmass Plagioclase

0.89 0.73 0.73

0.554 0.162 0.019

5.54E-12 1 .44E- 11 2.30E-12

55

10

252

4 of 9 750 1025 4 of 10 675 900 7 of 10 675 1250

64.0 63.8 71.1

1845650 1845650 1843150 1844325 1844325

7 5 1 1

103 173

5 0 4

10 of 5 of 6 of 6 of 3 of

11 11 10 9 9

550 675 800 675 750

Age

MSWD

(ka) 2580

0.69 2().79e 4.37e 4.99e

60

1160 6 1021 0 954 2 871 0 826 2 663 49 550 3 282 8 5

173 3 174 e 223 7 6

149 3 151 e 7

138 9

131

153 99

9 e 7 5 4 9 10 e

142 12

121 4 103 5 e 112 84 5 64 6 e

1052

8

9

of of of of of

11 10 11 9 10

39 13 38 8 32 4 24 2 3 24 2e 24 1 21 2 426 95

1.40 2.01d 0.93 1.01 0.15 2.48d 1.03 0.53 0.23 3.23 1.06 1.84 0.64 1.28 2.89d 1.13 2.83e 1.04 0.35 1.00 0.45 16.67d 1.96 0.52 8.45d 1.23 0.33 6.53d 15.9d 0.13 1.61 0.54 0.20 0.18 0.93 0.25 3.26d

Inverse isochron analysis SUMS (N-2)

Ar/36Ar

40

Age

intercept

2.92e 28.20e 5.39e 6.18e 1.59 1.98 0.88 1.07 0.14 3.03e 1.37

294.7 284.0 297.4 296.4 295.0 299.5 290.9 293.4 292.6 296.1

3 5 4 0 2 6 3 5 5

0.28

297.7

0

0

2637 5

994

786 667 294 170

f

367

(ka) 167 3 3 5 1 1 9 67 1 6 7 7 78 8

1.23

293.9

2

0.72 1.72 3.58e

298.0 296.5 296.4

9 6 9

0.38 1.32 0.14

302.4 294.4 293.6 288.0

9 1.0 6 7

1.44 0.82

292.9 294.5

6 5

129 6

2.21 3.80

295.2

9 3

114 66

20.55e 0.28 1.44 0.30 0.28

272.0 296.8 294.0 297. 8 295. 8

9 7.4 1.4 9 1.7

0.66 0.15 5.88e

299.0 294.3 297.1

1 7 5

f

0.79

0.3

109

0 4 3

130 5 123 10 1 174 0 181 3

8

9 30

129 9 26 40 29 6 19 6 22 5 17 65 346

9 32 58

All ages calculated relative to 27.92 Ma for neutron flux monitor Taylor Creek Rhyolite Sanidine 85G003. 1 bafd, block-and-ash-llow deposit; pfd, pumice-and-ash-flow deposit. 2 K/Ca molar ratio. 3 Atmospheric argon. 4 Number of heating increments used in regression. 5 Percentage 39Ar by mass. 6 Best eruption ages of samples highlighted in bold. a Experiment with excess argon, alternative regression. b Sample collected by W. J. Rea, Oxford University Mineral Collection. c Where multiple analyses completed for a sample the best age is the weighted average of the groundmass ages, also highlighted in bold. d MSWD (mean square weighted deviation) or e SUMS/(N-2) exceeds critical 95% confidence level of the F-variate statistic. This excess scatter is taken into consideration in the error estimate of the final age calculation. f Non-atmospheric isochron intercept at 95% confidence level.

Fig. 10. Age plateau diagrams for 26 experiments. errors shown. Asterisk indicates that MSWD exceeds critical 95% confidence level of the F-variate statistic. bafd, block-and-ash flow deposits; pafd, pumice-and-ash flow deposits.

Fig. 10. (continued)

EVOLUTION OF MONTSERRAT

excess argon is detected (Table 1). The isochron age of 130 5ka, obtained using the majority of steps, is preferred (see below). An age of 129 17ka was obtained for a basaltic andesite block-and-ashflow deposit (MVO830) from south of Roche's Bluff, interpreted as related to lava flow collapse. The similarities of ages and lithologies of the Roche's Bluff lava flow and block-and-ash-flow deposit to the lavas and pyroclastics of eastern South Soufriere Hills suggest that these young beds of Roche's Bluff are deposits of South Soufriere Hills. A basaltic block from the lower part of the exposed section of the scoria-fall sequence of western South Soufriere Hills (MVO1099) gave ages of 99 10 ka and 153 9 ka for two groundmass separates. These values do not agree at the 95% confidence level (c.l.), and give a weighted mean age of a with a high MSWD. A block-and-ash-flow deposit (MVO135) in the formerly submarine sequence (unit RBSV) underlying the Roche's Bluff lava gave an age of 1021 0 ka. The plateau MSWD exceeds the critical value, indicating that caution should be used in interpreting this age. This age is consistent with our observations of significant tectonic disruption and uplift of unit RBSV as compared to significantly younger overlying Roche's Bluff lava (130 5 ka) and adjacent Roche's Bluff block-and-ash-flow deposit 17 ka) of South Soufriere Hills-Soufriere Hills. It is inferred that unit RBSV formed at an earlier stage of the island's history, as part of an extensive submarine fan lying to the south of Centre Hills.

105

Samples from both the lava domes and the volcaniclastic stratigraphy of Soufriere Hills were dated. Reliable ages were obtained for andesitic Gages Dome and Perches Dome, and dacitic Galway's Dome. Chances Dome samples gave poor results due to alteration. Dating was not attempted on Castle Peak Dome, as its age is already constrained by radiocarbon dating of associated block-and-ash-flow deposits at 770 to 200 years BP (Roobol & Smith 1998), too young for effective 40 Ar/ 39 Ar geochronology. An age of 223 7 ka was obtained for a whole-rock core of Gages Dome (MVO152). The high temperature steps of this sample show evidence for xenocrysts and in fact define a good plateau with an age of 434 50 ka (Fig. 11). This can therefore be considered a minimum age for the xenocrystic component of the Gages Dome sample. For the medium-temperature plateau for this sample, the MSWD exceeds the critical value. This excess scatter can be attributed to a xenocrystic component affecting slightly the midtemperature plateau, and this age is considered a maximum eruption age. Groundmass separates of this sample gave ages of 158 6 and , which agree at the 95% c.l. The weighted mean of the groundmass ages of 151 4 ka is considered to be the best estimate of eruption age due to the evidence of xenocrysts in the whole-rock sample. Groundmass samples of Galway's Dome (18654 of W. J. Rea, Oxford University Mineral Collection) gave ages of 121 4 ka and 103 5 ka. These ages do not quite agree at the 95% c.L, and give a weighted mean of a with a high

106

C. L. HARFORD ET AL.

Fig. 11. Age plateau diagrams and inverse isochrons showing evidence of xenocrysts. On the inverse isochron plots, squares represent medium temperature age plateau increments, triangles represent high temperature increments, and crosses represent low temperature increments. errors shown. For MVO152. the high temperature increments form a secondary age plateau with MSWD of 0.23 and 40 Ar 36 Ar intercept of 295.8 0.9. giving a minimum age for the xenocrysts of . For MVO144. two high-temperature increments can be extrapolated to give ages of 11 and 32 Ma respectively, assuming an atmospheric 40 Ar/36 Ar intercept. Assuming these increments have geological significance. 32 Ma gives a minimum age for a xenocrystic component.

MSWD. An age of 32 4 ka was obtained for a whole-rock core of Perches Dome (MVO154), whereas groundmass separates of this sample gave a and . The weighted mean for the groundmass samples of 24 2 ka is not quite distinguishable from the whole-rock age at the 95% c.l. The weighted mean groundmass age is, however, regarded as the most reliable eruption age estimate, in view of the evidence for xenocrysts in other samples. Chances Dome sample MVO149 gave poor results. The age spectrum is incoherent, with considerable scatter above that which can be accounted for by analytical errors (Fig. 12d). The plateau diagram shows a pattern of decreasing age with increasing temperature, typical of irradiation-induced argon redistribution, suggesting a significant amount of fine-grained alteration phases. If all the argon is internally redistributed in the plateau steps, then the average plateau age represents the true sample age. An age of 200 ka is taken as a conservative upper limit for the sample age, but the age could be considerably younger. Chances Dome is a hornblende-hypersthene andesite and our geochronological data of other units indicate that this magma type was probably erupted after c. 110 ka. Lava blocks were dated from block-and-ash-flow deposits representing the basal (SH-I), mid-(SH-II) and upper (SH-III) levels of the Soufriere Hills pyroclastic stratigraphy (Fig. 5). Ages of

a and a were obtained from the block-andash-flow deposit (MVO819) at the lowest level of the stratigraphy exposed (SH-I). The two ages give a weighted mean of . From the mid-level (SH-II) an age of a was obtained (MVO777). Ages from the upper level of the stratigraphy (SH-III) of 38 8 ka (MVO127) and 24 1 ka (MVO775) are consistent with published radiocarbon ages of c. 32 to c. 17 ka. and also suggest they may be associated with the formation of Perches Dome.

Garibaldi Hill. A dacitic sample from a block-and-ash-flow deposit on the western sea cliffs of Garibaldi Hill (MVO785) gave an age of 282 8ka. This age suggests that the volcamclastic deposits of Garibaldi Hills may represent the initial products of the South Soufriere Hills Soufriere Hills centre.

Zero age experiments. To test for extraneous argon in Montserrat samples, step-heating experiments were carried out on both a groundmass and a plagioclase separate from lava erupted in November 1996 (i.e. zero age) (MVO104). The groundmass sample gave an apparent age of 21 22 ka. not significantly different from zero, and indicates that no extraneous argon is present in the

EVOLUTION OF MONTSERRAT

107

Fig. 12. Age plateau diagrams with graphs showing atmospheric argon partitioning. Atmospheric 40Ar ( 40 Ar at ) per effective wt% K released, is shown for each step: (a) low total 40 Ar at and (b) high total 40 Ar at both illustrate partitioning into low and high temperature steps, resulting in high precision for mid-temperature plateau steps; (c) poor partitioning resulting in poor precision age; (d) good 40 Ar at partitioning but altered sample (recoil) hence poor age constraint; (e) excess argon ( 40 Ar ex ) is also partitioned into low and high temperature steps, resulting in saddle-shaped age spectrum.

108

C. L. HARFORD ET AL.

genic 4() Ar compared to a relatively large amount of atmospheric Ar contamination. We have investigated the amount and distribution of atmospheric 40 Ar in different samples (Table 1, Fig. 12). As first noted by Hall & York (1978). the process of step-heating is effective in at least partially separating the radiogenic 40 Ar from the atmospheric 40 Ar in a sample. The former is released at midtemperature plateau steps, whilst the latter partitions very strongly into the low and high temperature increments, resulting in a higher precision for the plateau age than for the bulk fusion age (Table 12) Fig. 12). In addition, we note that some samples are better than others in terms of total atmospheric 40 Ar and in terms of distribution of release (Table L Fig. 12). Generally, samples with high total atmospheric 40 Ar contents give poor results, e.g. MVO104 groundmass. However, samples with high total atmospheric 40 Ar contents can give good age determinations if the atmospheric 40 Ar is very efficiently partitioned out of the plateau steps. For example MVO777 had one of the highest total atmospheric 40 Ar contents but yielded ages of good precision (Fig. 12b). Conversely, samples with relatively low total atmospheric 4() Ar contents generally give good results, but some give results with poor precision if the atmospheric 40 Ar is distributed more evenly in all heating steps, e.g. MVO809 (Fig. 12c). In addition, samples may give poor ages despite low atmospheric 40 Ar in the plateau steps for other reasons, e.g. MVO149 due to alteration (Fig. 12d). The data for the sample containing excess argon (MVO136) show strong partitioning of both the non-atmospheric and the atmospheric trapped components into the low- and high-temperature phases, resulting in lower precision for these steps (Fig. 12e). In addition, the partitioning of the excess argon results in higher apparent ages for the low- and high-temperature steps, and thus a saddle-shaped age plateau. The age of the base of the saddle approaches, but may not reach, the true age of the sample. We thus prefer the isochron age for this sample (see above). The distribution of release of the atmospheric 40 Ar operates a much stronger control on the precision of the age obtained than the total amount of atmospheric 40 Ar. Therefore it is not possible to select suitable samples on the basis of low total atmospheric 40 Ar prior to expensive and time-consuming irradiation. In addition, samples that gave poor age data were usually associated with other sample problems, e.g. alteration, recoil. 40

Fig. 13. Stratigraphic column outlining evolution of Montserrat based on new 40 Ar/ 39 Ar data. Black filled symbols represent new 40 Ar/ 39 Ar ages; grey filled symbols represent key radiocarbon data (oldest and youngest SH-III dates, and SH-G date; Roobol & Smith 1998); open symbols represent existing conventional K-Ar data from Briden et al. (1979) and LeGall et a!. (1983) on left. Note log-scale for age.

groundmass. However, the apparent age of 426 95 ka for the plagioclase experiment indicates that this sample contains extraneous argon. This extraneous argon could reflect the presence of either old xenocrysts incorporated from the country rock (inherited argon), or excess argon trapped in young crystals as they grew. The atmospheric value for the isochron 40 Ar/ 36 Ar intercept indicates that excess argon is not present. Therefore xenocrysts are significant in the plagioclase phenocryst population. The presence of xenocrysts indicates that the sample does not have a single crystallization age. Excess scatter in the age plateau, as indicated by the elevated MSWD, is thus to be expected. Evidence of xenocrysts indicates that groundmass separates are more suitable than whole-rock samples or mineral separates for 40 Ar/ 39 Ar geochronology of volcanic rocks. This is particularly important for porphyritic andesitic magma, such as that of Soufriere Hills, which have had long and complex histories. Inclusion of a xenocrystic plagioclase component in dated wholerock samples from Montserrat would have a relatively small effect on the date obtained because the potassium content of the plagioclase is about an order of magnitude lower than that of the groundmass. For example, we have obtained ages for whole-rock samples of less than c. 75 000 years older than the age obtained for a groundmass separate of the same sample (MVO 152, MVO154). In addition, if the xenocrystic component is of constant age, the effect on the date obtained will be reduced for older samples. Atmospheric 40Ar. A key difficulty in dating very young rocks, and in particular low-K samples, is measuring a small amount of radio-

Discussion The precision of the new 40 Ar 39 Ar data is much higher than that of the existing conventional K-Ar data (Fig. 13). In addition, all but two of the 40 Ar/ 39 Ar ages are significantly younger than K-Ar ages for the same units, highlighting the inaccuracy of the existing conventional K-Ar ages. All our ages are consistent with the known Stratigraphic order, giving additional confidence in our results. Previous workers (MacGregor 1938; Rea 1974) have suggested that volcanism on Montserrat oscillated from north to south and that it involved reactivation of old centres (Fig. 1). In contrast, our geomorphological observations of three discrete massifs with erosional maturity decreasing from north to south, suggest a southward migration of volcanism through time (Fig. 1). The new 40 Ar 39 Ar ages support this interpretation because the ages for the three massifs are non-overlapping and sequential (Fig. 13). The focus of volcanism on three discrete areas in turn suggests that a period of dormancy may have separated activity at the centres. The maximum duration of such a repose period between activity at Centre Hills and South Soufriere Hills-Soufriere Hills can be constrained using the youngest and oldest dated units of the older and younger centre respectively. Assuming the Garibaldi Hill age represents an early part of South Soufriere Hills-Soufriere Hills, the maximum duration of repose is 268 24 ka. The geochronology is therefore consistent with the suggestion that there was a repose period between the centres. Further sampling might, however, close the age gap estimated between the centres.

EVOLUTION OF MONTSERRAT

Silver Hills The two 40 Ar/ 39 Ar ages of 2580 60 ka and 1160 46 ka imply that Montserrat's subaerial history is significantly shorter than estimates based on K-Ar ages. The oldest reported K-Ar age for the island is 4.41 3 Ma (Briden et al. 1979).

Centre Hills The range of ages (954 12 to ) obtained for the Centre Hills implies a minimum duration of volcanism at this centre of 392 26 ka. Ages obtained from the top and bottom of the east coast cliffs of Centre Hills, are a and 826 12ka respectively, implying that the interval represented by this c. 150 m thick stratigraphic section is . Although correlation between the stratigraphic sections of the eastern and western flanks proved impossible, an 40 Ar/ 39 Ar age of 663 49 ka for a deposit from the top of the west coast section suggests that the deposits of the two flank areas formed concurrently. The massive lavas of Harris and Trant's (954 12ka and 871 12ka respectively) are significantly older than the dated flank pyroclastic deposits. This is consistent with the observation that these lavas are overlain in places by volcaniclastic deposits.

South Soufriere Hills—Soufriere Hills 40

a Ar/ 39 Ar ages for the Soufriere Hills area range from (SH-I) to 24 1 ka (SH-III) and a (Perches Dome), whilst lavas of South Soufriere Hills have ages in the range of c. 130ka. This indicates that South Soufriere Hills and Soufriere Hills were active either contemporaneously or alternately, and supports our treatment of these two areas together as one volcanic centre. However, geochemical differences between lavas of South Soufriere Hills and those of Soufriere Hills (Fig. 7b) (G. Zellmer, pers. comm.) indicate that the magma plumbing systems of these two areas may be discrete. The petrological and geochemical characteristics of the magma change with age in the South Soufriere Hills-Soufriere Hills complex. Samples with ages of c. 150 ka seem to be dominated by twopyroxene andesite, whereas all samples with ages of c. 1 lOka are hornblende-hypersthene andesites. This change coincides with ages for volcanism at South Soufriere Hills of c. 130 ka. The transition to hornblende-hypersthene andesites therefore appears to have happened after an interlude of mafic volcanism at South Soufriere Hills. There is also minor mafic volcanism recorded within the pyroclastic sequence in SH-II (Roobol & Smith 1998), and observed by us in SH-III. There is no significant difference between the ages of the lavas dated within South Soufriere Hills. The scoria-fall deposits on the western flanks and the lava flow deposits in the southern and eastern parts of the South Soufriere Hills massif may therefore have formed contemporaneously. Outcrop of the scoria-fall deposits may be controlled largely by wind direction: tephra in low-level plumes on Montserrat is transported dominantly to the west. Subordinate tephra-fall deposits on the southern and eastern flanks may represent their lateral equivalents. The 40 Ar/ 39 Ar age of 24 1 ka for a block-and-ash-flow deposit from the east coast section of Soufriere Hills implies that this deposit forms part of unit SH-III, from which comparable radiocarbon ages of c. 17 to c, 31 ka have been obtained (Baker 1985; Rea 1974; Roobol & Smith 1998; Wadge & Isaacs 1988). This facilitates a slight modification of the existing stratigraphic profile, such that MVO775 lies in SH-III (Fig. 5). The age of 38 8 ka obtained from the lower part of unit SH-III is also consistent with the radiocarbon ages (Fig. 5). The age of a for Perches Dome indicates that it is one of the younger domes of Soufriere Hills, in contrast to previous proposals that it is one of the older parts of the centre (Rea 1974), or indeed that it predated the three

109

major volcaniclastic units (Roobol & Smith 1998). Perches Dome may therefore have sourced block-and-ash-flow deposits in unit SH-III of the pyroclastic stratigraphy. Ages of the two other block-and-ash-flow deposits (75 10 ka and 174 3 ka) are, however, significantly different from the ages of Gages Dome (151 4 ka), Galway's Dome ) and Castle Peak Dome (c. 350 years BP), indicating that they were not sourced from these domes. Only one other dome is preserved, Chances Peak, indicating removal of at least one dome from the geological record. In fact, the dated block-and-ash-flow deposits represent only a small proportion of the eruptive periods of Soufriere Hills. Several distinct eruptive periods within unit SH-III are indicated by at least four well defined subunits, as well as at least four significantly different radiocarbon dates (Roobol & Smith 1998). Soil horizons within SH-III and the other major stratigraphic units also indicate time breaks between eruptive periods. The domes preserved may represent only a proportion of all the domes and eruptive periods of this system. Domes can be removed from the geological record by large-scale flank collapse. The sector-collapse event which produced English's Crater is thought to have occurred c. 4000 years BP. The horseshoe-shaped depressions at the summit of the South Soufriere Hills (Fig. 3b) are interpreted as sector-collapse scars. A sector-collapse event occurred during the ongoing eruption on 26 December 1997 resulting in loss of a total of 80-90 x 106 m3 of material from the flanks and actively growing dome (Sparks et al. 2002). Observation of large debris-avalanche deposits in the submarine bathymetry around Soufriere Hills provides additional evidence of sector-collapse events in the volcano's history (Harford 2000). We propose that large-scale collapse events play an important role in the evolution of the South Soufriere Hills-Soufriere Hills volcanic centre. Further support for incomplete dome preservation comes from the burial and erosion of the Castle Peak Dome by the 1995-1999 dome. The preservation potential of peripheral domes is much greater than domes erupted in the centre of the dome complex; the latter are likely to be removed during subsequent eruptions. For example, Gages Dome has probably survived for so long only because of its distance from the main focus of subsequent dome growth episodes. In addition, the ongoing eruption indicates that domes can self-destruct. Following the cessation of dome growth in March 1998, the dome gradually disintegrated through gravitational collapse events and small explosions (Norton et al. 2002) (Fig. 4a).

Uplifted areas 40

Ar/ 39 Ar ages for unit RBSV and for Garibaldi Hill are consistent with these areas being uplifted volcaniclastic sequences. An 40 Ar/ 39 Ar age of 282 8 ka from Garibaldi Hill suggests that this area may represent the early stages of Soufriere Hills. St George's Hill is also interpreted as an uplifted sequence of deposits, sourced from either Centre Hills or South Soufriere Hills-Soufriere Hills. The bedding of the three uplifted areas of Roche's Bluff, Garibaldi Hill and St George's Hill is broadly tilted away from Soufriere Hills. The observed uplift can be accounted for by footwall uplift of an approximately NW-trending normal fault (or faults) forming part of a major fault system that passes through the island (Harford 2000).

Geological synthesis Migration of volcanism. The new results show that volcanism on Montserrat has migrated southwards with time. This caused the long axis of the island to be orientated approximately north-south, and not parallel to the volcanic trench. Migration occurred at a time-averaged rate of c. 6 km Ma - 1 parallel to the island-arc trench and c. 2 km M a - 1 perpendicular to, and away from, the trench. These results are consistent with data from elsewhere in the arc (Wadge 1986). Migration rates parallel to the arc, both northwards and southwards, of 4-10 km M a - 1 , and perpendicular to the arc

C. L. HARFORD ET AL.

of c. 1 km Ma - 1 , have been estimated on St Kitts, Guadeloupe and St Vincent. Migration away from the trench has been interpreted to relate to the response of the downgoing slab to the growth of the accretionary prism (Wadge 1986). The longitudinal mobility is consistent with migration, relative to the overlying plate, of a single mantle melting anomaly. The pattern of migration could be interpreted as reflecting the presence of a single mantle source over a period of c. 2 Ma, and hence a mantle magma plumbing system stable for a remarkable length of time. Like the pulse-like volcanic activity at hot-spot volcanoes, such as that which has formed the Hawaiian island chain, the three discrete foci of activity may have formed as a result of the crustal incubation period required to allow eruption. If the magma source were to move relative to the plate, volcanism would remain focused on one area until the source region became too far removed from the volcanic centre to continue to feed it. A repose period whilst the magma source heated up the overlying plate would then follow before the next volcanic centre was formed. This hypothesis is consistent with geochronological data which suggest a dormant period of less than c. 250 ka between Centre Hills and South Soufriere Hills-Soufnere Hills. The between-centre spacing of c. 6 km corresponds to the proposed minimum depth of the magma chamber for the ongoing eruption of 5-6 km (Aspinall et al. 1998; Barclay et al. 1998). Time-averaged eruption rates and volcanic evolution. Time-averaged extrusion rates can be estimated using volumetric and geochronological data. Subaerial volumes represent only a proportion of total extruded volume because on an island such as Montserrat, much volcanic material is transported into the sea, through ash dispersion, pyroclastic flows, erosional reworking and large-scale collapse. On a longer timescale, fluvial and marine erosion remove large amounts of volcanic deposits. Therefore volcanic production rates can only be estimated from Montserrat's youngest centre, which is relatively unaffected by erosion. The preserved subaerial volume of the South Soufriere HillsSoufriere Hills volcanic centre is estimated at 12 km3 (using a digital elevation model of Montserrat) or 10km 3 dense rock equivalent (DRE; assuming average density of 2100 kg m-3 for the material forming the island prior to the ongoing eruption). The ongoing eruption of Soufriere Hills Volcano has produced a total of c. 400 x 10 6 m 3 DRE of magma by the time of writing (December 2000). A significant proportion of this material, >127 x 10 6 m 3 or >33%, has already been deposited in the sea. Sigurdsson et al. (1980) estimated from the volcanogenic sedimentation in the arc that 80% of the volcanic deposits produced over the past lOO ka have been deposited in the sea. The proportion deposited at sea depends on eruption style because large, energetic eruptions generally disperse material further. Considering that eruptions on Montserrat have been low-energy compared to those elsewhere in the arc (e.g. Dominica), a conservative figure of about 60% is estimated for the proportion of erupted volcanics that would have been transported to the sea on a timescale of 100-200 ka. Assuming an age of 170 ka for South Soufriere Hills—Soufriere Hills, the timeaveraged eruption rate for this centre is estimated at 0.15km 3 k a - 1 (0.005 m 3 s- 1 ) DRE. The time-averaged eruption rate during the ongoing eruption, assuming a total erupted volume of 400 x 10 6 m 3 DRE up to December 2000, has been c. 2 . 4 m 3 s - 1 DRE, around 500 times faster than that during the entire c. 170 ka history of the volcanic centre. In other words, a year of activity represents 500 years of dormancy and the five recent years of activity represent the average magma budget of the system over 2500 years. A further way to look at this comparison is that the estimated total volume produced by the centre over the last c. 170ka represents around 65 eruptions of the volume of the ongoing eruption to date. These volumetric comparisons are consistent with the proposal that the preserved domes do not represent all of the domes and eruptive periods of this centre. There may have been perhaps as many as an order of magnitude more eruptions than there are domes preserved.

Consideration of these volumetric estimations, together with geochronology data and stratigraphic observations, further implies that periods of volcanic extrusion are interspersed with significant periods of dormancy. Periods of elevated extrusion, marked by the three major stratigraphic units (SH-I. SH-II and SH-III), are separated by long repose periods represented by erosional unconformities in the stratigraphic record (Fig. 5). The erosional unconformity following SH-III represents a c. 17ka period of low activity lasting until 1995. based on a 14C age of 16 800 0 years BP (Roobol & Smith 1998) for the youngest dated deposit of SH-III. Within each of the major stratigraphic units, soil horizons mark shorter repose periods between eruptions. SH-III is the best-documented unit, and 14 C dates (Roobol & Smith 1998) indicate that the unit was deposited over a period of c. 15 ka from c. 32 ka to 17 ka. The 14 C dates and stratigraphy suggest at least four separate major eruptive periods within unit SH-III. The lack of significant erosion between deposits in unit SH-III suggests relatively short time gaps between their formation and hence probably significantly more than four eruptions during this 14 ka period. Periods of quiescence within this major eruptive period thus probably last less than c. 10 3 years. The time gaps between eruptions since the elevated activity represented by SH-III have been of the order of 102 to 104 years. No activity is recorded in the geological record from c. 17ka until c. 4ka. when the minor unit SH-G is interpreted as associated with formation of English's Crater. A subsequent break in deposition of c. 3500 years was followed by the small Castle Peak Dome eruption(s) probably in the seventeenth century (c. 350 years BP). A repose period of c. 350 years preceded the ongoing eruption, based on the lack of documented eruption since colonization of Montserrat in 1632. Comparison to other volcanoes. Montserrat's major volcanic centres have been long-lived (Centre Hills 400 ka: South Soufriere Hills-Soufriere Hills 170ka). in common with other documented volcanic systems. For example, in the Cascade arc, many systems are likely to have been active for >300ka. These include the Mount Baker, Mount Rainier. Hood and Mount Shasta systems, and the Mount Adams volcanic field (Hildreth & Lanphere 1994). The latter has been active since c. 940 ka. with the start of a major constructional phase at c. 520 ka. However, the time-averaged eruption rate of the South Soufriere Hills-Soufriere Hills system (0.15 km 3 k a - 1 ) appears lower than that of these systems. For example, the Mount Adams system has averaged at 0.25-0.4 km 3 k a - 1 over the last 940 ka (Hildreth & Lanphere 1994), and the TataraSan Pedro system, Chile (Singer et al. 1997), has averaged at about 0.2-0.3 km 3 k a - 1 over the last 200 ka. These systems therefore have time-averaged growth rates about twice that of the Soufriere HillsSouth Soufriere Hills system over the last 170ka. The reasons for this difference are likely to be complex; differences in magma production for whole arcs are not well understood (Wadge 1984). Factors influencing differences in time-averaged eruption rate may include a lower plate convergence rate in the Lesser Antilles than in the Cascades or the Andes, oblique convergence beneath Montserrat. differences in the crust underlying the volcanoes, variations in the ratio of intruded to extruded magma, and variations in differentiation mechanisms. The total volcanic production rate for the Lesser Antilles arc over the past lOO ka is estimated at 3 km 3 k a - 1 . based on estimation of volumes of both subaerial and submarine deposits (Wadge & Shepherd 1984). Therefore using our calculated time-averaged eruption rate for the South Soufriere Hills-Soufriere Hills, this volcanic centre accounts for c. 5% of the total volcanic production for the arc. The arc consists of only 12 active volcanoes, confirming that Montserrat's volcano is one of the less productive in the arc, commensurate with its size. Implications for ongoing and future activity. The ongoing eruption has been very similar in style to previous activity at Soufriere Hills, and in particular to that recorded by the upper major unit of the

EVOLUTION OF MONTSERRAT

pyroclastic section (SH-III). The only explosive eruption of Plinian intensity recorded in the geological record of the South Soufriere Hills-Soufriere Hills centre was an early (c. 174 ka), small event that produced a 1-m-thick pumice-fall deposit c. 3 km from the probable vent in the central Soufriere Hills. The vents of the Soufriere Hills have been confined to a WNW-trending zone. Hence, on the basis of the past 170 ka of activity, future activity at Soufriere Hills is likely to pose similar hazards to the ongoing activity. Both the stratigraphic records and geochronological data indicate that over its lifetime of at least 170 ka the Soufriere Hills has alternated between periods of elevated activity and dormancy with characteristic timescales of the order of 104 years. The stratigraphic packages documented by Roobol & Smith (1998) indicate at least two major pulses (SH-II and SH-III) since c. 11O ka. The youngest (SH-III) is the best documented in terms of age, and indicates a prolonged period of several dome eruptions from 31 ka to 17 ka. There was then a long period of dormancy until 4ka. The formation of English's Crater at 4 ka and eruption of the small volume Castle Peak Dome at 350 years BP (seventeenth century) preluded the ongoing eruption, which is similar in magnitude and style to the past periods of elevated activity. If the past is taken as a guide to the future then the ongoing eruption can be anticipated to mark the entry of the Soufriere Hills into a period of elevated eruptive activity that may last of the order off. 104 years. The stratigraphic record shows that this activity will not be continuous, but will occur as several discrete dome-forming eruptions interspersed with periods of quiescence lasting up to around c. 103 years.

Conclusions (1)

(2)

(3) (4)

(5) (6)

(7) (8) (9) (10)

New 40 Ar/ 39 Ar ages show that Montserrat has evolved as three distinct volcanic centres: Silver Hills (c. 2600 to c. 1200ka); Centre Hills (at least c.950 to c.550 ka); South Soufriere Hills-Soufriere Hills (at least c. 170 ka to present). Consistent southward migration of volcanism indicates migration of a single mantle magma source relative to the plate at c. 6 km Ma - 1 , with crustal incubation resulting in repose periods between activity at consecutive volcanic centres. Old xenocrysts are common in Montserrat's porphyritic andesites, indicating that reliable ages are best obtained by dating the groundmass. An interlude of mafic volcanism at South Soufriere Hills at c. 130 ka is probably represented by a major unconformity between two of the three major stratigraphic units of the Soufriere Hills. This seems to coincide with a change in petrology from dominantly two-pyroxene andesite to hornblende-hypersthene andesite. There is evidence of tectonic uplift of both early products of the South Soufriere Hills Soufriere Hills centre and older submarine volcanic rocks. Stratigraphy and age data indicate that only a proportion of the dome-forming eruptions are recorded as domes. Some older products are removed from the subaerial edifice by sector-collapse events, e.g. that which produced English's Crater at 4 ka. The time-averaged eruption rate of the South Soufriere Hills-Soufriere Hills centre is estimated at O.OO5 m3 s-1 (c.0.15km 3 ka- 1 )DRE. The ongoing eruption is very similar in style to previous activity at Soufriere Hills and future activity is likely to pose similar hazards. Soufriere Hills have been characterized by alternations of periods of enhanced activity and periods of dormancy, both lasting of the order of 104 years. Periods of elevated activity are characterized by several major dome-forming eruptions separated by quiescent interludes lasting up to around 103 years. Each of the three major units of the volcaniclastic stratigraphy of Soufriere Hills represents such a period of activity.

(11)

111

The ongoing eruption may mark the onset of a fourth period of enhanced volcanic activity at Soufriere Hills.

We thank the MVO for logistical support, MVO staff for field assistance, and pilots of St Lucia Helicopters and Bajan Helicopters for skilfully facilitating access to the field. We gratefully acknowledge T. Brewer for providing XRF analyses, P. Pearson and J. Young for examining microfossil samples, A. Smith for sharing unpublished stratigraphic information, and G. Zellmer for use of unpublished geochemical data. C.L.H. was supported by NERC studentship, R.S.J.S. by NERC Professorship, and argon analyses by NERC project IP/516/0997.

Appendix: 40Ar/39Ar geochronology methods

Analytical techniques Whole-rock samples of 130 to 480 mg consisted of 3 to 10mm long, 5mm diameter cores. Plagioclase and groundmass separates consisted of 100 to 300 mg of 250 to 500 um sieve fraction, and were prepared by magnetic and heavy liquid separation. Phenocryst phases were removed from the groundmass to better than 99% purity to avoid contamination with xenocrystic argon. Plagioclase and groundmass separates were loaded in 99.99% copper foil packets. Neutron flux monitor standards were loaded in aluminium foil packets, and loaded in 6 mm i.d. (internal diameter) quartz vials, at intervals of 14 to 31 mm, intercalated with the samples. The monitor mineral used was the 27.92 Ma USGS sanidine 85G003 from the Taylor Creek rhyolite (Dalrymple & Duffield 1988). The quartz vials were irradiated at 1 MW for between 30 minutes and three hours, depending on anticipated sample ages, in the Cd-shielded CLICIT facility at the Oregon State University Triga reactor. Samples were analysed at the Scottish Universities Research and Reactor Centre (SURRC) following standard procedures (Singer & Pringle 1996). Following a 12-24 hour bakeout at c. 200°C incremental step-heating experiments comprising nine to 17 steps were carried out in a double vacuum resistance furnace, attached to a small-volume gas-cleanup system with SAES Zr-Al C50 getters at 400°C and a zeolite finger. Temperatures reported for incremental heating steps are as calibrated by optical pyrometry. After 15 minutes heating, and 10 minutes further cleanup with an additional Zr-Al C50 getter usually operated at room temperature, isotopic analysis of the purified gas was carried out on an ultrasensitive rare-gas mass spectrometer (Mass Analyser Products 215), with a modified Nier source and variable slit. The instrument sensitivity varied between different experimental run periods from 2.7 to 8.5 x 10 - 1 4 mole 4 0 A r V - 1 , calibrated using expected total 39 Ar yields for samples of known K content. Each analysis was corrected for total system blank and mass discrimination. The blank was measured at a range of temperatures prior to analysis of each sample, and comprised two components. Firstly the cold blank consisted of a non-atmospheric contribution from the mass spectrometer and an atmospheric contribution from the line and typically comprised around 1 x 10 - 1 5 moles 40 Ar, 1 x 10 - 1 8 moles 39 Ar, 2 x 10 - 1 8 moles 37Ar and 10 x 10 - 1 8 moles 36Ar. We ensured that no residual sample remained in the crucible by carefully monitoring the 39 Ar and 37Ar at 1400°C between samples. The second component of the blank was temperature-dependent and increased exponentially with temperature in atmospheric ratios, with typically an additional 2 x 10 -15 moles 40 Ar at 1300°C. The spectrometer mass discrimination has particular significance for young samples, which contain large amounts of atmospheric argon compared to small amounts of radiogenic argon. The mass discrimination was therefore carefully monitored by regularly analysing aliquots of air ( 40 Ar/ 36 Ar = 295.5) from an on-line gas bulb and pipette system. The long-term running average is stable, with possible changes after major maintenance operations. The mass discrimination varied from 1.0047 to 1.0070 (per atomic mass unit), and is known to better than 0.1% for any given period. Based on previous work (Wijbrans et al. 1995), the reactor corrections for interfering neutron-induced reactions of 40K and 40Ca,

C. L. HARFORD ET AL.

are asfollows:[ 40 Ar/ 39 Ar] K = 0.00086, [ 36 Ar/ 37 Ar] Ca - 0.000264 and [ 39 Ar/ 37 Ar] Ca = 0.000673. The decay constants of Steiger & Jager (1977) were used in age calculations. J values were determined from the mean of five laser fusion analyses from each monitor packet. Comparison of the J curve to individual monitor estimates suggests a conservative error in J for samples is 0.3 or 0.5% depending on the irradiation. This error was propagated into the final plateau and isochron ages for each sample. We note, however, that for young samples the error in J makes only a very minor contribution to the total error in age. An uncertainty of c. 1 % in the absolute ages calculated from our data is introduced by the uncertainty in the age of the flux monitor. However, for studies such as ours on young volcanoes, this is of negligible importance for two reasons. Firstly, the relative ages of activity are of greater significance than absolute ages. Secondly, for ages young enough to compare to radiocarbon ages, this 1 % uncertainty is negligible compared to that introduced by other sources of uncertainty. Sample ages were calculated from both age plateau analysis and isochron analysis. Plateau ages were calculated as weighted means, where each age is weighted by the inverse of its variance (Taylor 1982), and involve a single-variable mean standard weighted deviation (MSWD) calculation to test for excess scatter. Isochron analysis was carried out using the cubic least-squares regression with correlated errors (York 1969). No significant difference between isochron and inverse isochron ages is found, as predicted by Dalrymple et al. (1988). All errors are reported as one standard deviation of the analytical precision, and all significance tests are done at the 95% confidence level. Errors on individual analyses include estimates of the standard deviation of analytical precision of the peak signals, the blank, the spectrometer mass discrimination, the reactor corrections and the half-life of 37 Ar. When determining the weighted mean age for splits of the same sample, an MSWD calculation is also carried out and taken into account in the final error.

Interpretation of results A key advantage of the 40 Ar/ 39 Ar step-heating technique lies in its ability to distinguish reliable ages from unreliable ones through a series of internal consistency tests for each experiment. Criteria used are adapted after Pringle (1993) and Singer & Pringle (1996). Each criterion uses a rigorous statistical test to assess whether an incremental heating experiment gave reliable geochronological information. An age is accepted as an accurate estimate of the eruption age if the following criteria are satisfied, (a) A well defined age spectrum plateau, as defined by the one-dimensional F-variate statistic, MSWD, being below the critical value (discussed below), exists for at least three contiguous steps which comprise >50% of the 39 Ar released, (b) A well defined isochron, as defined by the two-dimensional F-variate statistic for the York2 fit, SUMS/(N-2), being below the critical value, exists for the plateau points, (c) The plateau and isochron ages are concordant at the 95% confidence level, (d) The 40 Ar/ 36 Ar intercept on the isochron diagram is not significantly different from the atmospheric value of 295.5 at the 95% confidence level. For the F-variate test in (a) and (b), we use standard statistical practice, following Pringle (1993), using different cut-offs depending on the number of individual increments used (cf. Pringle 1993, table 2). The critical value test in (c) and (d) states that no difference in values can be detected if the difference in the values is less than 1.960 ( ) 1 / 2 , where and are the standard deviations of the two values (Dalrymple & Lanphere 1969). If such critical values are exceeded, significant scatter above analytical error exists. Sources of such excess scatter may be geological (e.g. alteration) or experimental (e.g. irradiation-induced 39 Ar recoil). Under such conditions useful age information may still be gleaned, provided such ages are treated with caution. The excess scatter is taken into consideration in the error estimate of the final age calculation. One geological source of excess scatter in the plateau age calculation is a

trapped non-atmospheric argon component (excess argon). This is indicated by an 40 Ar 36 Ar isochron intercept significantly above the atmospheric value of 295.5 and may result in an isochron age that is significantly different from the plateau age. Under such circumstances, the isochron age may be accepted as reliable, provided the SUMS (N-2) statistic does not exceed the critical value. Plateau ages are generally preferred over isochron ages. In our experience plateau analysis offers a more realistic means of error analysis for young volcanic samples than isochron analysis. The latter can indicate a very poor age precision (poor constraint on isochron slope) simply because all the steps are clustered at similar values of 39 Ar 40 Ar. Conversely, the precision may appear to improve when poor, high-atmospheric steps are included, simply because the points become less clustered. One of the advantages of the isochron analysis in the past was that it incorporated a measure of scatter about the mean that can be compared to the amount of scatter expected from analytical errors alone (SUMS (N-2)). However, now that an equivalent F-ratio statistical test has been incorporated into calculation of the error of the weighted mean plateau age (MSWD), this advantage has been eliminated. We acknowledge that using the age-plateau analysis may still underestimate errors, as this analysis essentially introduces a point with zero error of air composition, but this is not significant compared to other sources of error for young samples. The weighted average of the isochron intercepts for our samples of 295.0 . which is not significantly different from the atmospheric value of 295.5. supports our assumption that the non-radiogenic component for young subaerial volcanics almost always is air. In cases where isochron analysis indicates a non-atmospheric trapped component, as discussed above, the isochron analysis is used.

References ASPINALL. W. P.. MILLER. A. D.. Lynch, L. L.. LATCHMAN, J. L.. STEWART, R. C.. WHITE. R. A. & POWER. J. A. 1998. Soufriere Hills eruption. Montserrat 1995-1997: Volcanic earthquake locations and fault plane solutions. Geophysical Research Letters. 25. 3397-3400. BACON. C. R. & LANPHERE. M. A. 1990. The Geological Setting of Crater Lake. Oregon. In:. DRAKE. E. T.. LARSON. G. L.. DYMOND. J. & COLLIER. R. (eds) Crater Lake - An ecosystem Study. Pacific Division of the American Association for the Advancement of Science San Francisco. California 19-27. BAKER, P. E. 1984. Geochemical evolution of St Kitts and Montserrat. Lesser Antilles. Journal of the Geological Society of London. 141. 401-411. BAKER. P. E, 1985. Volcanic hazards on St Kitts and Montserrat. West Indies. Journal of the Geological Society, 142. 279-295. BARCLAY. J.. RUTHERFORD. M. J.. CARROLL. M. R.. MURPHY. M. D., DEVINE, J. D.. GARDNER. J. & SPARKS. R. S. J. 1998. Experimental phase equilibria constraints on pre-eruptive storage conditions of the Soufriere Hills magma. Geophysical Research Letters. 25. 3437-3440. BRIDEN. J. C.. REX. D. C.. FALLER. A. M. & TOMBLIN. J. F. 1979. K-Ar geochronology and palaeomagnetism of volcanic rocks in the Lesser Antilles island arc. Philosophical Transactions of the Royal Society of London. 291. 485-528. CARLUT, J., QLIDELLEUR. X.. COURTILLOT. V. & BOUDON. G. 2000. Paleomagnetic directions and K Ar dating of 0 to 1 Ma lava flows from La Guadeloupe Island (French West Indies): Implications for timeaveraged field models. Journal of Geophysical Research-Solid Earth. 105, 835-849. COLE, P. D.. CALDER. E. S.. SPARKS. R. S. J. ET AL. 2002. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano. Montserrat. In: DRITTT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21. 231-262. DALRYMPLE. G. B. & DUFETELD. W. A. 1988. High-precision Ar-40 Ar-39 dating of Oligocene rhyolites from the Mogollon-Datil volcanic field using a continuous laser system. Geophysical Research Letters. 15. 463-466. DALRYMPLE. G. B. & LANPHERE. M. A. 1969. Potassium-Argon Dating. W.H. Freeman. San Francisco.

EVOLUTION OF MONTSERRAT DALRYMPLE, G. B., LANPHERE, M. A. & PRINGLE, M. S. 1988. Correlation diagrams in Ar-40/Ar-39 dating - is there a correct choice. Geophysical Research Letters, 15, 589-591. DAVIS, W. M. 1926. The Lesser Antilles. Map of Hispanic America Publication 2. American Geographical Society. DEVINE, J. D. 2000. Petrological investigation of Soufriere Hills Volcano, Montserrat, eruptive products: April 1999 to March 2000. Montserrat Volcano Observatory, Open File Report. DEVINE, J. D., MURPHY, M. D., RUTHERFORD, M. J. ETAL. 1998. Petrologic evidence for pre-eruptive pressure-temperature conditions, and recent reheating, of andesitic magma erupting at the Soufriere Hills Volcano, Montserrat, WI. Geophysical Research Letters, 25, 3669-3672. DRUITT, T. H., EDWARDS, L., MELLORS, R. M. ET AL. 1999. Santorini Volcano. Geological Society, London, Memoir 19. DRUITT, T. H., YOUNG, S. R., BAPTIE, B. ET AL. 2002. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306. GILL, J. B. 1981. Orogenic Andesites and Plate Tectonics. Springer-Verlag, Berlin. HALL, C. M. & YORK, D. 1978. K-Ar and 40-Ar/39-Ar age of the Laschamp geomagnetic polarity reversal. Nature, 274, 462-464. HARFORD, C. L. 2000. The Volcanic Evolution of Montserrat. PhD thesis, University of Bristol. HILDRETH, W. & LANPHERE, M. A. 1994. Potassium-argon geochronology of a basalt-andesite-dacite arc system - the Mount Adams Volcanic Field, Cascade Range of Southern Washington. Geological Society of America Bulletin, 106, 1413-1429. LANPHERE, M. A. & DALRYMPLE, G. B. 1971. A test of the 40Ar/39Ar age spectrum technique on some terrestrial material. Earth and Planetary Science Letters, 12. LE GALL, B., BELLON, H., CARRON, J. P. & LE GUEN DE KERNEIZON, M. 1983. Donnees nouvelles sur le volcanisme de l'ile de Montserrat (Petites Antilles). Bulletin Societe Geologique de France, 7, 837-843. MACGREGOR, A. G. 1938. The Royal Society Expedition to Montserrat, B.W.I. The volcanic history and petrology of Montserrat with observations on Mt Pele, in Martinique. Philosophical Transactions of the Royal Society of London, B229, 1-90. MANIKEN, E. A. & DALRYMPLE, G. B. 1972. The electron microprobe and K-Ar dating. Earth and Planetary Science Letters, 17, 89-94. MANVILLE, V., WHITE, J. D. L., HOUGHTON, B. F. & WILSON, C. J. N. 1998. The saturation behaviour of pumice and some sedimentological implications. Sedimentary Geology, 119, 5-16. MURPHY, M. D. & SPARKS, R. S. J. 1999. Petrology and geochemistry of the Soufriere Hills magma. Montserrat Volcano Observatory, Open File Report. MURPHY, M. D., SPARKS, R. S. J., BARCLAY, J. ET AL. 1998. The role of magma mixing in triggering the current eruption at the Soufriere Hills volcano, Montserrat, West Indies. Geophysical Research Letters, 25, 3433-3436. MURPHY, M. D., SPARKS, R. S. J., BARCLAY, J., CARROLL, M. R. & BREWER, T. S. 2000. Remobilization of andesite magma by intrusion of mafic magma at the Soufriere Hills Volcano, Montserrat, West Indies. Journal of Petrology, 41, 21-42. NORTON, G. E., WATTS, R. B., VOIGHT, B. ET AL. 2002. Pyroclastic flow and explosive activity at Soufriere Hills Volcano, Montserrat, during a period of virtually no magma extrusion (March 1998 to November 1999). In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 467-481. PRINGLE, M. S. 1993. Age progressive volcanism in the Musicians Seamounts: A test of the Hot Spot hypothesis for the Late Cretaceous

113

Pacific. In: PRINGLE, M. S., SAGER, W. W., SLITER, W. V. & STEIN, S. (eds) The Mesozoic Pacific: Geology, Tectonics and Volcanism. AGU Geophysical Monograph 77, 187-215. REA, W. J. 1970. The Geology of Montserrat, British West Indies. DPhil thesis, University of Oxford. REA, W. J. 1974. The volcanic geology and petrology of Montserrat, West Indies. Journal of the Geological Society of London, 130, 341-366. ROBERTSON, R. E. A., ASPINALL, W. P., HERD, R. A., NORTON, G. E., SPARKS, R. S. J. & YOUNG, S. R. 2000. The 1995-1998 eruption of the Soufriere Hills volcano, Montserrat, WI. Philosophical Transactions of the Royal Society of London Series a - Mathematical Physical and Engineering Sciences, 358, 1619-1637. ROOBOL, M. J. & SMITH, A. L. 1998. Pyroclastic stratigraphy of the Soufriere Hills volcano, Montserrat - Implications for the present eruption. Geophysical Research Letters, 25, 3393-3396. SHACKLETON, N. J. 1987. Oxygen isotopes, ice volume and sea level. Quaternary Science Reviews, 6, 183-190. SIGURDSSON, H., SPARKS, R. S. J., CAREY, S. N. & HUANG, T. C. 1980. Volcanogenic sedimentation in the Lesser Antilles Arc. Journal of Geology, 88, 523-540. SINGER, B. S. & PRINGLE, M. S. 1996. Age and duration of the Matuyama-Brunhes geomagnetic polarity reversal from Ar-40/Ar-39 incremental heating analyses of lavas. Earth and Planetary Science Letters, 139,47-61. SINGER, B. S., THOMPSON, R. A., DUNCAN, M. A. ET AL. 1997. Volcanism and erosion during the past 930 ky at the Tatara San Pedro complex, Chilean Andes. Geological Society Of America Bulletin, 109, 127-142. SPARKS, R. S. J., BARCLAY, J., CALDER, E. S. ET AL. 2002. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 409-434. SPENCER-CERVATO, C. & THIERSTEIN, H. R. 1997. First appearance of Globorotalia truncatulinoides: cladogenesis and immigration. Micropaleontology, 30, 267-292. STEIGER, R. H. & JAGER, E. 1977. Subcommission on geochronology: convention on the use of decay constants in geo- and cosmochronology. Earth and Planetary Science Letters, 36, 359-362. TAYLOR, J. R. 1982. An Introduction to Error Analysis. University Science Books, Mill Valley, CA. WADGE, G. 1984. Comparison of volcanic production rates and subduction rates in the Lesser Antilles and Central America. Geology, 12, 555-558. WADGE, G. 1986. The dykes and structural setting of the volcanic front in the Lesser Antilles island arc. Bulletin of Volcanology, 48, 349-372. WADGE, G. & ISAACS, M. C. 1988. Mapping the volcanic hazards from Soufriere Hills Volcano, Montserrat, West Indies using an image processor. Journal of the Geological Society, 145, 541-552. WADGE, G. & SHEPHERD, J. B. 1984. Segmentation of the Lesser Antilles Subduction Zone. Earth and Planetary Science Letters, 71, 297-304. WIJBRANS, J. R., PRINGLE, M. S., KOPPER, A. A. P. & SCHEVEERS, R. 1995. Argon geochronology of small samples using the Vulkaan argon laserprobe. Proceedings Koninklijke Nederlandse Academic van Wetenschappen,98, 185-218. YORK, D. 1969. Least squares fitting of a straight line with correlated errors. Earth and Planetary Science Letters, 5, 320-324. YOUNG, S. R., SPARKS, R. S. J., ASPINALL, W. P., LYNCH, L. L., MILLER, A. D., ROBERTSON, R. E. A. & SHEPHERD, J. B. 1998. Overview of the eruption of Soufriere Hills volcano, Montserrat, 18 July 1995 to December 1997. Geophysical Research Letters, 25, 3389-3392.

Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano, Montserrat R. B. WATTS1, R. A. HERD2, R. S. J. SPARKS1 & S. R. YOUNG2 1 Department of Earth Sciences, Wills Memorial Building, University of Bristol, Queens Road, Bristol BS8 1RJ, UK (e-mail: Rob. [email protected]) 2 Montserrat Volcano Observatory, Mongo Hill, Montserrat, West Indies

Abstract: Eruption of the Soufriere Hills Volcano on Montserrat allowed the detailed documentation of a Pelean dome-forming eruption. Dome growth between November 1995 and March 1998 produced over 0.3km 3 of crystal-rich andesitic lava. Discharge rates gradually accelerated from <1 m3 s-1 during the first few months to >5m 3 s-1 in the later stages. Early dome growth (November 1995 to September 1996) was dominated by the diffuse extrusion of large spines and mounds of blocky lava. A major dome collapse (17 September 1996) culminated in a magmatic explosive eruption, which unroofed the main conduit. Subsequent dome growth was dominated by the extrusion of broad lobes, here termed shear lobes. These lobes developed through a combination of exogenous and endogenous growth over many weeks, with movement accommodated along curved shear faults within the dome interior. Growth cycles were recognized, with each cycle initiated by the slow emplacement of a large shear lobe, constructing a steep flank on one sector of the dome. A growth spurt, heralded by the onset of intense hybrid seismicity, pushed the lobe rapidly out, triggering dome collapse. Extrusion of another lobe within the resulting collapse scar reconstructed the steep dome flanks prior to the next cycle.

In recent decades, phenomena observed during the growth of lava domes have been closely monitored, the most notable examples being at Mount St Helens, USA, between 1980 and 1986 (Swanson et al 1987), Mount Pinatubo, Philippines, in 1991-1992 (Daag et al. 1996) and Mount Unzen, Japan, in 1991-1995 (Nakada et al 1999). As a result, many emplacement features and the processes controlling their formation have been described (e.g. Anderson & Fink 1992). The ongoing eruption of Soufriere Hills Volcano on Montserrat has involved the construction of an andesitic lava dome, with alternating phases of growth and gravitational collapse (Young et al. 1998). The eruption required an intense scientific monitoring effort, due to the associated hazards and their threat to the local population, and this has yielded extensive data records on the eruption. These records have been used to distinguish patterns of growth and to develop a better understanding of the mechanisms controlling lava-dome eruptions (Sparks 1997; Voight et al. 1999; Melnik & Sparks 1999, 2002; Wylie et al. 1999; Sparks et al. 2000). Because of near-daily helicopter observation flights during the eruption, an impressive photographic and video collection has accumulated, documenting the growth and collapse of the lava dome. Detailed monitoring of these morphological changes was also achieved using various surveying techniques, such that an accurate record of dome growth and magma production rate is available (Sparks et al. 1998). All of these data have been used, in conjunction with ground observations, theodolite, and electronic distance measurements, to produce maps detailing the complex development of the dome in time and space. This paper describes the chronological evolution of the dome throughout the first episode of dome growth (November 1995 to March 1998) with the aid of maps and photographs, showing examples of the different structures that were extruded at particular stages of the eruption. We discuss the relationships between the formation of these structures and the controlling mechanisms during magma ascent and emplacement of the dome. Understanding of changing rates and styles of dome growth is vital in successful hazard assessment during dome-forming eruptions. We demonstrate that different growth styles are intimately associated with the generation of pyroclastic flows and the inception of explosive eruptions. We also develop an interpretation of the observations that attributes much of the morphological variation and behaviour to rheological stiffening of the magma caused by degassing and associated crystallization and also to deeper processes in the magma chamber, which periodically supplies pulses of fresh, gas-rich magma. The terminology used is as follows. There was one lava dome extruded between 15 November 1995 and 10 March 1998. Extrusion of a second dome began in November 1999, but is not dealt

with in any detail in this paper. Individual extrusions during the November 1995 to March 1998 period are termed lobes, and each is named by its date of first appearance.

Geological setting Montserrat is a mountainous, diamond-shaped island, 16km long and 9 km wide, that lies towards the northern end of the Lesser Antilles island arc. It is almost entirely volcanic in origin and is dominated by three volcanic centres. Recent 40Ar/39Ar dating (Harford et al. 2002) highlights a southerly shift in the focus of magmatism with time, from the low-lying Silver Hills (c. 2600 to 1200 ka BP) in the north, through the Centre Hills (c. 950 to 550 ka BP) and the South Soufriere Hills-Soufriere Hills complex (c. 160 ka to the present) in the south. The most recent activity has focused on the Soufriere Hills region in the south-central sector of the island (Fig. 1). This activity has involved the sporadic growth and collapse of at least five andesitic lava domes and the formation of associated pyroclastic aprons surrounding these domes. The youngest and smallest dome, Castle Peak, was probably formed in a small eruption c. 350 years ago just prior to colonization of the island (Robertson et al. 2000; Harford et al. 2002). This dome nestled centrally within English's Crater (Fig. 2), a large structure breached to the east, which is believed to have formed by a c. 4 ka sector collapse (Roobol & Smith 1998). A moat thus circled Castle Peak dome on two-thirds of its circumference, with the final third facing into the eastward-trending Tar River valley. The 1995-1998 episode of dome growth involved the near-continuous extrusion of 300 x 106 m3 of andesitic lava, constructing a complex Pelean dome on top of Castle Peak dome and within English's Crater. A further 100 x 106 m3 of lava has been extruded since November 1999 up to the time of writing (October 2000). A consequence of this eruptive activity has been the partial destruction and burial of both Castle Peak and the rim of English's Crater.

Eruption chronology Since the onset of phreatic activity on 18 July 1995, the eruption has progressed in an atypical manner in comparison with other well documented historical dome eruptions (Newhall & Melson 1983). From the initial phreatic period continuing through to the present day, a sequence of distinct eruptive phases has been experienced (see Table 1). These phases highlight a fluctuating magma discharge

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 115-152. 0435-4052/02/S15 © The Geological Society of London 2002.

115

R. B. WATTS ET AL.

116

Fig. 1. Pre-eruption map showing the location of andesitic domes (light shaded areas) of the Soufriere Hills-South Soufriere Hills complex in southern Montserrat. Dashed line marks the outline of the rim of English's Crater and the sides of the Tar River valley. Darker shaded areas are older uplifted pyroclastic sequences. Inset map shows the entire island.

Fig. 2. Early January 1996. View of Castle Peak (CP) sitting within English's Crater (EC), a c. 1 km diameter structure, looking west from above the Tar River valley. Pale lava on top of Castle Peak is new dome growth (D), and brown-stained vegetation results from early phreatic activity. Gages dome (G) is seen in background to right.

Table 1. Growth rates and associated surface/extrusive phenomena observed during the 8 eruptive stages of the current eruption Stage

Time period

Growth rate (m 3 s - 1 )

Surface phenomena and extrusive features

I

18 July 1995 to 14 November 1995

Pre-dome

Phreatic explosions

II

15 November 1995 to 16 February 1996

0.1-0.5

Spines, whaleback structures

III

16 February 1996 to 30 September 1996

1-4 with daily spurts of >5 in July and August

Spines 4- megaspines. Type 1 + Type 1 shear lobes

IV

1 October 1996 to 12 December 1996

0.5-2

Blocky lava + spines, endogenous activity

V

13 December 1996 to 13 May 1997

2-4 with daily spurts of >5 in December and January

Megaspines shear lobes

VI

14 May 1997 to 10 March 1998

>5 with two phases of post-collapse explosive activity

Type 1 shear lobes, blocky lava

blocky lava. Type 14+ Type 2

VII

11 March 1998 to mid-November 1999

No surface extrusion

Sporadic dome collapse, sporadic explosions

VIII

Mid-November 1999 to date

2-5

Spines, Type 1 shear lobes + blocky lava

GROWTH PATTERNS AND DOME EMPLACEMENT

1100

117

ones sending clouds of steam and dark ash aloft, which were blown westward towards the capital town of Plymouth. The first large phreatic explosion on 21 August initiated the first full evacuation of southern Montserrat. Of significance were the explosion craters of July, August and November, which formed a NNW-SSE alignment across Castle Peak (Fig. 4). Much of the subsequent magma discharge was focused along this zone, which suggests a fundamental fracture control on magma ascent. Extrusion of the first fresh lava occurred on the SW flanks of Castle Peak (Fig. 4), although it was initially interpreted as a cryptodome due to its highly oxidized nature. This small dome extruded on 25 September 1995 and continued to extrude for the next four or five days, generating small rockfalls prior to the growth stopping. A central spine of less oxidized material was extruded last, and degradation of the surrounding dome during October 1995 increased the prominence of this spine.

Stage II: 15 November 1995 to 16 February 1996

Fig. 3. (a) Graph showing the change in height of the active focus of growth throughout the eruption. Day one represents the onset of dome growth on 15 November 1995. (b) Graph showing the change in dome volume throughout the eruption (after Sparks et al. 1998 with more recent updates) and the different eruptive stages of the 1995-1998 period.

rate or pulses in magmatic activity that overprint a gradual escalation in eruptive vigour and background magma discharge rate (Sparks et al. 1998). As each phase of activity occurred, lava was extruded in a variety of styles and morphologies. Figure 3 details the volume of the dome and height of the active area of dome growth with time in the 1995-1998 episode of dome growth. Here we have divided the history of dome growth into stages, which are identified on the basis of prominent changes in dome growth patterns or significant volcanic and/or seismic events (Table 1). The main shifts in dome growth are given in Table 2 with notes on the different styles of growth during these shifts. Below we describe the dome evolution in chronological order, drawing attention to major styles of growth and morphological development. This paper does not consider in detail the new episode of dome growth, which started in November 1999, although similar patterns of dome growth are being observed to those described here.

Stage I: 18 July 1995 to 14 November 1995 The start of the eruption on the 18 July 1995 was heralded by a vigorous phreatic vent opening at the site of the poorly defined Langs Soufriere on the NW flanks of Castle Peak (Fig. 4). Islanders were disturbed by a continuous roaring sound and the sight of a near-continuous jet of steam from within English's Crater. In later weeks, further new vents opened around Castle Peak (Fig. 4) and phreatic activity continued for over four months on a near-daily basis. Phreatic explosions were of a variable intensity, the larger

Phreatic activity continued throughout October and on until 15 November, when two small piles of fresh lava blocks were observed, one within the twin craters of the 18 July phreatic vent and the other between it and the September spine (Fig. 4). Growth of a new dome was confirmed, both by the presence of incandescent lava blocks and the onset of hybrid seismicity (Miller et al. 1998). The lava blocks were pale grey and generally <5 m in diameter, with larger, curving spines jutting out from the crater floor. The main growth occurred in the more southerly crater of July 1995, which was filled completely with new lava by the end of November. Spines were the main extrusive feature of this early stage of dome growth (Table 3), with typical widths of 30m and heights up to 35m. They exhibited a curved, outer surface covered by a thin breccia coating, with distinct subparallel grooves or striations running along their length. After several days of growth (typically at a few metres per day), spine collapse would occur due to gravitational stresses, forming piles of rock debris around the stubby base of the spine. Spines typically grew to an altitude of 810-860 m above sea level (a.s.l.) in this period of low magma discharge rates (0.1-0.5m3 s - 1 ) . Growth of these spines was focused along the NNW-SSE zone defined by the early phreatic vents, particularly in the vicinity of the July 1995 craters. In Stage II, many spines emerged (Table 3) and grew on the western flanks of Castle Peak, such that the new dome developed as coalescing piles of spine debris (Fig. 5a). Fumarolic activity commonly occurred at the edges of these piles with associated rockfall activity increasing as the dome grew. A prominent feature of the freshly extruded lava was the apparent lack of flow structures and nearsolid appearance of the blocks and spines. An interesting feature, formed between 24 January and 6 February, was the formation of whaleback structures coincident with a period of seismic tremor and slightly raised extrusion rates of around 1 m 3 s - 1 . These elongate bodies extruded individually as extremely viscous lava from the NNW-SSE vent pattern in different directions (Fig. 5b). Each whaleback was pushed across the surface of Castle Peak by continued magmatic pressure (in a manner analogous to toothpaste being squeezed out of a tube). Whaleback structures would reach up to 200m long, 30m wide and 35m high, and exhibit a smooth surface with a grooved appearance and striations aligned parallel with the direction of extrusion (Fig. 6a, b). Growth of each whaleback structure occurred sporadically at an estimated 20-30 m day - 1 and continued to move for up to a week prior to growth stagnating and the subsequent break-up of the structure into a chaotic jumble of blocks (Fig. 7). During growth, as the whaleback pushed forward, rockfalls spalled off its steep headwall, exposing incandescent lava within the interior. Whalebacks are previously undocumented, although they have been witnessed during other eruptions (e.g. Santiaguito dome in Guatemala, W. I. Rose, pers. comm.). Less distinct rubbly lava also extruded in January 1996, although poor visibility prevented detailed documentation.

R. B. WATTS ET AL.

118

Table 2. Chronology of activity in the 1995-1998 period of dome growth at the Soufriere Hills Volcano Date

Activity

25-29 Sep. 1995

First juvenile lava extruded as small pile of blocks and spines on SW flanks of Castle Peak, but extrusion rapidly stagnated.

15 Nov. 1995

Onset of continuous extrusion of juvenile lava as spines on NW flanks of Castle Peak.

24 Jan. 1996 to 3 Feb. 1996

Extrusion of northern whaleback over several days followed by southern whaleback.

15-18 Feb. 1996

Extrusion of southeastern whaleback.

Mid-Feb. to early April 1996

Repeated extrusion of large vertical spines in central area of Castle Peak, spine collapse produces first pyroclastic flows down Tar River valley.

Late April 1996

Growth spurt preceded by hybrid earthquake swarm, forming shear lobe of blocky lava (i.e. 25 April 1996 lobe) at summit directed to NE flanks.

1-3 June 1996

Extrusion of megaspine to form northern peak.

22-25 June 1996

Extrusion of megaspine to form southern peak.

12-15 July 1996

Extrusion of megaspine to form southwestern peak.

Mid-July to early Sep. 1996

Vigorous extrusion of shear lobes of blocky lava towards NE dome flanks during growth spurts triggering dome collapses on 29-31 July, 11-14 Aug. and 2-3 Sep.

17 Sep. 1996

Directed explosion following major dome collapse down Tar River valley removing one-third of dome volume.

1 Oct. 1996

Onset of renewed dome growth (i.e. 1 Oct. 1996 lobe) at base of explosion crater at low extrusion rate.

Early Nov. to early Dec. 1996

Intense seismicity accompanying minor dome growth, suggesting period of endogenous activity. Localized doming on southern dome flanks.

13 Dec. to 24 Dec. 1996

Rapid extrusion of 13 Dec. 1996 megaspine along shear zone in southeast sector. This megaspine is quickly overwhelmed by blocky lava generating dome collapse on 19 Dec.

25 Dec. 1996 to 10 Jan. 1997

Rapid extrusion of 25 Dec. 1996 lobe forming 'pancake' morphology in central part of dome following localized endogenous doming. This rapidly overwhelms 13 Dec. 1996 lobe and 1 Oct. 1996 lobe.

Mid Jan. to late Jan. 1997

Rapid blocky growth from central area directed across SE flanks, generating semi-continuous pyroclastic flow activity and large collapses on 16, 19 and 20 Jan.

Early Feb. to 26 March 1997

Blocky lava of 21 Jan. 1997 lobe directed down southeastern and eastern flanks from central area forming a conical dome morphology.

27 March to mid-May 1997

Growth of 27 March 1997 lobe guided southwards triggering collapse of southern flanks on 30-31 March and 11 April leading to inundation of White River valley.

16 May 1997

Large vertical spine extruded at dome summit.

17 May to early July 1997

Growth of 17 May 1997 lobe across northern flanks triggering dome collapse down northern flanks on 25 June followed by regrowth of 27 June 1997 lobe to north.

Mid-July to late July 1997

Stagnation of active lobe, growth slightly switched to NW flanks (14 July 1997 lobe).

Late July to 12 Aug. 1997

Repeated small collapses down western flanks lead to major collapse on 3 Aug. Entire 14 July 1997 lobe collapsed triggering period of 13 repetitive Vulcanian explosions.

13 Aug. 1997

Renewed blocky growth in crater followed by formation of active lobe to the west.

8 Sep. 1997

Stagnation of 13 Aug. 1997 lobe with fresh lobe growth directed northwards. Rockfall and pyroclastic flow activity completely fills Mosquito Ghaut.

21 Sep. 1997

Rapid extrusion of 8 Sep. 1997 lobe to the north triggers major dome collapse directed down Tuitt's Ghaut forming amphitheatre-shaped edifice. This triggers a period of 75 Vulcanian explosions (22 Sep. to 21 Oct.).

22 Oct. to 2 Nov. 1997

Renewed blocky growth in crater developing into northward growing 22 Oct. 1997 lobe.

3-6 Nov. 1997

Stagnation of 22 Oct. 1997 lobe contemporaneous with three collapses concentrated on the southern flanks attributed to switch in active lobe to the south.

Mid-Nov. to 25 Dec. 1997

Continuous growth of southerly directed 4 Nov. 1997 lobe constructing a large peak straddling former Galway's Wall area.

26 Dec. 1997

Major debris avalanche and sector collapse removing the entire southern flanks (i.e. 4 Nov. 1997 lobe) formed in previous two months. Violent pyroclastic density currents.

27 Dec. 1997 to late Feb. 1998

Redevelopment of southerly directed lobe (i.e. 27 Dec. 1997 lobe) rebuilding the large peak on the southern flanks.

1-10 March 1998

Cessation of first period of dome growth signalled by 50 m high spine at summit of 27 Dec. 1997 lobe.

3

3

81400

80500

Fig. 4. Map of English's Crater on 25 November 1995 showing the location of phreatic vents (with dates when they first appeared) and the initial extrusions of fresh andesitic lava in the early eruptive stages. Note that the 21 August phreatic vent opened up along a fracture-controlled line of smaller vents. Farrell's Wall, Gages Wall and Galway's Wall represent different sectors of the steep avalanche scar defined by English's Crater. The rim of the crater is marked by a dashed line. Peaks A B and C on this rim were used as prominent topographic features in surveys of the dome. Map co-ordinates are part of the Montserrat Grid System. Table 3. Theodolite data collected during Stages II and III of the eruption to monitor the height and growth rate of spines extruded at this time Date (eruption day)

Theodolite location

15/11/95(1) 5/12/95 (20) 10/12/95 (25) 17/12/95 (32) 20/12/95 (35)

Long Ground Long Ground Long Ground Whites Whites

24/12/95 (39) 26/12/95 (41) 28/12/95 (43) 30/12/95 (45) 3/1/96 (49) 8/1/96 (54) 11/1/96(57) 22/1/96 (68) 5/2/96 (82) 21/2/96(98) 11/3/96(117) 19/3/96 (125) 22/3/96 (128) 25/3/96(131) 26/3/96(132)

Whites Whites Whites Whites Whites Whites Long Ground Photo method* Photo method* Whites Observatory Whites Whites Whites Whites

29/3/96 (135) 30/3/96 (136) 5/4/96(142)

Whites Whites Long Ground

18/4/96(155) 20/4/96(157) 30/4/96 (167)

Long Ground Long Ground Whites

Height of spine (m a.s.l.)

Height of summit (m a.s.l.)

Spine growth rate (mday -1 )

760 814 805 805 825

>7 Cathedral spine 9 Highest of 3 spines 2.5

817 814 812 817 810 814 842

4 >9 823 825 821 866 847

857 874 885

9 5 11 Spine width = 35m 855

862 883

14 Spine width = 44m

905 903 929

* Measurements obtained by a technique used to calculate changes in dome volume through comparison of photographs taken from the same location. Dates in European format (5/12/95 = 5 December 1995). a.s.l., aboe sea level.

120

R. B. WATTS ET AL.

Fig. 5. Sequence of maps showing gradual development of the dome during the initial three months of dome growth leading to onset of pyroclastic flows at the end of March 1996. Shaded area marks the new dome growth. X-Y represents the line of section used in Figure 7.

GROWTH PATTERNS AND DOME EMPLACEMENT

121

Fig. 6. (a) 2 February 1996. View looking west above Castle Peak showing two whaleback structures on the new dome: the southern whaleback is marked S and the northern whaleback N (see Fig. 5b). Note the steep walls of English's Crater (EC) and the dome of Chances Peak (ChP) and telecommunications tower in the centre background. Gages Wall (G) is on the right with Plymouth Town (P) in the distance, (b) 1 February 1996. View looking east from crater rim near Chances Peak. Elongate structure in centre is the southern whaleback (S), which extruded subhorizontally away from the central area over several days (see Fig. 5b). Note the smooth outer surface of the whaleback, with rocks spilling off the leading edge on the right. Loose blocks sitting on its top are carried along during growth. CP is the main spine of Castle Peak dome.

Stage III: 17 February 1996 to 30 September 1996 In mid-February, a marked rise in the average background extrusion rate (c. 2 m 3 s - 1 ) was experienced at the start of this period, following a volcanotectonic earthquake swarm on 11 February 1996. There was also a steady increase in dome height, which reached 960m a.s.l. by the end of June 1996 (Fig. 3a). Spectacular spines with prominent vertical striations scored along their smooth, outer surfaces, and exhibiting characteristic curved-horn shapes

were commonplace throughout March and April (Fig. 8a, b). Growth rates of these spines averaged 6-7 m day - 1 , sometimes over 10m day - 1 , attaining heights of c. 40m before collapsing (Table 3). Spine growth was generally concentrated in the central area of Castle Peak, a slight southerly shift from the initial growth area of mid-November 1995. The continuous formation of spine debris enlarged the area covered by the new dome. By the end of March, the entire western half of Castle Peak had been overwhelmed by fresh lava (Fig. 5c) and blocky talus was rapidly

122

R. B. WATTS ET AL.

Fig. 7. Schematic diagram showing the development of whaleback structures in January 1996, along section X-Y in Figure 5b. (a) 15 January 1996. Dome growth occurring as spine extrusion and gravitational collapse, (b) 24 January 1996. Whaleback structure extrudes northwards away from central area averaging 30m day - 1 , (c) 29 January 1996. Growth of northern whaleback stops and another whaleback starts to drag across the dome surface southwards away from the centre at 20-30 m day - 1 . (d) 5 February 1996. Both whalebacks have stagnated and gradually break up. Spine growth and collapse recommences around summit area.

filling the moat of English's Crater. Rockfalls were no longer confined by the walls of English's Crater on the NE flanks, and they developed into small pyroclastic flows down the upper parts of the Tar River valley. The pulsatory character and most common style of dome growth can be illustrated by. observations in the second half of April. Two spines were extruded around 14-16 April from the July 1995 crater area coincident with a period of elevated hybridearthquake seismicity. The spines toppled over, one to the west and the other to the east. While the westerly spine remained stagnant, the easterly spine started to break up on 25 April as new lava started to extrude. This blocky lobe then expanded to the NE and formed a steep headwall at its advancing front from which rockfalls cascaded down into the Tar River valley. The flow front position

stagnated when it reached the steep edge of Castle Peak with generation of rockfalls at the flow front matching supply of lava from the lobe. The flow front developed a furrowed appearance of ridges and chutes related to repeated generation of erosive rockfalls spilling down from the summit of the lobe. This blocky lobe, here termed a shear lobe to highlight this process of directed extrusion (Fig. 9a), continued to produce lava until 2 May, when growth became more focused around the summit. This formation of shear lobes was repeated many times throughout the eruption although the style of lobe development was markedly different in the later eruptive stages (Fig. 9b, c), allowing a classification between earlystage Type 2 lobes and late-stage Type 1 lobes. A feature common to both types was that the summit of a lobe was commonly shifted away from the vent area during growth, giving an illusion of shifting vent positions as an individual lobe developed. A similar illusion occurred when a lobe stagnated and a fresh lobe was initiated, growing in a different direction from the previous lobe. Formation of Type 2 lobes was apparent after extended periods of stagnation, while Type 1 lobes predominated during periods of more steady growth, although fluctuations in discharge rate were evident during the development of both structures. From June to July 1996, noticeable shifts in the focus of dome growth followed the slow extrusion of broad features (up to 100 m wide) that are best described as fault-bounded megaspines (Fig. lOa, b). A megaspine is characterized by two contrasting parts. One side of a megaspine consists of a smooth, striated and curving wall which is interpreted as moving along a large fault in the dome. The other side consists of a headwall of massive, blocky material that breaks up during growth. A megaspine grows by upward or subhorizontal movement along the fault structure with lava blocks spalling off the main headwall as growth occurs (Fig. lOc). Emplacement of such a large structure often stopped after a few days, with renewed activity taking place in another localized part of the dome. The most notable examples of megaspine growth occurred during 1-3 June (to the north), 22-25 June (to the south) and 12-15 July 1996 (to the SW). Each of these extrusions formed a prominent peak on the dome and originated from the same central focus of growth. At this central focus, extrusion of fresh lava was guided along a curvilinear shear fault that directed the lava in a specific direction, sometimes over a hundred metres away from the previous growth area (Fig. 11). By mid-July 1996, the new dome was a substantial size (c. 30 x 10 6 m 3 ) and had multiple peaks due to the repeated formation of megaspines. Vigorous spurts of dome growth were also evident at this time in pulses lasting several hours, at estimated discharge rates of >5m s-1 . The focus of growth was located in the central area, with the 17 July 1996 lobe directing fresh lava down the NE flanks of Castle Peak and the upper reaches of Tar River valley. At these rapid growth rates, distinctive piles of blocky lava (c. 4-5 m diameter) were extruded, exhibiting a curvilinear shape with occasional larger spines projecting out. The coincidence of this vigorous growth and the focus of growth directing lava to the NE resulted in a series of major dome collapses producing large pyroclastic flows on 29-31 July, 11 August, 20-21 August and 2-3 September 1996. This involved the repeated collapse and reconstruction of the NE dome flanks (Fig. 12a, b) and the gradual inundation of the entire Tar River valley (Cole et al. 2002). These repeated collapses led to a decrease in the height of the active area of growth (Fig. 3a). Renewed growth was always focused at the base of the spoon-shaped scar formed by each collapse, often accompanied by a vigorous, semi-continuous ash plume from the central area. A new lava lobe, often bounded by a shear surface at the backwall, would then extrude as large, curved blocks that rapidly filled and engulfed the entire scoop (Fig. 12c). The lava lobe always expanded asymmetrically towards the open side of a previous collapse scar, guiding it away from the vent area. A particularly long episode of near-continuous collapse (c, 9 hours) occurred on 17 September, culminating in sub-Plinian magmatic explosive activity (Robertson el al. 1998). This sustained collapse removed over one-third of the new dome (c. 11.8 x 10 6 m 3 volume non-dense rock equivalent, all volumes quoted here are collapse volumes calculated from deposit volumes and or collapse scar

GROWTH PATTERNS AND DOME EMPLACEMENT

123

Fig. 8. (a) 9 April 1996. View looking north across English's Crater and the new dome, with the northeastern flanks and airport in the background. Large spine of fresh lava (P) is c. 40m high and c. 35 m broad. Broad spine in centre (Q) is the main spine of Castle Peak dome. Note Galway's Wall (GW) in foreground, (b) 12 April 1996. Typical curved-horn spine extruded in Stages II and III. Note the semi-cylindrical shape with smooth and striated outer surface. Other half of spine is massive lava of the spine interior that spalls off to form a basal skirt of debris around the base of the spine. This spine is estimated to be 40m high and 35m broad. Note Perches Mountain (PM) in background to the right.

volumes (Calder el al. 2002), and a substantial portion of the underlying Castle Peak dome, leaving a large steep-sided scar in the central area open to the east (Figs 13a and 14a).

Stage IV: 1 October 1996 to 12 December 1996 Extrusion of fresh lava did not resume for two weeks following the 17 September 1996 explosive eruption, which was estimated to have

reamed out the conduit to a depth of 4 km (Robertson et al. 1998). This event involved substantial widening of the upper conduit, with abundant ballistic ejecta of vent-wall breccia, hydrothermally altered rocks of Castle Peak and dense, juvenile blocks thought to have originated in the upper conduit (Robertson et al. 1998). The explosion was directed to the east as evidenced by strong asymmetrical distribution of ballistic ejecta, and this activity is believed to have flared open the upper conduit. Lava started to extrude at the base of the scar on 1 October with an initial discharge rate of 1.8 m3 s-1 The focus of upwelling was noticeably shifted c. 150 m to

124

R. B. WATTS ET AL.

Fig. 9. Characteristic features of Type 1 and Type 2 shear lobes shown in schematic form, (a) Development of a Type 2 shear lobe: following an extended period of stagnation, a fresh pulse of magma pushed out a viscous plug that was emplaced as a large spine or megaspine. The hotter magma that forced out the plug continued to ascend along another shear fault that detached from the conduit wall and magma extruded in the direction of least resistance (see Fig. 11). During extrusion, the lobe broke up into curving blocks with the highest point often displaced laterally away from the conduit, giving the illusion of shifting vent locations when a new lobe moved in another direction. The steep flow front advanced until it reached a steep slope (e.g. the margin of Castle Peak or the rim of English's Crater) where the front stagnated and generated rockfalls and pyroclastic flows as lava was supplied from behind, (b) Early stage development of a Type 1 shear lobe: following shorter periods of stagnation, a new shear lobe extruded with a large, coherent, smooth and striated upper surface with a broad headwall of massive lava. The upper surface developed a broad, semi-cylindrical shape supported by the surrounding dome flanks. This switch in the focus of activity and early stage development of a new lobe often triggered a dome collapse, e.g. 4 November 1997. (c) Late stage development of a Type 1 shear lobe: following stagnation of the viscous plug, hotter and more ductile lava would rise up along the same shear zone. Broad, curving spines extruded and broke up at the rear of the lobe and this activity alternated with the injection of magma into the core of the lobe. This latter process expanded the summit area and triggered rockfalls off the leading edge of the lobe. The repeated nature of these processes over several weeks developed a conical summit with a broad skirt of blocky talus. Formation of a Type 1 lobe was most apparent during periods of steady-state growth when only minor fluctuations in the average discharge rate were experienced.

the east in comparison to the location of spines and upwellings before 17 September 1996. From this time, the dimensions of lobes generally became substantially greater. This temporary shift in growth foci and the larger dimensions of subsequent lobes are attributed to the widening of the upper conduit asymmetrically to the east. Initially, the new lobe consisted of a slab of smooth lava overlying loose talus (Fig. 14a, b). The morphology of this lava exhibited a transition over several days from smooth (Fig. 14b) to an unusual darker, rubbly surface (Fig. 15a) and eventually to the blocky and spiny appearance that had previously characterized the dome. This period of activity shows an excellent example of the morphology formed by new growth infilling the scar after a major collapse. The early growth of the 1 October 1996 lobe was also the first example in this eruption of a lava morphology affected by lateral spreading (Fig. 15a). As growth continued, the lobe gradually filled the scar at a decreasing discharge rate and had apparently stagnated by 20 October 1996. Renewed extrusion, still at a reduced rate, took place on 22 October, but focused only on the central part of the lobe. This formed a central raised area of blocks and small spines surrounded by the lower abandoned sectors of the lobe (Fig. 13b). In early November 1996, intense shallow earthquake swarms occurred, dome growth rates dwindled to
height stagnated at 900m a.s.l. By mid-November, further earthquake swarms triggered landslides off the steep outer face of Galway's Wall on the southern rim of English's Crater (Fig. 15b). This apparent intrusion into the dome, causing localized uplift of over 30m on the southern dome flanks, was the first clear evidence for endogenous activity during the eruption. This raised levels of concern that the threat of sudden collapse of Galway's Wall might trigger rapid decompression of the dome interior (Young el al. 2002). Stage V: 13 December 1996 to 13 May 1997 The crisis related to the instability of Galway's Wall was temporarily relieved on 13 December 1996 when the southeastern margin of the 1 October 1996 lobe was uplifted along a major ductile shear zone bounded by a steep, striated fault with a trend of 110 a . The 13 December 1996 lobe (also known as the 'Venus' lobe) initially extruded as a megaspine near the zone of endogenous uplift. This fault-bounded feature extruded rapidly in a southeasterly direction overnight and continued growth along the same trend for several days, forming a single entity at least 150m long. 100 m wide and 100m high (Figs 16a and 17a). Within days, the megaspine had been

GROWTH PATTERNS AND DOME EMPLACEMENT

125

Fig. 10. (a) and (b) Maps highlighting the switches in the focus of activity during June and July 1996 following the emplacement of megaspines. Dotted lines represent approximate boundary between massive lava and loose talus blocks (for clarity, this distinction is not marked in later maps), (c) 19 August 1996, looking east from summit of Chances Peak. Megaspine extruded in early July 1996 (c. 40m high and c. 100m broad) showing a smooth, curved northern face to the left and massive, crystalline lava breaking off its broad headwall (H) as large blocks to the right. This large feature was subvertically emplaced in the northwest sector of English's Crater (see a) and remained a prominent peak on the dome for the following six months. Also in view is another megaspine (S) emplaced on the southern sector of the dome in late June 1996 (see b).

overwhelmed by the pulse of magma that had initiated its extrusion. Near-continuous rapid extrusion of blocky lava (c. 4-6 m3 s - 1 ) was guided along the same southeasterly directed shear fault that extruded the megaspine, and pyroclastic flows spilled off the SE dome flanks.

On 25 December 1996 a pronounced pulse of activity, heralded by localized endogenous doming and episodes of banded seismic tremor (Miller et al. 1998), led to the emplacement of the 25 December 1996 lobe, also known as the 'Santa' lobe (Fig. 16b). This new lobe initially punched through the central summit of the

126

R. B. WATTS ET AL.

Fig. 11. Schematic diagram illustrating the emplacement of a megaspine and the switching of activity that was apparent in June and July 1996. Diagrams represent a cross-section along the line X—Y in Figure lOb. (a) 30 June 1996. This was a period of very slow growth with only minor rockfall activity occurring at the summit. Lava within the upper conduit is crystallizing and forming a near-solid plug within the dome, (b) 13 July 1996. Increased seismic activity heralds a pulse of fresh magma through the conduit. This increased pressurization results in the lava plug being pushed out along a curved shear fault within the dome. This plug is emplaced as a megaspine at the summit producing rockfalls that break away from its headwall during growth, (c) 27 July 1996. Following emplacement of the megaspine, the hotter magma beneath is redirected along another shear fault that provides an easier pathway to the surface. The rapid rise from depth of hot magma at the head of the fresh magma pulse does not allow enough time for a plug to develop in the upper conduit. The 17 July 1996 lobe was a typical example of a Type 2 shear lobe comprising large blocks and stubby spines sourcing pyroclastic flows.

13 December 1996 lobe at a vigorous pace, with field observations suggesting a discharge rate between 6 and 9 m3 s - 1 . This fresh lobe spread out laterally across the summit to form pancake-shaped lava on top of the 13 December lobe. Notably, some of the most spectacular examples of incandescence during the night-time and even daytime were experienced during this period associated with rapid discharge rates (Fig. 17b). In the following two weeks, this lobe rapidly overwhelmed both the 1 October 1996 lobe and the 13 December 1996 lobe through lateral growth (Fig. 18). The 25 December 1996 lobe provides the best example of the more fluidlike behaviour exhibited by the lava when dome growth was more vigorous, The top third of the dome (c. 50-70 m) on 28 December 1996 consisted of a front of incandescent blocky lava which con-

tinually generated incandescent rockfalls down the lower two-thirds of the dome. The circular plan-form and slightly raised central summit of this lobe indicated that lava was extruded in the summit region but was able to spread laterally, forming an overall pancakelike morphology. Emplacement of this somewhat more fluid lava contrasts with the predominant extrusion of spines and lobes along ductile shear faults. The 25 December 1996 lobe initially spread symmetrically, spilling lava down the entire eastern flanks, but by mid-January its advance had become more focused towards the SE. The continued advance of the lobe guided lava down the SE dome flanks and triggered large pyroclastic flows down the SE side of the Tar River valley on 9 16 and 19-20 January 1997. With the northern half of the 25 December 1996 lobe now effectively abandoned, subsequent blocky growth then rapidly infilled the SE-facing scar, producing a series of small collapses (Figs 19c and 20a.b). By early February 1997. Castle Peak had been partially destroyed and completely buried by rockfalls and pyroclastic flows, although the discharge rate had declined to more moderate levels (2-4 m 3 s - 1 ). A slight switch in steady growth of the southeasterly directed 21 January 1997 lobe to a more easterly trend, focused rockfall activity down the eastern flanks throughout late February and early March. The conical dome that developed in the latter half of March illustrated well the effect of rockfalls on dome morphology. This period provided a very good example of the evolution of an asymmetric Type 1 shear lobe over an extended period of sustained growth. Large, curving spines of fresh lava would extrude away from the rear of the lobe, then gradually push forward and break up. spilling blocks onto the lobe summit. This activity alternated with magma pushing into the molten core of the lobe, expanding this area and triggering rockfalls off the lobe headwall (Fig. 9c). A swath of rockfalls from both collapse of spine debris and disintegration of the lobe headwall gradually formed a lobe with a furrowed appearance as rockfall chutes developed. This repeated process of alternating spine growth and magma injection into the core gradually formed a lobe with a broad, conical form (Figs 2l a and 22a). The end of March witnessed the extrusion of a remarkable structure, formed in the early development of the 27 March 1997 lobe (also known as the 'Easter' lobe). The growth of this lobe occurred at a time of good visibility and its development was well documented. The appearance of this structure immediately preceded a dome collapse on 30-31 March (c. 3.6 x 10 6 m 3 non-DRE deposit volume) focused on the southern flanks; this destabilization is attributed to the initial growth of the 27 March 1997 lobe. A broad mass of lava started to project out in a subvertical manner on 27 March, originating from about 50-70 m below the summit area of the conical dome lobe extruded throughout March. Its movement took place in a stick-slip fashion at estimated rates of 25-30 m d a y - 1 , with movement of the lobe accommodated by a southerly directed ductile shear fault in the dome. By 3 April, the 27 March 1997 lobe exhibited a smooth, yet striated, curved upper surface (c. 100m long and c. 120m wide) and a near-vertical headwall of massive lava almost c. 150m high (Figs 21b and 22b). Growth of this structure involved gradual rotation guided along by the shear fault surface forming a semi-cylindrical cross-section. Vigorous degassing emanated from around the horseshoe-shaped boundary zone between the lobe and the inactive parts of the dome. On initial extrusion, the smooth backwall was directed subvertically, and as growth continued this changed down into a more subhorizontal position (Fig. 22c). This mode of extrusion can be closely correlated to the shear lobes of blocky lava evident in mid1996. In the case of the 27 March 1997 lobe, however, the smooth, semi-cylindrical upper surface remained as one large coherent structure during extrusion instead of breaking up into smaller, curving blocks that gradually rotated forwards during growth (Fig. 9b). This difference may be partly attributed to the fact that the 27 March 1997 lobe was buttressed and supported by much larger surrounding dome flanks than the blocky lobes apparent in the earlier eruptive stages. The 27 March 1997 lobe continued to grow in the same manner, bulldozing through the southern flanks of the new dome and

GROWTH PATTERNS AND DOME EMPLACEMENT

127

Fig. 12. (a) and (b) Maps highlighting the scales and locations of collapse scars of 12 August 1996 and 2 September 1996, both formed as a result of the growth spurts prevalent in the Stage III period. Shaded area represents the new dome. Legend as in Figure 5. (c) 25 August 1996. View from point Y on (a) looking at NE flanks of new dome. Fresh lava blocks are extruding from the summit area (marked by arrow) and being directed down the NE flanks. Such growth was typical during spurts of activity that were commonly experienced during late July and August 1996. Note the ash-venting near the focus of extrusion at the rear of the lobe. Dashed line marks the scar rim from the 11 August 1996 collapse within which fresh lava of the 12 August 1996 lobe has filled up the scar in a two-week period. CP marks the prominent two-pronged spine of Castle Peak.

destroying the upper ramparts of Galway's Wall (Fig. 23a). Large blocks of massive lava spalled off the headwall of this lobe as it jerked forward to generate rockfalls and pyroclastic flows down Galway's Wail and into the White River valley. Within a week, dome talus and pyroclastic flows had overwhelmed what remained

of Galway's Wall. Following another dome collapse (c. 3 x 106m3 deposit volume) on the southern flanks on 11 April, a well developed lobe was observed extruding in a similar manner and trend as the 27 March 1997 lobe. The smooth backwall of this 13 April 1997 lobe, however, was steeply angled at c. 45° whilst the leading edge

128

R. B. WATTS ET AL.

Fig. 13. (a) and (b) Maps highlighting the scale and configuration of the 1 October 1996 lobe growth following the 17 September 1996 explosion. Light-shaded areas mark lava extruded prior to the September explosion; darker shaded areas mark the 1 October 1996 lobe. Legend as in Figure 5.

continued to push outward and break up during growth. Slow growth of this lobe continued for a few weeks until stagnation in mid-May, when the headwall of massive lava remained as the steep, upper ramparts on the southern dome flanks. A broad skirt of talus fanned away from these ramparts (Fig. 23b) onto the deposits from the associated pyroclastic activity, which had destroyed most of the White River valley (Cole et al 2002).

Stage VI: 14 May 1997 to 10 March 1998 This period commenced with a gradually accelerating background rate of extrusion (c. 4-5m 3 s - 1 ) and an increase in dome height to 1000m a.s.l. Between 14 and 17 May 1997, a distinct switch in the focus of growth took place, returning to the central area with an impressive c. 50-m-high vertical spine evident at the summit on 16 May (Fig. 24a, b). Within two days, this spine had collapsed and lava extrusion was guided in a northerly direction, with rockfalls spilling down a broad swath covering the southeastern-to-eastern sector and on around to the NW flanks (Fig. 24c). Despite limited visibility, the stark contrast between the shutdown of activity on the southern flanks and the broad spread of activity across the entire northern flanks indicated the development of the 17 May 1997 lobe directing lava to the north. This rapid switch in dome growth also indicated that vertical extrusion of lava was only evident between major switches in activity, with the preferred mode of emplacement taking place through subhorizontal shear lobes. This semicontinuous rockfall activity continued unabated at a steady rate for several weeks, broadening the dome flanks against the northern walls of English's Crater. By early June 1997, talus was level with FarrelFs Wall on the northern rim, and rockfalls tumbled directly down the northern dome flanks into Tuitt's Ghaut and Mosquito Ghaut. The ensuing weeks involved increasing pyroclastic flow activity down these ghauts, threatening the communities of the northeastern slopes of the volcano (Loughlin et al. 2002).

On 22 June 1997, a marked change in seismicity occurred, with an 8-hour cyclic pattern of intense hybrid earthquake swarms and associated cycles of ground deformation recorded by near-vent tiltmeters (Voight et al. 1999). This period of increased pressurization continued for several days and culminated in the rapid destabilization of the entire 17 May 1997 lobe. A major dome collapse (c. 6.4 x 106 m3 deposit volume) on 25 June resulted in the first fatalities of the crisis (Loughlin et al. 2002). Large pyroclastic flows were funnelled down Mosquito Ghaut, reaching about 6 km to the NE, lowering the dome summit by 100m and excavating a steepsided scar on the northern dome flanks. Within a few days the scar was refilling with blocky lava and stubby spines to construct the 27 June 1997 lobe. By 10 July, a broad headwall of massive lava had risen up within the scar indicating the northerly directed lobe had re-established itself (Fig. 25a). A gradual switch in activity then became apparent throughout late July, with rockfalls and pyroclastic flows initially coursing down Mosquito Ghaut. By 14 July rockfall activity was concentrated more to the west, directing pyroclastic flows down Gages valley. This shift is attributed to the stagnation of the 27 June 1997 lobe and the formation of the 14 July 1997 lobe directing an active headwall to the west. A further pulse in activity on 31 July was again marked by an abrupt increase in amplitude of tilt cycles in conjunction with intense hybrid earthquake swarms (Voight et al. 1999). This was associated with a growth spurt that produced a westerly directed dome collapse engulfing Gages valley and Plymouth Town in pyroclastic flow deposits on 3 August (Cole et al. 2002; Druitt et al. 2002). In a similar scenario to 25 June, the rapid influx of magma into the upper conduit had pushed out the entire shear lobe that was already actively growing to the west. This event produced a deep central crater in the dome with a breached, open scar facing to the west. The rapid removal of the 14 July 1997 lobe, caused by the collapse on 3 August, perturbed the magmatic system in the upper conduit and triggered a week-long series of repetitive Vulcanian explosions (Druitt et al. 2002). By mid-August, blocky lava was apparent within the crater and the emplacement of a westerly

GROWTH PATTERNS AND DOME EMPLACEMENT

129

Fig. 14. (a) 2 October 1996. Photo looking west from above the Tar River valley (see Fig. 13a). Castle Peak (CP) is the dark material in the left foreground surrounded by fresh, pale lava of the new dome, which appears as a rugged ridge defining the rim of the horseshoe-shaped collapse scar (CS) and explosion crater from 17 September 1996 collapse and explosive eruption. Small bun-shaped feature in the centre of this scar is the 1 October 1996 lobe (OL) extruded 15 days after explosive event had evacuated the upper conduit. Megaspine (M) of early July (see Fig. lOc) can be seen behind the central growth and Peak C on English's Crater is on far right. Trench in foreground is the main exit channel eroded by pyroclastic flow activity on 17 September 1996. (b) 2 October 1996. Close-up of new growth (1 October 1996 lobe, OL) seen in (a) showing smooth-topped massive lava (c. 20m thick) resting atop a mantle of rock debris. This feature, named informally 'the brain', gradually rose and infilled the 17 September 1996 explosion scar over the following month. Note the fumarolic activity around the perimeter of the fresh lobe.

directed 13 August 1997 lobe rapidly ensued. Elevated seismicity and extrusion rates (estimated at > 5m3 s - 1 ) continued with gradual collapse of the headwall continuing to generate pyroclastic flows towards Plymouth. By late August, the western dome flanks had been rebuilt (Fig. 25b, c), with the height of the dome nearly reaching 1000m a.s.l.

In early September 1997, the broad headwall of the 14 August 1997 lobe was widening its span to generate rockfalls down the northern flanks, as well as down Gages valley. By 8 September, rockfall activity inundated the northern flanks whilst activity had shut down on the western flanks, indicating stagnation of the 14 August 1997 lobe. Lava blocks spilled off the active headwall of

130

R. B. WATTS ET AL.

Fig. 15. (a) 10 October 1996. View looking west from above NE dome flanks showing the 1 October 1996 lobe (OL) infilling the scar (ES) formed by the 17 September 1996 explosive eruption. Small scar (S, in foreground) was formed by small rockfalls tumbling down the eastern flanks of the dome. Note the rubbly carapace (a few metres thick) that characterized the flat surface of the 1 October 1996 lobe. e. 160m in diameter here, (b) 3 December 1996. View looking WNW at the dome from above southern rim of English's Crater (EC). The 1 October 1996 lobe (OL) growing in the 17 September 1996 explosion crater and collapse scar can be seen in the centre. The steep rim (S) of the spoon-shaped scar of 17 September is well seen here. ED marks the approximate zone of localized doming noted during volume surveys at this time and interpreted as a site of endogenous growth. ChP marks the summit of Chances Peak dome (summit height c. 909m a.s.l.). Galway's Wall (GW) with fresh avalanche debris at its base is centre left, and a small remnant of Castle Peak (CP) is visible in central foreground.

the 8 September 1997 lobe now growing directly northward and debris completely filled Mosquito Ghaut at this time thus extending the limits of Farrell's Plain (Fig. 25c). The active lobe continued to spall away blocky lava onto the northern dome flanks and down Tuitt's Ghaut. The dome at this stage was voluminous (c. 85 x 106 m 3 ), with the northern flanks growing as a single lobe. On the morning of 21 September 1997, a swarm of hybrid earthquakes preceded a large dome collapse, An estimated 14.3 x 10 6 m 3

of material was funnelled down Tuitt's Ghaut, forming an amphitheatre-shaped scar in the dome, with a large crater open to the north (Figs 26a and 27a). This collapse was a further example of the rapid extrusion of a new shear lobe destabilizing the sector of the volcano in which it was actively growing, A second series of repetitive Vulcanian explosions followed the collapse, due to rapid depressurization of magma in the main conduit (Druitt et al. 2002). This month-long period produced only very minor changes to the overall

GROWTH PATTERNS AND DOME EMPLACEMENT

131

Fig. 16. (a) to (c) Maps highlighting extrusion of the 13 December 1996 lobe, the 25 December 1996 lobe, the 21 January 1997 lobe and the location of vigorous pyroclastic flow activity in mid- to late January 1997. Legend as in Figure 5.

dome morphology, although explosive activity reamed out a deep, funnel-shaped central crater, almost 300m in diameter. The onset of renewed dome growth on 22 October involved the slow extrusion of another lobe to the north. This feature initially extruded vertically, with the back wall of the explosion crater acting as the shear

zone accommodating its growth. As it continued to rise, the 22 Octoher 1997 lobe gradually projected to the north, partly infilling the scar produced by the 21 September collapse (Figs 26b and 27b). Growth of the 22 October 1997 lobe stagnated on 3 November and rockfall activity started to spill down the southern flanks of

132

R. B. WATTS ET AL.

Fig. 17. (a) 18 December 1996. View looking west from southern edge of Tar River valley marked by point Z on Figure 16a. Large structure in centre is the headwall of the 13 December 1996 lobe, around 150m high and 200m broad. VL marks the smooth, curved outer surface of the megaspine extruded in the early stages of this lobe. CS marks the scar rim of the September explosion crater and collapse scar; CP is the main spine of Castle Peak. Pale blocky talus in the foreground is rockfall debris produced by extrusion of the 13 December 1996 lobe, (b) 28 December 1996. Night-time view of glowing dome seen from the Whites area to the northeast, c. 2km from the dome (Fig. 1). Incandescent blocks of the 25 December 1996 lobe are evident, spilling down the eastern flanks of the dome. Note the silhouetted twin-pronged spine of Castle Peak in the left foreground.

the dome, coincident with a period of intense shallow hybrid earthquakes. Three major collapses (c. 8 x l 0 6 m 3 total volume of deposits) then occurred on 4 and 6 November, each concentrated on the southern flanks and sending pyroclastic flows coursing down the White River valley. These collapses are attributed to the 22 October 1997 lobe dislocating (hence stagnating), with redirected extrusion to the south (Fig. 28). The bulldozing effect of the 4 November 1997 lobe instigated collapse of older lava on the southern dome flanks (the two collapses on 4 November) and collapse of freshly extruded blocky lava from the headwall of the 4 November

lobe itself (the 6 November collapse; Figs 29a and 30a). Rapid advance of this lobe infilled the collapse scar formed by these events, pushing a broad headwall of massive lava southwards. Lava continued issuing from this southerly directed shear zone for many weeks. Throughout this period, activity alternated between the extrusion of broad, curving spines at the back of the lobe or rockfalls spilling away from the summit headwall. This latter activity suggested a pulse of magma feeding directly into the core of the shear lobe and triggering rockfalls off the summit headwall (Fig. 9c). This alternation between spine extrusion and headwall break-up, on a

GROWTH PATTERNS AND DOME EMPLACEMENT

133

Fig. 18. Schematic diagram illustrating extrusion of the 13 December 1996 lobe. Section is along line X-Y on Figure 16a. (a) Dome configuration on 3 December 1996, with earthquake swarms indicating endogenous activity but no clear evidence for dome growth at surface. Magma within conduit continues to degas and crystallize until remobilized. (b) Dome configuration on 13 December 1996. A pulse of fresh magma has uplifted part of the 1 October 1996 lobe, resulting in the extrusion of a megaspine, the initial growth of the 13 December 1996 lobe. This structure is pushed to the SE, bulldozing through the dome flanks generating rockfalls down the Tar River valley, (c) Dome configuration on 19 December 1996. The 13 December 1996 lobe is overwhelmed and broken up by the extrusion of fresh, hot, blocky lava. Pyroclastic flows inundate the Tar River valley until the waning stages of the pulse, (d) Dome configuration on 28 December 1996. The shear zone feeding the 13 December 1996 lobe starts to plug with crystalline lava. A further pulse of magma emplaces more fluid lava at the summit of the dome (i.e. the 25 December 1996 lobe), rapidly spreading over the 1 October 1996 lobe and 13 December 1996 lobe and spilling lava blocks down the eastern flanks.

near-daily basis, suggested short-term fluctuations in the discharge rate and alternating pulses of more viscous and less viscous magma. An immense peak was gradually constructed that straddled the Galway's Wall area (Figs 29b and 30b). By 21 December, the summit of the 4 November 1997 lobe was c. 1030 m high and the loading of this structure was undermining the strength of the Galway's Soufriere area, which was progressively buried. During a period of intense hybrid seismicity on 26 December, a debris avalanche and major dome collapse (known as the 26 December

1997 or 'Boxing Day' collapse) removed the entire 4 November 1997 lobe (c.55 x 106 m3 non-DRE of dome rock and talus) that had grown since early November (Sparks et al. 2002; Voight et al. 2002). Remarkably, the other sectors of the dome were unaffected by this major flank failure and dome collapse. Vigorous extrusion of blocky lava began to infill the resulting collapse scar immediately (Fig. 3la). Renewed development of a southerly directed lobe, the 27 December 1997 lobe became apparent, and within two months, the southern flanks were rebuilt

134

R. B. WATTS ET AL.

Fig. 19. Cartoon illustrating the development of the 25 December 1996 lobe (see Fig. 16b). Sketches on the left represent a NW-SE cross-section through the dome; sketches on the right represent plan views at the same time, (a) 28 December 1996. The 13 December 1996 lobe has stagnated. The 25 December 1996 lobe punches vertically through the 1 October 1996 lobe and proceeds to spread symmetrically across the summit, overwhelming the 13 December 1996 lobe, (b) 5 January 1997. Lava extrusion becomes more directed, guiding curved slabs to the east and overwhelming the 13 December 1996 lobe and spilling lava blocks down the eastern flanks. The western edge of the 25 December 1996 lobe has become abandoned from the rest of the lobe, (c) 17 January 1997. Oversteepening of the SE dome flanks triggers a dome collapse on 16 January, and ensuing blocky lava infills the horseshoe-shaped scar. Lava extrusion is focused within the scar and directed by a southeasterly shear lobe as a large sector of the 25 December 1996 lobe is abandoned.

(Fig. 30b), although the growth rate had noticeably waned as this period progressed. Development of this Type 1 shear lobe occurred as alternating spine extrusion and rockfalls off the summit headwall suggesting endogenous activity, a similar pattern to that observed in November and December 1997. By late February, several large spines had extruded prior to the formation, in early March, of a prominent 50-m-high spine (informally termed the Galway's Spine) that towered atop the 27 December 1997 lobe (Figs 31b and 32a). Coincident with extrusion of this feature was a marked reduction in rockfall activity, seismicity and ground deformation; all of

these factors signalled the cessation of the first episode of dome growth on about 10 March.

Stage VII: 11 March 1998 to mid-November 1999 A full account of this 20-month interim period between the two phases of dome growth is presented by Norton et al. (2002). Despite no extrusion of fresh lava at the surface throughout this period.

GROWTH PATTERNS AND DOME EMPLACEMENT

135

Fig. 20. (a) 22 January 1997. View of same area as Figure 17a but closer and one month later on, following a period of vigorous blocky growth and subsequent dome collapse. Flower-shaped structure (with radial cracks) is extrusion of massive lava (c. 30m high) of the 21 January 1997 lobe sitting within the 20 January collapse scar (CS). Note the spine of Castle Peak (CP) in right foreground and the nearly vertical fractures within this spine, (b) 25 January 1997. Similar view taken three days later than (a), with the same features marked. The 21 January 1997 lobe developed into a pile of curvilinear, massive lava blocks (typically c. 5 m in size) which has continued to infill and engulf the collapse scar (see Fig. 16c). This blocky morphology was characteristic of a Type 2 shear lobe formed during periods of relatively high discharge rate (>5 m 3 s~'). Many of these blocks tumbled down the SE dome flanks (in foreground) to generate moderate-sized pyroclastic flows down the Tar River valley.

there were still intermittent periods of increased activity, notably the 3 July 1998 dome collapse (Fig. 32b). This large-volume collapse followed three months of quiescence with only rare rockfalls, small pyroclastic flows and subdued seismicity. The collapse involved 20-25 x 10 6 m 3 of lava and formation of a 200-m-deep horseshoeshaped canyon through the dome on its southeastern flank. There was no seismic precursor to this event, which is believed to have been predominantly gravitationally influenced, although heavy rainfall may have been a contributory factor. The collapse was

initially focused in an area weakened by continuous intense fumarolic activity. The southeastern flank was also a steep ridge of loose, blocky lava that formed the eastern rim of the scar from the collapse of 26 December 1997, and was therefore prone to instability. Following the collapse there occurred sporadic degassing and small explosions from a crater at the central base of the scar. A smaller collapse on 12 November 1998 removed c. 3 x 10 6 m 3 of lava from the western dome flanks. The collapse eroded away part of the western dome flanks and joined the 3 July 1998 scar forming a deep

136

R. B. WATTS ET AL.

Fig. 21. (a) 1 April 1997. View looking west above Tar River valley showing the conical form of the dome that developed throughout March 1997. As blocks tumbled eastwards off the active area they eroded deep rockfall channels and chutes into the talus of the lower dome flanks. Note that Castle Peak has been completely overwhelmed by fresh lava. Four days previously, the easterly directed 21 January lobe (JL) stagnated and the 27 March 1997 lobe (EL) started to grow towards Galway's Wall (G). Summit height of dome c. 970 m a.s.l. (b) 6 April 1997, from near the same location as (a). This photo highlights the dramatic contrast between the conical-shaped dome from growth in February and March 1997 (JL) and the initial, smooth appearance of the active 27 March 1997 lobe (EL) sliding out southwards (to the left) from the summit area. This also highlights the contrast in morphology of a Type 1 shear lobe in the early stages (EL), and the later stages (JL) of lobe development (see Fig. 9b,c). In this view, the 27 March 1997 lobe was around 200m long, and 150m wide, with a c. 150m high headwall that generated dome collapses down the White River valley as it grew. The arrow indicates motion. Note the gas plume emerging from along the boundary between the two parts of the dome

corridor-shaped gorge trending ESE-WNW through the entire dome. A further collapse on 20 July 1999, this time directed to the east, removed c. 5 x 106 m3 of lava, mainly off the 27 December 1997 lobe. This event excavated a smaller canyon through the northern flanks that linked up with the main gorge (Norton et al, 2002). Sporadic explosions and minor gravitational collapses continued to occur until the onset of renewed dome growth in midNovember 1999 (onset of Stage VIII; Table 1).

Discussion Petrological and rheological variations Lava dome extrusion is profoundly influenced by magma rheology. We begin our discussion of the observations of the Soufriere Hills dome evolution by summarizing the petrological characteristics of the andesite. These features provide major constraints on the

GROWTH PATTERNS AND DOME EMPLACEMENT

137

Fig. 22. (a) and (b) Maps illustrating the dome configuration in March/April 1997 showing how extrusion of the 27 March 1997 lobe affected the morphology of the dome. Legend as in Figure 5. (c) 6 April 1997. Close-up view of the 27 March 1997 lobe taken from point X on (b). Note the smooth, upper, semi-cylindrical surface of this lobe (BW), and the prominent striations trending parallel to the direction of shear movement, with deep, cross-cutting fractures also evident. Visual observations showed this feature to be sliding out in a stick-slip manner at an estimated 25-30 m day - 1 . This lobe was c. 200 m long (from left to right) and c. 150 m wide with a c. 150 m high, steep headwall. Rockfalls were continually spalling off the near-vertical headwall (H) at the leading edge of the lobe (in the left foreground) as a result of intermittent endogenous/exogenous activity. This lobe represents the only example throughout the g1995-1998 period where the smooth, upper surface remained intact during growth, rather than breaking up into large, curving slabs. JL is stagnated lava from the 21 January 1997 lobe. Theological variations and on the principal factors that control dome extrusion. Throughout the 1995-1999 period, the dome andesite was sampled by collection of blocks from dome-collapse and fountaincollapse pyroclastic flow deposits. Petrological work (e.g. Devine et al 1998; Murphy et al. 2000) highlighted only a minor variation in

bulk composition (SiO2 58.5-62.4%) and phenocryst/microphenocryst content (55-65%). The eruption during 1995-1999 is interpreted to have been driven by an open-system magma body fuelled by the influx of mafic magma into a very crystal-rich magma body resident at shallow crustal levels. Both the thermal input and pressurization due to mafic magma influx (reflected in the 1992-1995

138

R. B. WATTS ET AL.

Fig. 23. (a) 6 April 1997. View looking NW from White River valley area highlighting the destruction of Galway's Wall by growth of the 27 March 1997 lobe. EL marks the steep headwall (c. 150m high) of the lobe in which arcuate cooling fractures and zones of intensely sheared lava are evident. WF is the western dome flanks and S the highest point on the dome (c. 970 m a.s.l.), composed of lava extruded in February and March 1997 (21 January lobe) and now inactive. Note the deep erosional gully in Galway's Wall formed by numerous rockfalls and small pyroclastic flows sourced from the 27 March 1997 lobe, (b) Same view as (a) but taken in mid-May 1997, showing the construction of a broad talus of lava blocks, slabs and small spines that has completely buried the former Galway's Wall and the early plug-type extrusion feature of the 27 March 1997 lobe seen in (a).

seismic crisis and seismic crises in the previous 100 years) was sufficient to remobilize the source region to form a crystal-rich andesite magma. The samples contain 35-45 vol% phenocrysts and c. 20 vol% microphenocrysts within a microlite-bearing high-silicarhyolitic glass matrix. The main difference between samples has been the degree of crystallinity and texture of the groundmass, both factors affected by the magma discharge rate. Between November 1995 and mid-February 1996, the erupted lava had a highly crystalline groundmass with only 5-15% residual rhyolitic glass and typically extruded as large spines. This early phase of activity is interpreted to be the extrusion of degassed lava infilling the conduit from previous injections of magma that triggered the seismic crises. This is supported by the identification of amphiboles with heterogeneous hydrogen isotope compositions in samples from this period

(Harford & Sparks 2001). Samples from periods of rapid dome growth from August 1996 to March 1998 have tended to include higher glass contents (up to 30%). although the glass content range is wide (5-30%). The major pyroclastic flows sampled the deep interior of the dome to depths of 100-200 m. well away from the influence of surface cooling. These samples still exhibit extensive groundmass crystallization that are attributed to degassing rather than cooling (Sparks 1997). The ascent of andesitic magma from the chamber to the nearsurface environment has been modelled (e.g. Sparks 1997; Melnik & Sparks 1999, 2002) and changes in Theological properties are attributed to two intimately linked mechanisms. Gas exsolution due to decompression during slow ascent causes a large increase in the melt phase viscosity (Dingwell et al. 1996). Degassing also triggers

GROWTH PATTERNS AND DOME EMPLACEMENT

139

Fig. 24. Schematic diagram illustrating the dramatic switch in activity that was experienced in mid-May 1997. Sketches on the left represent a N-S cross-section through the dome whilst sketches on the right represent plan views of the dome at the same time.

crystallization in response to the consequent undercooling of the melt phase and this process also increases magma viscosity. The efficacy of both mechanisms increases during ascent and reaches a peak in the uppermost several hundred metres of the conduit, where Melnik & Sparks (2002) predict a zone of large overpressures. Degassing-induced crystallization has already been noted by many workers (e.g. Sparks & Pinkerton 1978; Lipman et al 1985; Wolf & Eichelberger 1997; Gardner el al 1998; Blundy & Cashman 2001) and is invoked for the Soufriere Hills magma (Sparks et al. 2000). However, the potency of this mechanism in andesitic dome-forming eruptions had not been fully appreciated. Throughout most stages of dome growth in the Soufriere Hills eruption, a persistent gas plume was seen emerging from around the dome summit, indicating effective degassing of the magmatic system. The consequence of degassing was to cause a profound rheological stiffening of the magma so that the lava was better characterized as hot, crystalline material with considerable strength (Sparks et al. 2000).

The rheological properties of the Soufriere Hills andesitic magma can be constrained from petrological observations and uniaxial loading experiments on dome samples at high temperature (Sparks et al. 2000). Petrological estimates of magma properties in the magma chamber suggest temperatures of c. 860°C with c. 5% H2O dissolved in the rhyolitic melt phase; thus a viscosity of 7 x 10 6 Pas is estimated, based on experimental results (Dingwell et al. 1996; Pinkerton & Stevenson 1992). Fully degassed and highly crystalline dome samples show highly non-linear deformation behaviour under uniaxial loadings of 9-26 MPa at temperatures of c. 990°C (Sparks et al. 2000). Viscosities in an initial period of steady deformation are of the order of 10 14 Pas, but apparent viscosities decrease rapidly to c. 1011 Pas just prior to failure along shear zones. The Soufriere Hills andesitic magma was already rich in phenocrysts and microphenocrysts in the magma chamber prior to eruption. During periods of slow ascent, degassing-induced

140

Fig. 25. Sequence of maps illustrating the growth of the dome in the aftermath of the 25 June 1997 collapse and highlighting the development and subsequent infilling of the explosion crater throughout August 1997. Legend as in Figure 5.

crystallization of magma in the upper conduit reached the threshold when crystals formed a touching framework. At this stage the magma transformed from a Newtonian fluid to a much more viscous non-Newtonian fluid with mechanical strength (Lejeune & Richet 1995). In contrast, a sufficiently fast magma ascent rate reduced the

time for microlite crystallization to the extent that the critical threshold was not attained and magma could extrude in a more fluidlike manner. Thus, during a period of fluctuating discharge rates, the flow of crystalline magma may switch between a fluid nature and that of a near-solid, non-Newtonian nature. Such an oscillating

GROWTH PATTERNS AND DOME EMPLACEMENT

141

Fig. 26. Maps showing the configuration of the 21 September 1997 collapse crater and subsequent growth of the 22 October 1997 lobe from 22 October to 3 November 1997. Legend as in Figure 5.

state is predicted by models of the non-linear dynamics of conduit flow (Denlinger & Hoblitt 1999; Melnik & Sparks 1999, 2002; Voight et al. 1999; Wylie et al. 1999). A consequence of this variation is heterogeneous deformation during lava emplacement, with the formation of spines and megaspines at slow rates (for nonNewtonian lava) and the development of shear lobes emplacing more ductile lava as large blocks and smooth slabs or pancakeforms at faster rates.

Morphologic variation A spectrum of extrusive features was observed throughout the 1995-1998 episode of dome growth, and together these structures can be considered to represent a morphologic continuum (Fig. 33). Each structure was essentially composed of the same components (i.e. a smooth, semi-cylindrical backwall and a steep, blocky headwall), and their growth bounded along shear faults was accommodated in the same manner, with the exception of pancake forms at more vigorous rates. Periods of growth varied widely, however, with spine and megaspine formation occurring in a few days, whilst individual shear lobe evolution operated over several weeks and months. The main difference between these structures related to their size and there appeared to be a link between the formation of each structure and the level of eruptive activity (Fig. 34). Only occasionally did the lava morphology and behaviour suggest a more fluid-like emplacement, with axisymmetric lateral spreading

of pancake lobes at the summit area during periods of more vigorous discharge rates (Fig. 33). The activity of late December 1996 provided the best example of this, with emplacement of the 25 December 1996 lobe occurring at a time when seismic tremor was commonly experienced. A similar relationship, linking growth behaviour to discharge rate, has also been determined in experiments using a Bingham plastic analogue (Griffiths & Fink 1997; Fink & Griffiths 1998). These laboratory experiments reproduced many of the structures observed in the Soufriere Hills eruption, although their formation was attributed primarily to variations in viscosity and the thickness of the cooled dome carapace, whereas here we relate the variations to the combined effects of discharge rate, cooling and degassing and related changes in the rheological properties of the magma. On considering the morphologic continuum, a gradual transition from highly crystalline structures (at low discharge rates) to those exhibiting more fluid-like features (at higher discharge rates) exists. By far the most predominant style of activity was the formation of shear lobes at moderate discharge rates, although two distinct classifications have been observed. In Type 1 shear lobes (Fig. 33), growth develops a stable structure that may grow over many weeks and months, constructing a lobe with a core of massive lava and steep talus flanks from semi-continuous rockfall activity. Such activity is prevalent during periods of long-term average extrusion rates of 2-5 m3 -1. In contrast, during periods of fluctuating extrusion rates from < l m 3 s - 1 to > 5 m 3 s - 1 ( e . g . July to August 1996), emplacement of megaspines would be rapidly followed by

R. B. WATTS ET AL.

142

(a)

Fig. 27. (a) 28 September 1997. View looking south from the north (near Harris Village), showing amphitheatre in the dome following the 21 September 1997 collapse. Breach in the dome faces into the head of Tuitfs Ghaut (T). Y represents the headwall of the stagnated 17 May 1997 lobe and Z the headwall of the stagnated 13 August 1997 lobe (see Fig. 26a). The main crater is c. 300m in diameter, (b) 6 November 1997. High aerial view looking towards southwest at the northern flanks of the dome showing the 22 October 1997 lobe, with headwall of massive lava sitting within the breached part of the 21 September 1997 crater. X marks the smooth, curving upper surface of this lobe. T, Y and Z represent the same features marked in (a). Activity had completely stagnated on the northern flanks, with major collapses affecting the southern flanks of the dome.

Type 2 shear lobes of blocky lava (Fig. 33). At such periods, the rapid sequential emplacement of these very different structures commonly triggered dome collapses with major pyroclastic flow activity. As the eruption progressed, a steadier discharge rate became established, promoting the formation of Type 1 shear lobes. As a result, long periods of lobe construction, with associated accumulation of rockfall debris, would occur with only minor pyroclastic flow activity (e.g. January to March 1998). This latter style of dome growth has been particularly predominant in the second phase of dome growth that commenced in November 1999.

Another factor to consider when explaining the difference in activity between the earlier and later stages of the eruption may be the former presence of the Castle Peak dome and its gradual destruction and complete burial by early February 1997. In the early eruptive stages, emplacement was partly controlled by the topography of Castle Peak. The directed explosion of 17 September 1996 was also an important moment in the eruption. Not only was it the first magmatic explosive activity, it completely exposed the head of the main conduit to surface conditions, as well as widening the uppermost parts of the conduit. Prior to this event, magma had

GROWTH PATTERNS AND DOME EMPLACEMENT

143

Fig. 28. Schematic diagram illustrating the hypothesized growth of the dome during extrusion of the 22 October 1997 lobe and the 4 November 1997 lobe. The cross-sectional view is along the line X-Y on Figure 26a. (a) 23 October 1997 (see Fig. 26a). Renewed slow dome growth in the central crater following a month-long period of Vulcanian explosions, (b) 29 October 1997. Gradual extrusion of 22 October lobe, initially guided vertically by backwall of crater and subsequently projecting to the north, infilling the breached part of the crater and directing rockfalls down Tuitfs Ghaut. As the 22 October 1997 lobe extrudes subhorizontally away from the crater, two small hummocks of lava extrude at the central summit area (see Fig. 26b). (c) 4 November 1997. Growth of the 22 October 1997 lobe and central hummocks has stopped while the southern dome flanks are destabilized by intrusion of the 4 November 1997 lobe generating a dome collapse down the White River valley, (d) 6 November 1997. The southern dome flanks have been completely destroyed, whilst all the other flanks of the dome are unaffected. As the 4 November 1997 lobe continues to extrude along a southerly directed shear fault, fresh lava breaks off the lobe's leading edge causing a further collapse directed down the White River valley, (e) 21 December 1997. Growth of the 4 November 1997 lobe has continued for almost two months, constructing a large peak which now straddles the former Galway's Wall.

to force its way through the fractured body of the Castle Peak dome before extruding.

Structural control on emplacement The dominant mode of emplacement following 17 September 1996 was through the formation of shear lobes bounded by large arcuate faults. During emplacement, the lava extruded sporadically in a stick-slip manner along curved fault structures, which are interpreted as shear faults that are sourced from the sides of the conduit. Similar features have been produced in analogue experiments by Donnadieu & Merle (1998) investigating the deformation of a volcanic edifice through forced intrusion of viscous magma. As indentation proceeded, asymmetric deformation generated a curved shear fault from the base to the outer edge of the edifice, with the fault controlling the directed emplacement of the magma analogue. This mechanism was postulated by Donnadieu & Merle (1998) to explain the cryptodome intrusion in the build-up to the Plinian eruption of Mount St Helens in May 1980. Observations of the Soufriere Hills dome indicate that a similar mechanism may be responsible for the growth and development of individual lobes during construction and destruction of the dome. Shear faults have not been recognized in lava domes before. An explanation for this may be the poor preservation potential of the shear surfaces. In the Soufriere Hills dome, a lobe was commonly broken up in the late stages of emplacement or buried by blocks from a later eruptive episode. The structure of a shear lobe is sometimes preserved and, at the time of writing (October 2000),

the current configuration of the Soufriere Hills dome still exhibits part of the smooth backwall of the 4 November 1997 lobe extruded in October 1997. Domes at other volcanoes also have preserved examples, such as a 55-m-long, 25-m-wide striated lobe emplaced near the summit of the Chinois Dome that formed in the 1929-1932 eruption of Mont Pelee, Martinique. In some cases, a single shear lobe may form the entire preserved part of a dome; the 34-ka Perches dome in the Soufriere Hills complex on Montserrat and the Gros Piton on St. Lucia are both examples of near-vertical shear lobes.

Growth stages and cycles During the eruption the dome has increased in volume, notwithstanding the counteracting effects of collapses. Growth has also been characterized by cyclic patterns, with repeated switches in the direction of lobe extrusion and pulsations in discharge rate. Here we discuss the interplay of individual pulses of lobe extrusion with the overall construction of the dome and the nature of the cyclic growth patterns. The early stages of the first episode of dome growth involved the gradual destruction and burial of the previous construct, the Castle Peak dome. As eruptive vigour increased, the directed extrusion of megaspines and subhorizontal shear lobes occurred in a non-systematic radial pattern around the central conduit. These structures armoured the lower dome flanks, acting as a foundation to the large subvertical shear lobes that dominated extrusion in the later eruptive stages. Upon stagnation, the broad headwall of each

144

R. B. WATTS ET AL.

Fig. 29. Maps illustrating the collapse resulting from growth of the 4 November 1997 lobe and the extent of regrowth following this event. Legend as in Figure 5.

lobe would form the steep upper ramparts on a sector of the volcano. This process of shear-lobe growth and stagnation continued around all sectors of the dome, so as eventually to construct a central depression, which was further modified by explosive activity in August, September and October 1997. This depression developed as a result of this repeated lobe formation, with each lobe directed away from the central conduit. In effect, the remnants of the smooth outer lobe surfaces had stagnated and merged together to form the inner walls of the central depression. In association with this process, each of the headwalls of the lobes stagnated and merged together to form the steep upper flanks of the dome. As extrusion of individual lobes constructed steep flanks, extrusion would either continue in the same direction, partly stagnate and widen the headwall to one side of the lobe, or completely stagnate and switch the focus of activity to another sector. Major switches of dome growth direction can, in several cases, be clearly linked to cyclic behaviour of the volcano. Throughout most of the eruption, and particularly in the latter half of the 19951998 period of dome growth, a distinct five to seven-week cycle of activity was recognized, which was intimately linked to hybrid seismicity and a pronounced increase and decrease in the period of tilt cycles (Voight et al. 1999). Typically, the onset of a cycle was marked by a period of intense hybrid seismicity that commonly resulted in a dome collapse event and a pronounced increase in discharge rate, In the weeks that followed, the amplitude and period of tilt cycles decreased and increased respectively and seismicity declined. Aseismic growth of the active lobe infilled the collapse scar, initially as rapidly emplaced blocky lava and small spines that

developed into a broad shear lobe with oversteepened flanks. As predominantly aseismic growth progressed, the discharge rate would gradually wane over several weeks, until dropping below a critical threshold. At this point, degassing-induced crystallization could operate more effectively to form a near-solid plug in the upper conduit, contributing to the retardation of surface extrusion. The onset of a new cycle is then attributed to a pulse of fresh, less viscous magma, rising up from the deep source. The new magma would reach the base of the consolidated plug and the resistance to upward flow marked the onset of hybrid seismicity. Pressure build-up beneath the plug would continue over a few hours to several days, as highlighted by intensifying swarms of hybrid earthquakes until reaching a critical limit that pushed the plug into the dome. The discharge rate rapidly increased as less viscous magma filled the conduit and the plug material was pushed out of the way. The rapid increase in flow rate and pressurized conditions in the upper conduit were, in many cases, enough to rapidly destabilize a lobe head wall and trigger a major dome collapse. The fatal 25 June 1997 dome collapse illustrates a typical growth cycle and, in the weeks leading up to the event, aseismic growth of a northerly directed lobe had primed the northern flanks for a collapse (Fig. 35a). In the weeks following the collapse, rapid redevelopment of a fresh northerly lobe was observed (Fig. 35b), with hybrid seismicity disappearing, thus initiating the next cycle (Fig. 35c). In this interpretation, the fluctuations in discharge rate observed during each cycle were a consequence of pulsations of fresh, low-viscosity and probably gas-rich magma released from the deep source with hybrid seismicity being a signature for the extrusion of a near-solid

GROWTH PATTERNS AND DOME EMPLACEMENT

145

(a)

Fig. 30. (a) 8 November 1997. View looking north directly at the headwall (NL) of the 4 November 1997 lobe (c. 150m high). Dotted line marks the original southern dome flanks immediately prior to extrusion of the 4 November 1997 lobe. Note also its curved upper edge (BW) and the vigorous ash-venting emanating from the rear of this lobe, (b) 27 February 1998. This excellent view of the southern flanks, taken high above the White River valley, highlights the dome configuration almost as it was immediately prior to the 26 December 1997 collapse. This shot was taken many weeks following this collapse; however, subsequent growth infilled the collapse scar (BD) and rebuilt the southern flanks in a near-identical manner. Note the characteristic rockfall chutes leading down from the blocky summit and eroding into the talus. Pyroclastic flow deposits from the 21 September 1997 collapse can be seen in the background, with Chances Peak (ChP) and Galway's Mountain (GM) dome also in view.

plug into the dome, possibly by stick-slip movement along the walls of the upper conduit (Denlinger & Hoblitt 1999; Voight et al. 1999; Wylie et al. 1999). Models of dome growth The observations of the growth structures and morphological development of the Soufriere Hills lava dome are now discussed in

the context of concepts and models of dome growth. Two main concepts have been developed to explain dome growth. First, the role of surface cooling with formation of a resisting crust have been explored by Fink & Griffiths (1992) in a series of laboratory experiments and by scaling analyses of force balances (summarized in Griffiths 2000). Second, the linked roles of gas exsolution and degassing-induced crystallization have been considered in textural studies (Cashman 1992; Cashman & Blundy 2000; Hammer et al. 2000) and theoretical models of conduit flow during dome extrusion

146

R. B. WATTS ET AL.

Fig. 31. Maps illustrating the dome configuration in the aftermath of the 26 December 1997 collapse and the focus of regrowth following this event until the cessation of growth around 10 March 1998. Legend as in Figure 5.

(Sparks 1997; Melnik & Sparks 2002). In both concepts, rheological stiffening and the onset of non-Newtonian rheology with the development of a yield strength are important. Here we discuss the relative roles of cooling and degassing-induced crystallization in controlling the dynamics of the Soufriere Hills lava dome and morphological features. This discussion develops that by Sparks el al (2000). In Griffiths (2000) similar phenomena and range of dome morphologies to those reported here are attributed to development of a cooled crust. Fink & Griffiths (1998), for example, demonstrate a similar spectrum of structures to that observed at Soufriere Hills, from spiny lobate domes to smooth-spreading pancake morphologies, by changing a dimensionless parameter which reflects the relative importance of heat advection (and therefore discharge rate) and heat loss through a cooling crust. Here we invoke a different view that this morphological spectrum is a consequence of different amounts of both degassing-induced crystallization and cooling. For lava extruded at the lowest rates, the slowly rising magma had time to lose more gas and crystallize efficiently, such that solidification was largely completed (90-95%) within the upper conduit, resulting in the emplacement of a spine or a megaspine. Cooling played a negligible role in this case. For faster-rising magma, lesser amounts of microlite crystallization occurred during ascent, promoting the formation of shear lobes. At the fastest ascent rates, lava extruded in a more fluid-like manner and a pancake-type morphology resulted. In this case, emplacement could be controlled both by degassing-induced crystallization and external cooling.

Another argument for the importance of a cooling crust put forward by Griffiths (2000) is that many lava domes show an increase in height with time, as in the case of Soufriere Hills (Fig. 3a). This is attributed to growth of a crust of increasing thickness. By comparing the forces driving lateral spreading of a dome with the yield strength of a cooled crust, a scaling result is obtained whereby the height is proportional to time to the power of 0.25. Griffiths (2000) shows that several domes indeed exhibit a power-law dependence, with a power approaching 0.25. Such an analysis is problematic for Soufriere Hills because the dome collapsed many times (see Fig. 3a). It is more meaningful to take a single episode of growth. For the period mid-February to August 1996. height data yield a best-fit power law with the exponent of 0.36. In another example, the period 1 October 1996 to 5 November 1996, the height data give a power exponent of 0.44. In neither case are these values close to that predicted by the cooling-crust model of Griffiths (2000). There are several problems with applying the cooling-crust model to Soufriere Hills. Firstly, the overall increase in height with time does not result from the growth of a single entity with thickening and strengthening of a cooled crust. The dome collapsed and new lobes extruded many times, so the scaling analysis and the growth of a single, cooling dome structure cannot be applied to the whole dome growth of 1995-1998. Secondly, the cooling-crust model is a static one, whereas Stasiuk & Jaupart (1997) and Melnik & Sparks (2002) show that there are important dynamic controls on dome height. The increase in height with time is attributed to increasing magma chamber pressure by Melnik & Sparks (2002),

GROWTH PATTERNS AND DOME EMPLACEMENT

147

(a)

(b)

Fig. 32. (a) May 1998. View looking NW from South Soufriere Hills at the southern flanks of the dome, composed entirely of the 27 December 1997 lobe. Galway's Spine (GS) on top of the blocky flanks is 50m high and 45m wide and its extrusion in early March signalled the cessation of dome growth. Chances Peak (ChP) is to the left, and eastern flanks of the dome are on the far right. The scar rim of the 26 December 1997 collapse is marked BD. (b) View looking NW from above the southern rim of English's Crater in early February 1999. The entire SE dome flanks have collapsed away exposing a vertical section (c. 200m high) through the hummocks extruded as part of the 22 October 1997 lobe (OL). BD marks the same location as in (a), NL marks the remaining part of the 27 December 1997 lobe and the summit of the new dome at 977m a.s.l.

who consider it to be unrelated to surface cooling. Thirdly, consideration of height versus time for the 1 October 1996 lobe gives a power law with an exponent of 0.44, significantly larger than the 0.25 value expected by a cooling model. Likewise, the height versus time data can also be analysed by a dynamical model (Melnik & Sparks 2002) without invoking cooling. There is a more general problem in that several different models can give quite similar power-law behaviour, so that finding a power law exponent of 0.25 is not sufficient to demonstrate that cooling is the dominant effect. Indeed

a simple dynamical model, where discharge rate is a linear function of dome height, gives a height versus time relationship quite close to a power law with exponent of 0.25. Fourthly, the cooling model requires very high crustal strengths (>108 Pa; Griffiths 2000), which are hard to reconcile with laboratory data on rock strengths. For example, geotechnical measurements (Voight et al. 2002) and experiments simulating explosions (Alidibirov & Dingwell 2000) indicate strengths of about 10 6 Pa for dome samples. Fracturing of the cooled crust and development of tensional cooling stresses further

R. B. WATTS ET AL.

148

EMPLACEMENT FEATURE + ASSOCIATED SEISMICITY A. NEAR-VERTICAL SPINE -growth over 2-3 days -coincident with periodic hybrid earthquake swarms

B. WHALEBACK STRUCTURES -growth over 4-5 days -coincident with repetitive hybrid earthquake swarms

C. MEGASPINE -emplaced over 2-3 days -generally aseismic growth occasionally with hybrid seismicity D. SHEAR LOBE-TYPE 1 -growth (broad spines) over many weeks to months, in form of intermittent endogenous + exogenous pulses -generally aseismic growth, often intense hybrid earthquake swarms prior to lobe collapse

E. SHEAR LOBE-TYPE 2 -growth (blocky lava) over days to weeks, also with endogenous pulses -growth and collapse often coincident with repetitive hybrid earthquake swarms + tremor F. PANCAKE LOBE -growth over 4-6 days -emplacement coincident with repetitive hybrid earthquake swarms + tremor G. EXPLOSIONS -commonly occur following rapid, large dome collapses -see Druitt et al. (2002)

Fig. 33. Variation in the type of structure emplaced in relation to the average discharge rate, and the relative roles of degassing-induced crystallization and cooling. Note that the boundaries between eruption rates are arbitrarily defined.

GROWTH PATTERNS AND DOME EMPLACEMENT

149

Fig. 34. Graph showing the structures extruded throughout the eruption. This highlights a relationship between the type of structure extruded and average discharge rate suggesting that changes in dome morphology may be a crude proxy for estimating discharge rate during the eruption.

weaken a cooled crust, an effect inconsistent with the high strength indicated from the cooling-crust model. While we concur that cooling can have a major role in many instances of lava emplacement, we suggest that it only becomes a factor during the most rapid periods of dome growth at Soufriere Hills.

Comparison with other documented dome eruptions The growth of the Mount St Helens dacitic dome between 1980 and 1986 allowed the first intense scientific monitoring campaign of a lava dome this century. Swanson & Holcomb (1990) detailed the complex growth history of the dome and described distinct patterns throughout its duration. Episodic lobe extrusion generally lasting several days was the predominant style; however, a one-year period of continuous endogenous growth was also observed. Three distinct periods of episodic growth showed a linear pattern in relation to the long-term growth rate. Each extrusion was generally preceded by a one to three week period of endogenous activity that often faulted and fractured the dome carapace and caused deformation of the crater floor. Between growth episodes, the dome would slowly spread laterally and subside as its hot, ductile core deformed due to gravitationally induced stress. The generally consistent shape of the dome for most of the eruption suggested a possible controlling mechanism. Iverson (1990) modelled the dome morphology as a core of relatively low strength enclosed by a strong, brittle carapace. This cooled, outer surface skin and the net effective viscosity of the lava were believed to play an important role in determining the location and manner of lobe emplacement on the dome. Variations in the surface texture of the lava throughout the 1980-1986 eruption also highlighted a relationship between the underlying slope and the water content of the lava (Anderson & Fink 1990). Several types of lobes were distinguished, each type characterized by varying degrees of smooth and scoriaceous surface textures. Growth was often focused along a large, smooth fracture known as a crease structure that formed predominantly on shallow slopes. A progressive increase in the formation of smooth-surfaced lava observed during the eruption was attributed to more effective degassing of the magma during ascent and emplacement. The features formed on the Mount St Helens dacite dome and mechanisms responsible for their formation show only limited similarity to those observed at the Soufriere Hills andesite dome. The more recent 1991-1995 dome-forming eruption of Mount Unzen, Japan, provides another comparison (Nakada et al. 1999). This dome grew on a steep and unstable slope such that the dome sporadically partially collapsed generating pyroclastic flows. During this period of dome growth, magma extrusion was near-continuous, involving the exogenous extrusion of 13 distinct lobes. Initial growth

involved spine formation and subsequent collapse into blocks; however, later lobes commonly exhibited crease structures indicating plastic deformation. Endogenous growth was prevalent at periods of low discharge rate, notably in the later stages of the eruption, and the style of endogenous growth was compared to the formation of basaltic lava pillows on a much larger scale (Nakada et al. 1995). The overall discharge rate followed a gradually declining trend with time, overprinting a pattern dominated by two distinct pulses (Nakada & Motomura 1999). The style of growth, switching between exogenous and endogenous activity, broadly paralleled this trend. At times of relative quiescence, a viscous plug would develop in the upper conduit, probably in response to shallow degassing and microlite crystallization. Groundmass crystallization of the magma below this plug raised the overpressure beneath the plug to a critical threshold and extruded the viscous plug, marking the onset of a fresh pulse in activity (Nakada & Motomura 1999). A similar mechanism was invoked to explain the extrusion of a large spine onto the endogenous dome summit of Mount Unzen in December 1994, an event that signalled the end of the eruption. Thus degassing, groundmass crystallization and consequent rheological stiffening were also key influences at Mount Unzen, as proposed here for the Soufriere Hills lava dome. The short eruption in 1989-1990 of Redoubt Volcano, Alaska, involved the rapid growth and destruction of 13 silicic-andesite domes (Miller 1994). The remote and hazardous location of this volcano, perched precariously within a steep amphitheatre, hampered visual observations. However, the blocky nature of the domes, and vigorous eruptive degassing, indicate that similar ascent and emplacement mechanisms to those detailed here may have been operating at Redoubt. The 1951-1952 dome-forming eruption of Mount Lamington in Papua New Guinea involved the development of structures similar to those documented in the Soufriere Hills eruption. After an initial paroxysmal explosion, dome growth involved the rapid extrusion of near-solid lava constructing a broad dome (Taylor 1958). In the later eruptive stages, localized extrusions took place in a piecemeal fashion, forming multiple peaks on the dome. This process of directed extrusion, and the late-stage formation of large 'hogs-back' features, bears a distinct resemblance to the growth mechanisms described here. Observations by Ferret (1935) throughout the 1929-1932 eruption of Mont Pelee (Martinique) also highlights similarities to the formation of the Soufriere Hills dome. Ferret documented the gradual construction of a broad lobe at the head of the Riviere Seche through a combination of direct observation and photography. From close proximity, he was able to describe the growth of individual spines and detail their variable nature, with solid outer faces and viscous interiors. Eruptive activity did not involve any

150

R. B. WATTS ET AL.

(a)

(b)

Fig. 35. (a) Development of the dome in the build-up to the 25 June 1997 dome collapse. This schematic view is from high above the eastern dome flanks looking westwards. This period of activity was marked by poor visibility and only limited views of the summit area were possible. Shaded area is the active 17 May 1997 lobe, (b) Following the dome collapse on 25 June 1997. rapid development of the 27 June 1997 lobe (shaded) was apparent infilling the collapse amphitheatre. The steep flanks on the northern side had been reconstructed by 7 July 1997. as seen during a brief window of good visibility on that day. (c) Graph highlighting the anti-correlation between periods of rapid lobe growth with the frequency of hybrid earthquakes for the period May to July 1997. Note that the 25 June 1997 dome collapse occurred during a period of intense hybrid earthquake activity such that individual hybrid events were merging together and could not be recorded separately. For this reason, the collapse event does not appear to occur at a period of peak earthquake activity.

large dome collapses or switches in activity, as have characterized Soufriere Hills. However, the directed extrusion of crystal-rich andesitic spines and broad lobes on the southern flanks of Mont Pelee suggest similar controlling mechanisms on dome growth. Perhaps the closest analogy to Soufriere Hills is one of the most active lava domes in recent history, namely Mount Merapi on the island of Java, Indonesia. Throughout the twentieth century, intermittent activity involved the gradual effusive growth and partial collapse of the summit lava dome, occasionally causing significant loss of life. A wealth of descriptive data exists for its previous eruptive activity, as summarized in Voight et al. (2000). Merapi is noted for the formation of individual lobes, variable scales of dome COllapse 7 and its ability to switch the focus of activity to a different sector of the edifice. Observations at the summit indicate the presence of large, elliptical-shaped lobes of crystal-rich andesitic

lava nestled within horseshoe-shaped collapse scars. All of these features have been documented during the Soufriere Hills eruption.

Conclusions This eruption provided invaluable improvements to our understanding of the processes operating during the ascent and emplacement of crystal-rich intermediate lavas. It has also highlighted the need for further research to elucidate the links between dome growth and associated seismic manifestations (e.g. Hidayat et al. 2000), that may result in even better real-time diagnostic capability. The recognition of directed extrusion along shear zones helps to explain the common occurrence of dome collapses and debrisavalanche-forming sector collapses that are a prominent feature

GROWTH PATTERNS AND DOME EMPLACEMENT of andesitic volcanism. Indeed, theories described here have also advanced our understanding of the mechanisms that control dome collapse and the triggering of dome-collapse pyroclastic flows. The most significant observations are summarized below.

151

D. Swanson and T. Druitt for their constructive reviews. This paper is dedicated to the remarkable spirit and resilience of the Montserratian population and to the memory of the nineteen people who lost their lives on 25 June 1997. R.S.J.S. acknowledges a NERC Professorship and R.B.W. a University of Bristol studentship. Published by permission of the Director of the British Geological Survey.

Controls on ascent and emplacement of andesitic lava-degassing versus cooling The evidence from the Soufriere Hills eruption indicates that the processes operating during shallow magma ascent are so effective that the Soufriere Hills andesite extrudes as one of the most viscous forms of lava on the Earth's surface. Degassing and subsequent microlite crystallization of the crystal-rich andesitic magma feeding the dome operate such that significant rheologic stiffening of the magma occurs. Emplacement switches between hot, crystalline material with considerable strength (at slow ascent rates) and a viscous Newtonian fluid (at faster ascent rates). Cooling plays only a minor role in the emplacement of crystal-rich andesitic lava.

Morphologic variation during emplacement The growth of the dome was predominantly governed by extrusions of near-solid spines at low discharge rates and shear lobes at moderate discharge rates. On emplacement, these structures were bounded by ductile shear faults that possibly correspond to the sidewalls of the main conduit. Lava was emplaced at the dome summit in the form of these structures, which sporadically broke up into blocky areas during growth or at the post-emplacement stage. Infrequently, some lateral spreading of more mobile, fresh magma was evident, forming a pancake-like, smooth-surfaced lobe, such as the 25 December 1996 lobe. The higher gas content and lower crystal content of the magma at the head of a fresh pulse could be important factors in explaining more fluid-like behaviour.

Structural controls on emplacement The observations described support the stick-slip mechanisms of dome growth proposed by Denlinger & Hoblitt (1999), Voight et al. (1999) and Wylie et al. (1999). In essence, the shallow ascent of magma involves it being pushed to the surface in a piston-like manner by magmatic pressure. Surface extrusion of the lava was accommodated along shear faults, either in a vertical manner (as spines) or subhorizontally away from the central conduit (as shear lobes) or as blocky lava often during growth spurts. Mapping of these structures highlighted no clear pattern as to the direction in which emplacement occurred. A simple theory is that the extrusion takes place along the path of least resistance (e.g. emplacement of the 22 October 1997 lobe into the open breach of the crater). The orientation of the curved faults is governed by stress distributions, as proposed by Donnadieu & Merle (1998). A new direction of faulting developed when a pulse of fresh, less viscous magma was impeded by a plug of crystalline, stagnated lava. The imposing size of these structures, the complex manner in which they are emplaced, and their ability to stagnate and switch the focus of extrusion pose problems in forecasting hazards during the active stages of such eruptions. The destructive potential of shear lobes (as demonstrated by growth of the 27 March 1997 lobe and its demolition of the southern dome flanks) deserves consideration in monitoring future dome-forming eruptions. The authors would like to thank all staff of the Montserrat Volcano Observatory, especially members of Team Volume and Team Seismic throughout the eruption. Special mention goes to B. Darroux of the Montserrat Police Force who is responsible for the impressive photographic collection, particularly in the early eruptive stages. Thanks also to G. Skerritt of the Surveys Department who collected most of the theodolite data listed in Table 3, and to S. Powell at the Imaging Unit, Department of Earth Sciences, University of Bristol. We also acknowledge J. Fink,

References ALIDIBIROV, M. & DINGWELL, D. B. 2000. Three fragmentation mechanisms for highly viscous magma under rapid decompression. Journal of Volcanology and Geothermal Research, 100, 413-421. ANDERSON, S. W. & FINK, J. H. 1990. The development and distribution of surface textures at the Mount St. Helens dome. In: FINK, J. H. (ed.) Lava Flows and Domes, IAVCEI Proceedings in Volcanology, 2. Springer-Verlag, New York, 25-46. ANDERSON, S. W. & FINK, J. H. 1992. Crease structures - indicators of emplacement rates and surface stress regimes of lava flows. Geological Society of America Bulletin, 104, 615-625. BLUNDY, J. & CASHMAN, K. 2001. Ascent-driven crystallisation of dacite magmas at Mount St. Helens 1980-1986. Contributions to Mineralogy and Petrology, 140, 631-650. CALDER, E. S., LUCKETT, R., SPARKS, R. S. J. & VOIGHT, B. 2002. Mechanisms of lava dome instability and generation of rockfalls and pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKElaar, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 173-190. CASHMAN, K. V. 1992. Groundmass crystallization of Mount St. Helens dacite 1980-1986 - a tool for interpreting shallow magmatic processes. Contributions to Mineralogy and Petrology, 109, 431-449. CASHMAN, K. & BLUNDY, J. 2000. Degassing and crystallization of ascending andesite and dacite. Philosophical Transactions of The Roval Society, 358, 1487-1513. COLE, P. D., CALDER, E. S., SPARKS, R. S. J. ET AL. 2002. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 231-262. DAAG, A. S., DOLAN, M. T., LAGUERTA, E., MEEKER, G. P., NEWHALL, C. G., PALLISTER, J. S. & SOLIDUM, R. 1996. Growth of a postclimactic lava dome at Pinatubo Volcano, July-October 1992. In: NEWHALL, C. & PUNONGBAYAN, R. (eds) Fire and Mud: Eruptions and Lahars of Mount Pinatubo, Philippines. University of Washington Press, Seattle, 647-664. DENLINGER, R. P. & HOBLITT, R. P. 1999. Cyclic eruptive behaviour of silicic volcanoes. Geology, 21, 459-462. DEVINE, J. D., MURPHY, M. D., RUTHERFORD, M. J. ET AL. 1998. Petrologic evidence for pre-eruptive pressure-temperature conditions, and recent reheating, of andesitic magma erupting at the Soufriere Hills Volcano, Montserrat, West Indies. Geophysical Research Letters, 25, 3669-3672. DINGWELL, D. B., ROMANO, C. & HESS, K.-U. 1996. The effect of water on the viscosity of a haplogranitic melt under P-T-X conditions relevant to silicic volcanism. Contributions to Mineralogy and Petrology, 124, 19-28. DONNADIEU, D. & MERLE, O. 1998. Experiments on the indentation process during cryptodome intrusions: New insights into Mount St. Helens deformation. Geology, 26, 79-82. DRUITT, T. H., YOUNG, S. R., BAPTIE, B. ET AL. 2002. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306. FINK, J. H. & GRIFFITHS, R. W. 1992. A laboratory analog study of the morphology of lava flows extruded from point and line sources. Journal of Volcanology and Geothermal Research, 54, 19-32. FINK, J. H. & GRIFFITHS, R. W. 1998. Morphology, eruption rates, and rheology of lava domes: Insights from laboratory models. Journal of Geophysical Research, 103, 527-545. GARDNER, C. A., CASHMAN, K. V. & NEAL, C. A. 1998. Tephra-fall deposits from the 1992 eruption of Crater Peak, Alaska: implications of clast textures for eruptive processes. Bulletin of Volcanology, 59, 537-555. GRIFFITHS, R. W. 2000. The dynamics of lava flows. Annual Review of Fluid Mechanics, 32, 477-518. GRIFFITHS, R. W. & FINK, J. H. 1997. Solidifying Bingham extrusions: a model for the growth of silicic lava domes. Journal of Fluid Mechanics, 347, 13-36.

152

R. B. WATTS ET AL.

HAMMER, J. E., CASHMAN, K. V. & VOIGHT, B. 2000. Magmatic processes revealed by textural and compositional trends in Merapi dome lavas. Journal of Volcanology and Geothermal Research, 100, 165-192. HARFORD, C. L. & SPARKS. R. S. J. 2001. Recent remobilisation of shallowlevel intrusions on Montserrat revealed by hydrogen isotope composition of amphiboles. Earth and Planetary Sciences Letters, 185, 285-297. HARFORD, C. L., PRINGLE. M. S., SPARKS, R. S. J. & YOUNG, S. R. 2002. The volcanic evolution of Montserrat using 4 0 Ar, 3 9 Ar geochronology. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption ofSoufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London. Memoirs, 21, 93-113. HIDAYAT, D., VOIGHT, B., LANGSTON. C., RATDOMOPURBO, A. & EBELING, C. 2000. Broadband seismic experiment at Merapi Volcano, Java, Indonesia: very-long-period pulses embedded in multiphase earthquakes. Journal of Volcanology and Geothermal Research, 100, 215-231. IVERSON, R. M. 1990. Lava domes modeled as brittle shells that enclose pressurized magma, with application to Mount St Helens. In: FINK, J. H. (ed.) Lava Flows and Domes, IA VCEI Proceedings in Volcanologv. Spnnger-Verlag, New York, Vol. 2, 47-69. LEJEUNE, A.-M. & RICHET, P. 1995. Rheology of crystal-bearing silicate melts: An experimental study at high viscosities. Journal of Geophysical Research, 100,4215-4229. LIPMAN, P. W., BANKS, N. G. & RHODES, J. M. 1985. Degassing-induced crystallization of basaltic magma and effects on lava rheology. Nature, 317, 604-607. LOUGHLIN, S. C.. CALDER, E. S.. CLARKE, A. B. ET AL. 2002. Pyroclastic flows and surges generated by the 25 June 1997 dome collapse, Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 191-209. MELNIK, O. & SPARKS, R. S. J. 1999. Nonlinear dynamics of lava dome extrusion. Nature, 402. 37-41. MELNIK, O. & SPARKS, R. S. J. 2002. Dynamics of magma ascent and lava extrusion at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano. Montserrat, from 1995 to 1999. Geological Society, London, Memoirs. 21. 153-171. MILLER, A. D., STEWART, R. C., WHITE, R. A. ET AL. 1998. Seismicity associated with dome growth and collapse at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3401-3404. MILLER, T. P. 1994. Dome growth and destruction during the 1989-1990 eruption of Redoubt Volcano. Journal of Volcanology and Geothermal Research, 62, 197-212. MURPHY, M. D., SPARKS, R. S. J., BARCLAY, J., CARROLL, M. R. & BREWER, T. S. 2000. Remobilization of andesite magma by intrusion of mafic magma at the Soufriere Hills Volcano. Montserrat, West Indies. Journal of Petrology, 41, 21-42. NAKADA, S. & MOTOMURA, Y. 1999. Petrology of the 1991-1995 eruption at Unzen: effusion pulsation and groundmass crystallization. Journal of Volcanology and Geothermal Research, 89, 173-196. NAKADA, N., MIYAKE, Y., SATO, H., OSHIMA, O. & FUJINAWA. A. 1995. Endogenous growth of the dacite dome at Unzen Volcano, Japan 1993-1994. Geology, 23, 157-160. NAKADA, S., SHIMIZU, H. & OHTA, K. 1999. Overview of the 1990-1995 eruption at Unzen Volcano. Journal of Volcanology and Geothermal Research, 89, 1-22. NEWHALL, C. G. & MELSON, W. G. 1983. Explosive activity associated with the growth of volcanic domes. Journal of Volcanology and Geothermal Research, 17, 111-131. NORTON, G. E., WATTS, R. B., VOIGHT, B. ET AL. 2002. Pyroclastic flow and explosive activity at Soufriere Hills Volcano. Montserrat, during a period of virtually no magma extrusion (March 1998 to November 1999). In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hi//s Vocano, Montserrat, from 1995 10 1999. Geological Society, London, Memoirs, 21, 467-481. PERRET, F. A. 1935. The Eruption of Ml. Pelee 1929 1932. Carnegie Institution of Washington, Baltimore. PINKERTON, H. & STEVENSON, R. J. 1992. Methods of determining the rheological properties of magmas at sub-liquidus temperatures. Journal of Volcanology and Geothermal Research, 53, 47-66.

ROBERTSON, R. E. A.. COLE, P. D.. SPARKS, R. S. J. ET AL. 1998. The explosive eruption ofSoufriere Hills Volcano. Montserrat. West Indies. September 17, 1996. Geophysical Research Letters, 25. 3249-3432. ROBERTSON. R. E. A.. ASPINALL. W ? . P.. HERD. R. A.. NORTON. G. E.. SPARKS. R. S. J. & YOUNG. S. R. 2000. The 1995-1998 eruption of the Soufriere Hills Volcano. Montserrat. WI. Philosophical Transactions of the Royal Society. 358. 1619-1637. ROOBOL. M. J. & SMITH. A. L. 1998. Pyroclastic stratigraphy of the Soufriere Hills Volcano. Montserrat: implications for the present eruption. Geophysical Research Letters. 25. 3392-3396. SPARKS, R. S. J. 1997. Causes and consequences of pressunsation in lava dome eruptions. Earth and Planetary Science Letters, 150, 177-189. SPARKS, R. S. J. & PINKERTON, H. 1978. Effects of degassing on rheology of basaltic lava. Nature. 276. 385-386. SPARKS. R. S. J.. YOUNG. S. R.. BARCLAY. J ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills Volcano, Montserrat. West Indies: November 1995 to December 1997. Geophysical Research Letters, 25, 3421-3424. SPARKS, R. S. J.. MURPHY, M. D.. LEJEUNE. A. M.. WATTS. R. B., BARCLAY. J. & YOUNG. S. R. 2000. Control on the emplacement of the andesite lava dome of the Soufriere Hills volcano. Montserrat by degassing-induced crystallization. Terra Nova. 12. 14-20. SPARKS. R. S. J.. BARCLAY. J.. CALDER. E. S ET AL. 2002. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat. from 1995 to 1999. Geological Society. London, Memoirs. 21. 409-434. STASIUK, M. V. & JAUPART. C. 1997. Lava flow shapes and dimensions as reflections of magma system conditions. Journal of Volcanology and Geothermal Research. 78. 31-50. SWANSON, R. A. & HOLCOMB. R. T. 1990. Regularities in growth of the Mount St Helens dacite dome 1980-1986. In: FINK. J. H. (ed.) Lava Flows and Domes. I.4 I 'CEI Proceedings in I 'olcanology. Springer-Verlag, New York. Vol. 2. 3 24. SWANSON. D. A.. DZURISIN. D.. HOLCOMB. R. T. ET AL. 1987. Growth of the lava dome at Mount St Helens. Washington. (USA) 1981 1983. In: FlNK. J. H. (ed.) The Emplacement of Silicic Domes and Lava Flows. Geological Society of America. Boulder. Special Paper. 212. 1-16. TAYLOR. G. A. M. 1958. The 1951 eruption of Mount Lamington, Papua. Commonwealth of Australia. Department of National Development. Bureau of Mineral Resources. Geology and Geophysics. Canberra. VOIGHT. B., SPARKS. R. S. J.. MILLER. A. D. ET AL. 1999. Magma flowinstability and cyclic activity at Soufriere Hills Volcano. Montserrat. British West Indies. Science. 283. 1138-1142. VOIGHT. B.. CONSTANTINE. E. K.. SISWOWIDJOYO. S. & TORLEY. R. 2000. Historical eruptions of Merapi Volcano. Central Java. Indonesia, 1768-1998. Journal of Volcanologv and Geothermal Research. 100. 69-138. VOIGHT. B., KOMOROWSKI. J.-C.. NORTON. G. E. ET AL. 2002. The 26 December (Boxing Day) 1997 sector collapse and debris avalanche at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London. Memoirs. 21. 363-407. WOLF. K. J. & EICHELBERGER. J. C. 1997. Syneruptive mixing, degassing, and crystallization at Redoubt Volcano, eruption of December 1989 to May 1990. Journal of Volcanology and Geothermal Research. 75. 19-37. WYLIE. J. J.. VOIGHT. B. & WHITEHEAD. J. A. 1999. Instability of magma flow from volatile-dependent viscosity. Science. 285. 1883-1885. YOUNG. S. R.. SPARKS. R. S. J.. ASPINALL. W. P.. LLYNCH. L. L.. MILLER. A. D.. ROBERTSON. R. E. A. & SHEPHERD. J. B. 1998. Overview of the eruption of Soufriere Hills Volcano. Montserrat. July 18 1995. to December 1997. Geophysical Research Letters. 25. 3389-3392. YOUNG. S. R.. VOIGHT. B.. BARCLAY. J. ETAL. 2002. Hazard implications of small-scale edifice instability and sector collapse: a case history from Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21. 349-361.

Dynamics of magma ascent and lava extrusion at Soufriere Hills Volcano, Montserrat O. MELNIK 1,2 & R. S. J. SPARKS1 Department of Earth Sciences, University of Bristol, Bristol BS8 1RJ, UK (e-mail: [email protected]) 2 Institute of Mechanics, Moscow State University, 1 Michurinskii prosp., Moscow 117192, Russia 1

Abstract: Growth of the andesitic lava dome during eruption of Soufriere Hills Volcano, Montserrat, involved general increases in discharge rate and dome height with superposed fluctuations on timescales from hours to months. We have modelled magma conduit flow incorporating viscosity variations caused by degassing and crystallization. Gas loss is modelled by permeable flow, with variations of permeability constrained by measurements. Crystallization kinetics is modelled by an Avrami law. Observations and petrological studies constrain conduit diameter (c. 30m), magma chamber crystal content (c. 60-65%), melt water content (c. 5%), magma temperature (c. 850°C) and chamber depth (c. 5 km). Gas loss and rheological stiffening cause a maximum in magma overpressure (magma pressure minus lithostatic pressure) at depths of a few hundred metres below the dome. This maximum can explain ground deformation patterns, shallow seismicity, and short-lived Vulcanian explosions. Crystallization kinetics causes a strong feedback mechanism and multiple steady solutions for discharge rate. Small changes in chamber pressure, magma viscosity and conduit diameter strongly amplify discharge rate. Multiple solutions allow cyclic variation of discharge rate. Escalation of activity is attributed principally to influx of hot mafic magma into the chamber, resulting in chamber pressure increasing with time. Dome growth can be intrinsically unpredictable due to non-linear effects.

The purpose of this paper (and also Melnik & Sparks 2002) is to develop general fluid-dynamical models for the flow of andesitic magma in volcanic conduits, and to use these models to interpret extrusive and explosive activity at Soufriere Hills Volcano, Montserrat. About 0.3km 3 of crystal-rich andesitic magma was erupted between July 1995 and March 1998 (Sparks et al. 1998; Robertson et al. 2000). The principal products of the eruption were an andesitic lava dome, pyroclastic flow deposits with associated pyroclastic surge and ashfall deposits formed by collapse of the dome, and pyroclastic flow and fall deposits formed by sub-Plinian and Vulcanian explosive activity. Information has been published on growth of the lava dome (Robertson et al. 2000; Watts et al. 2002), volumes of products (Sparks et al. 1998), seismicity (Miller et al. 1998; Aspinall et al. 1998), ground deformation (Mattioli et al. 1998; Shepherd et al. 1998; Voight et al. 1999), pyroclastic flow activity (Cole et al. 1998), explosive activity (Robertson et al. 1998; Druitt et al. 2002) and petrology (Devine et al. 1998a; Barclay et al. 1998; Murphy et al. 2000). There are thus many constraints on eruptive conditions that were either directly observed or can be inferred from observations. Data on discharge rate and the height of the active dome (Watts et al. 2002) can be used to evaluate the results of conduit flow models, and the conduit flow models can be used to understand the eruption. Understanding of flow dynamics in volcanic conduits has advanced with development of sophisticated numerical models that take advantage of increases in computer power. Most attention has been towards simulations of steady conduit flows in high-intensity Plinian eruptions (e.g. Wilson 1980; Dobran 1992; Macedonio et al. 1994; Sparks et al. 1997; Papale 1999; Melnik 2000). Recent models have begun to include non-linear effects, such as viscosity variation due to gas exsolution (Dobran 1992; Sparks et al. 1994; Papale 1999), gas loss during magma ascent (Jaupart & Allegre 1991; Woods & Koyaguchi 1994; Jaupart 1998) and diffusion effects in exsolution (Alidibirov 1988). A notable feature of models produced in Russia is the idea that there can be several stable solutions to the system of equations that describe explosive conduit flows (Slezin 1983, 1984; Barmin & Melnik 1993; Melnik 2000). Less attention has been paid to conduit flow models for lava extrusions, with most models assuming constant viscosity (e.g. Stasiuk et al. 1993). Sparks (1997) made calculations for conduit flow in lava dome eruptions, which incorporated large viscosity variations caused by degassing during magma ascent. These ideas were developed by Melnik & Sparks (1999), who demonstrated the strongly non-linear behaviour of lava dome eruptions. The modelling studies reported here develop further conduit flow models for the extrusion of lava domes (Melnik & Sparks 1999). Data from Montserrat are used for comparison and application of the model. Account is taken of viscosity and density varia-

tions due to progressive gas exsolution and gas loss in conduit flow during dome extrusion. The model is more comprehensive than the approximate model of Sparks (1997), and uses a parameterization of the viscosity of the Soufriere Hills andesite as a function of melt water content, crystal content and pressure. This paper also incorporates a model of crystallization kinetics after Hort (1998), which is a more sophisticated treatment than that used by Melnik & Sparks (1999). The viscosity variation of the andesitic magma spans over eight orders of magnitude, from conditions in the chamber to fully degassed and highly crystallized lava in the dome. We investigate the effects of partial gas loss during magma ascent on the behaviour of dome extrusions. Gas loss greatly affects density variation and pressure distribution in the magma column. The eruptive activity has shown oscillatory and pulse-like behaviour on a short timescale, typically several hours, and on longer timescales of several weeks (Voight et al. 1999). Our model demonstrates that strong non-linear effects can cause cyclic patterns of behaviour. The dynamics of the transient explosive activity in andesitic dome eruptions is considered by Melnik & Sparks 2002. The Soufriere Hills eruption Following seismic unrest from 1992 to 1995, the eruption began on 18 July 1995 with the opening of steam vents and phreatomagmatic explosions (Young et al. 1998; Norton et al. 2002). The eruption developed within English's Crater, a horseshoe-shaped collapse scar (1 km diameter) open to the east. Extrusion of an andesite lava (58 to 60% SiO2) began about 15 November 1995. By March 1996 pyroclastic flows were generated by partial collapses of the growing dome. The sizes of dome collapses increased with time as the dome size increased. A major collapse of 35% of the dome (with a DRE volume of 27 x 106 m3) on 17 September 1996 was followed by a sub-Plinian explosive eruption (Robertson et al. 1998; Calder et al. 2002). Dome growth continued through the remainder of 1996 and further pyroclastic flows formed in December 1996 and January 1997. Up to this stage the pyroclastic flows were confined in the Tar River valley to the east of English's Crater (Cole et al. 1998). By April 1997 the dome overwhelmed the SW walls of English's crater, and pyroclastic flows moved down the White River valley. Discharge rate increased during May 1997 from about 2.5m s -1 to over 5m3 s-1 (Sparks et al. 1998). The dome overwhelmed the northern and western crater walls during the June to August 1997 period, with major dome-collapse pyroclastic flows on 25 June and 3 August overwhelming the northeastern and western flanks respectively. Explosive activity resumed in early August 1997 with a series of

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 153-171. 0435-4052/02/S15 © The Geological Society of London 2002.

153

154

O. M E L N I K & R. S. J. SPARKS

13 Vulcanian explosions (Druitt el al. 2002). The dome grew at about 7.5m 3 s -1 during August and September 1997, leading to a major dome-collapse pyroclastic flow of 13.6 x 10 6 m 3 on 21 September 1997. This collapse was followed by a series of 75 repetitive Vulcanian explosions (22 September to 21 October 1997) with an average repose period between explosions of 9.5 hours. The resulting explosion crater was then rapidly filled and dome growth switched to the south. Further large dome-collapse pyroclastic flows were generated on 4 and 6 November down the SW flanks and the dome continued to grow at 7-8 m3 s-1. By 26 December 1997 the dome had reached its greatest volume ( 1 1 3 x l 0 6 m 3 ) and height (1030m a.s.l.). Dome growth during November and December 1997 had been on the SW rim of English's Crater. This area had been unstable back to October 1996 (Sparks el al. 2002). There was an area of active fumaroles and hydrothermally altered rock known as Galway's Soufriere on the upper flanks of the volcano SW of the crater rim. Following a 24-hour intense earthquake swarm these flank rocks (and part of the dome talus) failed to form a large debris avalanche (Sparks el al. 2002). The failure was immediately followed (within minutes) by collapse of about 40% of the dome, which disintegrated to form a high-

velocity pyroclastic density current, which devastated 10km 2 of southwestern Montserrat. Following the 26 December 1997 activity the lava dome regrew quickly in the collapse scar, and by early March 1998 had reached almost the same volume and height as the dome prior to the flank failure. The dome then ceased growth, after which activity was characterized by a series of collapses of the dome, episodic ash-venting and minor explosions (Norton el al. 2002). Dome growth resumed in November 1999. Eruption parameters Dome growth The growth of the lava dome has been documented in detail (Sparks el al 1998; Watts el al 2002). Figure 1 shows (a) the variation of dome volume, the volume of pyroclastic products, and total extruded volume, and (b) data on the height of the actively growing summit of the dome from 15 November 1995 to March 1998. The model presented in this paper provides a framework for the

Fig. 1. (a) Volume of the dome, total magma production and pyroclastic products versus time, and (b) height of the actively growing summit of the lava dome versus time. The start date corresponds to the first appearance of the lava dome on 15 November 1995. The end date on 10 March 1998 corresponds to the cessation of magma extrusion prior to the resumption in November 1999. Note that inactive parts of the dome can be higher than active parts, particularly after a major dome collapse when new lava infills a collapse scar. Stages II-VI. discussed in the text, are shown. Stage I occurred from July 1995 to 15 November 1995 and is not shown.

DYNAMICS OF MAGMA ASCENT AND EXTRUSION

interpretation of such data, but the data also provide some information for input to the model. Changes in discharge rates and dome height, together with geophysical information, constrain the roles of magma chamber pressure, magma properties and conduit geometry in controlling the extrusion of the dome. We divide the eruption into seven main stages and provide here an interpretative account in terms of discharge rate, dome growth and observational constraints on the conduit dimensions and its evolution. In Stage I magma ascended from a magma chamber at about 5 km depth (Barclay et al. 1998) to the surface. We assume that this process started in July 1995 though the process of ascent could have begun in the pre-eruptive seismic crisis, which can be traced back to 1992 (Young et al. 1998; Robertson et al. 2000). Magma reached the surface four months later, so magma ascent at this stage was very slow, providing opportunities for extensive gas loss. The dimensions of early vents and explosion craters formed in July 1995 suggest a maximum width of any feeding conduit close to the surface of not more than 40 m. Stage II is marked by the appearance of magma at the surface around 15 November 1995, and lasted until mid-February 1996. Dome growth was slow (0.1-0.5m3 s-1) and was often characterized by formation of spines (Watts et al. 2002). Individual spines grew a few metres per day (Watts et al. 2002) and reached 30 to 70 m in height. They were unstable and collapsed after a few days, so the dome height varied irregularly. The spines had striated surfaces and were interpreted as extrusions of solid, crystalline plugs of lava from the uppermost parts of the conduit (Watts et al. 2002). The diameter of the major spines in the vicinity of the craters above the conduit suggested that conduit dimensions were a maximum of about 30-35 m. Magma erupted in Stage II is thought to represent the magma that had filled the conduit in Stage I. Such magma would have been thoroughly degassed due to its slow ascent. We assume that it took until early February 1996 to push out this degassed material. If we also assume that the conduit had a diameter of 30 m down to the magma chamber at 5 km, then the volume of magma occupying the conduit at the start of dome extrusion would have been about 3.5 x 106 m3, which is comparable to the amount of magma erupted in Stage II. Stage III started as a spurt of growth in mid-February 1996, with magma discharge rates of around 1.8m 3 s -1 and a steady increase in dome height. Ground deformation data (Jackson et al. 1998) do not show any change in the rate of deformation in early February when dome growth accelerated. This observation suggests that the increase in magma discharge rate was related to a change in magma properties rather than an increase in magma chamber pressure or conduit dimensions. The change in discharge rate is linked to more gas-rich and less viscous magma reaching the surface. By July 1996 there were signs of increasing vigour in the eruption and a notable and sharp change in deformation rate on about 20 July. The eruption then became more pulse-like and the dome started to shed large pyroclastic flows during late July and August 1996. This period of instability can be seen in the height data (Fig. Ib), which show the height decreasing towards the end of this period while the extrusion rate increased. Stage III finished with the major collapse and explosive activity of 17 September 1996 (Robertson et al. 1998). Stage IV began with the extrusion of a new lava lobe on 1 October 1996 in the collapse scar and explosion crater formed on 17 September 1996. The initial discharge rate was about 2 m 3 s - 1 , which was close to the value just prior to the explosive eruption. Volcanotectonic earthquakes and analysis of eruption dynamics (Robertson et al. 1998) indicate that the magma conduit had been drained to a depth of 4km during the explosive eruption. The twoweek hiatus before dome growth resumed is consistent with magma ascending at 2m 3 s - 1 from 4km depth along a conduit with an average diameter of 28 m. During dome growth between 1 October and the end of November 1996 discharge rate and the rate of dome height change both decreased with time (Fig. 2). If these changes were due entirely to an increase in overburden pressure as the dome height increased, with all other parameters (density, rheological properties, conduit dimensions and magma chamber pressure)

155

Fig. 2. Volume discharge rate (triangles) and dome height (dots) versus time for the dome that began growth on 1 October 1996. The discharge rate was estimated by a best-fit regression curve through volume versus time data presented on Figure 1.

remaining constant, then the relationship between dome height and discharge rate should be approximately linear. The trend is in fact non-linear (Fig. 2), an observation that will be discussed later in this paper. Stage IV merged into a period of very low discharge rates (<0.5m 3 s - 1 ), stagnation of the dome height, and strong deformation of the volcano's upper flanks in early December (Young et al. 1998). Stage V started in mid-December 1996, with reinvigorated growth shown both by increase in dome height and higher discharge rates, reaching over 5m3 s-1 in December 1996 and January 1997. Discharge rates declined somewhat in March and April 1997. Stage VI began in mid-May 1997 and is marked by a pronounced increase in discharge rate, eventually reaching values of 7 to 8 m3 s-1 by August 1997. These high magma discharge rates continued to early March 1998 when dome growth stopped. There were marked changes in ground deformation style and rate in May 1997 (R. A. Herd, pers. comm.), indicating some fundamental and quite rapid change in the system. The dome reached significantly greater heights than previous stages (up to 1030m a.s.L), but the dome was also highly unstable in two respects. First, in Stage VI there were a number of large collapses with generation of major pyroclastic flows. Second, the major collapses of 3 August 1997 and 21 September 1997 were each followed by series of repetitive Vulcanian explosions. The instability is manifested in the height data, which display large fluctuations in the height of the area of active growth (Fig. Ib). Stage VII constitutes the period of little or no growth activity from March 1998 to mid-November 1999.

Conduit dimensions and geometry Observations of the early vents, spines and focusing of the dome growth indicate that the conduit has a width of about 30m (Watts et al. 2002). The estimate of the ascent rate of the new dome that extruded on 1 October 1996 gave a diameter of 28m. Devine et al. (1998b) have used decompression reaction rims on hornblendes to estimate ascent velocities in the period December 1995 to August 1996. They observed that reaction rim widths increased as the discharge rate of the dome increased. Using experimental calibrations on hornblende stability in dacitic magma from Mount St Helens at 850°C, they estimated ascent velocities in the range 0.001 to >0.012ms - 1 . The samples studied by Devine et al. (1998b) were collected from pyroclastic flows and spent an unknown amount of

156

O. M E L N I K & R. S. J. SPARKS

time in the dome. Therefore these velocities are minimum values. In May and June 1996 the estimated velocity was 0.008ms -1 at a time when the discharge rate averaged 2.3 m3 s -1. This period is the most reliable for the use of this method, as it is within the experimental conditions of Rutherford & Hill (1993). If the average velocity is assumed to be two-thirds of the centreline velocity, as in laminar pipe flow 7 , then the conduit diameter is deduced to be a minimum of 24m. Two craters formed in July 1995 were the locus of initial dome growth (Watts et al 2002), and were aligned along a WNW to ESE direction over about 100m, perhaps indicating fracture control. Throughout the eruption much of the dome growth was largely focused over the area of the July 1995 craters. The uppermost parts of the conduit were widened asymmetrically to the east, following the 17 September explosive activity (Watts et al. 2002). Thereafter the focus of growth was 100-150 m to the east of the July 1995 craters. The Vulcanian explosions in 1997 occurred in this area, forming deep, symmetrical craters. In contrast, earthquake alignments in Stage I of the eruption indicate controls along a NE-SW lineament (Aspinall et al. 1998). Ground deformation data have been interpreted by Mattioli et al. (1998) in terms of two different directions of dyke formation. However, the simple radial pattern of ground deformation (Shepherd et al. 1998) is consistent with a cylindrical conduit with a shallow pressure source a few hundred metres beneath the dome. Overall these observations imply that magma ascent is focused in a narrow zone, which suggests that the conduit can be treated as approximately a cylinder of diameter D (D 30m).

Pressures We first define pressures in an erupting lava-dome system. Excess magma chamber pressure is defined as the total chamber pressure minus the magmastatic pressure caused by the weight of magma in the conduit and in the dome. The lithostatic pressure is defined as the pressure of the country rocks surrounding the conduit and overlying the magma chamber. The magma chamber can have a pressure that is higher or lower than lithostatic pressure. In a chamber with the pressure equal to lithostatic pressure, magma ascent is the consequence of buoyancy alone. A magma overpressure is defined at any height in the conduit as the magma pressure minus the lithostatic pressure. Observations place some constraints on pressures in the system. The dome creates a backpressure on the conduit, so that it might be expected that dome growth declines as the dome approaches a height where the magmastatic pressure balances the total magma chamber pressure. This behaviour can be seen in the dome-growth data from October to November 1996 (Fig. 2). On several occasions the dome approached some upper threshold to its height and growth slowed down (Fig. Ib). The overall pattern indicates an increase in this upper height limit with time, coinciding with increases in discharge rate. These data allow an estimate of the minimum excess pressure at the top of the conduit (i.e. base of the dome) and a minimum estimate of the excess chamber pressure when a new period of dome growth starts. In the case of growth starting on 1 October 1996 (Stage IV), the minimum excess chamber pressure is estimated at 3.1 MPa using a lava density of 2400 kg m- 3 and assuming that the top of the conduit is at 760m a.s.l. The dome eventually reached a height of 1030m or 270m above the top of the conduit, suggesting a minimum chamber excess pressure of 7.3 MPa. In the modelling results we will show that the height at which dome growth diminishes can be much less than the equilibrium height. Thus the chamber excess pressure could have been considerably above these minimum estimates. The increase in height with time suggests an increase in excess magma chamber pressure with time, but should not necessarily be taken as evidence of an increase in total magma chamber pressure. The weight of the magma column depends on the density of the magma, which will depend on vesiculation and gas loss. As an

example, for a constant total chamber pressure, the excess pressure will be a function of the efficiency with which gas is lost from the column during ascent. Magma that loses its gas continually and is always dense will exert a larger pressure than a magma which retains its gas and vesiculates as it approaches the Earth's surface. Thus the increase in height and discharge rate may reflect some change in the vesicularity profile in the magma column with time, with no change or even a decrease in total magma chamber pressure. We will develop numerical models that explore these effects below. Other indicators of pressure come from observations of ground deformation and explosive eruptions. Shepherd et al. (1998) interpret the radial swelling of the volcano during 1996 in terms of a pressure source in a cylinder at about 700 m below the dome with an overpressure of about 10 MPa. Voight et al. (1999) also inferred a shallow pressure source about 400m below the dome from cyclic patterns of inflation and deflation measured by tiltmeters. Explosive activity has implications of high gas pressures at shallow depth in the conduit system (Clarke et al. 2002; Druitt et al. 2002). Pumice fragments from explosions are platy and angular (Druitt et al. 2002) and are similar to fragments formed in laboratory experiments by Alidibirov & Dingwell (1996). who found that excess pressures of several megapascals were required to cause spontaneous explosions of natural dome rocks. Ejection velocities of up to 140ms - 1 (Druitt et al. 2002) also imply explosion pressures in excess of 10 MPa (Fagents & Wilson 1993). Finally. McLeod & Tait (1999) have estimated the magma chamber pressures required to initiate and drive dykes to the surface. They find that required magma chamber pressure is a strong function of magma viscosity. For magmas with viscosities of 106 to 107 Pas. as at Soufriere Hills, excess pressures of several megapascals are calculated.

Magma properties The Soufriere Hills andesite is a highly porphyritic magma containing between 40 and 45% phenocrysts of plagioclase. hornblende, orthopyroxene and oxide and about a further 20% microphenocrysts of plagioclase. orthopyroxene. clinopyroxene and oxide. Petrological and experimental studies (Devine et al. 1998a; Barclay et al. 1998. Murphy et al. 2000) indicate that the magma in the chamber had a temperature of about 850 C, a rhyolitic melt phase (75 to 77% SiO 2 ) and a dissolved water content in the melt phase of 4 to 5%. The ubiquitous presence in the andesite of mafic inclusions, formed by intermingling of hydrous basaltic andesite. has been taken to imply that the eruption was driven by influx of mafic magma into the chamber (Murphy et al. 2000), with heat and perhaps excess volatiles being transferred to the andesite. Knowledge of the rheological properties of very crystal-rich magma is limited, as most experimental data are available on melts, crystal-melt suspensions with low concentrations, and low-degree partial melts (Dingwell 1998). There are few experimental data on the transition between magmas with very high crystal contents and partially molten rocks. However. Lejeune & Richet (1995) have investigated high-crystal-content silicate systems with spherical and elliptical crystals. We estimated the viscosity of the Soufriere Hills andesite as follows. We have used recent experimental data on rhyolitic melt compositions (Hess & Dingwell 1996) to estimate the viscosity of the rhyolite melt phase at 850 C as a function of water content. Rhyolitic melt with 5 w t % dissolved water has an estimated viscosity of 3 x 10 4 Pas. and that with 1 \vt% water has an estimated viscosity of 6.4 x 10 6 Pas. The following relationship is commonly used to estimate the effects of crystal content

is the viswhere is the viscosity of the crystal-melt suspension. cosity of the Newtonian melt. 3 is the volumetric fraction of crystals

DYNAMICS OF MAGMA ASCENT AND EXTRUSION and E is an empirical constant. This relationship, after Marsh (1981), gives reasonable agreement with observations for E 1.67 and crystal contents up to about = 0.55. However, this relationship cannot be extended to the very high crystal contents observed in the Soufriere Hills magma ( B = 0.65-0.9), since viscosity becomes infinite in Equation 1 at B = 0.62. Lejeune & Richet (1995) report experimental results that show a very strong rheological transition at crystal contents where the solid fraction reaches the close-packing condition. For example, close-packing occurs at B = 0.55 for uniform spheres. However, the solid fraction at close-packing state depends strongly on crystal shape and crystal size distribution. For suspensions with a wide size range the close-packing condition can occur at (3 = 0.7 (Sohn & Moreland 1968), while irregular shape reduces the solid fraction at close-packing state. At crystal contents above the close-packed condition viscosity becomes again less sensitive to crystal content and is governed by crystalline deformation. Experiments reported in Sparks et al. (2000) have studied the creep behaviour of a natural sample of Soufriere Hills andesite with B = 0.92 and at 993°C. Initially the samples behave in an approximately Newtonian manner under load pressures from 4 to 24 MPa with an estimated creep viscosity of 10 1 4 Pas. However, the material responds in a highly non-linear manner, and deformation accelerates through to semi-ductile failure. The fast stage of deformation indicates a viscosity of order 10 11 Pas. The problem then involves interpolating the results of Equation 1 with the experimental estimates of the viscosity of the fully degassed and highly crystalline dome. There is clearly considerable uncertainty about the viscosity of the magma and how it might vary with crystal content and pressure. We have chosen to model viscosity as one of the variables using the following relationship:

where 9(B) is a function of volume concentration of crystals 0, and (C) is the viscosity of the pure melt phase, which depends on melt composition and melt water content c. Equation 2 is a variant of similar equations used in other models (e.g. Jaupart 1998) with being fitted to the experimental results of Hess & Dingwell (1996). Further discussion of the parameter (B) is developed below. Another effect of crystallization is eventually to give magma strength. Although the model developed below assumes a Newtonian approximation, there is ample evidence that highly crystalline lava develops a considerable strength. Voight et al. (1999) estimate a minimum strength of 1 MPa for the lava from the height of spines.

157

Magma permeability Gas loss is part of the dynamic process of the dome extrusion of an originally gas-rich magma. Magma permeability and its variation with porosity need to be determined. There are few data on the permeability of volcanic materials, and the two main studies (Eichelberger et al. 1986; Klug & Cashman 1996) give quite different results. We have measured the porosity and permeability of samples from the Soufriere Hills eruption (Table 1). The samples include dome rock collected from pyroclastic flow deposits, pumice samples from Vulcanian explosions, and samples of denser vesiculated ballistic blocks that are thought to originate from the upper parts of the conduit in the sub-Plinian explosive eruption of 17 September 1996 (Robertson et al. 1998). The data broadly cover the same range of permeabilities of 10-16 to 10-12 m2 as in previous studies (Eichelberger et al. 1986; Klug & Cashman 1996). Variation of permeability (k) with volume fraction of gas (a) (Fig. 3) can be approximated by the following formula (Table 1): log(k( ) o) = -10.2(100a)0.014/ ;

> 0.03

(3)

where KO is permeability coefficient. A k0 value of 1 is a best-fit for the observations, k0 = 10 is an upper bound to the data and K0 = 0.2 is a lower bound for most data. We demonstrate a posteriori that k0 < 1 is essentially impermeable for practical purposes and that results only weakly depend on the value of koThese (cold) samples have significant permeability at low porosities. This might suggest that permeability is partly caused by later processes, such as microfracturing during cooling of the sample, and that these measurements are not representative of magmatic conditions. However, examination of the samples in thin-section and with backscattered electron images shows that vesicles with clear interconnections make up almost all of the porosity. Although microfractures exist, they make up a negligible proportion of the porosity and, in agreement with Klug & Cashman (1996), these observations suggest that the room-temperature measurements are representative of the matrix permeability at high temperature. Glass compositions in the pumice samples indicate that the original magma before ascent had about 30 to 35% melt. During magma ascent and dome extrusion, the melt content was reduced variably to between 15 and 5% by groundmass crystallization (Murphy et al. 2000). The melt phase is completely distributed and interconnected between the crystals and all the vesicles are confined to the melt phase. Thus, if the vesicularity of the melt phase is considered alone, porosity values are much higher in the melt than they are in the bulk sample. For example, a typical dome sample

Table 1. Data on the porosity ana permeability of the Soufriere Hills andesite

Sample

Date

Description

Total porosity (%)

Permeability (10- 1 5 m 2 )

SSMon13 MVO697 MVO103 MMon5A MVO1072 MVO104 MVO1078 SSMon7 MVO627 SSMonl2 MMon21 MVO203 MVO204 SSMon9 MVO1003a MVO289 SSMon8

17 Sep. 1996 Sep./Oct. 1997 1 Oct. 1996 Mid-late Jan. 1996 16 Nov. 1998 1 Oct. 1996 12 Dec. 1998 17 Sep. 1996 17 Sep. 1996 17 Sep. 1996 2/3 Sep. 1996 6/7 Mar. 1997 6/7 Mar. 1997 17 Sep. 1996 8 Apr. 1999 25 Sep. 1997 17 Sep. 1996

Dense 'glassy' fragments Dense 'glassy' bomb Brown dome Early dome Lava block Grey dome Lava block Dome rock with tuffisite veins Lava block Vent wall (welded?) breccia bomb Vesicular dome Lava block Lava block Foliated vesicular lava Vesicular ballistic bomb Pumice Pumice

2.3 3.2 17.0 9.9 14.4 16.4 18.4 21.2 21.5 22.5 33.0 30.5 30.6 35.3 49.4 49.7 71.5

0.6 0.73 56.4 8.5 17 54.0 33 592.8 28 10.2 5092.9 644 658 417.2 808 530 4012.6

The measurements were carried out commercially using standard absorption techniques for porosity and gas flow measurements of sample cylindrical cores with fixed pressure difference.

158

O. M E L N I K & R. S. J. SPARKS

Here c is concentration of dissolved water (in wt%). Solidus and liquidus temperatures also depend strongly on chemical composition, which changes during the crystallization. We will consider that the effective liquidus temperature is: T m = Thql(1 - 3 ) + T s o l 3

(5)

The true dependence can be non-linear, but a linear model is a sufficient approximation. We introduce nucleation rate / (m s -1), which defines the number of newly nucleated crystals per cubic metre per second, and linear crystal growth rate U ( m s - 1 ) . The conventional approach (Rao & Rao 1978) describes these parameters as follows:

(6a)

(6b)

Fig. 3. Data on permeability and porosity of the samples listed in Table 1. The curves show calculations using equation (3) with values of k0 labelled on curves.

with 15% bulk porosity and 15% melt would in fact have a melt porosity of 50%. In a vesiculating magma, bubble interconnections start becoming significant at bubble concentrations in the melt phase of about 30 to 50% (Eichelberger et al. 1986; Blower et al. 2001). Thus a crystal-rich magma can develop relatively high permeabilties at low bulk porosities. Two factors may make the permeability in the magma conduit and dome deviate from the values found in cold samples. At depth, where the melt content is high and porosity is low, the permeability is likely to be even lower than the measurements suggest, since the low-porosity natural samples have acquired their current pore structure at a later stage by a combination of gas loss and groundmass crystallization. The relationship between porosity and permeability must go through irreversible changes during magma ascent and dome extrusion. Close to the top of the conduit and in the dome, fracture systems develop and gases are observed to escape preferentially along fractures and shear zones on the dome surface. Thus the permeability near the surface may be higher than indicated by our measurements of matrix permeability, as a consequence of fracture permeability.

Here AGt, and AGC are the activation energies of atomic diffusion and formation of a single nucleus, respectively; A/hc is the enthalpy difference between the melt and the crystal, T is temperature, R is the universal gas constant, c is surface tension, and U0 and I0 are constants. The first exponents (&Gt:RT) in Equation 6 determine the rate of diffusion of components to the growing crystal. At the liquidus temperature (Tm } ) these rates are equal to zero and increase with undercooling. For some critical undercoolings AT/ = Tm - T and ATU = Tm - T nucleation and growth rates reaches their maxima and decrease with further undercooling. There are no experimental data to determine the activation energies as a function of undercooling and chemical composition. Also Hort (1998) did not consider that a diffusion coefficient in silicate melts is a function of water content. We will multiply Equations 6 by (co) (c) where C0 is a magma water content in the chamber, because the product of viscosity to diffusion coefficient remains nearly constant in silicate melts (Navon & Lyakhovski 1998). For m(c) we use the formula obtained by Hess & Dingwell (1996). Using Equation 4 we can express Tm as a function of c with 3 = 0 and considering that (d dc]U T n = 0, we finally obtain: (7)

Here 0u depends on AT,, to satisfy the condition for the maxima in crystal growth rate, but the equation for this is rather complicated and will not be presented here. For I(c) using the same approach we can write the equation: (8)

Crystal nucleation and growth kinetics In Melnik & Sparks (1999) a simple crystal growth equation for the case of degassing-induced crystallization was proposed. A weak feature of this proposed equation is that crystal growth rate was linearly proportional to the degree of undercooling. In fact for high degrees of undercooling the growth rate slows down. The results of these earlier calculations could not be related to the observations of natural samples, as a single nucleation event was postulated, and all the crystals had the same size. Here we apply an approach modified after Hort (1998), for situations when crystal growth is induced by isothermal degassing. Crystallization starts when the magma temperature becomes lower then the liquidus temperature. Simakin et al. (1999) studied the influence of concentration of dissolved water (c) on the liquidus and solidus temperatures of rhyolitic melts similar to the melt phase of the Soufriere Hills andesite. Their data can be approximated by: (4)

Here 9i is a function of AT/, and (/ determines the width of the nucleation peak. Therefore to determine the kinetics of crystallization values of/o, UQ, AT1, ATu,/ and 0 should be known. Figure 4 shows values of U(c, 3) and I(c. 3) calculated using Equations 7 and 8, normalized by Uo, I0)Volume concentration of crystals is then determined by the following equation:

(9)

The outer integral (1) determines the amount of crystals that appear in the time interval [O.t]. The internal integral (2) shows the change of surface area for the same interval of time. By multiplying integrals by 3 (t), where is a shape factor, we will finally get the increase in volume.

DYNAMICS OF MAGMA ASCENT AND EXTRUSION

159

(lOf)

(10g) (lOh) (lOi)

(10j) (10k) (101)

0.01

0.02

0.03

0.04

0.05

Fig. 4. Normalized crystal growth (a) and nucleation rates (b) as a function of concentration of dissolved gas c. Description of the physical model for conduit flow We have modelled the ascent of magma along the conduit from the chamber with the following equations for conduit flow:

(lOa)

(lOb)

where pm, pc and pg are the melt, crystal and gas densities, respectively, p is a density of mixture, a and B are volume concentrations of bubbles and crystals, respectively (in the condensed phase), V and Vg are velocities of magma and gas, respectively, p is pressure, c is the mass concentration of dissolved gas, and ,g are magma, melt and gas viscosity, respectively, D is the conduit diameter, k( ) is the permeability, R is the gas constant, T is the temperature, x is the vertical co-ordinate, n is the number density of bubbles per unit volume, G and J represent mass transfer due to crystallization and gas exsolution, respectively,Ddif),///is a diffusion coefficient, Cf- is the solubility coefficient, a is the bubble radius and r is the radial distance from the centre of the bubble. The system in Equations lOa to lOh is a development of the model of Melnik & Sparks (1999). Equations lOa and lOb represent conservation of mass for melt and crystals and dissolved and exsolved gas, respectively; Equation lOc states the conservation of momentum for the mixture as a whole in which the inertial term is negligibly small; Equation lOd is Darcy's law; Equations lOe-h summarize mass fluxes in the system; and Equations lOi-1 describe physical properties of magma. For the mass transfer between melt and bubbles (Equation lOe) we assume that it happens quasistatically, so that an analytical solution for the concentration gradient in a bubble shell (Navon & Lyakhovski 1998) can be applied. The difference between the local pressure and internal pressure in the growing bubble is not large for slowly ascending magma. We therefore neglect pressure disequilibrium, although we recognize that pressure disequilibrium might arise because of other processes such as microlite crystallization (Stix et al. 1997; Sparks 1997). We also assume isothermal flow conditions. Equations 10 are solved numerically between the magma chamber and the top of the lava dome. Flow in the dome is represented by a continuation of the conduit, with the same diameter for the active zone of flow within the dome and extrusion of new lava at the summit, consistent with observations (Young et al. 1998; Watts et al. 2002). As the extrusion rate is subsonic, we assume that the pressure at the top of the dome is equal to atmospheric. For a particular choice of the chamber pressure and chamber crystal content, magma discharge rate can be found by means of a b cut and try' method (Melnik 2000). Table 2 summarizes the set of parameters used in the modelling. We also present a much-simplified analytical model showing the same behaviour in the Appendix.

(10c) (lOd)

Results

Influences of magma density and permeability on eruption behaviour

(lOe)

Our model of magma discharge differs from previous ones by incorporating magma permeability and crystallization kinetics.

160

O. M E L N I K & R. S. J. SPARKS

Table 2. Parameters for the Soufriere Hills andesite eruption used in modelling Parameter

Symbol

Value range

Information sources

Magma chamber depth

L

5 km

Earthquakes (Aspinall et al. 1998) and phase equilibria (Barclay el al. 1998)

Magma chamber overpressure Melt water content Magma temperature Magma crystal content Crystal growth rate Nucleation rate Max. growth undercooling Max. nucleation undercooling Conduit diameter Viscosity coefficient Viscosity coefficient Viscosity coefficient Permeability coefficient Density of melt Density of crystals Density of wallrock Solubility coefficient Gas viscosity

Pel,

co

T 30 Uo Io

ATr ATI

D 00

3 ko Pm PC Pr

cf

g

0-20 MPa 5% 850 C 0.6 1.2 x 10-11nms-1 - 6 - 3 lO m s-1 50 C 100 C 30 m 1.6 0.62 20.6 1 2300 kg m-3 2700 kg m- 3 2600 k g m 3 4.1 x I0 - 6 Pa- 1 :2 1.5 x 1 0 - 5 P a s

First we study the influence of permeability coefficient on eruption dynamics, assuming no crystallization in the conduit (U, I = 0). Figure 5a, b shows profiles of overpressure and porosity along the conduit for three different cases: (i) unvesiculated magma density equal to wallrock density (2600kgirT-3); (ii) unvesiculated magma density (2700 kg m-3) greater than wallrock density; and (iii) unvesiculated magma density (2500 kg m-3) less than wallrock density. Here the magma density is that of the solid and melt phases alone in the chamber prior to gas exsolution. Total chamber pressure is assumed to be equal to the lithostatic pressure. The exit pressure is atmospheric, which corresponds to the initial stages of dome extrusion. These assumptions isolate the influence of magma buoyancy and avoid consideration of dome growth. Results are displayed for several values of permeability coefficient k0In deeper parts of the conduit the pressure drop occurs mostly due to the weight of the magma column, because the relatively small magma viscosity makes conduit resistance small. In case (i), conduit pressure remains close to lithostatic in its deeper parts. In the upper part of the conduit, vesiculation reduces the mean density of magma so that pressure in the conduit declines more slowly than lithostatic pressure. Thus the magma overpressure increases with height, reaching a maximum at shallow depths. In the uppermost part of the conduit, a rapid pressure drop occurs as a result of increase in conduit resistance due to increase in viscosity and velocity of magma. When the initial density of the magma is

Range of crustal strengths Barclay et al. (1998) Barclay et al. (1998); Murphy et al. (2000) Murphy et al. (2000) From calculations From calculations From calculations From calculations Dimensions of spines and early crater: hornblende reaction rims (Devine et al. 1998/b) From calculations From calculations From calculations Regression through data from Table 1

Stolper (1982)

larger then the wallrock density (case ii). overpressure becomes negative at depth, and then positive due to vesiculation at shallow depth. In case (iii). overpressure increase with height is greater due to the larger density difference between magma and wallrock. and maximum overpressure is higher. Strong, shallow-level overpressures were also recognized in previous studies of volcanic flow : s \vith a strong dependence of viscosity on pressure (Jaupart & Allegre 1991: Dobran 1992: Sparks 1997: Melnik 2000). Increase in the permeability coefficient leads to a decrease of volume concentration of the gas phase (Fig. 5b). Therefore, the average density increases and maximum overpressure values decrease. If all the gas is retained in the magma (ko = 0). the volume fraction of bubbles reaches high values (near to 0.995) that are physically impossible. Explosive fragmentation can be expected well before such high porosities can develop. With an increase in permeability to values observed in the natural samples (A'o = 0.2-10). maximum porosity decreases to values observed for dome material (see Table 1). Maximum bubble fraction is reached at depths of several tens of metres. When the permeability coefficient is large (ko = 50100). porosity is small along the whole conduit. Magma discharge rate depends weakly on permeability coefficient when k () is small, then has a maximum and sharply decreases with increase in permeability (Fig. 6). This feature can be explained by considering the momentum Equation lOd. The pressure drop in the conduit is controlled bv the mixture weight and conduit

Fig. 5. Profiles of overpressure (a) and porosity (b) with depth in the conduit. Conditions for calculations are described in Table 2. Crystal content in the conduit is constant and equals 0.6. Viscosity coefficient (3) = 50. Unvesiculated magma density is equal to (i). bigger than (ii) and smaller than (iii) the density of surrounding rocks. Values of Ak0 are labelled on curves. Note that the vertical scale on the right is height above the magma chamber. The right-hand figure depicts just the uppermost 500m of conduit.

DYNAMICS OF MAGMA ASCENT AND EXTRUSION

Fig. 6. Relationship between discharge rate and permeability coefficient for the three different cases described in Figure 5. resistance. When k0 is small, the average density of the mixture is also relatively small, but the velocity for the same discharge rate is larger and, therefore, the conduit resistance is large. As k0 increases, the average density of the magma column also increases, but conduit resistance decreases. The interaction of these two effects explains the calculated relationship between magma discharge rate and permeability. As magma discharge rate and conduit resistance are proportional to velocity, the smaller the pressure drop due to the weight of the mixture, the higher the discharge rate.

The effect of microlite crystallization and dome growth on discharge rate As the dome increases in height the viscous resistance of the dome and the weight of the magma column increase. It can thus be expected that magma discharge rate will decrease, as is observed (Figs 1 and 2). Here we assume that as dome height increases, the length of the conduit increases. The lava is extruded at the summit of the dome. We assume that the zone of actively rising magma in the dome is cylindrical with the same diameter as the conduit. While some lateral spreading can be expected, this model captures the key features of the growth in which viscous resistance and magma weight control the total chamber pressure. The model is consistent with observations of the Soufriere Hills dome, where lava typically extruded in the summit area or as discrete spines with inferred dimensions comparable to the conduit (Watts et aL 2002). We take the data for dome growth from 1 October 1996 (Fig. 2) to constrain the rheological properties of magma. We first consider a model with fixed conduit diameter, with no crystal growth (U, I=0), and the rheological model given by Equation lOi. Calculations of discharge rate as a function of dome height show that the observed decrease of discharge rate in the lava lobe of 1 October 1996 cannot be simply explained by an increase of the dome height. To explain the decrease of discharge rate from 1.8 to O . l 5 m 3 s - 1 , the dome height needs to increase by more than 600m for fixed chamber pressure (dashed curve on Fig. 7). Substantial chamber underpressuring of about lOMPa is implied. Moreover, the rate of chamber pressure drop must increase with decrease in discharge rate. Therefore we hypothesize that crystallization in the conduit is the major factor in this dome extrusion. As the dome grows and discharge rate decreases, magma spends more time in the conduit, loses more gas, and therefore more crystallization occurs during ascent. This in turn leads to increases in

161

Fig. 7. Observations of magma discharge rate versus dome height for the lava lobe that appeared on 1 October 1996 (dots) and its approximation (solid curve) by means of calculation of Equations 10 with determination of the rheological properties of magma and crystallization rates. Dashed line on right is the calculation with no crystalization. Arrowed dashed line is unsteady path of growth. viscosity and density of the magma, with the consequence that discharge rate decreases further. We therefore calculated the viscosity coefficient 0(B) that is required to fit the initial dome growth rate in this period. We solve the whole system of Equations 10. Observational data (Murphy et al. 2000) show that for high discharge rates the volume fraction of crystals ( ), at the top of the conduit is around 0.7, whereas for lower discharge rates can be up to 0.95. The values of parameters in Equations lOg and lOh are chosen to fit these observations. Based on the concepts in Lejeune & Richet (1995), the shape of the viscosity coefficient 9(0) is given by: log

= arctan

(11)

Parameters 0, w; and must then be chosen to fit the data for dome height and discharge rate by minimization of the sum of squares:

mn

i

- Hc(Qi))2

where Hi and Qi represent the set of observational data and HC(Q) is the dome height calculated from Equations 10. Figure 7 shows the best-fit rheological model for a zero chamber overpressure (solid line). Incorporation of crystallization kinetics introduces a new type of dome growth instability. As discharge rate decreases, crystallization causes viscosity and therefore conduit resistance to increase. This leads to further decrease in discharge rate until the system comes to a critical point where further monotonic decrease in discharge rate is impossible. Slezin (1984) and Melnik (2000) showed that, when the solution of the steady boundary problem has several regimes represented by S-shaped curves, the real unsteady solution will follow one of the steady regimes (with the highest or the lowest discharge rate). There will be unsteady transitions between the regimes. Such a transition is shown by an arrow on Figure 7. The system cannot move down the curve as the viscous dome cannot respond by decreasing in height, so the system moves to very slow discharge rate conditions. Figure 8 shows the inferred viscosity dependence in terms of 9( ) and magma viscosity for atmospheric pressure on volume fraction of crystals. The crystal fraction in the chamber was taken

162

O. M E L N I K & R. S. J. SPARKS

rate the CSD is bimodal. representing the two nucleation events in ascending magma. The accurate study of the natural samples will allow future comparison of the model predictions with observations. Properly calibrated, the model can be used as a dischargerate meter.

General relationships between discharge rate and the governing parameters

Fig. 8. Calculated viscosity coefficient 0( ) and magma viscosity at atmospheric pressure as a function of crystal content. The calculation of 0(3) involved a best-fit to the data shown in Fig. 7.

as 0.6; a sharp increase in viscosity occurs for higher values of . The viscosity of the degassed lava dome can be estimated. The melt viscosity is estimated at 10 9 Pas and thus the dome will have a viscosity of 10 11 to 10 1 2 Pas which is consistent with experimental data on dome samples. The procedure described above essentially involves calibrating the values of 9(3) and crystallization parameters. While it would clearly be better to estimate these parameters independently, the paucity of experimental data mean that this cannot yet be done. Relationship between crystal size distribution and magma discharge rate We consider now the variation of crystal size distribution corresponding to different discharge rates. Figure 9 shows distributions along the conduit of (a) volume fraction of crystals ( ). (b) number density of crystals ( N ) , (c) average crystal radius ((R) = (4 /3 N 1/3 ), (d) magma undercooling (AT), (e, f) normalized crystal growth and nucleation rates for discharge rates of 1.8, 0.8 and O . l 5 n m 3 s - 1 . Results for higher discharge rates behave similarly. Crystal content, number density of crystals and degree of undercooling increase monotonically, and the average crystal radius decreases with progressive crystal nucleation. The behaviour for low discharge rate is different. As magma ascends slowly, the amount of crystals increases rapidly. This leads to drop of undercooling according to Equation 5. Nucleation stops and the growth of existing crystals continues. This leads to an increase in average crystal radius. When the magma reaches shallow levels in the conduit the degree of undercooling increases again as the gas exsolves rapidly. A second nucleation peak appears, and the average crystal radius then decreases. Having the distribution of crystal growth rate along the conduit we can obtain the relationship between the average crystal radius at the Earth's surface and the depth of nucleation (Fig. lOa). Knowing the distribution of number density of crystals with depth we can calculate the amount of crystals between radius R/ and R2 and finally the crystal size distribution (CSD):

Figure lOb shows CSDs for the cases corresponding to Figure 9. For higher discharge rates the distribution is unimodal as nucleation continues through all the ascent. For low discharge

We now study the dynamics of dome growth for a wider range of governing parameters using the calibration of the October 1996 dome growth to fix the parameters. Figure 11 shows how magma discharge rate varies with chamber overpressure for a fixed conduit diameter of 30m and a magma chamber crystal content of 0.6. Figure 12 shows the variation of discharge rate with variable magma chamber crystal content for a conduit diameter of 30m and for chamber overpressures of 0 and 5 MPa. Figure 13 shows the variation of discharge rate with conduit diameter for a fixed magma chamber crystal content of 0.6 and for chamber overpressures of 0 and 5 MPa. The results are shown contoured for the height of the dome. There are several interesting features of these results. The discharge rates ranged from 0.1 to about 10 m 3 s -1 or more during the 1995 1999 period of the Soufriere Hills eruption, and can be accommodated by changing magma chamber pressure alone in the range 0 to 20 MPa with fixed conduit dimensions. The large range of discharge rates might alternatively be explained by minor changes in magma properties (such as crystal content) or conduit diameter. This sensitivity is a consequence of the strong feedbacks in the system, which amplify the effects of small parameter changes. For example, simpler models of conduit flow, which do not consider crystallization, predict a two-fold change in magma discharge if the chamber overpressure or viscosity changes by a factor of two or if the conduit diameter changes by about 20%. Inspection of the plots shows that, for chamber overpressures <10 MPa. the same two-fold changes can produce much larger (over an order of magnitude) shifts in discharge rate as there is more than one steady-state solution. For magma chamber overpressures above about 10 MPa the discharge rates are too high for much microlite crystallization to occur, so the variation is monotonic and similar to the results for a simpler model without kinetic effects. Figure 11b illustrates the implications of the multiple solutions by considering possible eruption paths. Following Woods & Koyaguchi (1994) we assume that chamber pressure variation is related to the difference between influx of magma to the chamber and outflux from it (dp c h , dt = -(Qoutl - Qin) T. where r is a chamber pressure relaxation time). We assume Q in = 0.3m 3 s -1 for the representation of a possible eruption path. Suppose the dome starts growing from point A with a high discharge rate. Chamber pressure decreases, and the dome height increases until the system reaches point B. At this point the system has to move to point C to reestablish steady flow conditions. From point C further increase in dome height will be very slow and the system will move to point D with a slowly increasing discharge rate. The transition from C to D nearly follows the line of constant dome height. From point D the system must adjust to point E where further increase in dome height with decrease in chamber pressure is not possible. There are two possible paths of eruption from this point. First, if a dome collapse occurs, reducing the height substantially, the system adjusts to point G and eventually to H due to rapid pressure drop and increase in dome height. Second, the system can adjust to point F with a strong decrease in discharge rate. In this sequence of events discharge rate changes of up to an order of magnitude are predicted. Cycles of this kind can occur many times and thus can be a cause of pulses in dome growth. The results for different magma chamber crystal contents (Fig. 12) reflect the extreme sensitivity of the rheological behaviour in a crystal-rich magma close to the close-packed condition (Fig. 8). We emphasize that the location of this strong rheological transition

DYNAMICS OF MAGMA ASCENT AND EXTRUSION

Fig. 9. Distributions of (a) volume fraction of crystals ( ), (b) number density of crystals ( N ) , (c) average crystal radii ((R) = (4 /37 /N) l / 3 )), (d) magma undercooling (AT), and normalized crystal growth (e) and nucleation rates (f) along the conduit.

163

164

O. MELNIK & R. S. J. SPARKS

Fig. 10. (a) Relationship between the average crystal radius at the Earth's surface with the depth of nucleation and (b) crystal size distribution, for different magma discharge rates.

is not well constrained, which makes it hard to interpret the data in a fully quantitative way. Curves are strongly sigmoidal, which again allows for multiple steady solutions, and marked changes in magma discharge rate as a consequence of very small changes in dome height, chamber overpressure or magma chamber crystal content. Similar sensitive relationships can be seen for minor changes in conduit diameter (Fig. 13). These calculations are steady-state, but the adjustment from one steady-state to another will not be sudden. The flow system must adjust within an unsteady regime, which may be long due to the very high magma viscosity. This topic should be the future development of conduit flow models.

Discussion We have developed a model of magma ascent in lava dome eruptions that describes conduit flow from the chamber to the base of the dome. This model takes account of gas exsolution, gas loss by permeable flow to the surface, the kinetics of crystallization, and large changes in viscosity that accompany these processes. This model shows strong non-linear feedback between dome growth and conduit flow related to increase in dome height, gas loss and crystallization. We now interpret geophysical and eruptive phenomena at Soufriere Hills Volcano and other similar volcanoes in terms of the modelling results.

Fig. 11. (a) The relationship between discharge rate and magma chamber overpressure and (b) possible paths of unsteady eruption. Curves are contoured in terms of dome height from 0 to 300m. Conduit diameter is 30m and chamber crystal content 0.6.

DYNAMICS OF MAGMA ASCENT AND EXTRUSION

165

Fig. 12. Variation of discharge rate as a function of chamber crystal content for the rheological properties of magma fixed by calibration of data of Figure ! Conduit diameter is 30m. Contours are dome height (m).

Fig. 13. Influence of conduit diameter on the discharge rate for two different chamber overpressures. Magma chamber crystal content is 0.6. Contours are dome height (m).

Ground deformation Our results confirm previous results (Sparks 1997; Melnik & Sparks 1999) that the sharp rheological gradients caused by gas exsolution as magma ascends result in large magma overpressures in the conduit with an overpressure maximum. The model predicts that the overpressure maximum should be several megapascals in the uppermost parts of the conduit. The depth and magnitude of the pressure

maximum depend on the permeability of the magma, the density of the magma in comparison with wallrocks, and the discharge rate. Our predictions of the magnitude and depth of an overpressure maximum agree with observations at Soufriere Hills Volcano. Shepherd et ai (1998) observed a simple radial deformation pattern around the volcano that required a shallow pressure source below the dome. The model of pressure source of 700 m depth with overpressure of 10 MPa provided the best fit to the data, assuming

166

O. M E L N I K & R. S. J. SPARKS

elastic behaviour of the volcano. Voight et al, (1999) modelled deformation cycles measured by tiltmeter at Soufriere Hills in terms of pressure sources at about 400m below the dome. The calculations in Figure 14 show the sensitivity of overpressure maximum and its depth to changes in (a) magma crystal content and (b) conduit diameter for the same conditions as in Figures 12 and 13 with zero chamber overpressures. The tendency is for both overpressure maximum and depth to increase with increasing discharge rate, except at intermediate discharge rates where the curves are sigmoidal. These results form the basis for interpreting ground deformation patterns. There were some major changes in ground deformation during the eruption, which

coincided with increases in magma discharge rates. One change occurred around 20 July 1996 when outward rates of movement on network lines radial to the dome increased substantially (Shepherd et al. 1998; Jackson et al. 1998). These increases in deformation rate coincided with an increase in magma discharge rate in late July 1996 (Fig. 1), which was associated with a short-lived increase in discharge rate to over 10m 3 s -1 . and generation of some major dome collapses and pyroclastic flows (Sparks et al. 1998). During May 1997. discharge rate increased from about 3 m 3 s - 1 to about 7.5m3's-1 and was subsequently maintained at these high levels until the dome stopped growing in March 1998. May 1997 also saw a significant increase in the rate of outward movement of lines to

Fig. 14. Value and position of overpressure maxima (a) for variation of chamber crystal content (as in Fig. 12a) and (b) for variations in conduit diameter (as in Fig. 13a). Solid lines: for left plots are contours of overpressure (MPa); for right plots are depth of the pressure maximum (m) measured from the top of the lava dome. Dashed lines show the corresponding dome height (levels 0, 150 and 300m) as in Figures 12a and 13a.

DYNAMICS OF MAGMA ASCENT AND EXTRUSION

the north and east of the volcano (R. A. Herd, pers. comm.). These movements subsided when the dome stopped growing. Our model suggests that sudden changes in deformation rate reflect the changes in discharge rate causing relocation of overpressure maximum in the conduit and a change in its magnitude.

Seismicity and short-term cyclic behaviour Dome growth is associated with characteristic shallow earthquakes (Chouet 1996). Earthquakes with both high and low frequencies have been termed hybrid earthquakes at Soufriere Hills Volcano (Miller et al. 1998). Earthquakes dominated by monochromatic low frequencies are termed long-period earthquakes. The hybrid and long-period earthquakes at Soufriere Hills Volcano occur at depths between 0 and 2km and are numerous, with tens to hundreds of events per day. Our model shows that high overpressures develop in the uppermost several hundred metres of volcanic conduits. The values of overpressure calculated in our model are comparable to, or may even exceed, the typical strengths of the highly crystalline magma and conduit wallrock. Voight et al. (2002) give measurements of rock strength of the andesitic lava of a few megapascals. If the gas pressures are approximately equal to total pressure, then the rocks are further weakened. Conditions are thus favourable for hydrofracturing, with movements of high-pressure gases causing the hybrid earthquakes, with low frequencies due to resonance from gas flow (Chouet 1996; Neuberg et al. 1998). Other evidence for this interpretation of the shallow seismicity comes from tuffisite veins in samples of conduit wallrock, and from venting of gas and ash from fractures in lava domes, which occurs simultaneously with longperiod earthquakes (Gil Cruz & Chouet 1997; Sparks 1997). Hybrid earthquakes occur in swarms that typically last a few hours to a few days (Miller et al. 1998). They are associated with quasi-periodic ground deformation and eruptive cycles (Voight et al. 1999), with hybrid swarms building up in intensity during inflation and inferred magma pressurization. The hybrid swarm declines as the cycle passes through its peak and various eruptive phenomena develop in the deflation period, including accelerated dome growth, Vulcanian explosions and vigorous ash- and gasventing. Voight et al. (1999) attributed the cycles to build-up of overpressure, which reaches a failure condition manifested by a spurt of dome growth or an explosion or onset of ash- and gasventing when a pressurized fracture reaches the dome surface. Our model can contribute to understanding this short-timescale cyclic behaviour. The model predicts that high overpressures exceeding the strength of the system will be developed at steady-state. This is inherently unstable as the system will fail repeatedly as it attempts to build up overpressure to achieve equilibrium. True equilibrium may never be attained because the system is too weak to sustain the high overpressures predicted. We note that our current model does not yet incorporate features that are likely to contribute to the cyclic patterns, notably the non-Newtonian rheology and strength that develop in highly crystalline degassed magma. The possibility of stick-slip behaviour has also been discussed elsewhere (Voight et al. 1999; Denlinger & Hoblitt 1999; Wylie et al. 1999) in application to the short-period cyclic behaviour.

Porosity of eruption products The model suggests that there will be a porosity maximum in the uppermost part of the conduit. In one respect our model is likely to be unrealistic with regard to porosity and gas loss in the dome itself. We have assumed that the dome is effectively a continuation of the conduit. The model only considers vertical gas flow, but within the dome gas should be lost laterally, as a consequence of its fractured and permeable character. We thus surmise that the porosity will reach a maximum close to the top of the conduit. Robertson et al. (1998) present measurements of dome samples and ballistic blocks ejected during the sub-Plinian explosive activity of 17 September 1996. The ballistic blocks have angular shapes and are thought to have originated by brittle fragmentation of the upper-

167

most parts of the conduit, after dome collapse removed about 200 m of overburden above the conduit. Samples of the dome from the pyroclastic flows that immediately preceded the sub-Plinian eruption had densities of 2100 to 2400 kg m-3, equivalent to porosities of 10-20%. The ballistic clasts had densities of 1200 to 2000 k g m - 3 , indicating porosities of 20-50%. These observations support the existence of the predicted high-porosity zone at the top of the conduit.

Explosive activity The model has implications for the interpretation of explosive activity at Soufriere Hills. Overpressures are predicted in the upper conduit at values that may exceed the strength of the conduit wallrocks and lava dome. This pressure distribution accentuates the effects of unloading after a major dome collapse, which can easily lead to favourable conditions for explosive eruption (see also Melnik & Sparks 2002). All the periods of explosive activity have followed major dome collapses with removal of 100 to 200m of overburden dome. The system is thus critically poised for transitions in eruptive style. The pressure gradients in the upper parts of the conduit are steep and so explosive intensity is predicted to increase initially as the fragmentation wave propagates down the conduit into deeper, higher-pressure magma (Melnik & Sparks 2002). The Vulcanian explosions at Soufriere Hills Volcano typically lasted a few tens of seconds (Druitt et al. 2002). Eruptive velocities and calculated magma overpressure escalated rapidly over first the 10 seconds of each individual explosions. Pumice clasts are angular and often have a platy shape similar to the shapes in particles generated by experimental simulations of volcanic explosions (Adilibirov & Dingwell 1996). These observations support an interpretation that explosive eruptions involve brittle fragmentation of a high-porosity gas-pressurized zone in the upper conduit. We note that the non-lithostatic pressure conditions preclude using estimates of pressures from explosion dynamics for calculation of explosion depth.

Evolution of the eruption We now discuss the evolution of the eruption in the context of the model results. The main variables that control discharge rate and dome growth are chamber pressure, magma rheology and conduit dimensions. Other controls include permeability of the magma and conduit wallrock and crystallization kinetics. The major observables are magma discharge rate and dome height. The discharge rate varied by a factor of over 100 (0.1 to over 10m 3 s - 1 ) and dome height reached up to 300m. There were significant pulsations on a variety of timescales and some fairly abrupt changes in discharge rate, which were associated with changes in ground deformation. There are unfortunately more controlling variables than observables, so the model cannot be used to provide unique interpretations. Furthermore, in situations where microlite crystallization timescales are comparable to ascent timescales strong feedbacks develop, as discussed earlier. We now consider each parameter in turn and postulate that variation in magma chamber overpressure is the major cause of the evolution of the eruption, rather than large changes in conduit dimensions or the viscosity of the magma in the chamber. The andesite has remained similar in composition, phenocryst content and petrological characteristics throughout the eruption, with minor and unsystematic fluctuations in bulk composition (Devine et al. 1998a; Murphy et al. 2000). There is, however, also evidence for microscale (thin-section) heterogeneity, disequilibrium, recent reheating and involvement of basaltic andesite magma intrusion into the magma chamber (Couch et al. 2001). Constraints on eruption temperatures are poor due to the disequilibria, and the high phenocryst and microphenocryst content makes the magma viscosity sensitive to subtle variations in crystal content.

168

O. M E L N I K & R. S. J. SPARKS

Melt viscosity and density variations may have contributed to discharge rate fluctuations. There is, however, no evidence that variation of melt viscosity is systematic with time and could explain the overall increases of discharge rate and maximum dome height with time (Fig. 1). Discharge rate is a strong function of conduit diameter. Petrological and observational constraints suggest a diameter of about 30m with an uncertainty at about 5m. Changes in conduit diameter could account for some of the variation in discharge rate, but there is no independent evidence to indicate substantial changes took place. Widening or narrowing of the conduit might be expected to be gradual processes and so could not account for fairly abrupt observed shifts in discharge rate. The only major abrupt changes to the uppermost parts of the conduit occurred in association with explosive activity. For example, explosive activity of 17 September 1996 is thought to have enlarged the conduit exit (Watts el al. 2002), but this change did not result in any marked discharge rate change. Over the 1995-1999 eruptive period magma discharge rate varied by a factor of 100. If the magma viscosity and chamber overpressure remained constant during the eruption, then the diameter would have to have varied by a factor of about 2.7. There is no evidence that such a large change in conduit dimension took place, except at the shallowest levels related to explosions. The major changes in ground deformation observed in July and May 1997 were followed by a substantial rise in magma discharge rate and dome height. These observations are more consistent with a significant increase in chamber overpressure. We therefore attribute a major control on the overall evolution of the eruption to variations in magma chamber pressure. More speculatively, we discuss the possible role of influx of mafic magma into the chamber in increasing magma chamber pressure. There is evidence that replenishment by hydrous basaltic andesite magma triggered the eruption and that this magma continued to move into the chamber during the eruption (Murphy el al. 1998, 2000). The influx of mafic magma can cause an increase in chamber pressure in four ways. First, magma chamber pressure must increase as mafic magma invades into the shallow chamber (Blake 1984; Folch & Marti 1998; Snyder 2000). Second, heat transfer between the magmas causes remelting of the crystals in magma and wallrocks, which involves positive volume changes and pressure increases (Huppert & Sparks 1988). Third, the mafic magma must cool and crystallize. If the mafic magma is water-saturated, which is inferred to be the case from the stability of hornblende (Barclay el al. 1998), then pressure will increase (Tait et al. 1989; Folch & Marti 1998). Fourth, if emplacement of mafic magma results in convective motions of volatile-saturated magma (Couch et al. 2001), then uprise of bubbly magma can also result in pressure increase. When the feedback between crystallization in the conduit and chamber pressure is taken into account, it is plausible to explain the entire range of discharge rates by changes of a few megapascals in chamber overpressure. The main features of the volume and height data are that the eruption escalated and was punctuated by a number of fairly abrupt accelerations. The dome stabilized at some height following a period of increased dome extrusion rate (Fig. 1). Notably the dome regrew to this stable height quite quickly after a major dome collapse. Progressive chamber pressure increase can explain the general time-averaged increases in dome height and discharge rate. If mafic magma influx is the main reason for this postulated increase, then all of the dynamics discussed above can contribute to increase in chamber pressure. Heating of the andesite and volatile transfer could also have lowered magma viscosity, although we have no direct evidence that this was so. It is also possible that the increasing magma chamber pressure with time widened the conduit, although there is no independent evidence to indicate that this was substantial. Thus if magma viscosity or conduit diameter changed during the eruption they would be in the same sense as chamber pressure and caused ultimately by the same process, namely influx of mafic magma. The causes of pulsations of activity are inevitably more speculative. There are, however, plenty of suspects. We have shown that when microlite crystallization occurs there are inherent

instabilities and multiple stable states (Figs 11-14). Dome collapses can also trigger changes of state. However, several of the pulsations (such as those of February 1996. 20 July 1996. May 1997 and 21 June 1997) preceded, rather then followed, major dome collapse episodes and were associated with deeper volcanotectonic earthquake swarms, changes of cyclic deformation periodicity, and significant ground deformation changes. Our model suggests that ground deformation changes are ultimately caused by higher discharge rates and greater chamber overpressures. The processes causing the pulsations are therefore most plausibly located in the chamber. The various processes that can increase pressure in a chamber operate on various timescales (remelting. convection, volatile exsolution and transfer and convection) and can involve fairly abrupt events such as convective overturn, volatile saturation and crossing a crystallization phase boundary. Further discussion of chamber processes is beyond the scope of this paper, but is clearly a worthwhile avenue for future modelling research.

Conclusions We have demonstrated that the interpretation of lava dome growth in terms of conduit flow 7 dynamics becomes substantially more complex when the processes of gas exsolution, gas loss and crystallization are incorporated into the model. Steady-state solutions can be multiple for fixed system properties, making possible major fluctuations in magma discharge rate. Furthermore, the model indicates considerable sensitivity of magma discharge rates to slight changes in the physical system. A number of features of the model allow some new interpretations of the geophysical and eruptive phenomena observed at Soufriere Hills Volcano in the 1995-1999 period and at other similar volcanoes. (1)

(2)

(3)

The model predicts that large overpressures can develop in the uppermost parts of the conduit with a pronounced overpressure maximum. Values of overpressure can exceed the strength of wallrocks and the magma itself. The model provides a framework for interpreting ground deformation patterns, which are dominated by shallow-level pressurization in the conduit, rather than a deep chamber, and for understanding shallow-level seismicity and the propensity for short-lived Vulcanian explosions. The model also supports the interpretation that the escalation of the Soufriere Hills eruption during 1995-1999 was principally caused by increasing magma chamber pressure due to the influx of hydrous basaltic andesitic magma.

While dynamical models help improve understanding of monitored data they do not necessarily decrease uncertainty. Our model still lacks some aspects of the physics, yet it already shows strongly nonlinear and complex behaviours. Our model also investigates only steady-state solutions. In fact the high viscosity of lava-dome systems makes it likely that eruptions will be unsteady, with a richer range of behaviours when time dependency is investigated. The authors acknowledge a grant from the Royal Society for the visit of O. Melnik to Bristol University. NERC Research Grants GR3 11683. GR3-11020 and GR3 10679. R.S.J.S. acknowledges support from the Leverhulme Trust (F /182 AL) and NERC through a Research Professorbhip, 0,M, acknowledges support from the grant of Russian Foundation for Basic Research (99-01-01042). Comments by H. Huppert. H. Mader. A. Woods, A. Neri and T. Druitt were much appreciated.

Appendix

Analytical model for the conduit flow Here we will consider a simple conduit flow model to show qualitatively the influence of crystallization on magma discharge rate.

169

DYNAMICS OF MAGMA ASCENT AND EXTRUSION

We will model magma as a Newtonian fluid with viscosity depending on volume fraction of crystals in such a way that it has constant value for 0 < (where is a critical crystal fraction), — for > For simplicity we will neglect the presence of dissolved gas and bubbles and assume the density of magma to be constant. Constant crystal growth rate is assumed. The system can then be modelled by the set of equations:

density of crystals are also constant. We introduce the following dimensionless variables and simplify the momentum equation with the help of the mass conservation equation. Finally we have:

(A3)

(A2)

We will omit the prime in respect of dimensionless variables to simplify notations. Equations (A3) can be integrated with boundary conditions of Equation A2:

The notation is the same as for Equations 10 except that n is the number density of crystals. Equation Al consists of conservation of mass, number density of crystals and momentum in rising magma. With constant magma density, the velocity of magma and number

(A4)

Here x is the level in the conduit (measured from the magma chamber) where the critical concentration of crystals is reached. Combining Equations (A4) one can obtain the following equation for the magma ascent velocity:

(A5) Equation (A5) is a quadratic polynomial on V. Its solution is limited to velocities corresponding to .x* < 1. Also for Equation (A5) to have two real roots it is necessary that its discriminant is not negative. This condition leads to:

For ,x* < 1 it is necessary that:

Therefore the full solution of (I) can be represented as follows:

(A6)

Fig. Al. Variation of (a) dimensionless magma ascent velocity and (b) height, at which the critical volume fraction of crystals is reached, with dimensionless magma chamber pressure. Curves are labelled with the value of ijj which represents the crystal growth rate.

If — 0 or 2 = 1, Equations A6 has only one solution,V 3, which is a representation of the classic Pouiseille solution for constant magma viscosity. It is clear from Equations A6 that only two parameters ( 2 and w = /A0) determine discharge rate for fixed chamber pressure. Pressure p\ depends only on w therefore change in viscosity ratio will only shrink or expand lower branches of solution, leaving the transition point to the upper regime unchanged. Figure Al shows (a) dimensionless magma ascent velocity and (b) the value of x* as a function of chamber pressure (p c h ) for — 10. Parameter 1 is labelled on the curves. The relationship between chamber pressure and discharge rate has the same form as found numerically for Equations 10 and already discussed in relation to Figure 11. Therefore the existence of multiple steady-state solutions represents a fundamental feature of magma ascent with crystallization.

170

O. M E L N I K & R. S. J. S P A R K S

References ALIDIBIROV, M. A. 1988. Physical nature and mechanism of volcanic explosions. PhD thesis. Institute of Volcanology DVO. Russian Academy of Sciences. Petropavlovsk-Kamchatski. ALIDIBIROV. M. A. & DINGWELL. D. B. 1996. Magma fragmentation by rapid decompression. Nature, 380, 146-148. ASPINALL, W. P.. MILLER, A. D.. LYNCH. L. L.. LATCHMAN. J. L., STEWART. R. C. WHITE. R. A. & POWER, J. A. 1998. Soufriere Hills eruption. Montserrat: 1995-1997: Volcanic earthquake locations and fault plane solutions. Geophysical Research Letters. 25, 3397-3400. BARCLAY. J.. CARROLL. M. R.. RUTHERFORD. M. J.. MURPHY. M. D.. DEVINE. J. D.. GARDNER. J. & SPARKS. R. S. J. 1998. Experimental phase equilibria: constraints on pre-eruptive storage conditions of the Soufriere Hills magma. Geophysical Research Letters. 25. 3437-3440. BARMIN, A. A. & MELNIK. O. E. 1993. Features of eruption dynamics of high viscosity gas-saturated magmas. Fluid Dynamics. 28. 195-202. BLAKE. S. 1984. Volatile oversaturation during the evolution of silicic magma chambers as an eruption trigger. Journal of Geophysical Research, 89. 8237-8244. BLOWER. J. 2001. A three-dimensional network model of permeability in vesicular material. Computer Geosciences. 27. 115-118. CALDER, E. S., LUCKETT. R. & SPARKS, R. S. J. 2002. Mechanisms of lava dome instability and generation of rockfalls and pyroclastic flows at Soufriere Halls Volcano. Montserrat. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat. from 1995 to 1999. Geological Society, London. Memoirs. 21, 173-190. CHOUET. B. A. 1996. Long-period volcano seismicity - its source and use in eruption forcasting. Nature, 380, 309-316. CLARKE. A. B.. NERI, A.. VOIGHT, B., MACEDONIO. G. & DRUITT. T. H. 2002. Computational modelling of the transient dynamics of August 1997 Vulcanian explosions at Soufriere Hills Volcano. Montserrat: influence of initial conduit conditions on near-vent pyroclastic dispersal. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 319-348. COLE, P. C., CALDER. E. S.. DRUITT, T. H., HOBLITT. R.. ROBERTSON. R. E. A., SPARKS. R. S. J. & YOUNG. S. R. 1998. Pyroclastic flows generated by the eruption of the Soufriere Hills Volcano. Montserrat. Geophysical Research Letters. 25. 3425-3428. COUCH. S., SPARKS. R. S. J. & CARROLL. M. R. 2001. Mineral disequilibria in lavas explained by convective self-mixing in open magma chambers. Nature. 411. 1037-1039. DENLINGER, R. P. & HOBLITT, R. P. 1999. Cyclic eruptive behavior of silicic volcanoes. Geology. 27. 459-462. DEVINE. J. D.. MURPHY. M. D., RUTHERFORD. M. J. ET AL. 199&a. Petrologic evidence for pre-eruptive pressure-temperature conditions, and recent reheating, of andesitic magma erupting at the Soufriere Hills Volcano, Montserrat. West Indies. Geophysical Research Letters. 25. 3669-3672. DEVINE, J. D., RUTHERFORD. M. J. & GARDNER. J. C. 1998b. Petrologic determination of ascent rates for the 1995-1997 Soufriere Hills Volcano andesite magma. Geophysical Research Letters. 25. 3673-3676. DINGWELL, D. B. 1998. Recent experimental progress in the physical description of silicic magma relevant to explosive volcanism. In: GILBERT. J. S. & SPARKS, R. S. J. The Physics of Explosive Volcanic Eruptions. Geological Society, London, Special Publications. 145. 9-26. DOBRAN, F. 1992. Non-equilibrium flow in volcanic conduits and application of the eruption of Mt. St. Helens on May 18 1980 and Vesuvius in AD79. Journal of Volcanology and Geothermal Research. 49, 285-311. DRUITT, T. H., YOUNG, S. R., BAPTIE, B. ET AL. 2002. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306. ElCHELBERGER. J. C., CARRIGAN, C. R.. WESTRICH. H. R. & PRICE. R. H.

1986. Non-explosive silicic volcanism. Nature. 323. 598-602. FAGENTS. S. P. & WILSON. L. 1993. Explosive Volcanic Eruptions V I I . The range of pyroclastic ejecta in transient volcanic explosions. Geophvsical Journal International, 113, 359-370. FOLCH, A. & MARTI, J, 1998, The generation of overpressure in felsic magma chambers by replenishment. Earth and Planefurv Science Letters. 163. 301-314.

GIL CRUZ. F. & CHOUET. B. A. 1997. Long-period events, the most characteristic seismicity accompanying the emplacement and extrusion of a lava dome in Galeras Volcano. Colombia, in 1998. Journal of Volcanology and Geothermal Research. 77. 121 158. HESS. K. U. & DINGWELL. D. B. 1996. Viscosities of hydrous leucogranite melts: a non-Arrhenian model. American Mineralogist. 81. 1297-1300. HORT. M. 1998. Abrupt change in magma liquidus temperature because of volatile loss or magma mixing: effects of nucleation. crystal growth and thermal history of the magma. Journal of Petrology. 39. 1063-1076. HUPPERT, H. E. & SPARKS. R. S. J. 1988. The generation of granitic magmas by intrusion of basalt into continental crust. Journal of Petrology. 29. 599 624. JACKSON. P.. SHEPHERD. J. B.. ROBERTSON. R. E. A. & SKERRIT. G. 1998. Ground deformation studies at the Soufriere Hills Volcano. Montserrat I: Electronic distance meter studies. Geophysical Research Letters. 25. 3409 3412. JAUPART. C. 1998. Gas loss from magmas through conduit walls during eruption. In: GILBERT. J. S. & SPARKS. R. S. J. (eds) The Physics of Explosive v 'olcanic Eruptions. Geological Society. London. Special Publications. 145. 73-90. JAUPART. C. & ALLEGRE. C. J. 1991. Gas content, eruption rate and instabilities of eruption regime in silicic volcanoes. Earth and Planetary Science Letters. 102. 413-429. KLUG. C. & CASHMAN. K. V. 1996. Permeability development in vesiculating magmas: implications for fragmentation. Bulletin of I'olcanologv. 58. 87-100. LEJEUNE. A. M. & RICHET. P. 1995. Rheology of crystal-bearing melts: an experimental study at high viscosities. Journal of Geophvsical Research. 100. 4215-4229. McLEOD. P. & TAIT. S. 1999. The growth of dykes from magma chambers. Journal of Volcanology and Geothermal Research. 92. 231-245. MACEDONIO. G.. DOBRAN. F. & N E R I . A. 1994. Erosion processes in volcanic conduits and application to the AD79 eruption of Vesuvius. Earth and Planetary Science Letters. 121. 137-152. MARSH. B. D. 1981. On crystallmity. probability of occurrence and rheology of lava and magma. Contributions to Mineralogy and Petrology, 78. 85-98. MATTIOLI. G. S.. DIXON. T. H.. FARINA. F.. HOWELL. E.. JANSMA. P. E. & SMITH. A. L. 1998. GPS measurement of surface deformation around Soufriere Hills Volcano. Montserrat. Geophvsical Reseach Letters. 25. 3417-3420. M E L N I K . O. E. 2000. Dynamics of two-phase conduit flow of high-viscosity gas-saturated magma: large variations of sustained explosive eruption intensity. Bulletin of Volcanology. 62. 153-170. M E L N I K . O. E. & SPARKS. R. S. J. 1999. Non-linear dynamics of lava dome extrusion, \ature. 402. 37-41. M E L N I K . O. E. & SPARKS. R. S. J. 2002. Modelling of conduit flow dynamics during explosive activity at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & K O K E L A A R . B. P. (eds) The Eruption of Soufriere Hills v'olcano, Montserrat. from 1995 to 1999. Geological Society. London. Memoirs. 21. 307-317. M I L L E R . A. D.. STEWART. R. C.. WHITE. R. A.. ET AL. 1998. Seismicity associated with dome growth and collapse of the Soufriere Hills Volcano. Montserrat. Geophvsical Research Letters. 25. 3401-3405. MURPHY. M. D.. SPARKS. R. S. J.. BARCLAY. J.. ET AL. 1998. Role of magma mixing in triggering the current eruption at the Soufriere Hills volcano. Montserrat. West Indies. Geophysical Research Letters. 25. 3433-3436. MURPHY. M. D.. SPARKS. R. S. J.. BARCLAY. J.. CARROLL. M. R. & BREWER. T. S. 2000. Remobilization of andesite magma by intrusion of mafic magma at the Soufriere Hills volcano. Montserrat. West Indies. Journal of Petrology. 41. 21-42. NAVON. O. & LYAKHOYSKI. V. 1998. Vesiculation processes in silicic magmas. In; GILBERT. J. S. & SPARKS. R. S. J. (eds) Physics of Explosive Eruptions. Geological Society. London. Special Publications, 145. 27-50. NEUBERG. J.. BAPTIE. B.. LUCKETT. R. & POWER. J. 1998. Results from the broadband seismic network on Montserrat. Geophvsical Research Letters. 25. 3661 3664. NORTON. G. E., WATTS. R. B.. VOIGHT. B. ET AL. 2002. Pyroclastic flow and explosive activity at Soufriere Hills Volcano. Montserrat during a period of virtually no magma extrusion (March 1998 to November 1999). In: DRUITT. T. H. & K O K E L A A R . B. P. (eds) The Eruption of Soufriuru Hills Volcano, Momsvrrnt, from 199:* to 1999. Geological Society. London. Memoirs. 21. 467-481.

DYNAMICS OF MAGMA ASCENT AND EXTRUSION PAPALE, P. 1999. Strain-induced magma fragmentation in explosive eruptions. Nature, 397, 425-428. RAO, C. N. R. & RAO, K. J. 1978. Phase Transition in Solids. McGraw-Hill, New York. ROBERTSON, R. E. A., COLE, P., SPARKS, R. S. J. ETAL. 1998. The explosive eruption of Soufriere Hills Volcano, Montserrat 17 September 1996. Geophysical Research Letters, 25, 3429-3432. ROBERTSON, R. E. A., ASPINALL, W. P., HERD, R. A., NORTON, G. E., SPARKS, R. S. J. & YOUNG, S. R. 2000. The 1995-1998 eruption of the Soufriere Hills volcano, Montserrat, WI. Philosophical Transactions of the Royal Society of London, Series A, 358, 1619-1637. RUTHERFORD, M. J. & HILL, P. M. 1993. Magma ascent rates from amphibole breakdown: an experimental study applied to the 1980-1986 Mount St Helens eruptions. Journal of Geophysical Research, 98 19 667-19 685. SHEPHERD, J. B., HERD, R. A., JACKSON, P. & WATTS, R. 1998. Ground deformation measurements at the Soufriere Hills Volcano, Montserrat, 2. Rapid static GPS measurements, June 1996-June 1997. Geophysical Research Letters, 25, 3413-3416. SIMAKIN, A. G., ARMIENTI, P. & EPELBAUM, M. B. 1999. Coupled degassing and crystallization: experimental study at continuous pressure drop, with application to volcanic bombs. Bulletin of Volcanology, 61, 275-287. SLEZIN, Y. B. 1983. Dispersion regime dynamics in volcanic eruptions. 1. Theoretical description of the motion of the magma in the conduit. Vulkanologiya i Seismologiya, 5, 9. SLEZIN, Y. B. 1984. Dispersion regime dynamics in volcanic eruptions. 2. Flow rate instability condition and nature of catastrophic explosive eruptions. Vulkanologiya i Seismologiya, 1, 23-35. SNYDER, D. 2000. Thermal effects of the intrusion of basaltic magma into a more silicic magma chamber and implications for eruption triggering. Earth and Planetary Science Letters, 175, 257-273. SOHN, H. Y. & MORELAND, C. 1968. The effect of particle size distribution on packing density. Canadian Journal of Chemical Engineering, 46, 162-167. SPARKS, R. S. J. 1997. Causes and consequences of pressurisation in lava dome eruptions. Earth and Planetary Science Letters, 150, 177-189. SPARKS, R. S. J., BARCLAY, J., CALDER, E. S. ETAL. 2002. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of the Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 409-434. SPARKS, R. S. J., BARCLAY, J., JAUPART, C., MADER, H., PHILLIPS, J. & STURTEVANT, B. 1994. Physical aspects of magma degassing I. experimental and theoretical constraints on vesiculation. Mineralogical Society of America Review, 30, 413-446.

171

SPARKS, R. S. J., BURSIK, M. I., CAREY, S. N., GILBERT, J. S., GLAZE, L., SIGURDSSON, H. & WOODS, A. W. 1997. Volcanic Plumes. John Wiley, Chichester. SPARKS, R. S. J., YOUNG, S. R., BARCLAY, J. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills volcano, Montserrat: November 1995 to December 1997. Geophysical Research Letters, 25, 3421-3424. SPARKS, R. S. J., MURPHY, M. D., LEJEUNE, A.-M., WATTS, R. B., BARCLAY, J. & YOUNG, S. R. 2000. Control on the emplacement of the andesite lava dome of the Soufriere Hills volcano by degassing-induced crystallization. Terra Nova, 12, 14-20. STASIUK, M. V., JAUPART, C. & SPARKS, R. S. J. 1993. Variations of flow rate and volume during eruption of lava. Earth and Planetary Science Letters, 114, 505-516. STIX, J., TORRES, R. C., NARVAEZ, M. L., CORTES, G. P., RAIGOSA, J. A., GOMEZ, D. M. & CASTONGUAY, R. 1997. A model for vulcanian eruptions at Galeras volcano, Colombia. Journal of Volcanology and Geothermal Research, 77, 285-304. STOLPER, E. 1982. Water in silicate glasses: an infrared spectroscopic study. Contributions to Mineralogy and Petrology, 81, 1-17. TAIT, S., JAUPART, C. & VERGNIOLLE, S. 1989. Pressure, gas content and eruption periodicity of a shallow crystallizing magma chamber. Earth and Planetary Science Letters, 92, 107-123. VOIGHT, B., SPARKS, R. S. J., MILLER, A. D. ET AL. 1999. Magma flow instability and cyclic activity at Soufriere Hills Volcano, Montserrat. B.W.I. Science, 283, 1138-1142. WATTS, R. B., HERD, R. A., SPARKS, R. S. J. & YOUNG, S. R. 2002. Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 115-152. WILSON, L. 1980. Relationships between pressure, volatile content and ejecta velocity in three types of volcanic explosion. Journal of Volcanology and Geothermal Research, 8, 297-313. WOODS, A. W. & KOYAGUCHI, T. 1994. Transitions between explosive and effusive volcanic eruptions. Nature, 370, 641-644. WYLIE, J. J., VOIGHT, B. & WHITEHEAD, J. A. 1999. Instability of magma flow from volatile-dependent viscosity. Science, 285, 1883-1885. YOUNG, S. R., SPARKS, R. S. J., ROBERTSON, R., LYNCH, L., MILLER, A. D., SHEPHERD, J. & ASPINALL, W. A. 1998. Overview of the Soufriere Hills Volcano and the eruption. Geophysical Research Letters, 25, 3389-3392.

Mechanisms of lava dome instability and generation of rockfalls and pyroclastic flows at Soufriere Hills Volcano, Montserrat E. S. CALDER1, R. LUCKETT2, R. S. J. SPARKS1 & B. VOIGHT3 1 Department of Earth Sciences, University of Bristol, Bristol BS8 IRJ, UK (e-mail: [email protected]) 2 British Geological Survey, Murchison House, West Mains Road, Edinburgh EH9 3LA, UK 3 Department of Geosciences, Pennsylvania State University, University Park, PA 16802, USA

Abstract: Between 3 January 1996 and 1 December 1998 at Soufriere Hills Volcano, Montserrat, there were 27 150 seismic signals attributable to rockfalls and pyroclastic flows. Large (1-4 x 10 6 m 3 ) and major (>4 x 10 6 m 3 ) dome collapses began to occur after the extrusion rate reached 2m3 s-1 and the lava dome exceeded 30 x 10 6 m 3 in volume. Large to major collapses occurred on 26 occasions and were usually associated with periods of elevated extrusion rate (6-13m 3 s - 1 ), intense hybrid earthquake swarms and/or inflation-deflation cycles of crater-rim ground deformation. During these cycles, gas-rich pulses of magma, 140 000-320 000m3 in volume, were intruded into the dome and, soon thereafter, dome collapses from the headwalls of shear lobes were generated. Large dome collapses also occurred after 10 March 1998, when magma extrusion ceased, but these had no seismic precursors. These events represent structural failures from oversteepened canyon-like walls and were followed by intense degassing, suggesting that gas pressure build-up within the relict dome may have played some role. Rockfall counts and durations established from seismic data show variations that correlate with extrusion rate. Using pyroclastic flow runout and rockfall duration data as proxies for event magnitude, power law relationships between frequency (total number of events) and magnitude have been found. Seismic signals associated with rockfalls and pyroclastic flows commonly comprise both high-frequency and long-period components. Intense degassing from the dome is interpreted as the source for the longperiod component. These results indicate that rockfalls and pyroclastic flows generated by dome collapse at Montserrat are not simply the result of passive dome failure, but are intimately related to discharge of pressurized gas and pulses of magma extrusion. Pyroclastic flows were usually sourced from lobe headwalls, where lava was hot and gas-rich and where fragmentation of the microvesicular andesite lava occurred more readily.

Lava dome eruptions commonly generate pyroclastic flows by collapse and disintegration of hot, microvesicular and usually crystalrich lava. Pyroclastic flows generated in this manner are generally short-lived, occur in discrete pulses, and comprise blocks and lapilli of dense juvenile material. Relatively modest pyroclastic flows of this type have been observed in numerous eruptions: Mont Pelee in 1902 and 1929-1932 (Anderson & Flett 1903; Lacroix 1904; Perret 1937), Santiaguito in 1977 (Rose et al 1977), Mount Unzen in 1990-1995 (Yamamoto et al 1993; Nakada & Fujii 1993; Nakada et al 1999), Colima in 1991 and 1998-1999 (Rodriguez-Elizarraras et al 1991) and Merapi Volcano in 1984-1998 (Boudon et al 1993; Abdurachman et al 2000; Voight et al 2000a,b). Such flows have been observed to travel a few to over 10 km from the vent, at speeds of as much as 60ms - 1 . Continued lava dome growth can generate pyroclastic flows repeatedly over a significant period of time, typically a few months to several decades (e.g. Mount Unzen, May 1991 to February 1995; Montserrat, March 1996 to at least December 2001; Bezymianny 1956-present). Yet, despite widespread occurrence, generation of these flows involves processes understood at only a basic level. Such flows arise from: (a) passive failure of unstable dome faces (e.g. Mellors et al 1988; Ui et al 1999); (b) failure of oversteepened lava-flow fronts or lava lobes (e.g. Sato et al 1992; Yamamoto et al 1993; Ui et al 1999; Sparks 1997; Voight & Elsworth 2000); and (c) explosive expansion or diffusion of pore gas within the dome (Sato et al 1992; Fink & Kieffer 1993; Voight & Elsworth 2000). Detachment of a significant mass of material from the dome can then generate a pyroclastic flow. The processes by which cascading debris transforms into a pyroclastic flow are not well understood. This phenomenon is generally attributed to the production of abundant fines by spontaneous disintegration of the lava: an inherent property of hot, crystal-rich lava containing pressurized gas in vesicles and connected pore space (Rose et al 1977; Mellors et al 1988; Sato et al 1992; Voight & Elsworth 2000). The ongoing eruption of Soufriere Hills Volcano, Montserrat, began on 18 July 1995 with a four-month period of phreatic activity centred in English's Crater (Fig. 1). In November 1995, an andesite lava dome started to grow within the crater and by late March 1996 the first small pyroclastic flows were generated by lava dome collapse (Young et al 1998). High extrusion rates (2-6 m3 s-1) from July to September 1996 led to a series of large dome collapses,

culminating in a magmatic explosive eruption on 17 September (Robertson et al 1998). Dome extrusion continued and, as English's Crater was open to the east, until March 1997 all pyroclastic flows were confined to the Tar River valley (Fig. 1). Subsequent overtopping of the southern, western and northern parts of the crater wall led to formation of significant deposit fans on all sides of the dome (Fig. 1). A detailed account of the pyroclastic flow chronology can be found in Cole et al (2002). Two series of Vulcanian explosions (Druitt et al 2002b) occurred in August and SeptemberOctober 1997, and a large debris avalanche and dome collapse occurred on 26 December 1997 (Sparks et al 2002) removing about 55 x 10 6 m 3 of the lava dome and volcanic edifice. Dome extrusion continued until around 10 March 1998, at which time the total amount of magma erupted since 1995 was 300 x 10 6 m 3 dense rock equivalent (DRE). With cessation of magma extrusion, seismic activity decreased. A major dome collapse on 3 July 1998 marked the onset of a series of collapses and explosions (Norton et al 2002) although no renewed extrusion was apparent. In November 1999, after 20 months of residual activity, a second phase of lava extrusion commenced. This paper deals with the period November 1995, when lava extrusion commenced, to November 1999. During the eruption of Soufriere Hills Volcano, pyroclastic flows have been generated by a number of different mechanisms (Cole et al 1998, 2002): (1) pyroclastic flows caused by the collapse of the lava dome; (2) surge-derived pyroclastic flows (Druitt et al. 2002a); and (3) pyroclastic flows formed by collapse of vertically directed eruption columns (Druitt et al 2002b; Clarke et al 2002; Norton et al 2002). Activity has been dominated by dome-collapse pyroclastic flows, herein simply referred to as pyroclastic flows. The behaviour of these flows, and the nature of their respective deposits, are largely similar to those of dome-forming eruptions elsewhere (Anderson & Flett 1903; Yamamoto et al. 1993; Boudon et al. 1993; Cole et al. 2002). During the period November 1995 to December 1999, over 165 x 10 6 m 3 of rock debris was shed from the dome as pyroclastic flows, the deposits of which inundated over two-thirds of the volcano's flanks (Fig. 1). Understanding dome stability and the mechanisms of pyroclastic flow formation are key objectives in volcanology. Previous investigations (Mellors et al. 1988; Sato et al. 1992; Yamamoto et al 1993; Ui et al 1999) were based on detailed photographic

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 173-190. 0435-4052/02/ 15 © The Geological Society of London 2002.

173

174

E. S. CALDER ET AL.

Fig. 1. Map of southern Montserrat with distribution of total accumulated (November 1995-March 1998) pyroclastic flow deposits. Topography exerted a major control on flow paths and, therefore, on the spatial development of the depositional fans over time (Cole et al 2002). The volumes (x 10 6 m 3 ) of pyroclastic flow material accumulated within each of the main drainage systems are given. The deposit volumes within the Tar River and White River valleys are underestimates of the actual volumes discharged in those directions, as many of the flows there entered the sea. Key locations, such as the positions of selected seismic stations (black dots) and the Chance's Peak tiltmeter. are given. English's Crater and the locations of Galway's Wall and Castle Peak are shown in the bottom left-hand corner. The section line through English's Crater wall marks the position of the section in Figure 6c.

documentation of collapsing lava lobes and their spontaneous disintegration. Theodolite measurements of the advancing lobes at Mount Unzen Volcano constrained critical overhanging angles at which failure and detachment occurred (Yamamoto et al. 1993). Tilt and seismic precursors to dome collapse were recognized at Merapi Volcano (Voight et al. 2000a, b). In this paper, we consolidate data from several monitoring methods used at the Montserrat Volcano Observatory (MVO) to examine the mechanisms of dome collapse and pyroclastic flow generation. Measurements of magma extrusion rates, crater-rim ground deformation, and seismicity associated with the Montserrat pyroclastic flows have been an integral part of MVO monitoring. We investigate how these processes influence dome instability and, subsequently, pyroclastic flow generation, thus enabling some advances in understanding of the causative mechanism of dome collapses.

Lava dome instability Dome-collapse phenomena Relatively simple 'gravitational* collapse of lava domes commonly generates pyroclastic flows that are referred to as 'Merapi type' flows (Escher 1933; Voight et al. 2000a). These represent the large-scale end-member of the spectrum of lava-dome-collapse phenomena. Collapses of the Soufriere Hills lava dome range from conventional lava block rockfalls. through small, pyroclastic flows to major collapses involving as much as 55 x 10 6 m 3 of material. Rockfalls represent the smaller-scale end-member of collapse phenomena, and range from assemblages of discrete blocks (sometimes up to 20m diameter or more) that roll, bounce and slide downhill, to avalanches of blocks that produce substantial volumes

MECHANISMS OF LAVA DOME INSTABILITY

of fine ash. Rockfalls rarely travel far beyond the base of the talus apron 0.5-0.8 km from the dome summit, but they can involve moderately large quantities of material. On the basis of visual observations, rockfalls can be distinguished as either (1) passively generated, occurring during periods of little apparent growth, and relatively low activity, or (2) actively generated, occurring during periods of more intense, continuous or semi-continuous rockfall activity associated with new growth, or with recent changes in the locus of active growth. Passively generated rockfalls occurred typically as isolated events on inactive flanks. Relatively little elutriation of fine ash is produced during these events. Ash appears to be mostly surface ash kicked up by blocks rather than fresh ash formed by comminution and block disintegration. Small, gently convecting clouds that formed above passive rockfalls rapidly lost coherency and deteriorated into diffuse, laterally drifting clouds that travelled only a few hundred metres. Many of these rockfalls were initiated by discrete, falling blocks that descended whole with little fragmentation, remobilizing talus en route and initiating slow grain flow. In contrast, actively generated rockfalls are derived from active or rapidly growing parts of the dome. Associated ash clouds commonly convected vigorously and rose as small coherent plumes for 300-1000m (Fig. 2a). This hot, more buoyant ash was formed by fragmentation during avalanching. These actively generated rockfalls graded into pyroclastic flows when larger volumes of similar material collapsed. During the period of magma extrusion, from November 1995 to March 1998, 20-140 rockfalls were commonly generated every day. Pyroclastic flows are distinguished from rockfalls by their larger size, longer runouts ( 0.5km), greater production of fine ash and appreciable buoyant hot ash clouds. Qualitatively, pyroclastic flows form when there is a larger proportion of ash produced by rapid, spontaneous disintegration of the collapsing dome material (Sato et al. 1992). Pyroclastic flows comprise a basal avalanche or blockand-ash flow of dense blocks, lapilli and ash and an overriding, dilute pyroclastic surge component. The surge component of these flows is, at least in part, derived by elutriation from the underlying avalanche (Fig. 2b). The basal avalanche is generally incandescent and poorly sorted (Calder et ai 1999; Grunewald et al. 2000; Cole et al. 2002). The pyroclastic surges had a significant lateral component of motion, but those associated with small pyroclastic flows were commonly very weak: the upper, buoyant portions of the surge are referred here to as ash plumes (Fig. 2c). These were generated sequentially along the flow path and rose a few hundred metres to several kilometres or more during large collapses (Bonadonna et al. 2002), producing high-rising thermals (Woods & Kienle 1994). In practice, there is a continuum of collapse phenomena, and thus for small events it is difficult to define rigorously whether a particular event is a rockfall or pyroclastic flow (e.g. Fig. 2a). Between March 1996, when the first pyroclastic flows were produced, and February 1998, small (<1 km) pyroclastic flows occurred on an almost daily basis. Deposit volumes were about 103-104m3 for small pyroclastic flows and 104-106m3 for typical, medium-sized (runouts of 1-4 km) pyroclastic flows, such as that of 12 May 1996, which travelled 2.9 km down the Tar River valley and reached the sea (Cole et al. 2002). Large dome collapses on Montserrat, defined here as those with deposit volumes 1-4 x 10 6 m 3 , occurred on 18 occasions throughout the eruption (Table 1). Eight dome collapses had deposit volumes >4 x 10 6 m 3 and are defined as major dome collapses. Large to major collapses represented significant events in the eruption chronology and therefore were periods when the MVO was on heightened alert. They occurred as discrete single-pulse events (<5 minutes' duration) or as sustained events, where a major portion of the dome collapsed over a period of a few minutes to as much as several hours (Table 1). During the longer collapse episodes (e.g. 11 April 1997, 180 minutes; 3 August 1997, 120-150 minutes), retrogressive failures excavated spoon-shaped cavities as much as 200m deep in the dome. On collapse, each large slice of dome generated discrete, large-volume, energetic flow pulses producing discrete peaks in seismic energy. These main flow pulses were to some extent masked by the quasi-continuous generation of

175

a

Fig. 2. (a) The Soufriere Hills lava dome in February 1997. The upper third of the dome is composed of coherent, blocky andesitic lava, with the lower two-thirds being an apron of loose talus. A rockfall is seen travelling down the eastern side of the dome into the upper reaches of the Tar River valley. Rockfalls sourced from active growth areas on the dome tended to generate small, but coherent, ash plumes such as the ones shown in this photograph, (b) A pyroclastic flow in the White River valley in November 1997. The flow is confined, with a low snout and overriding ash clouds. The average velocity was 8-16ms - 1 . (c) The same flow as shown in (b), but a few seconds earlier (note the position of the flow front relative to the valley corner kX\ The buoyant ash plumes generated above the flow dominate the scene, while the front of the basal avalanche is rarely observed. Ash plumes rise several hundred metres to 14km in the largest collapse events. The scale of the thin basal avalanche, relative to the several-kilometre-high overriding ash plumes, is one of the most striking features.

smaller-scale rockfalls that preceded, accompanied and followed large collapses. Deposit volumes of these flows ranged between 1 x 10 6 m 3 and 55 x 10 6 m 3 (Table 1), with runout distances from 3 km to 6.7km.

Table 1. Summary of large to major dome collapwx ofthe eruption, as defined hy deposit volumes (106 m3 or runout distances over 3km Collapse date

29 J u l . 96*

Deposit volume (xl06m*)

2.8/3.0(1)

Extrusion rate m3 S 1 )

Percentage of dome collapse

Volume calculation based on

DRE collapse volume (xl()6m3)

Hybrid swarms

Deposit map

2.3

Y

24.9

5.2

9.2

DRE dome volume (x!06m3)

Collapse duration (min)

Relation to lava dome growth

180

Occurred shortly after switch in dome growth, new lobe directed to NE Occurred shortly after switch in dome growth, new lobe directed to NE

31 Jul. 96*

--1.0

Deposit estimate

0.8

Y

-23.0

4.6

3.5

4 Aug. 96* 1 1 Aug. 96* 21 Aug. 96* 2 Sep. 96* 3 Sep. 96* 17 Sep. 96* 19 Dec. 96* 9 Jan. 97* 13 Jan. 97* 16 Jan. 97* 20 Jan. 97* 30 Mar. 97

-3.0 3.2/3.5(1) 2.9/2.9 (l> -2.0 -3.0 11.8/12.3 -1 -4.0 -2.0 -2.0 -3.0 -3.0 2.6 ( 2 ) /2.6 { l )

Scar/fan survey Scar/fan survey /deposit map Deposit map Scar/deposit estimate Scar/deposit estimate Scar/Tar R survey No. flows/runout No. flows/runout No. flows/runout No. flows/runout No. flows/runout Estimate/survey Wht R

2.3 2.7 2.2 1.5 2.3 9.5 3.1 1.5 1.5 2.3 2.3 2.0

Y Y N N Y N Y Y Y Y Y N

-23.0 -25.0 -26.0 -27.0 -26.0 -27.0 22.9 -3 1 .0 -34.0 -34.0 34.3 42.3

2.1 4.8 2.4 3.2 3.2 2.3 4.7 11.1 9.9 9.3 6.6 1.5

10.0 10.8 8.5 5.6 8.8 35.2 13.5 4.8 4.4 6.8 6.7 4.7

31 Mar. 97 1 1 Apr. 97 25 Jun. 97 3 Aug. 97 21 Sep. 97

-1.0 2.9 (2> /3.0 m 5.5 (2) /6.4 (1) 8.8(2)/9.(1) 13.6(2/14(I)

Estimate Wht R Estimate/survey Wht R Survey Mosq;Trant/Bclhm Estimate/survey Frt G Survey Tuitts/ Wht G/Trant

0.8 2.3 4.9 7.0 11.0

N N Y Y N

-40.0 -43.0 -68.0 -81.0 61.6

1.3 3.4 6.4 11.4 2.8

2.0 5.3 7.2 8.6 17.9

180 25 120 150 20 30

4 Nov. 97

-2.0

Survey Wht R /Fan

1.5

N

64.1

3.4

2.3

45 70

6 Nov. 97 26 Dec. 97* 3 Jul. 98* 5 Nov. 98 12 Nov. 98* 20 Jul. 99*

-6.0

Survey Wht R 'Fan Scar estimate Fan survey/scar estimate Scar estimate Scar estimate Scar estimate

4.6 46 15 19 0.8 2.5 4.2

N Y

-65.0 94.9 -93.0 -77.0 -76.0 -64.0

12.9 4.8 0 0 0

7.1 49.6 20.4 1.0 3.3 6.6

35 15 30 150

55 (3.4)

20 25 1.0 (4) -3.0(4) -5.0(4)

0

60 180 240 480 540 60 70 140 60 60 45

Occurred after new lobe had filled up collapse scar of 28 and 31 Jul. Occurred after new lobe had filled up collapse scar of 1 1 Aug. Occurred after new lobe had filled up collapse scar of 1 1 Aug. Occurred after new lobe had filled up collapse scar of 2/3 Sep. Related to new lobe on 12 Dec. Related to new lobe in SE overwhelming 17 Sep scar. m

Occurred after new lobe had filled up collapse scar of 9 Jan. Occurred shortly after switch in dome growth, new lobe directed to S Related to new lobe on 30 Mar. Related to new lobe on 30 Mar. Related to 17 May lobe overwhelming northern crater wall Occurred after switch in dome growth, new growth directed to W Occurred shortly after switch in dome growth, new lobe directed to NE Occurred shortly after switch in dome growth, new lobe directed to S Related to new lobe on 4 Nov. Related to lobe which filled up the collapse scar of 6 Nov. Occurred after cessation of lava extrusion Occurred after cessation of lava extrusion Occurred after cessation of lava extrusion Occurred after cessation of lava extrusion

* Dome collapses d u r i n g which a significant portion of the flow entered the sea. Valley abbreviations: Tar R, Tar River valley; Wht R, White River valley; Mosq, M o s q u i t o Ghaut; Frt G, Fort G h a u t ; Tuills, T u i l t ' s Ghaut; Wht G, White's C ' h a u l ; Trants, Tranfs Yard and Belhm, Belham River valley. Deposit volumes represent non-DRE volumes, as measured in the field or estimated. Data include: (I) volumes of surge deposits calculated from mapped areas and estimated thickness. (2) Taken from Calder et til. (1999). (3) Total volume of dome rock (25 x l() 6 m 3 ) and lalus (30 x 1()6 m 3 ) that entered the debris avalanche and pyroclaslic density current, as estimated from the volume of the collapse scar. The volume of dome rock and talus involved in the pvroclastic density current alone is estimated as 35 45 x 10 6 m 3 by Sparks et al. (2002). (4) Scar volume rather t h a n deposit volume. DRE values are calculated by assuming the dense rock has a density of 2600 kg m 3 the pyrochistic How deposits have a bulk density of 2000 kg m 3 and the dome and scars have a bulk density of 2200 kg m 3 The DRE collapse volumes are therefore calculated by m u l t i p l y i n g the deposit volumes (including, the surge component) by 0.77 and the scar volumes (indicated by (4)) by 0.84. Collapse volume is calculated using DRE collapse and DRE dome volumes. The data have been rounded to one decimal place. "Hybrid swarms' indicates the occurrence of precursory hybrid earthquake swarms: Y, yes; N, no discrete e a r t h q u a k e swarms, although individual hybrid earthquakes may have occurred; , no seismic activity. DRE dome volumes are based on dome surveys or, where w r i t t e n in italics, estimated by extrapolation of extrusion rate data. Extrusion rates are given as the 7-day r u n n i n g average (Sparks el al. 1998).

S

C AL D ER ET CE

r-

MECHANISMS OF LAVA DOME INSTABILITY

Velocity estimates for some pyroclastic flows have been determined by analyses of superelevation effects (Druitt et al. 2002a), seismic and energetic constraints (Sparks et al. 2002; Loughlin et al. 20020, b) and video footage (Calder 1999). Average flow-front velocities were of the order of 3-10 ms-1 for small flows with runout < 1 km, 5-20 m s-1 for medium flows with runout 1-3 km, and about 20-30 ms-1 for large to major dome-collapse flows, with runouts >3km. Overriding pyroclastic surges varied from small, diffuse clouds to highly energetic pyroclastic surges that inundated large sectors of the volcano's flanks. Peak velocities of 70-90 ms-1 (Sparks et al. 2002) have been estimated for the 26 December 1997 energetic pyroclastic density current that devastated 10km 2 of southern Montserrat. Velocity estimates and valley cross-sectional areas were used to calculate a peak flux rate of about 60000m 3 s-1 for the largest pyroclastic flow of the dome collapse of 25 June 1997. Quotient flux rates often exceeded valley retaining capacities, causing overspills to occur (Calder 1999; Loughlin et al. 2002a, b). Deposition and erosion by the flows themselves also modified topography and subsequently the flow path of following flows. Early pyroclastic flow deposition (e.g. during 25 June 1997 dome collapse; Loughlin et al. 2002a, b) reduced channel cross-sectional area, and enhanced the prospect of overspill of succeeding flows. Seismicity and generation of pyroclastic flows The seismic signals on Montserrat have been monitored by a maximum of 11 seismometers, including, since October 1996, an array of

Fig. 3. Typical seismic signals associated with dome collapse, (a) Rockfall signal starting at 23:28 on 27 June 1996 from Gages (MGAT) short-period station, (b) Seismic signal of the onset of continuous pyroclastic flow activity in the Tar River valley, at 11:50 on 17 September 1996 from Gages (MGAT) short-period station, (c) Seismic signal of long-period/ rockfall at 20:19 on 5 May 1997 from St Patrick's (MSPT) short-period station. This signal comprises a long-period signal immediately preceding the higher frequency rockfall signal. On each record the lines are ten minutes apart, and the tick marks are at one-minute intervals. All times are local (UTC — 4 hours).

177

between one and five three-component broadband (0.03-30 Hz) instruments. Volcanic earthquakes have been categorized into the following earthquake types: volcanotectonic, hybrid, tremor and long-period (LP) earthquakes (Miller et al. 1998). Collapses of the lava dome (rockfalls to pyroclastic flows), explosions, crater-wall landslides and lahars also generated characteristic seismic signals (Miller etal. 1998). Rockfalls and pyroclastic flows were associated with similar emergent, cigar-shaped waxing and waning seismic signals that had frequencies of 1-10 Hz and differed only in their durations and/or amplitudes (Fig. 3a, b). Source mechanisms and signal properties have been determined from broadband seismometers by Luckett et al. (2002). Two source mechanisms contributed to these seismic signals. (1) The flowing debris acted as a moving, rather inefficient surface source, generating a high-frequency (2-8 Hz) seismic wavefield. The amplitude of this high-frequency component depended on the magnitude of the event (i.e. the volume of the moving debris), but also on the relative positions of the seismometer and the flow. The signal at any station became stronger, then waned as the pyroclastic flow approached and subsequently moved past. Event durations were generally between 10 seconds and several minutes and were dependent on the runout distance of the rockfall or basal avalanche component of the flow. The large majority of these seismic signals were caused by small rockfalls that did not extend further than the edge of the dome talus, and which produced signals lasting for 40-120 seconds. (2) A long-period (LP) component (peak in spectral amplitude 1-2 Hz) occurred near the onset of most rockfall

178

E. S. CALDER ET AL.

signals (Fig. 3c). These consisted of an LP earthquake followed by a rockfall or pyroclastic flow within 10-20 seconds. Most LP signals detected at Montserrat are interpreted as having a shallow source within the dome (Neuberg et al. 1998). Luckett

el al. (2002) have shown that most dome rockfalls were preceded by, or contain. LP components throughout the signals. Synchronized field and seismic observations show that the initial LP component was sometimes associated with intense jets of gas and

C

CO

DC

Fig. 4. (a) Ten-minute-average RSAM (real-time seismic amplitude measurement) plots of 14 large to major dome collapses down the Tar River valley for 24-hour periods, (b) Ten-minuteaverage RSAM plots often large to major dome collapses down the other valleys for 24-hour periods. Valleys are abbreviated as follows: WR. White River valley; MG. Mosquito Ghaut: FG. Fort Ghaut: TG. Tuitt's Ghaut: TR. Tar River valley. Shaded windows mark the main dome collapse periods. RSAM peaks generated by minor pyroclastic flow activity (PF). ash-venting (AV). hybrid (H). tremor (T) and volcanotectonic (VT) earthquake swarms are indicated. Several of the volcanotectonic earthquake swarms registered in 1996 were reclassified in 1997 as being hybrid earthquakes, RSAM has frequency-dependent clipping, so a limitation of this data is that clipping of high-amplitude signals is not easily observed graphically. All data were taken from Long Ground seismic station, just to the north of the Tar River valley (MLGT. Fig. 1). and the collapse events are listed in Table 1. Collapse events such as that of 3 August 1997 (9.1 x 10 6 m 3 non-DRE). which occurred on the western flank of the dome, register as less energetic than smaller events on the eastern side.

179

MECHANISMS OF LAVA DOME INSTABILITY

ash from the dome surface in the area where failure occurred. Observations of gas jets immediately preceding collapse have also been made at Mount Unzen Volcano (Sato et al. 1992; Ui el aL 1999). In contrast to these findings, Uhira et al. (1994) identified three phases in rockfall and pyroclastic flow signals at Mount Unzen. They observed wedge-shaped blocks of lava falling from a steep face and distinguished (1) block fall, (2) collision with the ground and (3) flow transport as essential components in exciting the seismic signals. Yamamoto et al. (1993) and Uhira et al. (1994) reported that the volumes of pyroclastic flows shed from the Unzen dome (estimated by visual observation) were proportional to the maximum amplitude of seismic waves generated by the pyroclastic flow. Likewise, Norris (1994) used maximum seismic amplitude to investigate single-block failures at Mount St Helens. However, the area of broadband seismic amplitude envelopes is considered to be a more appropriate measure where pyroclastic flows occur continuously over a period of several hours (Norris 1994; Brodscholl et al. 2000). For Montserrat, broadband analyses of this type have not yet been undertaken. However, clearly the relationship between volume of pyroclastic flow and signal amplitude is not simple. Here, we present seismic-energy data as measured by real-time seismic amplitude measurement (RSAM), which measures the average amplitude of seismic signals using a sampling rate of c. 60 per second (Endo & Murray 1991). Figure 4 illustrates 10-minuteaverage RSAM plots for 24-hour periods on the days of each of the large to major dome-collapse episodes from the Long Ground shortperiod seismometer (see Fig. 1). The collapses that descended the Tar River valley (Fig. 4a) are directly comparable as the flow paths are approximately constant, while those that travelled down the remaining drainages are shown in Figure 4b. Peaks in RSAM for the 29 July 1996 flows (deposit volume 3.0 x 10 6 m 3 ) are larger in amplitude than some volumetrically larger and wider-spreading flows later in the eruption (e.g. 3 July 1998, deposit volume 20-25 x 10 6 m 3 ). Later flows travelled over a smooth surface of newly accumulated, unconsolidated deposits, which may have dampened the seismic signal. For a given collapse episode, correlation of seismic RSAM data with visual observations suggests that short, sharp RSAM peaks represented discrete pulses of large-volume, energetic flows. Minute-scale pulses in collapse intensity are better identified using broadband seismic signals. Six main pulses ranging from 30 to 240 seconds in duration were identified from the 26 December 1997 collapse (Sparks et al. 2002, fig. 6). Three pulses, each less than 120 seconds in duration, were identified from the seismic signal of 25 June 1997 (Loughlin et al. 2002b, fig. 3). Moreover, these major pulses can be correlated with layers in the deposits (26 December 1997, Ritchie et al. 2002) and eyewitness accounts (25 June 1997; Loughlin et al. 2002a) respectively. Sequential flow pulses that occurred on these timescales are attributed to the retrogressive failure of individual portions of the dome.

Fig. 5. Comparisons of (a) RSAM with (b) cyclic tilt records and (c) the total number of triggered earthquakes per hour for the period 18-22 May 1997. The triggered earthquakes are dominated by hybrid events that subsequently produce the cyclic, low-amplitude RSAM peaks. The large spikes in RSAM are rockfalls and small pyroclastic flows, produced at the onset of deflation. Spikes on tilt records represent noise (modified from Voight et al. 1998).

Most of the large dome collapses have been directly preceded by swarms of hybrid earthquakes (Table 1, Fig. 4a, b) (Miller et al. 1998; Voight et al. 1998). These swarms (more than ten events per hour) have been accompanied by cyclic crater-rim ground deformation (Voight et al. 1998, 1999). Hybrid earthquakes, the most prominent type of dome-related seismicity, have impulsive first arrivals and a long-period coda. Within swarms, hybrid earthquakes were often highly repetitive and occurred at regular intervals (less than a second to several tens of minutes), with similar waveforms and magnitudes (Miller et al. 1998). Hybrid earthquakes were located directly beneath the crater, at relatively shallow depths of <2 km (Aspinall et al. 1998), and their waveform characteristics are explained by interface waves generated at the boundary of gasrich, bubbly magma embedded in a solid (Neuberg 2000). Lava extrusion appears to slow down or stop altogether during these shallow earthquake swarms. Swarms typically lasted for 3-30 hours, although on occasion swarms of a few days' duration occurred, notably in October-November 1996. Peaks in rockfall and pyroclastic flow activity generally occurred as hybrid activity waned (Fig. 5a).

Table 2. Summary of pre-collapse cyclic eruptive activity as manifested by earthquake swarms, rockfall and pyroclastic flow activity, explosions and/or crater-rim deformation cycles Period

Cycle period (h)

Number cycles (before failure)

Peak tilt amplitude (/Ltrad)

Extrusion rate (ms - 1 )

Extrusion/volume per cycle (m 3 )*

Jul./Aug. 1996 Dec. 1996 Jan. 1997 May 1997 Jun. 1997 Jul./Aug. 1997 Sep./Oct. 1997 Dec. 1997

4 6-8 6-8 12-18 8-12 10-12 9.7f 8-10

>16 8 >15 >7 9 13 >75 1

-

7-12 4-6 6-12 4-6 5-8 5-9 6-12 7-8

137 000 6 000 130 000 3 000 238 000 0 000 28 1000 8 000 245 000 1 000 284 000 4 000 314000 105000 245 000 3 000

1-2 1-2 10-25 30-40 20-35 -

* Time-averaged (7-day) extrusion rates (Table 1 and Sparks et al. 1998) for the periods listed are used to interpolate individual pulse volumes (volume/cycle = cycle period x extrusion rate). fThe intervals between explosions of September and October 1997 varied from 3 to 33 hours with an average of 9.5 hours for the 75 explosions (Druitt et al. 2002b).

180

E. S. CALDER ET AL.

Since July 1996, hybrid swarms have been characterized by a pronounced cyclicity (Table 2). The relationship of seismic swarms to large and major dome collapses is complex. Some collapses were preceded by one to several hybrid swarms in the hours and days before collapse (Fig. 4b). The 25 June 1997 collapse was preceded by the onset of intense seismic swarms starting on 22 June (Voight et al. 1998; Loughlin et al 2002b). Similarly, the 3 August 1997 collapse was preceded by swarms starting on 31 July (Voight et al. 1998) and the 4-6 November 1997 collapse was preceded by swarms starting on 2 November. The 26 December 1997 collapse was preceded by an intense hybrid swarm that peaked in the preceding 24 hours (Sparks et al. 2002). On the other hand, one of the longest and most intense swarms, eight days in early December 1996, did not lead to a collapse although major deformation of the crater wall was produced. Indeed, there have been hundreds of intense swarms over the course of the eruption, and only a few have been followed by large collapses. However, the examples illustrated in Figure 4 suggest that the onset of intense swarms after a period of relatively low seismicity is indeed a precursor to large collapses, although the collapse does not necessarily occur in the first of such swarms to develop after a long quiescence. Large dome collapses that took place after lava extrusion had ceased present an exception, and were clearly not preceded by increases in seismic activity (Fig. 4a, b). Particularly shallow or intense earthquake swarms can also trigger rockfalls by directly shaking the edifice. From December 1996 to March 1997, hybrid earthquakes shook the edifice and crater wall, clearly triggering both dome rockfalls and 'cold' landslides from the denuded area of the south crater (Galway's) wall. Although this generated many small rock avalanches, the maximum runout of these flows never exceeded c. 1 km. Moreover, the first large earthquake in any swarm produced the largest flows, dislodging the bulk of recently accumulated unstable material. Subsequent earthquakes in any swarm dislodged minor amounts of material, producing relatively small flows. Luckett et al, (2002) demonstrate that seismic signals generated by runout of the cold, crater-wall landslides are similar to the dome rockfall signals, but lack the long-period component. The generation mechanism of crater-wall landslides and some dome rockfalls may thus be attributed to seismic-shaking-induced instability. This mechanism is not, however, considered dominant with respect to most collapses of the growing dome. Indeed, during the cyclic patterns of ground deformation and hybrid seismicity described by Voight et al. (1999), dome collapses and rockfalls occurred predominantly in the deflation parts of tilt cycles, during declining seismicity.

Cyclicity and pyroclastic flow generation A high-resolution tiltmeter was installed in December 1996 on Chance's Peak, c. 150m from the crater rim (Voight et al. 1998, 1999). From January 1997 to August 1997, tiltmeter data displayed cyclic variations that correlated with seismicity and rockfall or pyroclastic flow activity (Table 2). The system was destroyed in mid-January 1997 but was re-established in May 1997 and operated until 5 August when the tiltmeter components were destroyed by Vulcanian explosions. The tiltmeters were high-gain, bubble-type biaxial-platform tiltmeters, with a resolution of O.lurad. Digital time-series data were telemetered to the MVO at 8-minute intervals and analysed in near-real-time along with seismic RSAM data. In Deeember 19967 the tiltmeter revealed pressuflration cycles with low-amplitude (l-2 r ad), inflation/deflation cycles superimposed on a steady, long-term drift. The cycle period was 6-8 hours and was clearly correlated with RSAM cycles. In May 1997, tiltmeters displayed a stronger rhythmic pattern of repetitive inflationdeflation cycles. Cycle amplitudes were in the range 10-25 rad and each cycle lasted 12-18 hours, with the deflationary part of the cycle occurring more rapidly than the inflation (Fig. 5). Hybrid earthquakes occurred in increasing numbers as the inflationary part of the cycles progressed, sometimes merging into continuous tremor as inflation approached its maximum and then declined (Fig. 5). The

peak in rockfall and pyroclastic flow activity occurred just after the onset of deflation during the waning hybrid swarm. The high peaks in RSAM are due to rockfalls and pyroclastic flows, and the loweramplitude peaks are produced by the swarms of hybrid earthquakes. Tilt amplitudes declined after the end of May. but marked cyclic activity commenced again with an abrupt inflation at 05:30 local time on 22 June 1997. and was followed an hour later by a sharp deflation coincident with sustained pyroclastic flows down Tar River valley. This tilt excursion was the first in a series of nine, high-amplitude (c. 30/mid), short-period (8-12 hour) cycles that culminated in the major dome collapse on 25 June 1997. The deflation that coincided with the collapse event was more pronounced than for the preceding cycles. By 10 July 1997. tilt amplitude had flattened and periods extended for 30 hours. High-amplitude cyclicity resumed on 31 July, and a major dome collapse occurred on 3 August. Vulcanian explosions started on 4 August 1997. and the tiltmeter was destroyed on 5 August (Druitt et al. 2002b). Since then, hybrid earthquake swarms have continued to serve as a proxy to define pressurization cycles. The volume of magma extruded during each cycle has been estimated from time-averaged dome-extrusion rates and cycle periodicity (Table 2) and ranges from 140000 to 320000m 3 . For initial dome volumes (given in Table 1). each new cycle increased the dome volume by 0.4 0.2%. Assuming a simple geometry of an inflating half sphere, these volume increases correspond to cyclic inflation of only several tens of centimetres at the dome surface, as radius increases would have been 0.07-0.21%. Lava extrusion since April 1996 was. however, more characteristically marked by growth of discrete lobes along ductile shear zones in the upper portions of the dome (Watts et al. 2002). Active growth typically occurred either at the summit (Fig. 6a. b) or by advance of the steep headwall of a shear lobe (Fig. 6a. c). The headwalls of shear lobes had cross sections in the order of 10 4 m 2 . so a cyclic pulse of dome extrusion of order 1-3 x 105 m3 resulted in an advance of 10-30m of the flow front in a few hours. Lobe headwalls were critically unstable, so a pulse-like advance inevitably resulted in elevated rockfall and pyroclastic flow activity. Lobe advancement rates, measured by theodolite, were about c. 30m per day (on 6 March 1997) when the extrusion rate was 1 3m 3 s - 1 and cyclic hybrid earthquake swarms were occurring. Pyroclastic flow and rockfall activity occurring during the deflation stages of cycles normally comprised relatively small events, e.g. in May 1997 (Fig. 5). Large to major volume dome collapses occurred as the culmination of a period of between eight and 15 activity cycles, e.g. December 1996; January and June 1997. Major dome collapses, which occurred during magma extrusion, followed dome volume increases of 1-5 x 106 m3 over timescales of a few days. There are, however, some exceptions, e.g. the major 26 December 1997 collapse event occurred at the end of the first and only cycle in that period (although it was 37 hours long). During the August 1997 explosive period, Vulcanian explosions occurred at the end of each individual earthquake and tilt cycle (Druitt et al. 20026). The period of the September October 1997 explosions was marked by little evidence for cyclicity other than the explosions themselves. Hybrid earthquake swarms were absent and tilt data were unavailable.

Magma extrusion and dome size in relation to pvroclastic flow generation The rate of magma extrusion and dome growth has had a primary effect on eruptive style and has played a fundamental role in hazards assessment throughout the eruption (Sparks et al. 1998). For the first 11 weeks following the onset of dome extrusion in November 1995. lava extrusion rates were low (c. 0.5m 3 s-1r). After February 1996. extrusion rates increased to 2.1m 3 s - 1 (Fig. 7a) (Sparks et al. 1998). A further, marked increase occurred at the end of May 1997, with extrusion rates ranging from 4 to 1 3 m 3 s - 1 . with substantial fluctuations to December 1997. The tendency during the November 1995-March 1998 period has been for extrusion rate to increase with time (Sparks et al. 1998). although this is

MECHANISMS OF LAVA DOME INSTABILITY

181

Fig. 6. (a) A shear lobe on 15 April 1997, moving southwards towards the White River valley. The steep platey upper surface is a slickensided ductile shear zone of highly crystalline lava, with the headwall represented by the ragged surface of jointed lava (left). The shear zone within the dome controls the geometry of the extruded shear lobe, with the rate of shear lobe advance controlled by magma extrusion rate. Most pyroclastic flows originate from instability of the headwall, and the slight concavity of the surface shown is a collapse scar. Collapse of the headwall led to numerous pyroclastic flows in this sector during April-May 1997. Topographic profiles through the headwall are shown in (c). (b) Nine dome growth profiles determined photographically from a position 3 km north of the dome (Fig. 1, Harris Lookout) for the period 1 December 1996 to 27 March 1997. This period shows regrowth of the dome after the 17 September 1996 explosive activity. Summit growth occurred by the sequential extrusion of spines and vertically to southeasterly directed shear lobes (25 December 1996 and 21 January 1997 lobe; Watts et al. 2002). (c) N-S profiles through the southern wall of English's Crater (Galway's Wall) and growth of the 27 March and 13 April 1997 lobes taken from global positioning system/laser binocular data. The original topography of English's Crater is shaded (note that Galway's Wall deteriorated significantly during this period allowing rockfalls to be shed into the White River valley by 30 March) and the maximum level of excavation of the explosive activity of 17 September 1996 is indicated. Collapses occurred on 30 March and 11 April 1997 with subsequent growth occurring in the scar of the previous collapse. The headwalls of these southerly directed shear lobes represented large unstable masses perched above the Galway's Soufriere area. considered unusual in other dome-forming eruptions (Nakada et al. 1999). Extrusion rates for the Soufriere Hills dome (1-13 m3s-1) are, however, similar to those of many other dome-forming eruptions as reported by Newhall & Melson (1983). The c. 0.3 km3 of andesite erupted by March 1998 was partitioned into: 113 x 106 m3

(DRE) volume of the dome on 10 March 1998 when its growth ceased; 1 5 0 x l 0 6 m 3 volume of accumulated pyroclastic flow deposits; and 28 x 10 6 m 3 of explosive ejecta. Extrusion-rate pulses have occurred on timescales ranging from a few hours to months (Fig. 7a). Three long-term pulses

182

E. S. CALDER ET AL.

Fig. 7. Dome growth in relation to pyroclastic flow runout and deposit volumes, (a) Weekly running-average dome-extrusion rate (dashed line is shown where volume estimates are spaced at more than one month intervals), (b) Runout distances of pyroclastic flows. The runouts of major collapses that entered the sea have been approximated (dashed arrows), (c) Cumulative pyroclastic flow deposit volume. Modified from Cole et al. (1998).

lasting a few months have been identified with extrusion rates of 3-8 m3 s -1 These pulses reflect processes at depth, such as influxes of new magma into an open-system chamber (Sparks et al. 1998). Short-term pulses lasting a few days reach extrusion rates of over l O m 3 s - 1 , although 5-8m 3 s -1 represents a more realistic average. Pulsations of six to seven week cycles have been well established since late 1997 (Voight et al. 1999). The weekly running-average extrusion rate has been compared with pyroclastic flow runout and cumulative pyroclastic-flow volume (Fig. 7b, c). The runout distances of the collapses on 29 and 31 July, 4, 11 and 21 August, 2, 3 and 17 September and 19 December 1996 and on 9, 13, 16 and 20 January 1997 are minimum values, because these flows entered the sea at the end of Tar River valley. Recent submarine surveys suggest that 40-60 x 10 6 m 3 of 1995-1999 pyroclastic material lies in deep-water lobes extending 4km off the Tar River coast (S. Carey pers. comm.). This indicates that these flows were substantial in volume and that they and their derivative submarine flows had large runouts. Small pyroclastic flows occurred frequently throughout the eruption and were typically associated with periods of low to average extrusion rates (l-3m 3 s - 1 ), but account for only a small fraction of the total volume erupted. Large to major dome collapses were preceded and followed by periods of elevated extrusion rate (4-8 m 3 s- 1 ), (e.g. 29-31 July, 11 August 1996). In particular, the periods of late July and December 1996 were characterized by a closely spaced series of large dome collapses, which continued for a month. Dome growth was focused in localized areas by easterly directed shear lobes. These lobes rapidly produced oversteepened.

unstable flanks and generated relatively large collapses. Each collapse scar was infilled quickly with several million cubic metres of new lava within a few days, and the failure cycle was then repeated. A few periods may have reached less substantial extrusion rates. Only moderate extrusion rates (c. 2.1 m 3 s - 1 ) appear to have preceded the 17 September 1996 explosive eruption, although data during this period are not well constrained. Likewise, the MarchApril 1997 collapses occurred in association with magma discharge rates of 1-3m 3 s - 1 . These collapses were associated with the development of a particularly unstable shear lobe over Galway's Wall (Fig. 6b), and were preceded by increased numbers of longperiod earthquakes, suggesting increased gas and pressure build-up within the dome. The capacity for generating large-volume collapses clearly depends on the dome volume. Estimated collapse volumes have been normalized to respective initial dome volumes in Table 1 and Figure 8. Most large (1-4 x 10 6 m 3 ) and major (>4 x 10 6 m 3 ) collapses represented < 18 vol% of the dome with no systematic change (in actual collapse volume or percentage collapse volume) over eruption duration (Fig. 8a. b). Clearly, however, as dome volume increases, a collapse of c. 18 vol% will mean successively larger volume collapses. The explosive eruption of 17 September 1996. the debris avalanche and dome collapse of 26 December 1997 and the collapse of 3 July 1998 (post-extrusion collapse) were clear exceptions, representing outliers in the data with collapses of c. 35. c. 50 and 20vol% respectively (Fig. 8c; Table 1). In general, volume percentage collapse is therefore independent of dome volume (Fig. 8c) and average extrusion rate (Fig. 8d) for these ranges.

183

MECHANISMS OF LAVA DOME INSTABILITY

Fig. 8. (a) Measured collapse volume (DRE x 10 6 m 3 ) against time (eruption day) for the 25 dome-collapse events in Table 1. (b) Percentage collapse volume (calculated using collapse volumes and initial dome volumes from Table 1) against time. Percentage collapse volume plotted against (c) dome volume and (d) seven-day running average lava extrusion rate. The zone under the dashed line in (b-d) marks percentage collapse volumes of <18 vol%, within which most dome collapses lie.

For periods of up to several weeks, low cloud cover over Soufriere Hills obscured direct visual observations of the dome. Rockfall activity, as interpreted from seismic data, has thus been used as the principal indicator of the status of the volcano. There have been several periods, however, when rapid dome growth occurred without accompanying seismic activity (Sparks et al. 1998; Miller et al. 1998). Daily rockfall counts from the seismic records over a three-month period from 1 May to 1 August 1996 (Fig. 9) show marked fluctuations over a relatively short period, and a decline in rockfall frequency in July 1996. During the same period, however, rockfall amplitudes steadily increased. Dome volume estimates suggest that during that time the magma extrusion rate was increasing markedly (from c. 2 to c. 6m3 s - 1 ). The relationship between rockfall activity and extrusion rate is therefore complex

Fig. 9. The number of rockfalls per day over the period 10 May-1 August 1996. Rockfall duration data are largely constant, with a sharp increase at the end of July when many small pyroclastic flows were produced. The data show increasing rockfall amplitudes since late June, even though rockfall activity (as measured by event frequency) appears to have been low from late June to late July. Data shown in (b) and (c) are running averages of the raw data.

and rockfall counts need to be considered in conjunction with rockfall amplitude and duration data for meaningful interpretation. Local topography can also be important: when a shear lobe reaches a steep slope, rockfall activity is enhanced although, in some cases, rockfalls may occur less frequently while involving larger volumes.

Pyroclastic flow magnitude-frequency

relationships

It is instructive to characterize pyroclastic flow activity in terms of frequency (in number of events) versus magnitude (volume dislodged). For larger-scale dome collapses this is relatively straightforward and achieved by event recording and by surveys of deposit volumes. An indirect method of estimating pyroclastic flow volume

184

E. S. CALDER ET AL. Table 3. Flow runout for selected medium to major dome collapses Date

Runout (km)

Volume (x!06m3)

3 Apr. 96 12 May 95 31 Mar. 97 5 Jim. 97 17 Jun. 97 25 Jun. 97 30 Mar. 97 11 Apr. 97 3 Aug. 97 21 Sep. 97

1.6 2.9 2.5 3.1 3.9 6.7 3.6 4.1 5.6 6.0

0.15 0.33 0.16 0.38 0.77 5.5 2.6 2.9 8.8 13.6

Selected from Calder et al. 1999. Volumes of deposits of pyroclastic flows (non-DRE), excluding surge components (Table 1). The runout for 31 March 1997 is for flows shed down in the Tar River valley in the afternoon of that day.

from runout relationships can also be utilized (Calder et al. 1999). This scaling analysis, based on empirical data, suggests that runout length is proportional to volume to the one-third power (Dade & Huppert 1998; Calder et al. 1999). Selected, well constrained, volume and runout data for Montserrat dome collapse flows are summarized in Table 3 (and illustrated in Calder et al. 1999, fig. 2).

For rockfalls and small pyroclastic flows, the frequency-magnitude relationship is more accurately investigated using seismic data. Relative event magnitudes can be determined using parameters such as signal duration and or amplitude. Automatic triggering by seismic software ensures that even small events are accurately counted and that records are complete for night-time events, as well as for those when observation conditions are poor. Between 3 January 1996 and 1 December 1998. 27 150 rockfalls triggered the short-period seismic network. During triggering, magnitudes of events were not differentiated, so that larger pyroclastic flows were also incorporated. Their numbers are comparatively small, however, and do not significantly influence the trend. Rockfalls and pyroclastic flows with runouts of less than 1 km were commonly generated 20-150 times a day (Fig. 10). although during certain periods of the eruption rockfalls were almost absent. Low rockfall counts (<20 per day) were characteristic of periods when extrusion rate was either low or zero (before March 1996 and March-July 1998), or when the dome was small and the rockfalls produced were too small to trigger the seismic network (pre-1996 and 17 September-December 1996). For the period November 1995-September 1997, the maximum amplitude and duration of rockfall and pyroclastic flow signals were measured by MVO staff manually from the paper records of the short-period seismometers. Maximum-amplitude data were frequently clipped (true maximum not recorded) but rockfall and pyroclastic flow durations exhibited variations that could clearly be correlated to observed volcanic activity. The first three and a half

Fig. 10. (a) Number of rockfalls per day throughout the eruption, as counted by triggering of the broadband seismic network. Dates for the peak numbers are labelled above the figure, and the troughs below. Periods when rockfall frequency is low are 22 November-1 March 1996 (extrusion rate 0.5-2m 3 s -1). after the 17 September 1996 explosive activity (dome growth did not recommence again until 1 October and significant rockfall activity recommenced around 10 December), during the 22 September-21 October 1997 Vulcanian explosions, and after lava extrusion ceased (10 March 1998) until 30 June 1998. Periods characterized by elevated rockfall activity have occurred prior to large collapse events (e.g. 6-12 January and 11-26 December 1996). The major collapses of 25 June and 21 September 1997 were not preceded by rockfall spikes. The collapses of 28 July 1996 and 26 December 1997 occurred immediately after periods of anomalously low rockfall activity, (b) Weekly running-average extrusion rate for the same period.

MECHANISMS OF LAVA DOME INSTABILITY

185

Fig. 11. Rockfall durations in seconds for the period December 1995 to September 1997, as measured manually from the seismic paper records by MVO staff. Pulses in rockfall activity are labelled 1-4 and a-e, and gaps are labelled G1-G3. The periods of explosive activity of 17 September 1996 and August and September-October 1997 are also indicated.

months of dome extrusion were characterized by relatively few, short-lived rockfalls occurring in four discernible pulses, each of two to four weeks' duration (Fig. lla, 1-4). On around 16 April 1996, the rockfall record became more continuous, with significant numbers of rockfalls beginning to exceed 100 seconds in duration. These represent the first rockfalls and pyroclastic flows out of English's Crater, and they coincided with the development of the first shear lobe of the lava dome (25 April 1996 lobe), which by the end of April had advanced over the steep front of the 350-yearold Castle Peak Dome (Watts et al. 2002). An abrupt increase in rockfall activity occurred on around 8 May 1996 and four days later the first pyroclastic flows reached the sea 2.7km from the dome. Between the end of July and mid-September 1996, five distinctive peaks in activity occurred spaced at one to two week intervals (Fig. lla-e). This phase culminated in the 17 September 1996 sub-Plinian explosive activity. Subsequent growth of the 1 October lobe was characterized by sparse rockfalls until around 15 December 1996. The remaining period, January to September

Fig. 12. (a) Frequency histogram of rockfall duration data for the period December 1995 to September 1997 (Fig. 11). Durations are categorized in 10-second intervals, (b) Log plot for the rockfall duration (20-500 seconds) frequency data, (c) Frequency histogram of pyroclastic flow runout distances throughout the eruption. Runouts have been estimated to the nearest kilometre and are categorized in 1-km bins, (d) Log plot for the pyroclastic flow runout-frequency data. These data include the number of rockfalls taken from the seismic broadband triggers. All rockfalls are assumed to have runouts of <500m.

1997, was characterized by relatively high activity with subtle pulsations in rockfall numbers and durations. During this later period, conspicuous gaps in the data exist for around 10 March, 24 June and 3 August 1997 (Gl, G2, G3 respectively in Fig. 11b). A frequency histogram of rockfall and pyroclastic flow durations has been compiled from these data for the 22-month period December 1995 to September 1997, and the plot suggests an approximately normal distribution (Fig. 12a). For a subset of rockfalls and pyroclastic flows ranging in duration between 60 and 500 seconds (considered more realistic values), a linear relationship is generated on a log-log plot (Fig. 12b). Pyroclastic flow runout distance data were compiled from a number of sources (Fig. 12c). Commonly, runouts of observed medium to large flows were recorded in the MVO observations book and broadcast in daily reports. Flows that occurred at night were only observed seismically, so that runouts were ascertained by inspection of the deposits the following day. For small to medium flows, reported observations were augmented by information from

E. S. CALDER ET AL.

186

Signal Duration, D (s)

Fig. 13. Rockfall and pyroclastic flow frequency (in number of events. N) in relation to flow magnitude as determined by runout. R, for pyroclasuc flows and seismic signal duration. D, for rockfalls. the seismic paper records, which provided an accurate account of the number of events where "several' or "continuous small scale pyroclastic flow activity' was otherwise recorded. We also used the assumption that where 'several" flows were reported, five occurred and that flows recorded as being in the upper Tar River valley' had runouts of between 0.5 and 1 km. Figure 12d includes rockfalls as obtained from seismic-trigger event counts. Low-amplitude rockfall signals with durations of <3 minutes have been correlated visually with rockfalls travelling <0.5km. Thus, the majority of these signals (c. 27100) can be reasonably attributed to the <0.5km category. The results suggest a linear relationship between frequency and runout between 0.5 and 7km on the log-log plot. Vertical error bars for rockfall numbers are smaller than the scale of the symbol, as these were counted automatically by seismic software. The data show that moderate events (1-3 km runout) occurred, on average, c. 15 times a month, while large dome collapses occurred on average once every one to two months. The plot also suggests, as to some extent anticipated by recording bias, that the number of 0.5-1 km flows may be underestimated by as much as 75%. Small pyroclastic flows were frequent enough to defy systematic reporting. Vertical-axis errors for the moderate- to largescale events are small, as these events are less numerous and are accurately documented. The relationships shown in Figure 12b and d can be used to illustrate frequency-magnitude relationships of dome-collapse phenomena throughout the eruption. Event durations should be related to runout distances for collapses generated as discrete pulses and with similar velocities. Multiple-pulse collapses have a more complex relation, and some larger pyroclastic flows exhibited this characteristic. Therefore, although different graphical approaches have been used for rockfalls and larger pyroclastic flows, the data quantify the relative frequencies of various sizes of events in relation to both duration, D and runout, R (Fig. 13).

Generation of pyroclastic flows

Role of active growth areas The locus of dome growth has been a major control on both dome stability and on the temporal and spatial distribution of rockfalls and pyroclastic flows. The first significant pyroclastic flows developed when extrusion rates increased to 2 m3 s-1 or more, and dome growth became dominated by exogenous 'shear lobe' extrusion, as described in detail by Watts et al. (2002). Lava was extruded

asymmetrically from the central vent and broke out through the old dome carapace to form a lobe with a blocky. steep headwall from which rockfalls and pyroclastic flows were generated (Fig. 6a. b. c). The steep headwall often developed where the lobe met a steep slope. In this case, the front stagnated and further advance was accommodated by avalanches and rockfalls. The flow front could thus stay relatively stationary for periods of days to weeks, although it was continually fed by flow of lava behind. The area below the headwall commonly became incised by deep erosion chutes due to repeated passage of rockfalls. A major feature of dome-growth evolution at Montserrat was the sudden switching of extrusion direction. A given lobe would be active for many weeks and then stagnate, and a new lobe would develop in a different direction. In many cases, large to major collapses and pyroclastic flows were associated with a switch in active area and establishment of a new lobe (Table 1). New lobes were often extruded within the scar of a previous collapse, and substantial new collapses occurred once the scar had refilled. Large collapses of older parts of the dome commonly occurred during switches in extrusion direction due to push-up of older lava by the incipient formation of a new lobe below (e.g. 4 November 1997. Table 1), The generation of pyroclastic flows with runouts > 1 km was commonly associated w i t h collapses of fresh lava and newly extruded parts of the dome, and was therefore intimately linked with shear-lobe extrusion.

The fragmentation of vesicular crystal-rich andesite The Soufriere Hills magma is a crystal-rich andesite (58.5-60.6 wt% SiO2). which has shown no significant variation in bulk chemistry since the beginning of the eruption (Murphy et al. 1998, 2000). The phenocryst assemblage of the dome lava consists of plagioclase (28-30 vol%), amphibole (3-10vol%). orthopyroxene (2-5vol%) and oxides. The total crystal content of the rapidly erupted lava is 65-75 vol% with 35-50 vol% phenocrysts and 20-25 vol% groundmass microlites (<80 m). The remaining 25-35 wt% magma comprises high-SiO: rhyolite glass (76-80 wt% SiO 2 ). partly devitrified (Murphy et al. 1998). In slowly erupted lava, the glass content can reach as low as 5-15vol% due to additional microlite crystallization in the conduit. The initial melt phase of the source magma contains 4-5% water at 5-6 km depth with a bulk magma water content of about 1.6 wt% (Barclay et al. 1998: Murphy et al. 2000). Magma rheological properties vary from a crystal-rich magma (25-35 vol% rhyolitic melt with 4-5 wt% H 2 O) with viscosity 10 6 Pas to a degassed crystalline lava (5-15vol% melt) with an apparent viscosity as much as 10 1 4 Pas and considerable strength (Sparks et al. 2000). Spontaneous autobrecciation of incandescent blocks of lava was observed during dome rockfalls and pyroclastic flows, especially at night (e.g. December 1996). Large blocks fragmented during initial stages of movement, and after impact with the ground surface disintegration produced small, fines-rich pyroclastic flows. On two occasions, discrete blocks projected from the basal flow front of small pyroclastic flows were seen to exfoliate rapidly (from the outside inwards). generating a trailing plume of ash behind the projecting block. This process may also help to explain the commonly observed rounding of relatively dense juvenile blocks, which occurred even in some of the smallest-volume pyroclastic flow deposits. Autobrecciation of hot microvesicular lava has also been recognized at Mount Unzen (Sato et al. 1992) and Mount St Helens (Mellors et al. 1988). Sato et al. (1992) postulated that the lava blocks had high internal pore pressure and low tensile stresses, facilitating disintegration. They considered the landing shock of blocks on the slope, and the shearing at the base of the flow, to be the main triggers for fragmentation. Autobrecciation has been interpreted as an inherent property of these lavas above a critical temperature (where tensile strength is relatively low). The influence of internal pore pressure is also of critical importance. The gas within hot dome rock with a porosity of 10vol% and an internal pressure of 1 MPa will expand to ten times its volume when released to atmospheric pressures during collapse

187

MECHANISMS OF LAVA DOME INSTABILITY

(Sparks 1997). The actual gas pressure in the Montserrat dome may have greatly exceeded this pressure (Voight & Elsworth 2000). Pressurized gas can be trapped in vesicles (which can burst when confining pressure is removed) and can also occur in pressurized volatiles diffusing through crack-permeable magmas (Voight & Elsworth 2000). This gas-expansion mechanism is a major factor in the generation of energetic pyroclastic currents associated with the large Montserrat dome-collapse pyroclastic flows (Woods et al. 2002). Observations suggest that, in most pyroclastic flows, the process of fragmentation and pyroclastic surge generation is highly dynamic and continues to occur during disintegration of blocks in the moving pyroclastic flow, as well as at the collapse onset.

Discussion The question 'did it fall or was it pushed?' is one that has preoccupied workers concerned with the generation of dome-collapse pyroclastic flows. It appears that both of these scenarios can, and did, occur during 1995-1999 at Soufriere Hills Volcano. Although the details surrounding each individual dome collapse can vary significantly, some generalizations can be made. To facilitate this, dome-collapse phenomena have been categorized into three generic types (Table 4). Passively generated rockfalls Passively generated rockfalls occurred during periods of low activity (slow dome growth) and were associated with inactive flanks and older carapace material, and not with active shear lobes or headwalls. Comminution was less complete, and fewer fines were generated. Source material was thus probably largely degassed and spontaneous fragmentation of vesicular rock was not so active. Rockfall runout was therefore limited, with rockfalls rarely reaching past the break in slope at the base of the dome talus. At Montserrat, dome growth appears to have been mainly exogenous (Watts et al. 2002) but these rockfalls produced by collapse of unstable, largely degassed, carapace material may be comparable to the smallscale collapses produced by endogenous growth, as recognized by U i e t al. (1999). Actively generated collapses (rockfalls and pyroclastic flows) Most dome collapses (rockfalls and pyroclastic flows) were generated from actively growing portions of the dome, or shear-lobe

headwalls. Rockfalls that were actively generated were commonly voluminous due to rapid development of large, metastable, oversteepened areas associated with rapid growth in a localized region of the dome. Larger collapses also occurred due to the inherently greater instability of the lava mass, and the tendency of the failure to progressively grow in volume. Where substrate slope changed, as at the edge of English's Crater, or over the edge of the old Castle Peak Dome, lobe fronts became inherently unstable and generated numerous rockfalls and pyroclastic flows. Periods when actively generated rockfalls were prevalent (e.g. May 1997; Fig. 5) were more often associated with the formation of longer-runout pyroclastic flows (when sufficiently large volumes of material were involved). Pyroclastic flows are therefore visualized as the larger-volume equivalent of actively generated rockfalls, as it was consistently from this activity that the longer runout (>1 km) pyroclastic flows were generated. There is evidence that most, but not all, large to major dome collapses at Montserrat were related to pulses in extrusion rate. Dome extrusion rates were 4-6 m3 s-1 in the two or three days before the 29 July 1996 collapse and most other large to major collapses were preceded by one to five days of enhanced dome-growth rates (Table 1: e.g. December-January 1997 collapses; 25 June 1997; 3 August 1997). This is established by volumetric data for some cases (Figs 7 and 10), and interpreted indirectly for others; for example, by changes in tilt or by the onset of intense swarms of hybrid earthquakes. For much of the eruption, the occurrence of hybrid earthquakes appears to have been directly associated with the onset of instability of the lava dome. The tilt data identify significant pressure pulses spaced five to seven weeks apart, with the best examples beginning 22 June 1997 (just prior to the 25 June collapse) and 31 July 1997 (Voight et al. 1999). The timing of several other collapses (e.g. 21 September, 4 November and 26 December 1997) also fit this pattern, although tilt data were then unavailable. The tilt cycles have been interpreted in terms of shallow pressurization (<0.6km below the base of the dome), at a position in the conduit influenced by a dramatic rise in magma melt viscosity. The change in viscosity is believed to be related to melt degassing and microlite crystallization (Sparks 1997; Voight et al. 1998, 1999; Melnik & Sparks 1999; Sparks et al. 2000). The hybrid earthquakes are further indicators of shallow pressurization, but require a threshold value of pressure. The pressure rise causes an eventual extrusion of a slug of magma, pressure is relieved in the upper conduit, and deflation occurs. Analysis of rockfall seis- mic records (Luckett et al. 2002) provides further evidence that actively generated rockfalls are intimately coupled with degassing of the dome. Vigorous hybrid earthquake swarms can also provoke rockfall activity by directly shaking the edifice. This mechanism is conzsidered as a minor, but recognizable, contributor to lava-dome instability during dome-growth periods.

Table 4. Summary of dome-collapse phenomena Passively generated rockfalls

Actively generated collapses

Post-extrusion collapses

Factors leading to collapse Locus on dome of collapse Extrusion rate prior to collapse Apparent role of rainfall Gas content of material Volume of collapse Relation to tilt cycles and/or hybrid earthquakes Ash-venting Explosive component

Distributed on cold carapace 0-2 m3-s-1 May provoke rockfalls Largely degassed <10 3 m 3 Not directly related but may be provoked by shaking Not observed Not observed

Shear lobes and push-up zones >2m-V Possible effect (rarely) Gas-rich 10 3 -10 7 m 3 Commonly associated

Unstable areas Om3s-1 Provokes collapse in some cases Degassed carapace, gas-rich core 103-107m3 No precursory seismic activity or crater-rim deformation Following collapse (commonly) In some cases

Characteristics of flows Runout Degree of fragmentation Flow types

<0.5km Fine ash poor Slow grain flows

0.5-7 km Fine ash rich Small pyroclastic flows to violent pyroclastic density currents

Following collapse (rarely) In some cases associated with pulses of gas-rich magma

0.5-7 km Fine ash rich Small pyroclastic flows to major dome collapses associated with violent pyroclastic surges

188

E. S. CALDER ET AL.

Given the Soufriere Hills Volcano extrusion rates of 1-13 m3 s -1, and dome volumes that reached up to 113 x 106 m3 during the 199599 period, most collapses of the lava dome removed no more than 18% of the dome (Fig. 8). The volumes of collapses, and the morphologies of the scoop-shaped scars they created, may have been influenced by the volume and structure of shear lobes. Shear lobe volumes were commonly in the range l(P-10 6 m 3 (Watts et al. 2002), and therefore comparable to that of most medium to large dome collapses. Rockfall counts, amplitude and duration data (as determined seismically) reflect extrusion rate and growth locus variations and can thus be used (together) as a proxy for extrusion rate (Figs 9-11). A few large to major dome collapses present exceptions. The 17 September 1996 and 26 December 1997 collapses were both exceptionally large (collapse of 35 and 50vol% of the dome respectively), reflecting more complex generation mechanisms. The 17 September 1996 collapse volume included lava excavated from the crater during the subsequent sub-Plinian explosive activity (Fig. 6c; Robertson et al. 1998). The 26 December 1997 collapse was generated by failure of the hydrothermally altered wall of English's Crater, and the volume of dome rock and dome talus involved in this complex failure were 25 x 106 m 3 and 30 x 106 m3 respectively (Sparks et al, 2002; Table 1). The 30 March and 11 April 1997 collapses were unusual in that they occurred during relatively low extrusion rates (<3 m 3 s-1) although they were clearly related to shear-lobe extrusion. These events were also associated with the overtopping and partial destruction of Galway's Wall, and growth of the lobe headwall over the weakened, erodible, steep crater rim. The distinction between passively and actively generated rockfalls leads to insights into the factors responsible for the generation of pyroclastic flows from fragmentation of the collapsing lava. Shear-lobe headwalls provide a rapidly replenished source of fresh lava only hours to days old, in a localized steep region. Within actively generated rockfalls the comminution is more complete, and this probably reflects the spontaneous disintegration of gaspressurized vesicles and cracks. This is consistent with observations of autobrecciation, which suggest that wholesale fragmentation occurs when a pervasively microfractured rock-mass with overpressurized pore space is critically stressed. The excavation level of collapse within the dome core is also of primary importance in the degree of fragmentation and in the subsequent generation of pyroclastic surges and buoyant ash plumes by rapid expansion of gases within the collapsing material (Voight & Elsworth 2000; Woods et al. 2002). Further insights into the mechanism of spontaneous disintegration proposed for Montserrat are given by shock-tube fragmentation experiments on crystal-rich andesite. Such experiments suggest three fragmentation mechanisms - elastic unloading, a fragmentation wave and gas-filtration flow - which can contribute, depending on lava structure and decompression style, to the fragmentation of the magma (Alidibirov & Dingwell 2000). Experiments on Merapi Volcano (Indonesia) andesite at temperatures of 100-900°C and pressures up to 20MPa show that the degree and character of fragmentation are controlled by density, crystal content and vesicle shape and distribution, with gas permeability the most important physical parameter (Spieler & Dingwell 1998). Post-extrusion collapses The collapses that occurred during the period of residual volcanic activity, after dome growth had ceased in March 1998 and before it commenced again in November 1999, present contrasting collapse mechanisms. The large collapses of 3 July 1998, 5 and 12 November 1998 and 20 July 1999 were preceded by neither hybrid earthquake swarms nor by crater-rim deformation. These collapses were associated with massive destruction of steep portions of the old, partially hydrothermally altered dome and they are distinctively different phenomena from the collapses associated with dome growth that are the main subject of this paper. The collapses initiated

suddenly (Fig. 4), without build-up of rockfall events or other seismicity in the preceding minutes or hours. The onset of the seismic signals of some of these collapses (e.g. 3 July and 12 November 1998) has long-period components, suggesting a possible explosive triggering component. Unfortunately, visual confirmation of an explosive character is not available, although for the 12 November 1998 collapse, pyroclastic flows were produced simultaneously in three valleys. These collapses were also characterized by periods of intense degassing or ash-venting in the hours following collapse. The degree to which gas pressure build-up within the dome plays a role in their generation mechanisms is not yet well constrained. In these cases, however, it seems that gravitational instability of the precipitous rock walls, coupled with introduced gas pressure, was sufficient to produce flow deposits as voluminous as 20-25 x 10 6 m 3 , and associated pyroclastic surges that inundated areas of >6km~. Especially heavy rainfall occurred before and during the collapse of 3 July 1998 (Norton et al. 2002). and this has been recognized elsewhere as an important collapse-triggering mechanism (Yamasato et al. 1998; Ratdomopurbo & Poupinet 2000; Voight et al. 2000b). Although rainfall-triggered collapses may also occur during active growth periods, they are likely to represent a more important mechanism during the months to years following cessation of extrusion. These different collapse types currently fall under the broad term 'gravitational collapse". At Montserrat. passively generated rockfalls, to some extent like post-extrusion collapses, occurred by development of internal weaknesses or pressure build-up, in situ, until the point at which gravity collapse could take over. Conversely, actively generated collapses were pushed to the point at which gravitational collapse could take over. The generation of pyroclastic flows (runout 0.5km) from the collapsing material was determined by collapse of sufficiently large volumes of readily fragmented material either by actively generated collapse or during post-extrusion collapses. Conclusions We have investigated; (1) the factors that lead to instability of a lava dome, and that facilitate or trigger collapse episodes; and (2) how these characteristics promote the formation of rockfalls and pyroclastic flows. Rockfall or pyroclastic flow generation by collapse of unstable portions of the growing lava dome occurred on an almost daily basis from March 1996 to March 1998. Large to major (>10 6 m 3 ) dome collapses began to occur once the volume of extruded andesite exceeded 30 x 10 6 m 3 . Large to major pyroclastic flows were generated during periods of elevated extrusion rate( 6 - 1 3 m 3 . s - 1repetitive hybrid swarms, and cyclic crater-rim ground deformation. Material was shed from the dome at the onset of each deflation cycle, soon after a slug of magma (typically c. 250000m 3 ) had been extruded into the base of the dome. Most large to major collapses were produced after eight to 15 inflation-deflation cycles, but did not usually remove more than 18vol% of the dome. To a first order, the volume of material involved in a collapse depends on the volume of the dome. The time-averaged rate of lava extrusion for Montserrat was typically on the order 4-7 m3 s-1. For this range of extrusion rates, a log-linear relationship has been established for frequency versus magnitude of pyroclastic flows (using runout us a proxy for magnitude) and rockfalls (using event duration as a proxy for magnitude). The frequency data suggest a deficiency in recording of flows with 0.5-1 km runout. The characteristics of lava extrusion are determined by extrusion rate, which can vary from week to week, and cyclic activity that can occur several times a day. For much of the eruption, lava has been extruded in shear lobes, which have played an important structural role in the generation of rockfalls and pyroclastic flows. The direction of the extruding lobe, and the location of the active headwall, have controlled the locations and directions of subsequent collapses. The timings of lobe extrusion determined variations in

MECHANISMS OF LAVA DOME INSTABILITY

rockfall and pyroclastic flow activity, both on a month-to-month scale and within individual hour-scale cycles. Collapse volumes and collapse-scar shape were probably also controlled by lobe structures. Importantly, active shear lobes provided the source material of pyroclastic flows, where material was hot and gas-rich, and disintegration of the microvesicular andesite lava occurred readily. Collapse events occurring since the cessation of dome growth in March 1998, during the period of 'residual volcanic activity' (i.e. 3 July 1998, 5 and 12 November 1998 and 20 July 1999), were not related to extrusion rate or to shear-lobe formation, nor were they apparently associated with precursory seismic activity or ground deformation. In this case, catastrophic failures occurred from steep, eroded canyon-like walls, along inherent weakness planes, and were possibly triggered by environmental factors (rainfall) and facilitated by diffusing gas pressure within the dome. Seismic evidence suggests that there may have been an impulsive onset to some of these events, so that the assumption of a fully passive, gravity-driven collapse mechanism may not always be appropriate for these post-extrusion collapses. MVO is supported by the UK Government (Department for International Development) and the Government of Montserrat. E.S.C. was supported by a NERC studentship GT4/95/35/E while at the University of Bristol. R.S.J.S. was supported by a NERC fellowship. The participation of B.V. was partly supported by NSF. We thank staff at the MVO who were responsible for the daily processing of the seismic data, in particular V. Bass. M. James is thanked for drafting the original version of Figure 6c, and G. Thomson for resolving a number of problems regarding seismic data. Reviews by C. Newhall and S. Nakada and thorough editing by T. Druitt helped to improve the manuscript.

References ABDURACHMAN, E. K., BOURDIER, J. L. & VOIGHT, B. 2000. Nuees ardentes of November 22 1994 at Merapi Volcano, Indonesia. Journal of Volcanology and Geothermal Research, 100, 345-361. ALIDIBIROV, M. & DINGWELL, D. B. 2000. Three fragmentation mechanisms for highly viscous magma under rapid decompression. Journal of Volcanology and Geothermal Research, 100, 413-421. ANDERSON, T. & FLETT, J. S. 1903. Report on the eruptions of the Soufriere St. Vincent, in 1902, and on a visit to Montagne Pelee, in Martinique Part 1. Philosophical Transactions of the Royal Society London, A-200, 353-553. ASPINALL, W. P., MILLER, A. D., LYNCH, L. L., LATCHMAN, J. L., STEWART, R. C., WHITE, R. A. & POWER, J. A. 1998. Soufriere Hills eruption, Montserrat 1995-1997: Volcanic earthquake locations and fault plane solutions. Geophysical Research Letters, 25, 3397-3400. BARCLAY, J., RUTHERFORD, M. J., CARROLL, M. R., MURPHY, M. D., DEVINE, J. D., GARDNER, J. & SPARKS, R. S. J. 1998. Experimental phase equilibria constraints on pre-emptive storage conditions of the Soufriere Hills magma. Geophysical Research Letters, 25, 3437-3440. BONADONNA, C., MAYBERRY, G. C., CALDER, E. S. ET AL. 2002. Tephra fallout in the eruption of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 483-516. BOUDON, G., CAMUS, G., GOURGAUD, A. & LAJOIE, J. 1993. The 1984 nueeardente deposits of Merapi Volcano, Central Java, Indonesia: Stratigraphy, textural characteristics and transport mechanisms. Bulletin of Volcanology, 55, 327-342. BRODSCHOLL, A. L., KIRBANI, S. B. & VOIGHT, B. 2000. Sequential domecollapse nuees ardentes analysed from broadband seismic data, Merapi Volcano, Indonesia. Journal of Volcanology and Geothermal Research, 100, 363-369. CALDER, E. S. 1999. Dynamics of small to intermediate volume pyroclastic flows. PhD Thesis, University of Bristol. CALDER, E. S., COLE, P. D., DADE, W. B. ET AL. 1999. Mobility of pyroclastic flows and surges at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 26, 537-540. CHOUET, B. A. 1996 Long-period volcano seismicity - its source and use in eruption forecasting. Nature, 380, 309-316.

189

CLARKE, A. B., NERI, A., VOIGHT, B., MACEDONIO, G. & DRUITT, T. H. 2002. Computational modelling of the transient dynamics of August 1997 Vulcanian explosions at Soufriere Hills Volcano, Montserrat: influence of initial conduit conditions on near-vent pyroclastic dispersal. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 319-348. COLE, P. D., CALDER, E. S., DRUITT, T. H., HOBLITT, R., ROBERTSON, R. E. A., SPARKS, R. S. J. & YOUNG, S. R. 1998. Pyroclastic flows generated by gravitational instability of the 1996-97 lava dome of the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3425-3428. COLE, P. D., CALDER, E. S., SPARKS, R. S. J. ET AL. 2002. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 231-262. DADE, W. B. & HUPPERT, H. E. 1998. Long runout rockfalls. Geology, 26, 803-806. DRUITT, T. H., CALDER, E. S., COLE, P. D. ET AL. 2002a. Small-volume, highly mobile pyroclastic flows formed by rapid sedimentation from pyroclastic surges at Soufriere Hills Volcano, Montserrat: an important volcanic hazard. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 263-279. DRUITT, T. H., YOUNG, S. R., BAPTIE, B. ET AL. 2002b. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306. ENDO, E. T. & MURRAY, T. L. 1991. Real-time seismic amplitude measurement (RSAM), a volcano monitoring and prediction tool. Bulletin of Volcanology, 53, 533-545. ESCHER, B. G. 1933. On a classification of central eruptions according to gas pressure of the magma and viscosity of the lava. Leidsche Geologische Mededeelingen, 6, 45-49. FINK, J. H. & KIEFFER, S. W. 1993. Estimates of pyroclastic flow velocities from explosive decompression of lava domes. Nature, 363, 612-615. FISHER, R. V. 1995. Decoupling of pyroclastic currents: hazards assessment. Journal of Volcanology and Geothermal Research, 66, 257-263. GRUNEWALD, U., SPARKS, R. S. J., KEARNS, S. & KOMOROWSKI, J.-C. 2000. Friction marks on blocks from pyroclastic flows at the Soufriere Hills volcano, Montserrat: implications for flow mechanisms. Geology, 28, 827-830. LACROIX, A. 1904. La Montagne Pelee et ses eruptions. Masson et cie, Paris. LOUGHLIN, S. C., BAXTER, P. J., ASPINALL, W. P., DARROUX, B., HARFORD, C. L. & MILLER, A. D. 2002a. Eyewitness accounts of the 25 June 1997 pyroclastic flows and surges at Soufriere Hills Volcano, Montserrat, and implications for disaster mitigation. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21,211-262! LOUGHLIN, S. C., CALDER, E. S., CLARKE, A. B. ET AL. 2002b. Pyroclastic flows and surges generated by the 25 June 1997 dome collapse, Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 191-209. LUCKETT, R., BAPTIE, B. & NEUBERG, J. 2002. The relationship between degassing and rockfall signals at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 595-602. MELLORS, R. A., WAITT, R. B. & SWANSON, D. A. 1988. Generation of pyroclastic flows and surges by hot-rock avalanches from the dome of Mount St. Helens Volcano, USA. Bulletin of Volcanology, 50, 14-25. MELNIK, O. & SPARKS, R. S. J. 1999. Nonlinear dynamics of lava dome extrusion. Nature, 402, 37-41. MILLER, A. D., STEWART, R. C., WHITE, R. A. ET AL. 1998. Seismicity associated with dome growth and collapse at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3401-3404. MURPHY, M. D., SPARKS, R. S. J., BARCLAY, J. ET AL. 1998. The role of magma mixing in triggering the current eruption at the Soufriere Hills Volcano, Montserrat, West Indies. Geophysical Research Letters, 25, 3433-3436.

190

E. S. CALDER ET AL.

MURPHY. M. D., SPARKS, R. S. J., BARCLAY, J.. CARROLL. M. R. & BREWER, T. S. 2000. Remobilisation origin for andesite magma by intrusion of mafic magma at the Soufriere Hills Volcano, Montserrat. W. I., a trigger for renewed eruption. Journal of Petrology, 41, 21-42. NAKADA, S. & FUJII, T. 1993. Preliminary report on the activity at Unzen Volcano (Japan), November 1990-November 1991: Dacite lava domes and pyroclastic flows. Journal of Volcanology and Geothermal Research, 54, 319-333. NAKADA, S., SHIMIZU, H. & OHTA, K. 1999. Overview of the 1990-1995 eruption at Unzen Volcano. Journal of Volcanology and Geothermal Research, 89, 1-22. NEUBERG, J. 2000. Characteristics and causes of shallow seismicity in andesite volcanoes. Philosophical Transactions of the Royal Society of London, A358, 1533-1546. NEUBERG, J., BAPTIE, B., LUCKETT, R. & STEWART, R. 1998. Results from the broadband seismic network on Montserrat. Geophysical Research Letters, 25, 3661-3664. NEWHALL, C. G. & MELSON, W. G. 1982. Explosive activity associated with the growth of volcanic domes. Journal of Volcanologv and Geothermal Research, 17, 111-131. NORRIS, R. D. 1994. Seismicity of rockfalls and avalanches at three Cascade Range volcanoes: Implications for seismic detection of hazardous mass movements. Bulletin of the Seismology Society of America, 84. 1925-1939. NORTON, G. E., WATTS, R. B., VOIGHT, B. ET AL. 2002. Pyroclastic flow and explosive activity at Soufriere Hills Volcano, Montserrat, during a period of virtually no magma extrusion (March 1998 to November 1999). In: DRUTTT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London. Memoirs, 21, 467-481. PERRET. F. A. 1937. The eruption of Mt. Pelee 1929-1932. Carnegie Institute, Washington. Publication No. 458. RATDOMOPURBO. A. & POUPINET, G. 2000. An overview of the seismicity of Merapi volcano (Java, Indonesia) 1983-1994. Journal of Volcanology and Geothermal Research, 100 193-214. RITCHIE, L. J., COLE, P. D. & SPARKS, R. S. J. 2002. Sedimentology of deposits from the pyroclastic density current of 26 December 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT. T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 435-456. ROBERTSON, R. E. A.. COLE, P. D., SPARKS, R. S. J. ET AL. 1998. The explosive eruption of Soufriere Hills Volcano, Montserrat, 17 September 1996. Geophysical Research Letters, 25, 3429-3432. RODRIGUEZ-ELIZARRARAS, S., SlEBE, C., KOMOROWSKI, K-C, ESPIDOLA,

J. M. & SAUCEDO, R. 1991. Field observations of pristine block-andash flow deposits emplaced April 16-17 1991 at Volcan Colima, Mexico. Journal of Volcanology and Geothermal Research, 48, 399-412. ROSE, W. I, PEARSON, T. & BONIS, S. 1977. Nuee ardent eruption from the foot of a dacite lava flow, Santiaguito Volcano, Guatemala. Bulletin of Volcanology, 40, 23-38. SATO, H.. FUJII, T. & NAKADA, S. 1992. Crumbling dacite dome lava and generation of pyroclastic flows at Unzen Volcano. Nature, 360, 664-666. SPARKS, R. S. J. 1997. Causes and consequences of pressurisation in lava dome eruptions. Earth and Planetary Science Letters, 150. 177-189. SPARKS, R. S. J., YOUNG. S. R., BARCLAY, J. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills Volcano, Montserrat, West Indies: November 1995 to December 1997. Geophysical Research Letters, 25. 3421-3424.

SPARKS. R. S. J.. MURPHY. M. D., LEJEUNE. A. M.. WATTS. R. B.. BARCLAY. J. & YOUNG, S. R. 2000. Control on the emplacement of the andesite lava dome of the Soufriere Hills Volcano. Montserrat by degassing-induced crystallization. Terra Nova. 12. 14-20. SPARKS. R. S. J.. BARCLAY, J.. CALDER. E. S. ET AL. 2002. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano. Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21.409-434. SPIELER. O. & DINGWELL. D. B. 1998. Experimental fragmentation of vesicular basaltic andesite from Merapi (abstract). AGU 79. No 45. UHIRA. K.. YAMASATO. M. & TAKEO. M. 1994. Source mechanism of seismic waves excited by pyroclastic flows observed at Unzen volcano. Japan. Journal of Geophysical Research. 99. 17757-17773. U1, T.. MATSUWO, N.. SUMITA. M. & FUJINAWA. A. 1999. Generation of block and ash flows during the 1990-1995 eruption of Unzen Volcano. Japan. Journal of Volcanology and Geothermal Research, 89. 123-137. VOIGHT. B. & ELSWORTH, 2000. Instability and collapse of gas-pressurised lava domes. Geophysical Research Letters. 27. 1-4. VOIGHT, B.. HOBLITT. R. P.. CLARKE, A. B.. LOCKHART. A. B.. MILLER. A. D.. LYNCH. L. & MCMAHON. J. 1998. Remarkable cyclic ground deformation monitored in real-time on Montserrat. and its use in eruption forecasting. Geophysical Research Letters. 25. 3405-3408. VOIGHT. B., SPARKS. R. S. J.. MILLER. A. D. ET AL. 1999. Magma flowinstability and cyclic activity at Soufriere Hills Volcano. Montserrat. British West Indies. Science. 283(5405) 1138-1142. VOIGHT. B.. YOUNG. K. D.. HIDAYAT. D. ET AL. 2000a. Deformation and seismic precursors to dome-collapse and fountain-collapse nuees ardentes at Merapi Volcano. Indonesia 1994-1998. Journal of Volcanologv and Geothermal Research. 100, 261-287. VOIGHT. B.. CONSTANTINE. E. K.. SISWOWIDJOYO. S. & TORLEY. R. 2000/b Historical eruptions of Merapi Volcano. Central Java. Indonesia. 17681998. Journal of Volcanology and Geothermal Research. 100. 69-138. WATTS, R. B.. HERD, R. A.. SPARKS. R. S. J. & YOUNG. S. R. 2002. Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano. Montserrat. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21. 115-152. WOODS. A. & KIENLE. J. 1994. The dynamics and thermodynamics of volcanic clouds: theory and observations from the April 15 and April 21. 1990 eruptions of Redoubt Volcano. Alaska. Journal of Volcanologv and Geothermal Research. 62. 273 299. WOODS, A. W., SPARKS, R. S. J.. RITCHIE. L. J.. BATEy. J.. GLADSTONE. C. & BURSIK. M. 2002. The explosive decompression of a pressurized lava dome: the 26 December 1997 collapse and explosion of Soufriere Hills Volcano, Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano. Montserrat. from 1995 to 1999. Geological Society, London. Memoirs. 21. 457-466. YAMAMOTO. T., TAKARADA. S. & SUTO. S. 1993. Pyroclastic flows from the 1991 eruption of Unzen Volcano. Japan. Bulletin of Volcanology. 55. 166-175. YAMASATO, H., KITAGAWA. S. & KOMIYA. M. 1998. Effect of rainfall on dacitic lava dome collapse at Unzen volcano. Japan. Papers in Meterology and Geophysics. 48(3). 73-78 [in Japanese]. YOUNG, S. R., SPARKS. R. S. J.. ASPINALL. W. P.. LYNCH. L. L.. MILLER. A. D.. ROBERTSON. R. E. A. & SHEPHERD. J. B. 1998. Overview of the eruption of Soufriere Hills Volcano. Montserrat. 18 July 1995 to December 1997. Geophysical Research Letters. 25(18). 3389-3392.

Pyroclastic flows and surges generated by the 25 June 1997T dome collapse, Soufriere Hills Volcano, Montserrat S. C. LOUGHLIN 1 , E. S. CALDER2, A. CLARKE 3 , P. D. COLE4, R. LUCKETT 5 , M. T. MANGAN 6 , D. M. PYLE7, R. S. J. SPARKS2, B. VOIGHT3 & R. B. WATTS2 1 British Geological Survey, West Mains Road, Edinburgh, EH9 3LE, UK (e-mail: [email protected]) 2 Department of Earth Sciences, University of Bristol, Bristol, BS8 1RJ, UK 3 Department of Geosciences, Pennsylvania State University, University Park, PA 16802, USA 4 Centre for Volcanic Studies, University of Luton, Park Square, Luton, LU1 3JU, UK 5 British Geological Survey, Keyworth, Nottingham, NG12 5GG, UK 6 United States Geological Survey, Menlo Park, California, USA 7 Department of Earth Sciences, University of Cambridge, Cambridge, CB2 3EQ, UK

Abstract: On 25 June 1997, an unsteady, retrogressive, partial collapse of the lava dome at Soufriere Hills Volcano lasted 25 minutes and generated a major pulsatory block-and-ash flow, associated pyroclastic surges and a surge-derived pyroclastic flow that inundated an area of 4km 2 on the north and NE flanks of the volcano. Three main pulses are estimated to have involved 0.78, 2.36 and 2.36 x 106 m3 of debris and the average velocities of the fronts of the related block-and-ash flow pulses were calculated to be 1 5 m s - 1 , 16.1ms - 1 and 2 1 . 9 m s - 1 respectively. Deposits of block-and-ash flow pulses 1 and 2 partially filled the main drainage channel so that material of the third pulse spilled out of the channel at several places, inundating villages on the eastern coastal plain. Bends and constrictions in the main drainage channel, together with depositional filling of the channel, assisted detachment of pyroclastic surges from the pulsatory block-and-ash flow. The most extensive pyroclastic surge detached at an early stage from the third block-and-ash flow pulse, swept down the north flank of the volcano and then climbed 70 m in elevation before dissipating. Rapid sedimentation from this surge generated a high-concentration granular flow (surge-derived pyroclastic flow) that drained westwards into a valley not anticipated to be at high risk. Observations support the hypothesis that the interior of the Soufriere Hills Volcano lava dome was pressurized and that pyroclastic surge development became more substantial as deeper, more highly pressurized parts of the dome were incorporated into the pyroclastic flow. Surge development was at times so violent that expanded clouds detached from the block-and-ash flow within a few tens of metres of the lava dome.

The partial dome collapse on 25 June 1997 was one of the largest during the eruption and it was one of the best documented. The pyroclastic flow and surges generated by the collapse swept down the northern flanks of the volcano, killing 19 people and seriously injuring seven others. From a sociopolitical point of view, this was the most important event to occur during the eruption, and it exerted a strong influence on hazard management and decisionmaking throughout the subsequent course of the eruption. This paper describes the 25 June 1997 dome collapse, the dynamics of the consequent block-and-ash flow and pyroclastic surges, and the deposits formed. The hazards and damage caused by the flow, surges and surge-derived pyroclastic flows are emphasized. This paper is based on analysis of monitoring data, field study of deposits, observational data collected from a time-lapse video camera based at W. H. Bramble Airport (5 km to the NE of the volcano; Fig. 1), photographs and eyewitness accounts (Loughlin et al 2002). Other papers in this volume provide additional information on the 25 June deposits (Bonadonna et al 2002; Cole et al 2002; Druitt et al 2002) and the actions of the MVO scientists (Aspinall et al. 2002; Kokelaar 2002).

Construction of a lava dome in English's Crater (Fig. 1), at the summit of Soufriere Hills Volcano, began in November 1995. Pyroclastic flows generated by partial dome collapse began to develop and travel to the east in late March 1996, and first reached the sea on 12 May 1996. Dome growth continued throughout 1996 and into 1997 (Sparks et al. 1998). On 10 February 1997, pyroclastic flows overtopped the south crater wall for the first time, and during March and April 1997 flows travelled down the White River valley to the south and SW (Cole et al 1998). By the end of May 1997, the northern flanks of the volcano were threatened by dome collapse pyroclastic flows for the first time. The lava dome had reached a volume of 65 x 10 6 m 3 (Dense-Rock Equivalent; DRE) and was growing at a rate of 3 to 5m3 s-1 (Sparks et al 1998).

Terminology

Activity in May-June 1997

Collapse of the lava dome at Montserrat typically generated three components: a high-particle-concentration basal part containing abundant blocks and termed a pyroclastic flow or block-and-ash flow; a low-particle-concentration pyroclastic surge that overlies the block-and-ash flow and may detach from it and move independently; and a lofting buoyant ash plume. Rockfalls are distinguished from dome collapse pyroclastic flows by runouts < 1 km and the failure of many blocks to disintegrate completely and produce copious ash. The 25 June 1997 collapse produced an extremely unsteady, or pulsatory, block-and-ash flow. We refer to three main flow pulses, rather than three separate flows, mainly because of the continuous

Between 14 and 17 May the focus of dome growth switched from the SW to the NE side of the dome. Lava extrusion was directed to the north, and rockfalls began to spill across a broad swath from the east side to the NW side of the dome. This broadly distributed activity suggested the development of a subhorizontal shear lobe directed to the north (see 'active 17 May 1997 lobe' in Watts et al. 2002, fig. 2a). The lobe grew almost continuously through late May and into June. Swarms of hybrid earthquakes (each with a highfrequency start and long-period coda; Miller et al. 1998) accompanied this growth daily between 13 May and 27 May. Each swarm was followed immediately by enhanced rockfall activity, mainly on

nature of the seismic signal. Each pulse was capable of forming a flow front where it advanced rapidly over just-deposited or stillmoving material.

Summary of precursory volcanic activity

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 191-209. 0435-4052/02/$15 © The Geological Society of London 2002.

191

192

S. C. L O U G H L I N AT AL.

Fig. 1. Map showing the extent of the 25 June 1997 pyroclastic flow deposits. Sampling sites 1-3 are marked.

the north and NE flanks of the dome. The daily number of earthquakes during this period was typically about 100. Rockfall debris rapidly built up in the moat between the dome and the northern crater walls and, by 19 May, material began to spill over the low points of the northern crater rim into the top of Tuitt's Ghaut (Fig. 1). On 27 May, small pyroclastic flows reached 200m down Tuitt's Ghaut. On 5 June, small pyroclastic flows travelled 3.1 km down Tuitt's Ghaut, and rockfall debris began to spill into Mosquito Ghaut (Fig. 1). A relative lull in activity occurred for about a week after 6 June, during a period of poor visibility. By 14 June the level of rockfall activity increased on the northern side of the dome, and deposits extended 500 m down Mosquito Ghaut. On 16 June a significant dome collapse sent pyroclastic flows 1.5km down Mosquito Ghaut and 1.6km down the upper reaches of Fort Ghaut to the west. On 17 June pyroclastic flows travelled 4km down Mosquito Ghaut and 1.8km down Fort Ghaut. A small pyroclastic surge detached from the block-and-ash flow in Mosquito Ghaut and climbed above the channel, causing extensive tree knockdown on the steep walls of the valley. It also generated a broad singe zone that extended 400m west of Mosquito Ghaut (in the prevailing wind direction). A scar developed within the lava dome above Mosquito Ghaut as a result of the collapses on 16 and 17 June (Fig. 2a). The seismicity was relatively low between 27 May and 22 June and the daily number of earthquakes (mainly hybrid and long-period earthquakes) remained below 40. Starting on 18 May 1997 a tiltmeter installed on Chances Peak (Fig. 2) showed a regular pattern of inflation and deflation of the dome. The inflationary part of the cycle usually correlated with the hybrid earthquake swarms and the deflationary part correlated with enhanced rockfall and pyroclastic flow activity. Between 5 and 14 June the periodicity was about 12-16 hours and amplitude was 16-18 rad (Voight et al. 1998, 1999). By 16 June the tilt amplitude

had flattened to only 5-10 rad and the number of hybrid earthquakes declined. At about 16:00 LT (all times are local time, LT = GMT minus 4 hours) on 17 June, inflation increased steeply, peaked at 21:00 and was followed by rapid deflation. A significant collapse at 23:00 then sent pyroclastic flows into both Mosquito Ghaut and the upper reaches of Fort Ghaut. High-amplitude tilt cycles lasted a further two days until 19 June when low-amplitude cycles returned, although the deflationary part of the cycles continued to correlate with heightened rockfall and fumarolic activity. At 05:30 on 22 June a sharp increase in the rate of inflation occurred and the subsequent sharp deflation was coincident with sustained pyroclastic flows that travelled about 1 km to the east of the dome. This was the beginning of a series of nine inflationdeflation cycles that culminated in the partial dome collapse of 25 June 1997 (Fig. 3; Voight et al. 1999). The periodicity reduced to about 8-12 hours and the amplitude increased to as much as 30 rad. Between 22 and 25 June, hybrid earthquake swarms increased in both duration and numbers of earthquakes. Each period of inflation and the associated seismic swarm was attributed to an increase in magma pressure in the upper conduit caused by the supply of gas-rich magma from depth (Voight et al. 1999). When pressure exceeded a threshold value, a plug of viscous magma was injected into the dome, conduit pressure relaxed and deflation began. Although enhanced rockfall and some pyroclastic flow activity occurred during deflation, the intensity of these did not always correlate with the intensity of the preceding earthquake swarm. However, on 24 and 25 June the hybrid earthquake swarms were intense, reaching the state whereby repetitive events merged into continuous tremor (Miller et al. 1998). Poor visibility throughout June hampered efforts to ascertain the dome volume and growth rate. During the morning of 23 June several small spines on the northern summit area of the dome were

PYROCLASTIC FLOWS OF 25 JUNE 1997

Fig. 2. Configuration of the lava dome (a) before and (b) after the 25 June 1997 partial dome collapse.

Fig. 3. Histogram of the number of triggered low-frequency (hybrid) earthquakes per hour compared to the radial tilt measured at Chances Peak in late June 1997. The low-amplitude tilt cycles before 22 June had no associated hybrid seismicity, whereas high-amplitude cycles that started on 22 June were accompanied by hybrid earthquake swarms during inflation. The partial dome collapse occurred at the onset of a rapid deflation (arrowed).

193

194

S. C. L O L ' G H L I N ET AL. Fig. 4. Seismogram from seismometer at Windy Hill site (3.5km NNW of lava dome) showing precursory hybrid earthquakes merged into tremor, the onset of the pyroclastic flow and three pulses of high-intensity seismic activity corresponding to three successive dome collapses and associated pyroclastic flow pulses. Local time = GMT — 4 hours. Assuming the seismic waves travelled at about 1 k m s - 1 . the seismic signal at Windy Hill began 3-4 seconds after the collapse began: this time lag is accounted for within the error. The peaks on this seismogram probably relate to the flow fronts of each pulse passing closest to the seismic station i.e. near the junction of Mosquito Ghaut and Paradise River.

Fig. 5. (a) Map showing the probable extent of deposits formed by pyroclastic flow pulse 1 based on eyewitness evidence. (b) Map showing probable extent of deposits formed by pyroclastic flow pulse 2 based on analysis of deposits and eyewitness evidence.

PYROCLASTIC FLOWS OF 25 JUNE 1997

observed (from the airport) to grow by several metres over a period of six hours (Fig. 2a).

Precursory activity on 25 June 1997 At 03:00 on 25 June a hybrid earthquake swarm began, comprising four to five moderate-amplitude events per minute, and this was accompanied by tilt inflation of the summit of the volcano (Voight et al. 1998, 1999). The tilt levelled off at about 05:20 and the summit began to deflate at about 06:10 (Fig. 3). By 06:15 the hybrid earthquake swarm had also declined in intensity and by 07:05 had given way to low-level tremor. Between 06:15 and 07:15, semicontinuous pyroclastic flows travelled down Mosquito Ghaut with runouts as far as 1.5 km. There were also some rockfalls and small pyroclastic flows from the SE and east faces of the dome during this time. Inflation of the summit area then recommenced at about 09:00 and a hybrid earthquake swarm began at 10:50. The swarm intensified rapidly, reaching about six events per minute between 11.30 and 12:30. The earthquakes were all of moderate amplitude, similar to the earlier swarm. The inflation trend flattened out at 12:00 and by 12:45 the seismic activity was dominated by tremor and individual earthquakes were no longer distinguishable. A dilute steam-and-ash cloud issued from the volcano at about 12:30 and drifted to the west at an altitude of 1500m.

Dome collapse of 25 June 1997 Between 12:40 and 12:50 the tiltmeter registered the onset of a sharp deflation (Fig. 3). A strong, continuous seismic signal began at about 12:55 (Fig. 4; Baptie et al. 2002), with more intense pulses of seismic activity commencing at 12:57:15 0 s), 12:59:55 0 s) and 13:08:20 0 s). We interpret the strong continuous signal as a single unsteady pyroclastic flow and the three pulses as major dome avalanches that formed major pulses within the flow. Assuming the seismic waves travelled at about 1kms - 1 , the seismic signal at Windy Hill began 3-4 seconds after the collapse began; this time lag is accounted for within the error. A time-lapse video camera installed at W. H. Bramble Airport, 5 km to the NE of the volcano, recorded the fronts of the three main flow pulses as they emerged from Paradise River by Harris Lookout (Fig. 1). The arrival times of the three flow fronts at various distal locations have been estimated from the video footage. The following section describes the three block-and-ash flow pulses, based on eyewitness, video and observational data. Using data from the seismic signals and the video, we estimated mean velocities for parts of the travel path of each of the flow pulse fronts. Maps showing the extent of the deposits from pyroclastic flow pulses 1 and 2 are shown in Figure 5. The extent of deposits from the third pulse corresponds to the whole area of impact shown in Figure 1.

Pyroclastic flow pulse 1 We assumed that the initiation of the first main flow pulse was related to a rapidly emergent rockfall-type seismic signal (Miller et al 1998) at 12:57:15 0 s) (Fig. 4). Eyewitness accounts indicated that the first flow pulse was confined to the steep-walled channel of Mosquito Ghaut (Loughlin et al. 2002). It travelled 4.7 km and stopped in the Paradise River valley just below Bramble village (Fig. 5a). The time taken for the flow front of pyroclastic flow pulse 1 to travel the 3.9km between the source and the sharp bend at Harris village, where it was first recorded by the airport camera, was 230 20s. The average velocity to this point was 16.7ms -1 (+1.6, -1.9). The flow front stopped near Bramble village about 80 seconds later, giving an average velocity for the 800m distance between Harris and Bramble village of l O m s - 1 . The average veloc-

195

ity of the flow front from the dome to Bramble village, a total distance of 4.7km, was 1 5 m s - 1 (+1, -0.9). Pyroclastic flow pulse 2 Eyewitnesses observed the second pyroclastic flow pulse in the upper parts of Mosquito Ghaut as the first flow front approached Harris village (Loughlin et al. 2002). The precise onset of pyroclastic flow pulse 2 on the seismic records was not clear because there was interference between the dying signal of flow pulse 1 and the emerging, slightly higher amplitude seismic signal of flow pulse 2 (Fig. 4). The best estimate start time was 12:59:55 0 s). Eyewitnesses observed a pyroclastic surge moving down the northern flanks of the volcano as the block-and-ash flow travelled down Mosquito Ghaut. The surge reached the main road near Riley's Yard where it lost momentum and stopped its lateral advance (Fig. 5b). Eyewitnesses on the road near Farrell's Yard saw ash flowing west down the main road at a high angle to the original flow direction. They commented on how the ash was confined to the road, was hugging the ground, was 'boiling' and was 'moving round bends like a vehicle' (Loughlin et al. 2002). This was probably a high-concentration flow derived from the rapid sedimentation of ash near the surge front (surge-derived pyroclastic flow; Druitt et al. 2002). Pyroclastic surges detached at bends along Mosquito Ghaut and reached Mandy's gas station 2.5km north of the lava dome (Fig. 5b). Eyewitnesses heard the gas station explode as they made their escape up Windy Hill (Loughlin et al. 2002). At 13:03:10 the eastern seismic stations stopped transmitting data and we assumed this was caused by the destruction of the Bethel telephone exchange cable, either by the block-and-ash flow or a pyroclastic surge. Unfortunately, it is not clear exactly where the damage took place. The cables crossed the valley at a bridge near Bethel (Fig. 5a), but then they followed the road alongside Paradise River to the west. The bridge was 5.4km from the dome and the time between onset and cut-off was 195 30 s. If the cables were destroyed at the bridge, the flow travelled this distance at an average of 26.7m s-1 , -3.6). This is high compared to other estimates so it seems more likely that the cables were destroyed further upstream, in Paradise River valley. The block-and-ash flow continued to Trant's village (Fig. 5b), confined to Pea Ghaut the whole way. Beyond Trant's, the pulsatory advance of the block-and-ash flow was described by eyewitnesses, who saw successive small pulses burst through the stalling flow front (Fig. 6; Loughlin et al. 2002). The flow eventually reached to within 50m of the coastline, a total of 6.8 km from the dome. The flow front of pyroclastic flow pulse 2 was recorded near Bramble village after 233 seconds (a total distance of 4.7km), giving an average velocity to this point of 20.2ms - 1 , -2.3). It continued for a further 2km and then stopped just beyond Trant's village, giving an average velocity of 11 m s - 1 for this later stage. The overall average velocity from the dome to Trant's village (a total distance of 6.7km) was therefore 16.1ms - 1 , —1.1). The higher average velocity to Bramble village, the greater runout distance and the higher amplitude of the seismic signal (Fig. 3) all indicate that pyroclastic flow pulse 2 was significantly more energetic than pulse 1. Pyroclastic flow pulse 3 The onset of the third pyroclastic flow pulse was evident on the seismogram (Fig. 4) and was estimated to have started at 13:08:20 0 s). The early stages of this were not observed, so deposits and effects were used to reconstruct the event. Scouring of the walls of Mosquito Ghaut indicated that the block-and-ash flows were highly erosive and almost filled the channel during transport, thus facilitating the rapid lateral expansion and separation of the overriding pyroclastic surges. It was a surge from the third flow pulse that extended furthest to the north and

196

S. C. L O U G H L I N ET AL.

Fig. 6. Photographs taken from the main road north of the airport of advancing pyroclastic flow pulse 2. (a) Pyroclastic flow pulse 2 moving along Farm River valley, by Trant's village, (b) An additional small pulse is shown punching through the stalling flow front (note airport to left of photo) (photos taken by R. B. Watts).

NW and formed the main surge-derived pyroclastic flows that drained westwards (Druitt et al. 2002). Erosion of the walls on both sides of the channel was extensive (but not continuous) for a distance of 2km along the flow path from the dome, particularly on the outside of bends where superelevation of the flow had occurred. The block-and-ash flow pulse continued to the NE along Mosquito Ghaut and into a steep-sided section of Paradise River with several bends (Fig. 1). Pyroclastic surges detached along the north side of the channel, scoured the outside of the bends and impacted the lower parts of Harris village (see fig. 10 in Loughlin et al. 2002). Block-and-ash flow deposits high on the outside of bends showed that superelevation occurred. Several houses on the edges of the ghaut were destroyed and the main road was buried (see fig. l0a in Loughlin et al. 2002). The superelevation of a flow as it rounds a bend reflects an approximate balance between cross-channel gradient in pressure gh/w and centrifugal foree where £ is gravity, h is elevation difference from one side of the flow to the other, w is flow width, u is flow speed and r is radius of curvature of the channel bend. The last parameter must be estimated and is somewhat problematic. On rearrangement, a simple relationship is obtained: (1) Calculations based on the highest level of the deposits on the outside of bends suggest velocities of 2 0 m s - 1 at Harris. Farther down slope, at a bend near Bramble village, eyewitnesses saw part of the block-and-ash flow spill out of the

drainage and continue for almost 1 km in a NE direction, causing extensive damage to buildings in Bethel village (Figs 7a and 8: see fig. 6 in Loughlin et al. 2002). The channel was partially filled with deposits so superelevation of the flow at the bend enabled it to escape. A surge also detached along this bend, scouring the side of the channel (Fig. 8) and barely outrunning the block-and-ash flow. The flow pulse continued northward along Pea Ghaut and part of it spilled NE onto the coastal fan at the next sharp bend to the north, near Bethel (Fig. 1 and Fig. 7b). From this point, blockand-ash flows spilled out along much of the main drainage, which had been partially filled with debris from flow pulse 2. and across the coastal fan. The distal part of the flows on the eastern coastal fan formed distinct, relatively slow-moving, ground-hugging lobes (Fig. 9a. b). These lobes had low. tapered flow fronts easily blocked by low obstacles such as garden walls. As these flows came to a halt. surges outran them by just a few metres (Fig. 11. Farm village was located on the outside of a sharp bend alongside Pea Ghaut and was partially buried by the block-and-ash flow deposits associated with pyroclastic flow pulse 3 (Fig. 10). Between Farm and Trant's the drainage channel (now Farm River) is deflected sharply towards the east by a steep hillside that cuts across the flow path at right angles. Deposits on the hillside show that superelevation occurred. Using Equation 1 the velocity of the flows here was calculated to be 8 m s - 1 . At Trant's. deposits of pyroclastic flow pulse 2 had probablyfilled Farm River valley, because flow pulse 3 spilled out and

Fig. 7. (a) Oblique aerial view SW across Bethel village with Harris Hill on the right. Block-and-ash flow pulse 3 partly escaped from the Paradise River valley at the bend (far centre of photograph) and destroyed much of Bethel village. The flow remained confined within a shallow stream valley (block-rich deposits), but an associated surge spread across a much broader area. Trees in this area were still standing but badly damaged, (b) Damaged house partially buried by block-and-ash flow deposits on the coastal fan. Flow was from left to right (photos taken by P. D. Cole).

198

S. C. LOUGHLIN ET AL.

Fig. 8. Oblique aerial view south across Bethel village with Pea Ghaut to the right showing block-and-ash flow deposits where flows escaped Paradise River valley. Part of block-and-ash flow pulse 3 spilled out of Pea Ghaut at the bend near the confluence with Tuitt's Ghaut (arrow 1 see also Fig. 7) and part at the next bend near Bethel (arrow 2). These flows escaped the drainage channel as a result of superelevation at bends. Note erosion of the outside of the bend in Pea Ghaut. Fine-grained surge deposits surround Bethel church and the house in the lower left foreground, where a family survived indoors (photo taken by S. C. Loughlin. BGS NERC).

inundated the remaining part of Trant's village (Fig. 11). Eyewitnesses observed several secondary pulses associated with flow pulse 3 (e.g. Fig. 1 11c). Pyroclastic flow pulse 3 was not observed on the video until 265 seconds after its onset, when the flow front was clearly visible approaching Trant's village. The poor visiblity up to this point was caused by the lingering ash cloud from flow pulse 2. It is estimated that the flow front had travelled 5.8 0.1 km to that point and therefore that the average velocity was about 21.9ms - 1 (+3.2, -2.6). Assuming the body of the block-and-ash flow had a flow front velocity of 2 0 m s - 1 near Harris village (see above), its velocity on the upper volcano flank may have been much greater. The flow front decelerated in the final stages before it came to a halt beyond Trant's village, 6.7km from the dome. Knowing the velocity, the peak flux of flow pulse 3 can be calculated. If we assume (based on evidence of erosion and surge detachment), that at peak flux the block-and-ash flow filled Mosquito Ghaut from source to bend 'X' (Fig. 12) and the minimum velocity of block-and-ash flow pulse 3 was within the range 20-30 m s - 1 , then the peak flux would have been 6-8 x 1 0 4 m 3 s - 1 . Dynamics and effects of surges Trees on the north flank of the volcano were flattened by surges that detached from the block-and-ash flow in Mosquito Ghaut. There were three bends from which significant surge detachment

occurred along Mosquito Ghaut (shown in Fig. 12). Surge transport indicators on the eastern side of Mosquito Ghaut imply a northerly motion and the implication is that a surge detached from a gentle bend about 0.7 km from the dome (Fig. 12). It appears that this north-directed surge drained back into Mosquito Ghaut near its confluence with Paradise River (Fig. 1). The second bend is along the flow path about 1.1 km from the crater rim. where the direction of drainage changed from NW to north. The outside of the bend was scoured and there was no evidence that these eroded slopes were created by undercutting and collapse. It is assumed that the surge continued on a northwesterly trajectory across the north flank of the volcano. This is confirmed by scouring of trees on the eastern margin of Tyre's Ghaut to the west (Fig. 1); trees were scoured on their SE sides, suggesting that they were impacted by NW-directed surges. The major bend in Mosquito Ghaut ('X' in Figs 12, 13 and 14) is about 1.38km from the crater rim along the flow path. At this bend, the drainage swings from NNW to east and the outside of the bend was strongly scoured (Fig. 12). The surge that detached here impacted Streatham and Windy Hill, where transport direction was indicated by scouring and charring of standing trees and telegraph poles on the upcurrent side, bent reinforcement bars and by ash shadow zones behind some houses (see flow direction arrows on Fig. 1). A corrugated iron sheet was wrapped around a tree at the foot of Windy Hill (Fig. 15a). Most indicators suggested a flow direction towards the north and NW but in the headwaters of Dver's River, flow direction was to the west.

PYROCLASTIC FLOWS OF 25 JUNE 1997

199

Fig. 9. (a) Oblique aerial photograph showing several lobes from block-and-ash flow pulse 3 advancing through Spanish Point, (b) The front of a flow lobe (central lobe in (a)) following a shallow ditch at Spanish Point (photos taken by P. D. Cole).

200

S. C. LOUGHLIN ET AL.

Fig. 10. Aerial view of the block-and-ash flow and surge deposits on the coastal plain. Pea Ghaut ran along the foot of the hills to the right but was filled with block-and-ash flow deposits. Part of Farm village was not buried and can be seen at the right margin of the flow deposits. At the very distal limit of the surge deposits, thin lobes extended at high angles to the flow direction. These were minor surge-derived pyroclastic flow deposits, probably formed by rapid suspended load deposition as the surge ran out of momentum and stalled (photo taken by G. E. Norton, BGS r NERC).

Surges detached at bends partly as a result of centrifugal forces when the block-and-ash flow changed direction; the continued motion of the surge in a straight line would be driven by momentum. However, these particular bends also corresponded with a constriction of the channel, a reduction in cross-sectional area and a decrease in slope from about 30° to 17° (Fig. 14). Deposits from flow pulses 1 and 2 would have accumulated in the shallow channel thus reducing its depth, and this, combined with the natural constriction, would have caused flow pulse 3 to bank up, thicken and fill the channel. The surge cloud would then have been unconfined and easily detached from the block-and-ash flow. Momentum carried the surge that detached at bend 'X' up Windy Hill to an elevation of 70m above the main road. Its velocity, in the vicinity of Farrell's Yard and Riley's Yard (Fig. 1), can be estimated at 3 7 m s - 1 by converting kinetic to potential energy: (2)

The calculation neglects frictional effects, but assumes that the base of the surge cloud rose 70m; the possible influence of surge cloud thickness on upslope sedimentation is not considered. On the upper flanks of the volcano the surges felled many trees (Fig. 12), but north of the main road the damaged trees remained standing (see fig. 9 in Loughlin et al. 2002). The ability of a surge to topple trees depends on the dynamic pressure (Valentine 1998): (3)

At Farrell's Yard and Riley's Yard (Fig. 1), leaves and some branches were stripped off trees but few trees were blown down. According to Valentine (1998), such light to moderate damage

requires dynamic overpressures of about 1000 Pa. Assuming this was the case, and assuming a velocity of 3 7 m s - 1 , the maximum surge density was therefore approximately 1 . 6 k g m - 3 . Valentine (1998) suggested a 90% tree blowdown, as observed on the proximal flanks of Mosquito Ghaut, requires dynamic overpressures of 2000 to 2400 Pa. Assuming the density of the surge was higher near source (a minimum of 1 . 6 k g m - 3 ) , the maximum proximal velocity required to produce the damage was 50-55 m s - 1 . Larger assumed average densities would require lesser surge velocities to achieve the same dynamic pressure. In reality, the vertical distribution of cloud densities is non-linear as discussed by Clarke & Voight (2000). Despite such complexities, it is clear that a rapid reduction in dynamic overpressure of surges occurred between the proximal region and Farrell's Yard. This was caused by the rapid decrease in velocity as a result of slope reduction, and a decrease in density as a result of rapid suspended-load fallout. A 6m diameter water tank was transported about 250m to the WNW, across the lower slopes of Windy Hill, and came to rest on top of an overturned car (Fig. 15a), implying that the dynamic pressure was still significant as the surge moved uphill. In Streatham, bitumen roof tiles were stripped off and some windows were cracked and broken, but structural damage to concrete block walls was minimal. Fires ignited inside concrete buildings with broken windows, but at the margins of the surge these fires did not necessarily spread throughout the house (see fig. 11 in Loughlin et al. 2002). It was hot enough to ignite wooden structures in the Streatham area including doors, rafters and huts. Most damage in Streatham and Windy Hill was caused by heat. The temperature of deposits in Dyer's River valley five days after the event was 410 C (Druitt et al. 2002), which is therefore the minimum temperature of the sediments in the surge. Eyewitnesses on Windy Hill described

PYROCLASTIC FLOWS OF 25 JUNE 1997

201

Fig. 11. Photographs from the main road above the airport showing the advance of pyroclastic flow pulse 3. (a) Pyroclastic flow pulse 3 emerged from the dissipating ash of flow 2 and spilled northwards out of Farm River, (b) The block-and-ash flow inundated Trant's village and spread across surrounding fields, (c) A further minor pulse escaped Trant's River and is clearly visible here at the break of slope (photos taken by R. B. Watts). how paint blistered on houses. Grass was singed in a sear zone <25 m beyond the distal margin of the surge deposits (Loughlin et al. 2002). Surges impacted the area east of Windy Hill and travelled 50 m uphill to a low point on a ridge at elevation 370m (immediately north of Mandy's garage; Fig. 1). A narrow singe zone extended about 50m down the drainage on the far side. This north-directed

surge detached at a gentle bend about 2km along the flow path where the channel changes direction from north to NE. The surge that detached from block-and-ash flow pulse 2 and impacted Mandy's gas station may also have detached at this location. As the surges that detached from flow pulse 3 swept across the northern flanks of the volcano, suspended-load fallout occurred across a broad area and generated high-concentration granular

202

S. C. LOUGHLIN ET AL.

Fig. 12. View of Mosquito Ghaut from the north. Arrows show sites of substantial surge detachment from block-and-ash flow pulse 3 and the approximate directions of surge transport. Significant surge detachment occurred at bend 'X' facilitated by constriction of Mosquito Ghaut, decrease in slope, and build-up of deposits (see also Figs 13 and 14). Note erosion of the walls of Mosquito Ghaut on the outside of the bends. Surge deposits both east and west of Mosquito Ghaut mark areas of severe tree blowdown although some trees are standing at the margins of deposits and some damaged trees can be seen standing in the foreground of the picture (photo taken by S. C. Loughlin, BGS @ NERC).

Fig. 13. Plan view of the drainage showing the main locations of surge detachment (s) from block-and-ash flow pulse 3. 'X' marks the detachment area of a significant surge that impacted Streatham village (see also Figs 12 and 14). Block-and-ash flow pulse 3 (b and a) partially escaped from the drainage in more distal areas.

flows of ash and lapilli. On the western side of the watershed, these drained westwards into Tyre's Ghaut and northwestwards into the headwaters of Dyer's River valley. These flows merged into a single 'surge-derived pyroclastic flow' that travelled a further 3 km along the Dyer's River valley (see Druitt et al (2002) for a detailed description of this phenomenon and its hazard implications). The flow came to a halt in the inhabited zone just 50m topographically below Cork Hill. Similar processes occurred to the east of the watershed, as indicated by deposition in minor gullies. However, in the main channels there is no evidence of this because the ash would have drained back into the block-and-ash flow path. Surges detached near Harris where superelevation of the blockand-ash flow occurred at bends. In the affected area, trees were stripped of some foliage but they remained standing, wooden buildings were burned to the ground and wooden window frames were singed. The dynamic pressure of a surge travelling at 2 0 m s - 1 (see calculation above) would be about 320 Pa. assuming a flow density of c. 1 . 6 k g m - 3 , and. as observed, this would be too low to cause blowdown of mature trees (Valentine 1998). Superelevation of the block-and-ash flow at a bend near Bethel caused the flow and associated surge to escape the drainage. A fortunate family at Bethel survived indoors as the margin of the surge impacted the north side of their home (fig. 6b in Loughlin et al. 2002). The side door was left open and they reported intense heat that lasted only a few seconds but was intense enough to seriously burn a child, who was nearest the door, and to singe paper in the house (Loughlin et al. 2002). The block-and-ash flow spilled across the coastal fan inundating more houses (e.g. Fig. 7b). On the coastal plain, small, narrow flows of ash spread from the distal margins of the stalling surge lobes and followed ditches, furrows or low friction surfaces (such as roads)

PYROCLASTIC FLOWS OF 25 JUNE 1997

203

Fig. 14. (a) A series of five cross-sections down Mosquito Ghaut based on pre-emption topography. Contours in feet (xO.3048m). The prominent bend (X) (see also Figs 12 and 13) occurs c. 1.38km downstream from the crater wall and represents the zone of greatest spillage of the 25 June pyroclastic surges, (b) This shows the Mosquito Ghaut-Pea Ghaut profile, the positions of the five sections in the upper valley (filled circles), cross-sectional areas at each section (open squares) and the position of the bend 'X' In the vicinity of sections 3 and 4 the valley narrows and shallows, considerably reducing the cross-sectional area. The slope also decreases rapidly from c. 30° between sections 1 and 2 to c. 17° between sections 2 and 4. Part of Mosquito Ghaut was infilled prior to 25 June, further increasing the potential for later flows to spill out of it.

downhill, even if they were at a high angle to the flow direction (Fig. 10). These minor flows were small-scale, surge-derived pyroclastic flows. Deposits and grain-size analysis

Block-and-ash flow deposits Inspection of deposits showed that significant block-and-ash flow deposition began about 0.5 km from the lava dome where the slope of Mosquito Ghaut changed from 58° to 30° (Fig. 14). Deposition from each successive flow pulse clearly must have affected the topography and contributed to filling of the channel. Beyond the confluence of Mosquito Ghaut and Paradise River, the block-and-ash flow deposits thickened significantly (<20m) and were rich in dome-lava blocks >1 m in size. At the confluence with Tuitt's Ghaut (Fig. 1) the angle of slope decreases suddenly from 21° to 11° and there is a sharp bend. This was the distal limit of flow pulse 1, but substantial deposition from flow pulse 2 probably occurred here as a result of deceleration, decreasing the depth of the valley. A distinct, but thin lobe of coarse debris with dense blocks up to 5 m in diameter extended about 800 m NE of the bend (Figs 1 and 8) through part of Bethel village. The distal margins of the block-and-ash flow deposits tapered to thin deposits with scattered blocks, then to narrow lobes of finergrained, well-sorted, block-poor surge and/or surge-derived pyroclastic flow deposits. The block-and-ash flow deposits comprised mainly dense andesitic dome rock. Large blocks >5 m were particularly prevalent in the Harris-Bethel area where slope was reduced and the flow slowed down. Hydrothermally altered blocks in the block-and-ash

flow deposits were probably entrained from the substrate (see Cole et al. 2002), but may have included blocks of dome rock altered by fumarolic activity. Slightly vesicular andesite blocks were present, particularly in the thin margins of the coastal block-and-ash flow deposits. Low-density debris, such as tree trunks and domestic gas canisters, were also abundant at flow margins. Grain-size analyses of three typical block-and-ash flow deposit samples show the poor sorting ( > 2.5) and the coarse nature of these deposits (median diameter < 1 ), compared to the other deposits (Fig. 16). Further accounts of block-and-ash flow deposits, including 25 June deposits, are given by Cole et al. (2002).

Pyroclastic surge deposits Extensive pyroclastic surge deposits (Fig. 1) defined a broad fanshape emanating northwards from the dome. Fine-grained surge deposits were also observed in the upper part of Tyre's Ghaut, a few hundred metres west of Mosquito Ghaut. These presumably were deposited by a surge cloud that expanded rapidly almost from source and surmounted an intervening ridge. Site 1 (Figs 1 and 17) shows two units of broadly equal thickness each comprising massive, fines-rich ash and lapilli. It is believed that each unit represents the surge deposit from flow pulses 2 and 3. A very thin layer of ash at the base may have been deposited from the poorly developed surge and/or lofting ash clouds associated with flow pulse 1. Alternatively, the ash layer may have been deposited from earlier pyroclastic flows, for example the flow on 17 June. The foot of Windy Hill (site 2; Figs 1 and 17) was affected only by a surge associated with block-and-ash flow pulse 3 and two layers were identified. The deposits varied in thickness, filling minor

204

S. C. L O U G H L I N ET AL.

PYROCLASTIC FLOWS OF 25 JUNE 1997

205

On the coastal plain, distinctive fine-grained deposits occurred at the distal margins of the block-and-ash flow lobes. They were typically up to 0.5m thick, thinning gradually towards the margins, and they lacked blocks or other coarse debris. They did not extend farther than 100 m beyond the flow deposits. It could be argued that these distal deposits were simply from fine-grained block-and-ash flows, the inevitable result of slow transport of fine-grained material, but the well sorted nature of the deposits suggests that they were surge-related.

Surge-derived pyroclastic flow deposits

Fig. 16. Median grain size (Md ) versus sorting ( =( 84- 16)/2) for the 25 June 1997 pyroclastic deposits. Tie lines link samples from the same location. Filled squares = block-and-ash flow deposits. Circles = fines-poor basal layer of the surge deposits. Open squares = fines-rich upper layer of the surge deposits. Triangles = surge-derived pyroclastic flow deposits.

depressions (e.g. furrows and hollows tens of centimetres deep) in the pre-existing topography. Sections here and at site 3 (Figs 1 and 17) typically showed a lower fines-poor layer of friable medium to coarse ash and lapilli that thickened into depressions and thinned onto highs. In ploughed fields this lower layer thinned and pinched out onto the ridges between furrows oriented approximately perpendicular to the flow direction. This layer could locally be subdivided further into a lower, finer-grained part rich in sheared soil clasts and vegetation fragments, and an upper, coarser part, rich in small fragments of dome rock (rare clasts up to 10cm). The upper layer was a brown, fines-rich ash layer of almost constant thickness that mantled the topography. Gas-escape structures were common within the upper layer (pipes of fines-poor ash and lapilli) and were usually rooted on fragments of carbonized organic debris or coarse clasts. Plumes of smoke and gases could be seen rising from small vents in the surge deposits in the days after the 25 June event, and were probably caused by burning vegetation. The deposits were typical of pyroclastic surge deposits formed following other dome collapses during this eruption (Cole et al. 2002). The upper fines-rich layer usually mantled the ground and normal grading of coarse clasts occurred locally (e.g. at site 3; Figs 1 and 17). Fines-poor samples had a median diameter (Md ) of -0.5 to 3 whereas fines-rich surge deposits had a Md of 2.5 to 4 (Fig. 16). Blocks <0.5m in diameter accumulated on the upcurrent side of some houses at Streatham. All the surge deposits were well sorted (1-2 ). In the drainage channel just NW of site 3, two massive, finegrained surge layers were separated by a parting of very fine ash, probably of fallout origin (Fig. 17). These two layers were interpreted to represent sedimentation from pyroclastic surges associated with both flow pulses 2 and 3. The lower layer was poorly sorted with localized normal coarse-tail grading of clasts, whereas the upper layer was fines-rich. The lower layer may represent the approximate limit of deposition from the surge associated with pyroclastic flow pulse 2. From eyewitness accounts, this layer may have been a surgederived pyroclastic flow deposit.

At the head of Dyer's River valley, distinctive surge-derived pyroclastic flow deposits formed a single layer 0.5-1 m thick (for a detailed description see Druitt et al. 2002). The deposits retained fluidity and high temperatures for several weeks afterwards. Surge-derived flow deposits have a median grainsize in the range 1-2.5 (Fig. 16) and vary from well sorted to poorly sorted. They were fine-grained, being composed entirely of material originally suspended in, and then deposited from, the detached surges.

Ash plume deposits Part of the material in the ash clouds generated by the 25 June pyroclastic flows convected rapidly to a height of about 10km and strong winds blew the clouds westwards. Western Montserrat was subjected to prolonged ashfall and up to 2-3 mm of ash rich in accretionary lapilli fell on some inhabited villages (Bonadonna et al. 2002). Eyewitnesses in the Harris area reported dark and choking conditions for at least 20 minutes after the passage of pyroclastic flow pulse 2 whereas the upper parts of Windy Hill, at the surgecloud limit, were unaffected by ash fall (Loughlin et al. 2002).

Volume of the deposits and individual flow pulses Surveys of the block-and-ash flow deposits after 25 June gave an estimated volume of 5.5 x 106m3 (non-compacted) (Calder et al. 1999). Surge deposits had a total volume of 0.8 x 10 6 m 3 and the surgederived pyroclastic flow deposit volume was 0.09 x 10 6 m 3 (both non-compacted), making a total of 6.4 x 10 6 m 3 , non-DRE. The relative volumes of the three block-and-ash flow pulses can be estimated following the empirical study by Calder et al. (1999) on the relationship between runout distance, L, and flow volume, V\ (4)

The coefficient A depends on the properties of the block-and-ash flow, the detailed topography along the flow path and the frictional characteristics of the flow path. Here we assume that each of the pulses had similar properties and flow paths, at least down to Bramble village at 4.7km. Thus, if A is assumed a constant for all three flows then the volume proportions can be estimated from the runout lengths by: (5)

The runout lengths are 4.8, 6.8 and 6.7 km for flow pulses 1, 2 and 3 respectively, and so the volume proportions are approximately 1:3:3. The volumes of the three major flow pulse deposits are thus

Fig. 15. Oblique aerial views of areas affected by the pyroclastic surges that detached from block-and-ash flow pulse 3. (a) Empty water tank transported by the pyroclastic surge in Streatham (note vehicle beneath the tank). Bent trees singed on one side and a corrugated iron sheet wrapped around the tree in the left foreground are indicators of the surge transport direction towards the NNW. Steel reinforcement bars on the roof of the house are also bent (arrowed), (b) View up Windy Hill from Streatham. Most of the small trees were still standing but lost leaves and some branches implying a relatively low dynamic pressure. Linear features on the ground are ridges and furrows of ploughed fields, not surge dunes (photos taken by P. D. Cole).

206

S. C. LOUGHLIN ET AL.

Fig. 17. Measured sections and grain size ( ) histograms of pyroclastic surge deposits at three sites (see Fig. 1 for location of sampling sites).

roughly estimated at 0.78, 2.36 and 2.36 x 106 m3. The substantially larger volumes of flow pulse deposits 2 and 3 are also supported by their higher average velocities compared to flow pulse 1. These estimates are subject to the caveats that there may have been differences in the coefficient A between the flow pulse deposits. Flow pulses 2 and 3 travelled over topography smoothed by the previous flow pulses and pulse 3 also spread out over unconfined areas with low topographic relief in distal areas. Some empirical evidence (Calder et aL 1999) suggests that the coefficient A does not vary substantially

in block-and-ash flow deposits, so the volume estimates are considered satisfactory to a first approximation. Assuming that the pyroclastic deposits have a bulk density of 2000 kg m-3 and the lava dome has a bulk density of 2200 kg m-3 (Calder et al. 2002), the minimum volume of the scar left by the partial dome collapse can be calculated: Total deposit volume x 2 2.2 = scar volume This gives a minimum scar volume of about 5.8 x 10 6 m 3 .

PYROCLASTIC FLOWS OF 25 JUNE 1997

Discussion

Generation and behaviour of pyroclastic surges The effects of the pyroclastic surges, such as flattened and broken trees, can be seen on the margins of Mosquito Ghaut and on the interfluves in proximal regions (Fig. 12). They suggest that rapid lateral expansion and generation of the surges from flow pulse 2, and especially flow pulse 3, began almost from source. The northerly directed shear lobe that disintegrated and collapsed on 25 June had been recently emplaced and was the active sector of the dome for the six weeks prior to the collapse (Watts et al. 2002). It is proposed that disintegrating rock from deeper levels in the dome had a high pore-fluid pressure, leading to rapid expansion of magmatic gases during rock fragmentation (Woods et al. 2002). This interpretation supports the idea that the violence of gas and rock emission, and rock disintegration, increases with the depth of excavation into the dome by collapse (Fink & Kieffer 1993; Calder et al. 1999; Voight & Elsworth 2000). The horizontal shear lobe that disintegrated and collapsed on 25 June was recently emplaced (Watts et al. 2002). Pyroclastic flow pulse 1 probably comprised the relatively degassed headwall of the shear lobe. For this pulse, surge development was relatively minor; there was no surge detachment and the surge remained confined in the drainage valley. In contrast, pyroclastic flow pulses 2 and 3 excavated the interior of the lobe, where more recently emplaced lava was more likely to have high gas pressures. The high gas pressure of fragmental material from deeper within the dome strongly influenced the generation of surge clouds. The violence of fragmentation and gas expansion propelled the pyroclast-gas mixture beyond the confines of the topography from source, and the steep upper reaches of the drainage did not confine this energetic surge. Woods et al. (2002) have modelled the pressure distribution within a dome of 300 m radius as a function of dome permeability and vent exit pressure, and they found that pressure may rise rapidly to more than 1 MPa within a few tens of metres of the dome surface. For example, for a permeability of 10-13 m2 and exit pressure of 8 MPa, the pressures at 7, 20 and 60m depth are approximately 0.3, 0.6 and 1 MPa respectively (calculations from Woods et al. 2002). Here we have chosen these depths to model collapse of three successive slabs of dome over an area of 104 m2. The product of this area and these depth intervals gives the estimated volume of the scar after 25 June. The internal pressure is thus predicted to have increased with each retrogressive collapse, resulting in more vigorous gas expansion and more substantial surge clouds with time. The pyroclastic surge deposits at sites 2 and 3 show bipartite layering with a lower fines-poor layer and an upper fines-rich layer (Fig. 17). Similar deposits were produced repeatedly during the eruption (Cole et al. 2002) and have been documented elsewhere during similar eruptions (e.g. at Unzen; Nakada & Fujii 1993). Fines-poor basal deposits of pyroclastic density currents have been attributed to gravitational segregation, vaporization of water held in pre-existing soils/deposits (Boudon & Lajoie 1989) and flashburning of vegetation (Boudon & Lajoie 1989; Charland & Lajoie 1989). Cole et al. (2002) have demonstrated that flash burning of vegetation or vaporization of water can be ruled out as an origin for the fines-poor lower layer. They favour a hydrodynamic model similar to that of Druitt (1992), where the paucity of fines is related to a combination of residual turbulence and gas escape as particles sediment out. The lack of stratification within the surge deposits suggests that rapid deposition of most material may have occurred with little or no traction, but the presence of sparse blocks >0.5m indicates that outsized blocks moved by rolling and saltation. The 'outsized' blocks may have been picked up from substrate, or they may be dome rock fragments that escaped the confines of Mosquito Ghaut. Local soil clots within the surge deposits are evidence of tractive erosion. High rates of suspended-load fallout from turbulent suspensions may generate high-concentration pyroclastic flows (Druitt 1992, 1998; Druitt et al. 2002). The unusual runout capability of surge-derived flows may be attributed partly to fluidization by

207

escaping magmatic gases released by particle abrasion or vesicle rupture during transport. Rapid sedimentation could lead to high transient pore pressures encouraging flow and reducing friction (Calder et al 1999; Druitt et al. 2002). As the surges on the coastal plain ran out of momentum, minor high-concentration flows travelled in directions at a high angle to the flow direction, typically along sloping surfaces of low friction such as paved roads. It is assumed that these minor surge-derived pyroclastic flows developed as a result of rapid sedimentation at the surge front.

Comparison with other dome eruptions The sustained but highly unsteady dome collapse on 25 June generated a series of pyroclastic flow pulses that rapidly modified the topography of the volcano flanks and devastated a broad area due both to the detachment of pyroclastic surges and to the escape of block-and-ash flows from the drainage channel. Pyroclastic surges are a recognized serious hazard at dome-building volcanoes (e.g. Fisher 1995). Similar lethal phenomena occurred at Unzen in 1991 (Nakada & Fujii 1993; Yamamoto et al. 1993; Ui et al. 1999; Fujii & Nakada 1999). Pyroclastic flow pulse 1 of 25 June had a similar runout to several flows at Unzen in 1991, but the average velocity of the Unzen flows was estimated to be slightly higher: between 26 and 28ms - 1 for runout distances of about 3.2km (Nakada & Fujii 1993). Dome-collapse pyroclastic flows and surges at Merapi, Indonesia (Boudon et al. 1993; Abdurachman et al. 2000; Voight et al. 2000a, b), were similar in many respects, but also showed higher velocities. At Merapi, trees were felled right across surge-affected areas (Kelfoun et al. 2000), whereas at Montserrat there were marginal areas within the surge zones in which damaged trees were still standing. This depends to some extent on species of tree, age and size, but also suggests that the dynamic pressure of the surges at Montserrat was rapidly reduced during transport. Surge deposits of Mont Pelee, Martinique (Fisher & Heiken 1982; Boudon & Lajoie 1989), showed a bipartite layer similar to the 25 June pyroclastic surge deposits, with a lower fines-poor layer and an upper fines-rich layer. The inferred velocities of surges at Mont Pelee were significantly higher than estimates derived from data presented here. Lacroix (1904) suggested velocities in the range 100-150ms-1.

Hazard implications The pyroclastic flow deposits resulting from the partial dome collapse on 25 June filled parts of the main drainage channel within minutes. Each successive pulse and progressive aggradation from each pulse contributed to the deposits and caused the drainage channel to become shallower. The risk of surge detachment and escape of subsequent pyroclastic flows from the channel therefore increased rapidly over several minutes. The runout distance of the second and third block-and-ash flow pulses was high and may have been enhanced by a high juvenile gas content released by clast attrition during transport. The substrate of hot, just-emplaced blocks and ash may also have aided transport by being readily sheared and by smoothing the channel. The gradually increasing runout and valley-confined nature of the pyroclastic flows at Soufriere Hills Volcano throughout May and early June 1997 suggested a steady increase in activity. The devastating and lethal 3 June 1991 dome collapse pyroclastic flows at Unzen were also preceded by several days of small flows of increasing runout distances (Nakada & Fujii 1993). Crisis management officials should be aware that sudden dramatic increases in the intensity and impact of lava dome eruptions can occur, and in some instances are to be expected. The surge-derived pyroclastic flow followed an unexpected drainage and far exceeded runout expectations even for the 'worstcase scenario'. Such a phenomenon was not anticipated as a primary hazard at Soufriere Hills Volcano. Surge-derived pyroclastic

208

S. C. LOUGHLIN ET AL.

flows generated by rapid sedimentation from a surge cloud have been recognized before (e.g. Hoblitt et al. 1981; Belousov 1996) but they have been an underestimated hazard near lava domes (Druitt et al. 2002). Their considerable runout capability and tendency to follow drainages that need not be sourced near the lava dome and can be at high angles to the surge transport direction means that particularly careful consideration needs to be given to the potential hazards from these phenomena. Knowing the potential volatile content and state of pressurization of the collapsing section of the lava dome is of critical importance to hazard assessment (Fink & KiefTer 1993; Voight & Ellsworth 2000). Collapse of a volatile-rich pressurized segment of dome is violent and will affect a far larger area than a simple gravitational collapse of unpressurized dome rock. At Soufriere Hills Volcano, the active shear lobe on the dome was pushed out at an elevated rate after 22 June due to pressurization caused by a fresh pulse of magma from a deep source. This caused destabilization of the headwall and the break-up and ultimate collapse of the entire lobe (Watts et al. 2002). It was known that part of the northern side of the dome was recently emplaced and possibly volatile-rich, but, even so. the violence of the third flow pulse and the extent of the deposits were considered remarkable. The potential to generate violent surges almost from source and the ability of the surges to segregate and transform into surge-derived pyroclastic flows has perhaps been underestimated at many dome-building volcanoes. Fink & Kieffer (1993) recognized that the course of future domecollapse pyroclastic flows cannot be reliably determined solely by tracing the paths of previous and historical flows. The collapse of different parts of the lava dome with different emplacement histories and the varying depth of excavation will generate flows with widely differing dynamic properties. Mapping the dome surface and monitoring lobe emplacement and degassing is therefore an important monitoring approach (Fink & Kieffer 1993; Watts et al. 2002). Sharp bends, rapid changes in slope, or constrictions in channels draining from a volcano should be considered as potential locations for detachment of pyroclastic surges. Additionally, the potential for channel floor topography to change rapidly during a single dome collapse should not be overlooked. Any area at risk from a detached pyroclastic surge (even beyond topographic barriers) should be examined for further drainage channels that could carry surgederived pyroclastic flows. We owe many thanks to all members of the MVO staff who worked particularly hard during this period of intense volcanic activity. The manuscript was improved by comments from B. Dade. R. Waitt and P. Kokelaar. R.S.J.S. acknowledges a NERC Professorship. B.V. acknowledges support from US National Science Foundation. S.L. and R.L. publish with the permission of the Director, British Geological Survey (NERC). References ABDURACHMAN, E. K., BOURDIER, J. L. & VOIGHT. B. 2000. Nuees ardentes of November 22 1994 at Merapi Volcano, Java, Indonesia. Journal of Volcanology and Geothennal Research, 100, 345-362. ASPINALL. W. P.. LOUGHLIN. S. C.. MICHAEL, F. V. ET AL. 2002. The Montserrat Volcano Observatory: its evolution, organization, role and activities. In: DRUITT. T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs, 21. 71-91. BAPTIE. B.. LUCKETT, R. & NEUBERG, J. 2002. Observations of lowfrequency earthquakes and volcanic tremor at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat. from 1995 to 1999. Geological Society, London. Memoirs. 21, 611-620. BELOUSOV, A. 1996. Deposits of the 30 March 1956 directed blast at Bezymianny Volcano, Kamchatka, Russia. Bulletin of Volcanology, 57, 649-662. BONADONNA, C.. MAYBERRY, G. C.. CALDER, E. S. ET AL. 2002. Tephra fallout in the eruption of the Soufriere Hills Volcano, Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London, Memoirs. 21. 483-516.

BOUDON. G. & LAJOIE. J. 1989. The 1902 Pelean deposits in the Fort Cemetary of St Pierre. Martinique: a model for the accummulation of turbulent nuees ardentes. Journal of Volcanologv and Geothennal Research. 38. 113-129. BOUDON. G., CAMUS. G.. GOURGAUD. A. & LAJOIE. J. 1993. The 1984 nueeardente deposits of Merapi volcano, central Java. Indonesia: stratigraphy, textural characteristics and transport mechanisms. Bulletin of Volcanology, 55. 327-343. CALDER. E. S.. COLE. P. D., DADE. W. B. ET AL. 1999. Mobility of pyroclastic flows and surges at the Soufriere Hills Volcano. Montserrat. Geophysical Research Letters. 26(5). 537-540. CALDER. E. S.. LUCKETT. R., SPARKS. R. S. J. & VOIGHT. B. 2002. Mechanisms of lava dome instability and generation of rockfalls and pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London. Memoirs. 21. 173-190. CHARLAND. A. & LAJOIE. J. 1989. Characteristics of pyroclastic deposits at the margin of Fond Canonville. Martinique and its implications for the transport of the 1902 nuees ardentes of Mt Pelee. Journal of Volcanology and Geothermal Research. 38. 97-112. CLARKE, A. & VOIGHT. B. 2000. Pyroclastic current dynamic pressure from aerodynamics of tree or pole blow-down. Journal of Volcanology and Geothennal Research. 100. 395-412. COLE. P. D.. CALDER. E. S., DRUITT. T. H.. HOBLITT. R.. ROBERTSON. R., SPARKS. R. S. J. & YOUNG. S. R. 1998. Pyroclastic flows generated by gravitational instability of the 1996-1997 lava dome of Soufriere Hills Volcano. Montserrat. Geophysical Research Letters. 25(18). 3425-3428. COLE. P. D.. CALDER. E. S.. SPARKS. R. S. J. ET AL. 2002. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21. 231-262. DRUITT. T H. 1992. Emplacement of the 18 May 1980 lateral blast deposit ENE of Mount St Helens. Washington. Bulletin of Volcanology. 54. 554-570. DRUITT. T. H. 1998. Pyroclastic density currents. In: GILBERT. J. S. & SPARKS. R. S. J. (eds) The Physics of Explosive Volcanic Eruptions. Geological Society. London. Special Publications. 145. 145-182. DRUITT, T. H.. CALDER. E. S. COLE. P. D. ETAL. 2002. Small-volume, highly mobile pyroclastic flows formed by rapid sedimentation from pyroclastic surges at Soufriere Hills Volano. Montserrat: an important volcanic hazard. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21. 263-279. FINK. J. H. & KIEFFER. S. W, 1993. Estimate of pyroclastic flow velocities from explosive decompression of lava domes. Nature. 363. 612-615. FISHER. R. V. 1995. Decoupling of pyroclastic currents: hazards assessment. Journal of Volcanology and Geothennal Research. 66. 257-263. FISHER. R. V. & HEIKEN. G. 1982. Mt Pelee Martinique: May 8 and 20 1902 pyroclastic flows and surges. Journal of Volcanology and Geothennal Research. 13. 339-371. FUJII. T. & NAKADA. S. 1999. The 15 September 1991 pyroclastic flows at Unzen Volcano (Japan): a flow model for associated ash-cloud surges. Journal of Volcanologv and Geothennal Research. 89. 159-172. HOBLITT. R. P.. MILLER. C. D. & VALLANCE. J. W. 1981. Origin of stratigraphy of the deposit produced by the May 18 directed blast. In: LIPMAN. P. W. & MULLINEAUX. D. R. (eds) The 1980 Eruptions of Mount St Helens, Washington. US Geological Survey. Professional Papers. 1250. 401-419. KELFOUN. K.. LEGROS. F. & GOURGAUD. A. 2000. A statistical study of trees damaged by the 22 November 1994 eruption of Merapi Volcano (Java. Indonesia): relationships between ash-cloud surges and block-and-ash flows. Journal of I'okanology and Geothermal Research. 100. 379-394. KOKELAAR, B. P. 2002. Setting, chronology and consequences of the eruption of Soufriere Hills Volcano. Montserrat (1995-1999). In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999, Geological Society. London. Memoirs. 21, 1-43. LACROIX. A. 1904. La Montagne Pelee et ses eruptions. Masson et cie. Paris. LOUGHLIN. S. C.. BAXTER. P. J.. ASPINALL. W. P.. DARROUX. B.. HARFORD. C. L. & MILLER, A. D. 2002. Eyewitness accounts of the 25 June 1997 pyroclastic flows and surges at Soufriere Hills Volcano, Montserrat and implications for disaster mitigation. In: DRUITT. T. H. & KOKELAAR.

PYROCLASTIC FLOWS OF 25 JUNE 1997 B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 211-230. MILLER, A. D., STEWART, R. C., WHITE, R. A. ET AL. 1998. Seismicity associated with dome growth and collapse at the Soufriere Hills volcano, Montserrat. Geophysical Research Letters, 25(18), 3401-3404. NAKADA, S. & FUJII, T. 1993. Preliminary report on the activity at Unzen Volcano (Japan), November 1990-November 1991: Dacite lava domes and pyroclastic flows. Journal of Volcanologv and Geothermal Research, 54, 319-333. SPARKS, R. S. J., YOUNG, S. R., BARCLAY, J. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills Volcano, Montserrat: November 1995 to May 1997. Geophysical Research Letters, 25(18), 3421-3424. UI, T., MATSUWO, N., SUMITA, M. & FUJINAWA, A. 1999. Generation of pyroclastic flows during the 1990-1995 eruption of Unzen Volcano, Japan. Journal of Volcanology and Geothermal Research, 89, 123-137. VALENTINE, G. A. 1998. Damage to structures by pyroclastic flows and surges, inferred from nuclear weapons effects. Journal of Volcanology and Geothermal Research, 87, 117-140. VOIGHT, B. & ELSWORTH, D. 2000. Instability and collapse of hazardous gas-pressurized lava domes. Geophysical Research Letters, 27, 1-4. VOIGHT, B., HOBLITT, R. P., CLARKE, A. B., LOCKHART, A. B., MILLER, A. D., LYNCH, L. & McMAHON, J. 1998. Remarkable cyclic ground deformation monitored in real-time on Montserrat, and its use in eruption forecasting. Geophysical Research Letters, 25(18), 3405-3408.

209

VOIGHT, B., SPARKS, R. S. J., MILLER, A. D. ET AL. 1999. Magma flow instability and cyclic activity at Soufriere Hills Volcano, Montserrat, British West Indies. Science, 283, 1138-1142. VOIGHT, B., YOUNG, K. D., HIDAYAT, D. ET AL. 2000a. Deformation and seismic precursors to dome-collapse and fountain collapse nuees ardentes at Merapi Volcano, Java, Indonesia 1994-1998. Journal of Volcanology and Geothermal Research, 100, 261-288. VOIGHT, B., CONSTANTINE, E. K., SISWOWIDJOYO, S. & TORLEY, R. 2000b. Historical activity at Merapi Volcano, Java, Indonesia, 1768-1998. Journal of Volcanology and Geothermal Research, 100, 669-138. WATTS, R. B., HERD, R. A., SPARKS, R. S. J. & YOUNG, S. R. 2002. Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 115-121. WOODS, A. W., SPARKS, R. S. J., RITCHIE, L. BATEY, J., GLADSTONE, C., & BURSIK, M. I. 2002. The explosive decompression of a pressurized lava dome: the 26 December 1997 collapse and explosion of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 457-466. YAMAMOTO, T., TAKARADA, S. & SUTO, S. 1993. Pyroclastic flows from the 1991 eruption of Unzen Volcano, Japan. Bulletin of Volcanology, 55, 166-175.

Eyewitness accounts of the 25 June 1997 pyroclastic flows and surges at Soufriere Hills Volcano, Montserrat, and implications for disaster mitigation

2

S. C. LOUGHLIN1, P. J. BAXTER2, W. P. ASPINALL3, B. DARROUX4, C. L. HARFORD 5 & A. D. MILLER 1 . 1 British Geological Survey, West Mains Road, Edinburgh EH9 3LE, UK University of Cambridge Clinical School, Addenbrookes Hospital, Cambridge CB2 2QQ, UK 3 Aspinall and Associates, 5 Woodside Close, Beaconsfield HP9 1JQ, UK 4 Montserrat Volcano Observatory, Mongo Hill, Montserrat, West Indies 5 Department of Earth Sciences, University of Bristol, Bristol BS8 1RJ, UK

Abstract: Eyewitness and survivor accounts allow reconstruction of the sequence of events on 25 June 1997, when a sustained partial collapse of the lava dome occurred leading to the death of 19 people. An unsteady pyroclastic flow was generated with three distinct pulses. The third flow pulse caused most of the damage to infrastructure and most, if not all, of the casualties. Pyroclastic surges detached along most of the path of the third flow pulse, and one travelled 70m up an adjacent hillside. Observations were made that will be important for the development of mitigation measures at future events involving hightemperature flows and surges. Temperatures remained high (300-400°C) at the periphery of the most voluminous and extensive surge, even though dynamic pressure and velocity were low, causing the death of seven victims. Some people survived at the margins of the surge zone but suffered serious burns when they were forced to walk across the hot surge deposits to safety. Deflagration of buildings and vegetation was immediate within the pyroclastic surge and intense fires burned long after the volcanic activity had ceased. Fires could be a serious secondary hazard in an urban area. Search-and-rescue efforts were hampered in the immediate aftermath of the pyroclastic flows and surges by smoke and ash in the atmosphere. The hot, locally gas-rich surge deposits posed a major hazard to search-and-rescue workers and volcanologists for days afterwards. Despite the efforts of officials, scientists and concerned members of the public, about 80 people were in Zones A and B of the Exclusion Zone on 25 June 1997. Our findings suggest that many had become accustomed to the pyroclastic flows and had become overconfident in their own ability to judge the threat by observing repeated flows that had gradually increased runout but remained restricted to valleys. Many people had contingency plans and believed that there would be observable or audible warning signs from the volcano if the activity were to escalate significantly. However, there were no such discernible warnings and individual contingency plans proved inadequate. Public education should concentrate on correcting such public misapprehension of hazardous phenomena and attendant risks in future volcanic crises.

In this paper we include descriptions of the events of 25 June 1997 provided by survivors, onlookers and search-and-rescue workers. The information is of value in the scientific analysis of eruptive events (e.g. Loughlin et al. 2002) and it provides insights into the effects of volcanic phenomena on the public (Bernstein et al. 1986). The descriptions and comments of the survivors and their perceptions of risk and response to warnings should be useful in devising better mitigation measures and in emergency planning in areas of active volcanism (Baxter et al. 1998). Such accounts are sparse in the volcanic literature. Notable exceptions are the accounts by Anderson & Flett (1903) of the eruption of the Soufriere of St Vincent volcano, in 1902, and by Rosenbaum & Waitt (1981) of the events at Mount St Helens in 1980. As in the eruption at Montserrat, deaths in both these cases were caused by pyroclastic flows and pyroclastic surges in particular. Anderson & Flett's vivid descriptions provide dramatic evidence of survival inside buildings in the periphery of a flow that killed more than 1500 people. At Mount St Helens, on 18 May 1980, 58 people were killed but 17 survived inside the area impacted by the pyroclastic surge (Bernstein et al. 1986). On 3 June 1991, 41 people, including three volcanologists, died in a pyroclastic surge from a dome collapse at Mount Unzen, Japan; four persons survived with minor burns (Shimabara Onsen Hospital 1993). As a consequence of dome-collapse pyroclastic flows and surges at Merapi, Indonesia, in 1994, there were 63 deaths and 23 persons who survived and required hospital treatment for burns (Baxter et al. 1998). The lessons of all these events were used in the application of disaster mitigation plans on Montserrat. On Montserrat, about 6000 people were evacuated from the Exclusion Zone in the months preceding the 25 June 1997 dome collapse. Volcanic risk management maps issued by the Government of Montserrat were used in conjunction with an alert system to control access to different zones on the island. The Montserrat Volcano Observatory (MVO) assisted the Government of Mont-

serrat with the development of the alert system and the drafting of the maps (Aspinall et al. 2002). The scientific team disseminated information about the volcanic activity in a wide variety of ways, including twice-daily radio reports, live radio interviews, newspaper and magazine articles, public meetings, school visits and poster boards (Aspinall et al. 2002). This was in addition to the extensive education campaign carried out by the Government Information Service. As a result, volcanic terminology was widely used in public debate and well understood by the local people who, after two years' experience, had become familiar with dome-collapse pyroclastic flows.

Terminology At Montserrat, collapse of the lava dome typically generated (1) a high-particle concentration basal part containing abundant blocks and termed a pyroclastic flow or block-and-ash flow, (2) a low-particle concentration pyroclastic surge that overlies the block-and-ash flow and may detach from the main flow and move independently and (3) a lofting buoyant ash plume. Rockfalls are distinguished from dome collapse pyroclastic flows by runouts < 1 km and the failure of many blocks to disintegrate completely and produce copious ash. The 25 June 1997 partial dome collapse produced an extremely unsteady, or pulsatory pyroclastic flow. Ghant is the local term for a deep stream valley.

Methodology Official lists of rescued persons, and of the dead and injured, together with accounts of eyewitnesses, indicate that there were

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 211-230. 0435-4052/02/$15 © The Geological Society of London 2002.

211

212

S. C. L O U G H L I N ET AL.

Fig. 1. Volcanic Risk map revised on 6 June 1997; the alert level was orange. Zones A and B, No access; Zone C, Access limited to short visits by residents and workers with rapid means of exit. Zone D, Daytime occupation for essential services and agriculture, residents allowed access in daytime. Essential services operate with standby transport and evacuation plans in place. Zone E, Full occupation by residents with contingency plan for evacuation. Zone G, full occupation. Major roads are shown as dashed lines.

80 people within Zones A and B (Fig. 1) at 12:55 LT (all times given are local time, L T = G M T minus 4 hours) on 25 June 1997. Of these, 19 lost their lives. Out of the remaining 61 people, we recorded the experiences of 37 in interviews and, although 24 people could not be traced, some of these had given interviews to local media and these were taken into account. There were also approximately 30 official wofkers in Zone C. They were based at W. H. Bramble Airport and Tram's quarry, in the east, and in Plymouth to the west (at the Port Authority, Monlec electricity company and the rice mill). In the weeks and months after 25 June 1997, 33 eyewitnesses were interviewed, some of them twice, to provide observational details for scientific analysis of the timing of eruptive activity and of the effects of the block-and-ash flows and pyroclastic surges. Locating eyewitnesses was difficult as many of them were displaced from their homes and were living in shelters, or with friends and family in the north of the island. Attempts to find those who went

to the United Kingdom were only partly successful. Recounted observations of MVO staff, officials and taxi drivers at the airport, and search-and-rescue workers, and our personal observations are also included. Although eyewitness accounts are subjective, they can be useful for constraining the sequence of events and the timing of the activity, especially when combined with information from scientific monitoring (e.g. Loughlin et al. 2002). For instance, as a result of poor visibility scientists could not know the exact sequence of events in the areas of Farrell's Yard. Streatham. Harris, Bethel and Farms (Fig. 2) without the help of eyewitnesses. Descriptions of the effects on vegetation and buildings were also useful in constraining temperatures and dynamics of the pyroclastic flows and surges. Some early interviews were brief and restricted to observational details and the evasive action taken, but many eyewitnesses commented later on broader issues. These included their response to warnings, their understanding of the hazards and risks of the

EYEWITNESS ACCOUNTS OF 25 JUNE 1997

213

Fig. 2. Map showing the full extent of the 25 June 1997 pyroclastic flow deposits.

volcano, and their perceptions of scientists and officials. These comments and their implications are also discussed here. Interviewers allowed eyewitnesses to speak freely as much as possible and avoided loaded or leading questions (Oppenheim 1992). The original descriptive words of the eyewitnesses are retained and editorial comments, for clarification, have been added in italics. Some accounts differ greatly, especially in relating times and velocities, although this is to be expected given the stress due to the experience. Overall, however, the accounts are coherent and present a clear picture of the sequence of events on 25 June 1997, from a number of vantage points. In the following sections, we provide overview summaries before the specific observations are reproduced. Each eyewitness or group is identified by a number and by location at 12:55 on 25 June (Table 1, Fig. 3). It should be remembered when reading these transcripts that Montserratians had become familiar with volcanological terminology and used it regularly in conversation. Summary of precursory volcanic activity, May-June 1997 From 14 May 1997, the focus of lava dome growth switched to the northern summit area of the lava dome (Watts et al. 2002; fig. 2). Rockfalls immediately began on the north side of the dome. By 19 May, material was spilling over the northern crater walls into the top of Tuitt's Ghaut. On 27 May, the first dome-collapse pyroclastic flows entered Tuitt's Ghaut and the size of subsequent flows increased steadily. On 6 June 1997, a revised risk map (Fig. 1) was issued by the Government of Montserrat, following advice from the MVO that risk on the northern flanks of the volcano had increased. Throughout early June, a tiltmeter installed at Chances Peak recorded a regular pattern of inflation-deflation of the dome with

a periodicity of 12-16 hours and amplitude of 16-18 rad (Voight et al. 1998, 1999). The inflationary part of each cycle was accompanied by an intense hybrid earthquake swarm, responding to an increase in magma pressure in the upper conduit (Voight et al. 1999). Enhanced rockfall/pyroclastic flow activity and ash-venting occurred during the deflationary phase of each cycle. On 16 June pyroclastic flows travelled 1.6km down Gages Ghaut to the west, and smaller flows occurred in Mosquito and Tuitt's Ghauts to the north. As a result of the steadily increasing runout of the flows in the northern valleys, a revised risk map was drafted; it was due for release on 25 June. On 17 June, pyroclastic flows travelled 1.8 km down Gages Ghaut and 4 km down Mosquito Ghaut. A pyroclastic surge surmounted the valley walls and several hundred square metres of land to the west (downwind) of Mosquito Ghaut was scorched (Fig. 2). Cyclic inflation and deflation of the crater area continued with renewed intensity on 22 June, with a reduced periodicity of 8-12 hours and an increased amplitude of as much as 30 rad. Rockfall and pyroclastic flow activity occurred regularly, coinciding with each deflationary period. For further discussion of precursory activity see Voight et al. (1998) and Loughlin et al. (2002).

Precursory activity on 25 June 1997 A hybrid earthquake swarm began at 03:00 and, continuing the cyclic activity, was accompanied by relative inflation of the lava dome. Deflation commenced at about 06:10. Between 06:15 and 07:15 pyroclastic flows in Mosquito Ghaut were almost continuous and reached maximum runouts of about 1.5km. Re-inflation of the dome began again at 09:00. Between 12:40 and 12:50, the tiltmeter

S. C. LOUGHLIN ET AL.

214

Table 1 . Location of eyewitnesses at 13:00 LT on 25 June 1997 (see Fig. 3) No.

No. in party

Location

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29

2

Harris, Zone A Trant's, Zone C Trant's, Zone C Bramble, Zone A Bramble, Zone A Tuitt's, Zone B Harris, Zone A Farm, Zone C FarrelPs, Zone A Trant's Police checkpoint, C/F Spanish Point, Zone B Dyer's, Zone B Dyer's, Zone B Harris, Zone A Harris, Zone A Streatham, Zone B Streatham, Zone B Harris, Zone A Airport, MVO staff Cork Hill, Zone D Airport, Zone C Airport, Zone C Airport, Zone C Harris, Zone A Dyer's, Zone B/C Trant's, Zone C St Peter's, Zone G Fox's Bay, Zone E Airport

1 1 2 1 1 2 1 7 2 3 5 3 2 2 2 1 2 3 1 1 1 2 1 2 1 1 1 1

No. at nterview

,2 1 1 22 2

showed the onset of the deflationary phase of the cycle. A strong continuous seismic signal began at 12:55, with intensified pulses of activity at 12:57, 13:00 and 13:08 (see fig. 4 in Loughlin et al. 2002: Baptie et al. 2002). This signal is interpreted to record a sustained and unsteady pyroclastic flow with three major pulses, each of which developed a distinct flow front (see Loughlin et al. 2002) From just after dawn through to mid-morning on 25 June 1997, scientists saw several people working in the fields at Farrell's Yard (Fig. 2), a regular practice that had continued almost uninterrupted throughout the crisis. With good visibility, the cyclic nature of the volcanic activity could be observed and some farmers used these cycles to their benefit, working only when the dome activity was low (i.e. during the inflationary part of the cycle). However, those farmers who arrived in the Exclusion Zone very early in the morning on 25 June (06:00) started work at the end of a period of low activity and continued working throughout the morning, despite particularly poor visibility (due to low cloud) and gradually increasing volcanic activity. By 07:00, people arriving at the St George's Hill police check point (Fig. 1) noticed a blue haze over the northern slopes of the volcano, a sulphurous smell and the sound of regular small dome-collapse pyroclastic flows in Mosquito Ghaut. These phenomena, in addition to the poor visibility, caused many people to return to the north. Those already farming in the Exclusion Zone were less aware of the sulphurous gas, probably because it built up gradually while they were there, but they did notice the increased number of rockfalls and small pyroclastic flows during the morning. Those people still living in the Exclusion Zone were mostly located to the north and NE of the volcano and were rarely affected by ash or gas emissions because of the easterly prevailing wind direction. They, and the farmers visiting the Exclusion Zone on a daily basis, had become used to seeing the regular, small pyroclastic flows in Mosquito Ghaut, but they rarely heard them.

Fig. 3. Location of eyewitnesses at 13:00 LT on 25 June 1997, and location of recovered bodies (A-E) within the surge zone. The locations of witnesses 27 and 28 lie off the map area. For further details see Table 1.

EYEWITNESS ACCOUNTS OF 25 JUNE 1997

215

Airport

It seems likely from the accounts below that by 12:55 on 25 June most farmers had left the fields and had either returned north or were in villages such as Farm (Zone B Fig. 1) for their lunch. Seven people are known to have been at Farrell's Yard when the collapse began, all having arrived at about 06:00. Four of them escaped by car after the second pyroclastic flow pulse, but the other three were killed on Windy Hill by the surge that detached from flow pulse 3 (location B Fig. 3). They had attempted to escape up Windy Hill on foot. Like many others, they had a contingency plan, but it was based on their previous experience of the activity, not on the more extreme scenarios that scientists had described were possible.

Eyewitness 22. 'Basically, there were some small pyroclastic flows early in the morning, but these were obscured by the low-lying clouds around the whole volcano. ... There were points when the cloud base actually lifted a bit so that we got some glimpses as to what was happening a little further up the mountain but at no point in time was it possible to see the mid- to upper sections of the dome. Given the frequency of pyroclastic flows I think that everyone was a bit on edge. Given the fact that we couldn't see and the frequency of the flows.1

Precursory activity: summary of eyewitness accounts

St George's Hill

Farrell's Yard

Recounted by the friend of an eyewitness. They went that morning to go to Lee's so they had to go up St George's Hill and there was a police checkpoint. When she reached that police checkpoint well she was so near. She say, ''What happened to the mountainside, why it so blue?" The officer say it was gas and if she stayed there for half an hour she would see at least two pyroclastic flows going down Mosquito Ghaut. They say they could hear the roaring of the mountain.'

Eyewitness 27. That morning I went down to Farms but it wasn't looking right. It was grumbling a lot and it was hazy and foggy that day. I didn't like it so I decided to leave. I left at about midday.' [Farm village is commonly referred to as Farms.] Eyewitnesses 9. That morning, when we got in, we had small constant ash flows ... something that would come down and drift away in a matter of 2 or 3 seconds. But they were much more constant that morning. It went on and on. We got in there at about a quarter after six. We saw constant pyroclastic flows all morning from 06:50; they were all small ones that reached the base of the mountain. It was going like that all morning. There were two or three heavy fogs in the morning when the rain was coming. We stayed in the rain picking carrots.'

The pyroclastic flow and accompanying surges Video footage, seismic evidence and eyewitness accounts all indicate that there were three major flow pulses (Loughlin et al. 2002). Maps showing the probable extent of deposition from pulses 1 and 2 are shown in figure 5 of Loughlin et al. (2002), but deposition was probably semi-continuous throughout. The extent of all 25 June deposits is shown here in Figure 2.

Streatham Flow pulse 1 Eyewitness 16. 'I didn't hear nothing at all [in the morning] ... After the thing happened many people talking about it - the blue haze. But I didn't hear it and me right down under the mountain, nothing to hear and not studying nothing.'

The first flow pulse remained confined within Mosquito Ghaut and reached Paradise River below Bramble village; it travelled 4.7km and had an average velocity of 1 5 m s - 1 .

Tuitt's

Flow pulse 2

Eyewitness 6. 'I saw a pyroclastic flow coming down Tuitt's Ghaut. Just a bit below where it came down first ... I kept a watch for most of the day but it did not give any sound. If I had heard some sound I would have moved away earlier.'

The second flow pulse, approximately two minutes later, travelled up to 6.8 km with an average velocity of 16.1 m s - 1 . Between the dome and Bramble village (4.7 km) it had an average velocity of 20.2 m s -1 . It remained confined to the valley and reached to within 50m of the

Fig. 4. Front of flow pulse 2 passing along Farm River, just south of Trant's village (note houses in centre of view) (photo taken by R. B. Watts).

216

S. C. LOUGHLIN ET AL.

coast where its front was observed to move in a pulse-like manner (Fig. 4), A pyroclastic surge detached from the flow as it travelled down Mosquito Ghaut and this swept across Farrell's Yard and Riley's Yard (Eyewitnesses 9), moving north as far as the main road. The north side of the road near Windy Hill was not affected, but the surge did cross the road and impact the southern part of Streatham village. Ash deposited from the surge was observed draining westwards down the narrow tarmac road surface towards Dyer's (Eyewitness 9). This has been interpreted as the generation of a surgederived pyroclastic flow (see Druitt et al. 2002). Flow pulse 3 The third and fastest flow pulse initiated eight minutes after the second. The block-and-ash flow remained confined in Mosquito Ghaut, but impacted property at Harris village that was in the valley. It spilled out of the river valley at a bend near Bramble village, beyond which parts of Bethel and Spanish Point were inundated (Figs 5 and 6). The flow continued down Pea Ghaut, spilling out of the shallow drainage and burying Farm village then Trant's village, where the flow front arrived as several pulses in quick succession (Fig. 7). Up to a point 5.8km from the dome, the flow had an average velocity of 2 1 . 9 m s - 1 . Pyroclastic surges detached from flow pulse 3 along Mosquito Ghaut and swept across the fan on the north flank of the volcano. The main surge crossed the main road and climbed 70m up Windy Hill. Streatham village was devastated (Figs 8 and 9) and seven

people were killed in this area (Fig. 3). Directional indicators in Streatham, such as bent fence posts, bent reinforcement bars, telegraph poles and trees singed on one side, suggest that the main surge was directed to the north and NW and detached from a sharp bend about 500m south of Farrells' Yard. In Streatham, temperature effects were extreme; aluminium window frames and glass melted and wooden material burned, whereas on Windy Hill window panes were cracked and wooden doors singed. The scorched zone downwind (west) of Windy Hill was up to 250m wide but it was as little as 5m on the upwind (east) side (Fig. 8). In the lower parts of Mosquito Ghaut a detached pyroclastic surge climbed 50m up a hillside to the east of Windy Hill. Rapid sedimentation from the main surge generated a high-concentration granular flow (surgederived pyroclastic flow) that drained westwards into a valley not anticipated to be at high risk (Druitt et al. 2002). Pyroclastic surges also detached from flow pulse 3 as it travelled along Paradise River and impacted the lower part of Harris village (Fig. 10). A wooden rum shop at the edge of the surge zone was burned to the ground and bottles within it melted to form contorted shapes. One person was killed in the Harris area. The flows and pyroclastic surges were apparently quite noiseless. Many people were not aware of the approach of the flows or surges until they noticed a sudden darkness or heard nearby items exploding. For example, Mandy's gas station (Fig. 2) exploded when it was hit by a pyroclastic surge that detached from flow pulse 2. Most people had personal contingency plans that relied on precursory audible warnings from the volcano itself, but it gave no

Fig. 5. Block-and-ash flow deposits in the village of Bethel, with Paradise River valley to the right of the photograph. Part of flow pulse 3 escaped from the confines of Paradise River valley as a result of superelevation at the bends (arrowed) and spilled through Bethel village, causing extensive damage. (Flow direction from upper right towards lower left.) The home of Eyewitnesses 11 is in the bottom left corner of the photograph (photo taken by S. C. Loughlin, BGS @ NERC).

EYEWITNESS ACCOUNTS OF 25 JUNE 1997

217

Fig. 6. (a) Bethel church burnt out as a result of fires ignited by the heat of pyroclastic surges that travelled from right to left. The home of Eyewitnesses 11 is on the left of the photograph (photo taken by P. D. Cole), (b) Bethel church and the home of Eyewitnesses 11 (bottom right). The main channel is shown to the rear of the photograph in which the main flow travelled from left to right. Block-and-ash flows and pyroclastic surges that escaped from the drainage travelled from top left to bottom right in this photograph (photo taken by S. C. Loughlin, BGS © NERC).

such warning. People deep in the Exclusion Zone, at Streatham, Windy Hill and parts of Harris (Zones A and B), could not hear the sirens that were sounded in Plymouth and at the airport (Zone C, Fig. 1) for the benefit of official workers. However, even as the flow pulses travelled towards the airport, some people close by who heard that siren did not believe the situation was especially serious and delayed their escape to turn off cookers, collect belongings or turn dogs loose. Several people survived inside buildings at the margin of the surge zone, but were then forced to walk across hot ash to escape. A couple and their daughter at Bethel were particularly fortunate to survive when a surge passed within a few metres of their house (Eyewitness 11, Fig. 6a). The two lower floors of their three-storey home were built into the slope (Fig. 6b) so they sought refuge in an interior room on the second floor, but some south-facing windows and one door had been left open (at right-angles to the flow direction) and hot ash entered. Their rapid retreat to within the house saved their lives although the 4-year-old daughter suffered burns. They were then able to escape without further injury along an unaffected strip of slightly raised ground leading to the coast. The wooden roofs of larger concrete homes near the margins of the surge were damaged only on the side impacted by the pyroclastic surge (e.g. Fig. 11), and concrete walls invariably survived the impact. While such walls, when impacted by block-and-ash flows or the main part of a surge, typically remained standing, the roofs, windows and interiors were generally destroyed (Fig. 12). Wooden homes burned rapidly, even at the surge margins and in the singe zone. Eyewitnesses described raging fires in all houses that were surrounded by surge or block-and-ash flow deposits. Although many witnesses described explosions in houses that they ascribed to exploding gas cylinders (used for domestic cooking), only whole gas bottles were found in the aftermath (Fig. 12). Pyroclastic flow and accompanying surges: summary of eyewitness accounts Farrell's Yard Eyewitnesses 9. 'The fog had just cleared 5 or 6 minutes before that. And it is a habit of mine, whenever I'm going through the Farrell's gate, to always look back. As I looked back, I realized that the pyroclastic flow was going down a ghaut that it has never gone down before [flow pulse 1 in the lower part of Mosquito Ghaut]. We knew at some point in time it would have been coming down

there but we didn't hear any rumbling that morning. As I put my hand on the gate to open it and looked back I saw it had already passed the gas station [Mandy's] ... But by the time we'd closed the gate there was a big avalanche, it all happened in seconds. Part of it had already got to houses [detached surge from flow pulse 2]. It went across Riley's towards Streatham and a small drift of mainly ash came onto the road ... We never hear any sound ... When it hit the road it went down the road [westward towards Dyer's], it didn't

218

S. C. LOUGHLIN ET AL.

Fig. 7. The front of flow pulse 3 escaped from Farm River and is shown advancing on the village of Trant's, which it destroyed. Further pulses are shown in a sequence of photographs in Loughlin et al. (2002) (photo taken by R. B. Watts).

bounce over on the other side ... The road was boiling and the ash moved round the bends on the road like a vehicle.' [They then drove up Windy Hill.]

Windy Hill Eyewitnesses 9. 'But then the other one that came down [detached surge from flow pulse 3], that one was so big that when it hit the

road, and it started spreading, it engulfed both sides ... The houses that we could see that weren't burning [on the edge of the singe zone on Windy Hill], the shingles on some of them was twisted, telephone wires was fallen, blowing around in the wind. The telegraph poles was not burning, just vegetation. There was one house at the end there that I watched. The paint just started caking up, blistering and the shingles just singed, they started melting while I was watching. That was right on the edge. But there was no ash up there at all/

Fig. 8. An aerial view looking south from above Windy Hill showing the extent of the surge deposits on the north flank of the volcano. Streatham village is to the right of the picture. (G. E. Norton. BGS NERC).

EYEWITNESS ACCOUNTS OF 25 JUNE 1997

219

Fig. 9. Looking north across Streatham in 1998, when much of the ash had washed away. Arrow 1 shows the house in which Eyewitness 16 survived the surge; arrow 2 shows the site of the burned wooden house where his wife was killed (A, Fig. 3). Note the house in Figure 11 to right of photograph (photo taken by S. C. Loughlin, BGS © NERC).

Eyewitness 17. [at the margin of the surge-impacted zone]. 'I went up there and saw the mountain all turn black, getting dark all over the place. Getting fire. I run back into me home ... it's a shed, a little house [wooden]. When I was in the house I feel the house getting hot. So me say "What is this?" and I see a lot of smoke. I recollect it was the ash. So it got in me nose, stifling. Me hold on about 4 or 5 minutes, in that time the roof of my little dwelling start to catch fire ... I get the front door open. So when I look outside [looking north], outside just as you see ... it just like nothing ever happened. Outside cool and calm and everything but when I got out of the house and I looked all around I saw all the rest homes all burned down and all the trees them and everything burned flat.'

rolling over and I fell over a bank [and by luck found his son's house]. I go inside the house and it full of ash... [He sheltered there for some time] 'I come out, I stood by my boy's house and watch houses there burning ... then I think I'll go and see what happened to my house. I go and as soon as I reach the top of the hill I see that my house is burned [Fig. 9]. The house burned flat, everything I had gone. We had two cutlasses in the bedroom, so I see two cutlass there and I go to pick them up and as I look around I see that my wife get burned up. 'You can see here I don't get no burns [on his legs] I have long pants [trousers] on but I didn't have nothing on my foot. So it from that I get damaged - if I had on my shoes I wouldn't have got damaged. I lose me toes and spend six months in hospital.'

Dyer's Eyewitness 12. 'When I reach Dyer's corner rain shelter, me feel heat and see stuff going down by Molyneux, stuff was black and running on the road. Smelt something like when you start to boil an egg and it rotten ... Stuff coming down towards me, still could hear nothing, felt no wind ... Me feel the heat all over, me say "see how me foot burn down", me feel like me been in the fire meself so me turn back. Me go back to Lee's. Got a lot of ash, shirt was full of ash, couldn't see the colour of the red shirt. Ash was white first in fine bits like rain ... I could see but not very plain, and me nose felt like it start to burn me when I feel the heat. Walked all the way back to Lee's, and ash was still falling on me.'

Streatham Eyewitness 16 [at the margin of the surge-impacted zone]. 'I was doing some work at the door and me back was turned. And my wife was at the kitchen [a separate outhouse]. Then my friend called out and asked me if I don't see what's happened. Then as soon as I look out the whole place covered up [lofting ash from the pyroclastic surge that detached from flow pulse 2]. Couldn't see at all ... I couldn't see no place at all that I could go. So I fly back in me house to get a searchlight and come out and put on the light, I can't see no way at all. So I was just trying to escape [he ran uphill to the west as a detached surge from flow pulse 3 impacted Streatham] rolling over,

Eyewitness 13. ' . . . and me run back through the bush to Molyneux ... Me go by Molyneux River, and me see fire come up. Fire seem serious, hot, so me turn back and me go by Molyneux ... Me no get no burn, but the ash me warm ... ash fall on me and cover me right down ... me couldn't see me fingernail ... me couldn't see one thing when the ash come over me ... Me went straight to Cork Hill [on foot] ... all around me dark for about one hour.' Harris Eyewitnesses 14. 'Ash come up on the school field and cover the whole of Harris's. Was grey and black, one mighty cloud [lofting ash from flow pulse 2 and detached surge]. I ran up and stood by the school [Fig. 10]. No hear a thing before I saw it. After, I hear hard cracking sound and thundering. Saw lightning. Start to feel hot, hot, hot, like I was in an oven [detached pyroclastic surge from flow pulse 3]. The whole place turn black for about 20 minutes, heavy ash falling ... black like black cloth. It smelt funny, like burning ash, bad ... Smelt strong, a bit like lighting matches smell ... like when people burn a motor tyre. Fire was pitching on the pavement by the school. Fire was pitching in the air as well. The whole of my body felt hot. There was a heavy rushing wind with it. That is what took out some of the house roofs. The building was shaking so I jammed myself on the balcony. The houses on the side of the road next to the ghaut were burnt ... It came down sly and sneaky.

220

S. C. LOUGHLIN ET AL.

Fig. 10. (a) An aerial view, looking east, of Harris village, Harris Lookout and Paradise River valley with the coastal plain behind. The airport runway is in the top left corner of the photograph. Survivors were airlifted from the summit of Harris Lookout on 25 June and on the following days. The prevailing wind was to the west so Harris was badly affected by ash in the aftermath of the pyroclastic flows. Harris school is in the right foreground (see Eyewitnesses 14) (photo taken by G. E. Norton. BGS NERC). (b) Looking south from the Montserrat Volcano Observatory (near Salem) at the rapidly advancing ash clouds. Ash was deposited across much of the western part of the island, especially on Corkhill and Salem (photo taken by W. P. Aspinall).

It travelled faster than a car, No car could escape, The pieces of fire were small and it was there before the place turned black. I think it must be the end of the world. Gas bottles blew and went "pow pow pow pow". Pyroclastic flow take over the whole place like a mighty sea. The heavy winds were like a hurricane.' Eyewitnesses 7. 'I was in Harris in Scott's shop. We heard a boom [probably Mandy's gas station blowing up] but by the time we were out of the shop it was past the Police Station and past the shop [pyroclastic flow 2] ... I went straight to my car ... I just drove like a daredevil up, up on the hill [Harris Lookout].' [The wooden shop was in the singe zone of the detached surge from flow pulse 3 and burned to the ground.] Harris Lookout and Smoky Hill Eyewitnesses 7. [en route from Harris to Smoky Hill]. 'Very high rolling clouds hung over Farms and Trant's. These clouds started

to virtually reverse on reaching close to the sea [due to the strong easterly wind]. The clouds were white, black and red from burning but mostly blackness. It stood up and seemed to roll back. This time I was fearful. There was wind from the sea. It was pitch black for about 15 minutes. Hot ash was raining down and there was lots of toxic gas ... There was no oxygen, you could hardly breath, you have to breath at intervals, it was like blocking your lungs. P. was breathing hard. I felt I was suffocating ,,, There was about 5 seconds of light then it was black again for 5-10 minutes. The atmosphere was still really hot. I decided to turn back to Harris. There was a little more light for 15 seconds and we started crawling back up the track.' Eyewitness 1. 'Ash engulfed Harris after the pyroclastic flow travelled to Farms [pyroclastic flow 2], and then there was a quarter of an hour of darkness. Ash was coming down and across in all directions, mainly from the Farms direction [flow pulses 2 and 3]. We thought it was just ashfall but then we heard gas explosions in the houses. There was thunder and lightning in the clouds but the pyroclastic flow made no sound.'

EYEWITNESS ACCOUNTS OF 25 JUNE 1997

221

Fig. 11. Partially burned building at the margin of the surge-affected area at Streatham (also shown in Fig. 9). (Surge transport direction from lower right to top left.) Note the evident implosion of upper floor window with combustion of the roof above (see text). Survival would have been possible if the window had remained intact. Boarding up the window might have prevented this damage (photo taken by P. D. Cole).

Tuitt's Eyewitness 6. 'While in the yard I saw ash hovering over Tuitt's at about 700ft [213m]. It was all blues and brown. I did not hear anything at that time ... I heard thunder twice northwest of where I was. I said to myself, "This is no thunder from a storm". I stood up a second and watched what was happening in the air and the direction it was travelling in. The cloud at that time was dark. It was covering part of Tuitt's; Bramble village had already been covered, and part of Spanish Point [lofting ash from flow pulse 3] ... All that I saw from Bramble village to Spanish Point was real smoke and fire. Apart from the thunder I heard no sound, felt no shake. There was no sign to tell me anything was happening before I saw the cloud.'

Bramble village Eyewitness 5. 'When I was in there, in the bush, I start to hear the trees go "ch ch ch". Then I look back, and when I look back I see the pyroclastic flow about 100 ft [30m] away from me [in Paradise Ghaut]. Then I tell myself, "It seem like I'm trapped today". I go inside my friends' house with the intent that it was safe. Then I open the window and look out and I see all the houses on the other side [Bethel] burning, coming back at me ... Well, it is lots of smoke with fire spinning like a top, from top to bottom, there is fire spinning something like when you throw gasoline somewhere then throw a match and it's moving very quick. So then I tell myself, "I still have to put my foots on the ground." So I run as far as White's [White's Yard], that is about half a mile [800m], the pyroclastic flow is still speeding, bouncing.'

side door we wouldn't have had any damage in the house. The side door was open and heat blasted in through the door. I was probably 8 ft [c. 1.5m] from her and I didn't feel anything. Just when the blast hit, it filled the whole house full of ash, it was dark and you couldn't see your hand in front of your face ... The blast was so fast ... The surge that went by us ... it was just like a whoosh and that was i t . . . as quick as it came, it began to dissipate ... I believe that it lasted a second ... but in that short time, we had orange trees that were probably 5-6 ft [c. 1.5-1.8m] tall and an inch to an inch and a quarter [c. 2.5-3.2 cm] in diameter that were singed off until they were just little stubs sticking out about a foot [c. 30 cm] high. The ones that were a little bit smaller you couldn't see a trace of them. They were at the side of the house. Everything from the road up was burning. There were some houses to the south that were burning, I noticed that the ones right next to us were not. Anything to the north of us looked like it was on fire. It was mostly the roofs [burning]. The surge did not knock the roofs in, it caught the roof on fire ... The church got burnt [Fig. 6b] ... I could see some smoke up there on the volcano side ... and the fire spread through the whole of the church. Somebody inside the church, with the church closed up, probably would have been perfectly safe at that point and then afterwards have come out [but it was surrounded by surge deposits]. 'The house was three level. There was a garage level, then there was the living quarters that had no windows to the volcano side because it was ground level, against the ground completely on the uphill side [Fig. 6b]. The upper level was completely burned out. When the blast came by we still had the windows open. The windows did stand up to the blast: it busted a few of them, they cracked, but I don't believe any of the glass fell out after that initial blast. It melted the rubber seal, it singed that o f f . . . The aluminium stuff stood up pretty well to the blast. It melted the door knob and the door stood up pretty well.'

Bethel

Trant's

Eyewitnesses 11 [Fig. 6]. 'I was on the roof of the house and I saw it coming down in the valley to the north of us [flow front of pulse 2 in Pea Ghaut] ... Then it seemed like maybe it was slowing a little bit, then another came [flow front of pulse 3]. That's what came down and burned Bethel [Figs 6 and 7]. My guess is it was 80-100mph [c.35-45ms - 1 ] ... That's when I decided we'd better hide in the bedroom, because we had a room there that was pretty well sealed off. Our daughter was burned. If I would have shut the

Eyewitness 3. That first pulse [actually flow pulse 2] followed the ghaut and didn't destroy Trant's [Fig. 4]. The second pulse [flow pulse 3] veered round to the north from the north edge of Farms, and took out the houses in Trant's one by one [Fig. 7].' Eyewitnesses 10. 'We first saw the pyroclastic flow [flow front of pulse 3] as it was starting to enter Trant's. First, a flow went close

222

S. C. LOUGHLIN ET AL.

Fig. 12. (a) A house near Streatham devastated by fires ignited by a pyroclastic surge. The standing trees imply that the dynamic pressure of the surge was relatively low here, but the temperature was high. Note the scattered, intact gas (LPG) cylinders in the right foreground (photo taken by P. D. Cole), (b) The shell of a house in block-and-ash flow deposits on the coastal fan (flow from top right to bottom left) showing the complete combustion of furnishings and other flammables due to the intense heat (photo taken by P. D. Cole).

by the south side of the village, confined to the ghaut but did not touch or damage the village [see fig. 6 in Loughlin el al. 2002]. Another came within less than 1 minute of the first and spread further northwards, this time taking down houses and trees [an additional secondary flow pulse; see fig. 11 in Loughlin et al. 2002]. The flow front looked like it was about 2-3 ft [c.().6-().9m] high with a tapered front. Surge was burning the sides right after the flow front passed. It took about 1 to 1.5 minutes from the point first seen to the end of the destruction. It appeared to sweep houses off their foundations and travelled in pulses or bursts.' Airport Airport staff and three MVO scientists began evacuation of the airport at c. 12:58 (see Appendix 1). They made observations from

the airport, during their evacuation and later from high ground to the north of Tram's. Evewitness 21. 'It was only monitoring the radio that made me know what was going on. The cloud base at the summit was low ... At one o'clock I saw this blanket of ash and hot material just over the other side of Harris [flow pulse 1]. Once I saw that, I knew it something big ... It took time to get to the airport because of the winding valleys [flow pulse 2]. It seemed to speed up somewhere in the Farms area to reach down there quicker than I thought [the river valley has many bends between Harris and Bethel but is straight between Bethel and Farm]. The ash move up to 30000ft [c. 9144m] in less than a minute. It was maybe something around 40000ft [e. 12200m] when I went. But it was something that was real devastating ... boiling as it was going down. It was travelling to me

EYEWITNESS ACCOUNTS OF 25 JUNE 1997

at that time at about 40 mph [c. 1 8 m s - l ] . When the flows broke off at a little point coming down to Pea Ghaut [flow front of pulse 3 escaped the drainage channel at Bethel; Fig. 5] it just blast over that side, it just ramp up. 'I watched it all the way down towards Farrell's Yard and by Bethel. In that area I saw it exploding, I don't know if it was gas bottles or what it was, but it was exploding. When it came down Spanish Point it was moving in all directions. It had no direct path, it was erratic basically.' Eyewitness 22. 'When the siren sounded it had just started at the top of the ghaut [flow pulse 1]; I think that was probably one of the reasons why nobody hesitated ... But after it got beyond Guadeloupe bends [between Harris and Bramble village] it took a very short time to get to Trant's [flow pulse 2]. There was a strong breeze on that day, so the wind was blowing the ash plume back on to Harris hill. Even though the flows were coming towards the airport, the angle of the plume was probably between 30 to 45 degrees because of the wind blowing the plume. It was very dark and ominous with a rolling movement and you could see the pulse-like movement at the base. At one point I got very scared coming up the dirt road because I thought it would roll over the area of the Trant's quarry and catch us before we went to the main road. "When we went up onto the hill, there were actually three or four pulses which we saw coming through Trant's village [flow pulse 3; see fig. 11 in Loughlin et al. 2002]. You could see subsequent explosions, maybe associated with gas propane bottles and many fires from buildings or whatever burning.' Eyewitness 23. 'Well that thing just come down there with no noise at all. It just come down on you and just do what it have to do. You know so at that point a lot of people get killed because they didn't hear nothing ... That thing come down there so fast it's like when you saw the puff of cloud here, it already taken out down there, it already under the bridge. So when you see the other flow it came down there so f a s t . . . You have very little chance to get out of that thing there man. That thing there just destroyed the whole place man. All lightning and thunder. And when it get the gas bottles you could hear the explosions when it went through the houses ... It just run through the houses, I tell you something ... When you look at the surge man, and see that thing just build up, it was something, man. It was really something to see, you know. It's the most beautiful thing to watch but it's so deadly.' Eyewitnesses 19 (MVO staff (A. Clarke, R. Watts & L. Pollard) evacuated through Trant's). 'I first saw the nose of the flow [at Trant's, flow front of pulse 2] ... The flow was following the Farm River; at this time the nose was just upstream of Trant's village ... 'As we were driving up the hill another coignimbrite cloud could be seen behind Quarry Hill. We interpreted this as another pyroclastic flow pulse [flow front of pulse 3]. This coignimbrite cloud was visible approximately 4 minutes after the first. The nose of the initial pulse of the flow was continuing down the Farm River towards the sea ... the plane took off as the initial pulse of the flow had nearly reached the part of the river channel that passes by the end of the runway. 'We drove to a viewpoint higher on the road ... The [third] pulse did not stay confined to the Farm River. Instead, after passing Quarry Hill, it continued to travel north (rather than east), where it left the river channel and spread out, engulfing and setting fire to Trant's village [Fig. 7]. The edge of this flow travelled as far as the Trant's police checkpoint ... 'A third coignimbrite cloud was seen in the same location as the others ... [secondary pulse associated with flow pulse 3; see fig. 11 in Loughlin et al. 2002]. The flow followed a path similar to that of the second, spreading to the north and this time completely burying Trant's village. Only burning telegraph poles and the bus shelter protruded above the deposits.'

223

Eyewitness 29. 'It's like burning hell down there, all the buildings burning and everything down there's like a burning hell. If anybody had told me this would happen to me I would never believe it right.' Observations by search-and-rescue staff following dome collapse Eyewitnesses 30-35 belonged to the Montserrat Police Force and the Montserrat Defence Force. They entered the exclusion zone after the dome collapse and so are not included in Table 1 or Figure 3. 25 June: Dyers Eyewitness 30. 'Went to Dyers almost 2 hours after eruption. Reached Dyers-Molineux road junction and had to stop because of ash on the road. I got out of vehicle and took a green small stick and pushed it through the material on the road - it burned immediately. This was at the edge of the material [surge-derived pyroclastic flow] where it went in about 3 inches [c. 8 cm]. I went up further and pushed it into the flow. It did not come back out and it burned completely. We only felt heat when we reached near to it.' 26 June: Windy Hill Eyewitness 30. 'There was not much material deposited on the ground, no stones, only ash. I can't be sure but there was about 2-3 inches [c. 5-8 cm] of ash, bare earth could be seen on little high points because it is a windy area.' 26 June: Streatham Eyewitness 31 [in the most badly affected part of south Streatham]. 'Went into the devastated area at about 15:00 on 26 June 1997 in the French chopper [helicopter]. We landed at Streatham. The whole villages of Streatham and Windy Hill was devastated with pyroclastic flow. Some houses had the walls and everything destroyed. Some were still standing but the walls were all cracked up - they would just fall down if pushed hard enough. Aluminium window shutters were melted and galvanized [steel] sheets were blackened and twisted. We could not see any wood. Vehicles were burned. Utility poles were burnt. All the insulation was gone [from the wires], leaving the copper bare like after a storm. Ash was still hot and white. We couldn't see any road [surface], the ash was like a blanket, it was over 10 inches [c. 25 cm] thick in places. All the ash was very fine with no rocks in it, and there was no vegetation left at all. Vegetable beds were bare of plants but still in original form [furrowed soil]. The ground was baked hard like clay.' Eyewitness 32 [in north part of Streatham at margin of surge deposits]. 'We went to a wooden house ... it was burnt flat [A in Fig. 3, and Figs 8, 9]. There was a lot of ash in the area of the house: it could cover my boots in some areas. There were no stones, only very fine ash.' Eyewitness 33. 'Farms was [buried and] unrecognizable and blue smoke was rising from several areas there. From Farms around to the Belham Valley there was burning (blue smoke) in the valleys and around the edges of the flows. In the Windy Hill area there was burning of grass and bush which extended up the hill about 100 yards [c. 91 m] immediately north from the heavy ash deposits. Timber and wood was still burning under the [surge] deposits. In Windy Hill there was heavy ash [almost] up to the road junction. Further north [in the singe zone] the houses were singed or blasted with the heat... If the three people who died had got about 150 yards further [B in Fig. 3], they would be alive. East of the flow [surge deposit] the area of burnt or singed grass was reduced to 25m or so.'

224

S. C. LOUGHLIN ET AL.

27 June: S treat ham Eyewitness 34. There was only light ash in Molyneux but the trees were burned. There were no big rocks in the Streatham area. Some houses looked like a fire just burned them then left the frame standing; some houses were destroyed completely. We felt the heat through the soles of our shoes when walking on the ash and had to move away from some areas or retrace our steps. We could feel heat from all around.'

28 June: Streatham Eyewitness 35 fin south Streatham]. The buildings were burnt, some of the wall structures were breaking, some were flat. There was twisted galvanized metal. and dead animals all over the place ... His house [victim C, Fig. 3] looked like it was bombed. The glass window [panes] had melted like plastic, whereas the windows at Windy Hill looked more like they were blasted out. Even metal looked bent at Streatham. The door of the house in Windy Hill where three bodies were found [victims B, Fig. 3] was wood and was charred, but it was still standing and intact [they would probably have survived if they had managed to get inside]. There was total devastation in Streatham. China ornaments were melted, it was like the whole village was bombed. We saw a lamp pole burning on the ground in Streatham. We could not see the original ground surface in Streatham ... there was an acrid smell. Houses close to the edge of the flow [about 1O m] were also just as damaged as houses in the flow. All houses where ash reached were totally consumed [by fire]. The fine ash was fluidized when walked upon.'

Search and rescue On 25 June 1997, the MVO helicopter began picking up people inside the Exclusion Zone at Trant's, within 15 minutes of the onset of the dome collapse. Video footage taken at the time, from the helicopter, shows the area impacted by the pyroclastic flows and surges engulfed in smoke and resuspended ash, with almost every building burning to some degree. Because of the strong easterly wind that blew the smoke and ash over the impacted area (for tephra description see Bonadonna et al. 2002), access by helicopter was possible only in areas on the eastern margin of the flow deposit. At 14:00, the helicopter picked up survivors at Spanish Point. At 14:45 about 12-15 survivors were sighted at the top of Harris Lookout, an elevated and undamaged area surrounded lower down to the south and east by pyroclastic deposits (Figs 1 and l0a). These uninjured people were flown out three or four at a time and included a 90-year-old lady who had refused to leave her home. There were few convenient landing places for the helicopter and it had to hover to rescue her. As the ground was covered in fine ash and visibility reduced by the pall of fine ash and smoke, such helicopter flights were risky. At about 16:00 the helicopter set down for houses to be searched at the margin of the flow in Bethel. The heat radiating from the deposit 10m away could easily be felt and the air was filled with the pungent smell of sulphur dioxide outgassed from tephra even though the ground had only a light covering of ash. Ash had also reached inside houses that proved to be empty apart from furniture. Indications are that sulphur dioxide and possibly other gases in the air had not reached lethal levels since a goat tethered in one of the gardens quite close to the flow margin was alive. Access to the whole area was possible by helicopter on 26 June, when fires were no longer burning and levels of smoke and resuspended ash were lower. Of the 27 people flown out of the Exclusion Zone by helicopter in the following five days, three had been seriously injured. Six were people who ignored warnings and had returned to the area after 25 June to check their homes. Others rescued included people too scared to come out of their homes on 25 June itself, and some who still refused to move, even later in the

immediate aftermath. Several of these people were elderly and/or confused. A lady of 86 years of age was admitted to hospital after being rescued from her home at Harris on 28 June. Deaths and injuries Seven people were killed by the surge in the Streatham and Windy Hill area (Fig. 3). Five intact bodies were retrieved during the morning of 26 June 1997 and another was recovered three days later. The last remaining body could not be reached because of the intense heat of the deposits. Six of the victims were close to the edge of the area impacted by the surges and were found outside houses, the individuals evidently having been caught whilst attempting to seek shelter. The bodies showed the effects of exposure of brief duration to extreme heat at the time of death, such as a combination of superficial (non-life-threatening) traumas with partial and full thickness burns of the skin accompanied by evidence of carbonization of the fingers. The limbs were flexed into the 'pugilistic attitude', as adopted by burns victims in general, with the arms drawn up against the body as a result of the heat (>200 C) coagulating the muscles (Baxter 1990). The remnants of burnt underwear adhering to the bodies, and the absence of hair, indicate that the clothes and hair had caught fire when the individuals were engulfed by the pyroclastic surge. Death would have been near-instantaneous. The surge would have been slow moving in this area and temperatures measured in the surge-derived flow deposits in Dyer's Valley suggest that the peak temperature of the surge was at least 400cC (Druitt et al. 2002). The dismembered remains of two bodies were recovered some weeks later from the pyroclastic flow deposits near Trant's, where they had been deposited at a flow front. Most of the remaining missing persons are thought to have been in the village of Farm, which was buried by deposits several metres thick, and were never recovered. Four of the seven injured persons walked out of the Exclusion Zone and were taken to the hospital by road. One of these drove out of the pyroclastic surge at Dyer's by car. with the treads of his tyres on fire. Inside the vehicle, feeling the heat of the enveloping surge, he turned on the ventilation fan, only to receive hot ash in his face. He received partial thickness burns to his feet when he eventually stepped out onto hot ash. At Streatham. one man survived (Eyewitness 16) by running uphill away from his home (Fig. 9). In shock, he too walked on hot deposits in his bare feet and as a result his toes had later to be amputated. His wife, who remained in their home, was killed (Figs 3 and 9). The two other people who walked to safety also had severe (partial and full thickness) burns to the feet from walking on the hot ash deposits to escape. The most severely burned survivor (Eyewitness 17) was rescued by helicopter on 26 June. He was found wandering on Windy Hill with partial thickness burns to the soles of his feet, right arm and shoulders, and superficial burns to the face, nostrils and around the mouth; he was diagnosed in the hospital as having sustained an inhalation injury. His initial survival was attributed to the fact that he was inside his wooden house at the time of the surge incursion, but as the roof caught fire and collapsed he was badly burned and forced outside where he had to walk on hot deposits. The 4-year-old daughter of the family who survived at Bethel (Eyewitness 11) received a superficial burn around her right wrist. Another victim was an inquisitive child who. several hours after the dome collapse, ventured onto the surge-derived flow deposits in Dyer's River Valley and sustained a superficial burn. Discussion of environmental impact

Flow and surge velocities and dynamic overpressures Valentine (1998) drew attention to the fact that blast damage from nuclear explosions is similar to damage caused by the dynamic

EYEWITNESS ACCOUNTS OF 25 JUNE 1997

overpressure of pyroclastic flows and surges. The human and environmental impacts of flows and surges are determined by their dynamic pressure, temperature and concentration of particles (Baxter et aL 1998). The ability to topple trees depends on dynamic pressure, Pdyn (Valentine 1998):

225

that small flows travelled along tracks and ditches at high angles to the main direction of travel. These pathways invariably had a steeper downhill slope and a low-friction surface so gravity was the driving force.

Damage to infrastructure where is the flow density and v is the flow velocity. Loughlin et al (2002) used the extent of tree damage to ascertain dynamic overpressure of the pyroclastic surges, using data collected by Valentine (1998). There was almost total blowdown of trees in proximal areas (upper flanks of Mosquito Ghaut) by pyroclastic surges, implying dynamic pressures of 2000-2400 Pa. Assuming a surge cloud density of about 1.6 k g m - 3 , the maximum velocities of the surges at these localities could have been up to 50-55 m s - 1 (Loughlin et aL 2002). Trees were still standing at a distance of 2 km from the dome, implying that the dynamic overpressure of the surges decreased rapidly over this distance. This was caused by a decrease in velocity as a result of slope reduction, and by a decrease in density as a result of rapid suspended-load fallout. Flow pulse 2 had an average velocity of 2 0 m s - 1 between the lava dome and Bramble village (4.7km), but beyond this point it slowed considerably due to the reduced gradient. The average velocity of flow pulse 3 between the lava dome and Trant's village (5.8km) was about 2 2 m s - 1 . Both pulses were similar in volume, but flow pulse 3 travelled faster, possibly because the substrate of hot, recently emplaced pyroclastic flow deposits aided transport by smoothing the channel, or because it was aided by a high juvenile gas content released by attrition during transport (Loughlin et al 2002). An empty water tank was moved about 250m by detached pyroclastic surges from fields north of the main road near Farrell's Yard to the lower slopes of Windy Hill (fig. 15 in Loughlin et al 2002). The tank was crumpled on all sides, consistent with being rolled or saltated as part of the surge bed load. At Mount Unzen, Japan, a car was similarly transported 120m by the dynamic pressure of the pyroclastic surge on 3 June 1991; it was severely crumpled and left behind it a trail of small parts (Fujii & Nakada 1999). The environmental impacts on Windy Hill were mostly attributable to the heat rather than the dynamic pressure of the surge, and this is consistent with the reduced velocity and density of the surge as it climbed the hill. Houses and outbuildings began burning as soon as they were engulfed by the surge.

Effects of topography Deposits showed that the pyroclastic surges were very highly responsive to even the slightest changes in topography. Medial to distal surge deposits were all channelled along valleys, shallow topographic lows or ditches. This behaviour probably saved the family at Bethel from worse injury. Their house stands on an almost imperceptible rise causing the surge to pass to one side of them in a very slight topographic low. A similar minor change in topography saved the survivors at Farrell's Yard. There, the northern flanks of the volcano slope down to the cross-island road that follows the break of slope at the foot of the Centre Hills (Fig. 2). The Farrell's Yard and Riley's Yard area is a watershed with drainages going west near Riley's Yard and NE near Farrell's Yard. The surge that detached from flow pulse 2 swept across the northern flank of the volcano but it divided, part flowing west and part flowing NE, and the highest point on the road was not hit (see fig. 5b in Loughlin et al 2002). The eyewitness observations suggest that the surge reached the road and ran out of momentum. The ash in the cloud deposited rapidly on the road and this material then flowed down the road to the east, at a high angle to the original transport direction. This behaviour appears to be that of a surgederived pyroclastic flow (see Druitt et al 2002). An eyewitness farther west, at Dyer's, also reported ash running along the road. Deposits at the distal end of surges on the coastal plain also show

We found no evidence that the reported explosions were caused by the sudden rupture of liquefied petroleum gas (LPG) bottles but instead speculate that the high temperatures caused the houses with their flammable contents to catch fire instantly and combust, as in a deflagration. Most of the houses had been left furnished. Several intact LPG bottles were found at the margins of the flow deposits (Fig. 12), although these may have been empty before the event. A detailed study of buildings damaged by the 26 December 1997 event showed that hot ash from a surge penetrated the buildings with walls of masonry or reinforced concrete through their windows, causing furnishings to ignite into a fire that could extend throughout the house. A single room in a house at Streatham, facing towards the oncoming surge, ignited in this way on 25 June (Fig. 11). Simpler wooden buildings rapidly burned to the ground, either from interior fires or by combustion of their outer walls by the hot ground deposits. Combustion of the roofs of buildings (rafters, boards) from the direct heat of a moving pyroclastic surge was a much less common finding. In the Streatham area the roofs of sturdy structures tended to be burnt out only on the side impacted by the surge (Fig. 11). Buildings struck by block-and-ash flows suffered complete combustion of flammables and only the damaged shells of structures occasionally survived (Fig. 12). An eyewitness reported on the cracking of windows by the heat rather than the force of the surge, but the loss of a single window resulted in hot ash getting inside the house and igniting the furniture and then the wooden roof immediately above (Fig. 11). The explosive destruction of the petrol filling station was remarked upon by several witnesses and highlights the hazard of these and other stores of flammable materials in populated areas if struck by a flow or, as in this case, the margin of a surge. Survival near the margins of a surge-affected area is normally possible in a solid building, but secondary explosions could jeopardize this. The most striking feature of the scene was the numerous fires as buildings and vegetation were rapidly consumed in the heat of the block-and-ash flows and surges. We speculate that the atmosphere inside a pyroclastic surge would be devoid of oxygen, except at the margins where mixing with air would take place. Combustible materials and vegetation would initially ignite, but then might go out as these were engulfed by the pyroclastic surge, only to reignite immediately as the rapidly convecting cloud dissipated and drew air back in. Smoke and its toxic constituents presented a hazard to those survivors at the margins of the devastated area and also to rescuers.

Injuries to survivors Severe burns to the feet occurred in six of the seven injured survivors as a result of walking on ash deposits only about 1-2 cm deep. The seriousness of this type of injury, which can lead to loss of toes or amputation of the feet, can be overlooked by medical attendants, who may be unaware that the contact temperature can be two or three times higher than that of boiling water. Surge deposits remained hot for several days where ash had accumulated more than a few centimetres deep, e.g. by the side of buildings or in hollows. Thick surge and block-and-ash flow deposits remained hot and fluidized for many months, often in irregular patches indistinguishable from otherwise firm deposits. At volcanoes where dome-collapse pyroclastic flows may occur, this hazard should be considered and rescue workers need to be made aware of the extreme danger presented by even very thin ash deposits when they are extremely hot.

226

S. C. L O U G H L I N ET AL.

Judging by the strong smell of sulphur dioxide in the impacted area soon after emplacement, this gas and hydrogen chloride (the most abundant gases emitted at the crater apart from H 2 O) were outgassed from the particles. Hydrogen sulphide is considered to have been a minor constituent of the emissions, but its characteristic rotten egg smell is not always discernible at volcanoes when mixed with other acid gases.

Survivors' comments on hazards and risks

when flows started coming over Tuitt's. I started looking for a place, but found nowhere. I'd been trying to secure a place for over a month'.

Eyewitness 16. They tell me that I must move and I wouldn't go by what they said because I have my ground and my stock to look after. I moved twice or three times. So I said "Well I'm not moving [again] because you're walking and walking every morning" [back to his fields in the Exclusion Zone] ... I had my work to do so I say well I'm not going to move 'til when things happen."

Reasons for living in the Exclusion Zone An estimated 15 people were living and sleeping in the Exclusion Zone before the collapse on 25 June, and nine of these were interviewed. All of those interviewed were aware that they were taking a risk by entering the Exclusion Zone, but they considered it essential and unavoidable for social and/or economic reasons. There was a serious shortage of housing in the north (Clay et al 1999a). Many had moved north into shelters or other emergency accommodation at least once, but, for various reasons, had returned to their homes in the Exclusion Zone. Some people, especially the elderly, refused to move to the overcrowded shelters, because of the lack of privacy, overcrowding, poor sanitation and because they were aware that they would be there for a long time while new housing was planned and built. The conditions and length of time people had to endure living in shelters was considered 'unacceptable in terms of British and industrial country standards of social well-being' (Clay et ai 1999a, b). Farmers were unwilling to leave their livestock. The judgement of others with mental health problems, young and old, may have been clouded by their illness. Eyewitnesses 15. (This early interview was not recorded on cassette; the interviewers took notes and this is a summary of those notes.) This elderly couple in Harris wanted to move as they realized they were at risk, but could not find anywhere to go. They did not want to live in a shelter, having heard stories about bad conditions there, although they had packed their car several times and been on the verge of moving. Having watched the volcano for 14 months, they thought they would be relatively safe on higher ground, and that the flows would remain channelled in the valleys. Eyewitness 12. 'Me stay because me wanted to stay, not because me couldn't move. Me say me nah believe what they say, me had to see meself.'

Eyewitnesses 11. 'We were going to be leaving in just a few weeks. Before this we had checked around in the north and there was absolutely no place to go. There was no place to rent or anything at that time ... That is why we were still there.' Eyewitness 14. The reason I went back to Harris was to get some peace. I was going to return to the shelter, because I notice the volcano started to look bad. I spent five months in England but came back because my brother said nothing was going to happen.' Eyewitness 3. 'I was living in Trant's because I had no other place to go. It was a combination of lack of provision for my cattle and housing further north ... My father looked after our approximately 35 cattle, but since he died in January I have had to do that. Just before January I lived in a rented place in the north, but when the owners wanted it back, me and my friends from Trant's were forced to move back. It's very hard to find places to rent, and rent is very high, particularly since the crisis. I needed a place in the north so that I could tend to my cattle in the morning for an hour, drive around the north road and still get to work by 8 am in Salem. From

Reasons for visiting the Exclusion Zone It is estimated that at least 65 people entered the Exclusion Zone during the morning of 25 June. Their reasons were varied, but included feeding and caring for animals and seeking peace, quiet and privacy. The people interviewed were aware of the heightened activity of the volcano in the preceding days. Many visitors checked radio updates in the morning before they went into the Exclusion Zone. According to official lists, nine people were killed in Farm (the bodies were not recovered because of the thickness of deposits) and. apparently, most of these were visiting rather than living in the Exclusion Zone. Of those who regularly returned to the Exclusion Zone, some were farmers who produced crops to feed evacuees in the north of the island, believing that they were helping their country in crisis. Pressure on officials to allow general access for farming and recovery of belongings was strong. Despite roadblocks, access to the Exclusion Zone was easy for those who knew paths and tracks through the bush.

Evewitness 2. (This eurlv interview was not recorded on cassette; the interviewers took notes and this is a summary of those notes.) A teacher at St John's kept her school books at her house in Trant's because there was no room anywhere else. She stressed that she rarely went back to the house and did not usually stay for long, because it was in the Exclusion Zone. However, on this day she had schoolwork to do and she required space and peace and quiet so she decided to take the risk.

Eyewitness 5. 'I only went to feed the animals. When police stopped people I used to say it is foolishness, but the Government was doing the right thing. I used to take the chance and walked on the shore to get home away from the police. I would get there by taking a ride close to checkpoint, then I walked to the shore and got a ride on the next side after the checkpoint.'

Evewitness 6. at Tuitt's.'

'At 6.30 am I went to look after some animals

Eyewitnesses 9. 'Well, livelihood, that is why we were there. We don't have much farming or much farmers because with the relocation a lot of people had left . .. We had quite a bit of agriculture to produce there and we used to supply the shelters with food. So most times when they would not let in other farmers they would let us in because we usually supplied the shelters.'

Perception of risk and response to warning By June 1997, the eruption had been ongoing for two years and some of those who had lived for prolonged periods near the volcano tended to use their own experience rather than official advice as a guide to future activity. Pyroclastic flows had gradually increased

EYEWITNESS ACCOUNTS OF 25 JUNE 1997

in their runouts over this period, and they had consistently been confined to valleys. Furthermore, cycles of activity had developed and were recognized. Persons familiar with this activity expected the pattern to continue and were unwilling to believe that there could be a sudden dome collapse considerably bigger than anything they had seen before. The possibility of a detached surge capable of surmounting topography was foreseen by scientists and accounted for in the risk management maps. However, some people remaining in Harris village (Zone A) believed that they would be relatively safe, even from surges, on high ground. Similarly, most farmers had a contingency plan to run or drive up Windy Hill when they saw or heard something alarming. Eyewitness 25. 'Most farmers used the patterns in the activity to plan their farming. They would go in during a lull and aim to finish before activity began again [see section on Precursory Activity for summary of cyclicity]. People think they can guess by patterns, that they had enough warning of the dangers and the activity. If there had been no market for their products, then they would not have been there.' Eyewitness 27. 'I was living back in St Peter's by that stage but I used to go back to Farms. I had a contingency plan that if anything happened I would go up to Windy Hill.' Eyewitness 2. 'At 1pm I looked at the clock and got up to get lunch. I heard a siren and opened the back door but saw nothing, it was just overcast. I'd tried to get the volcano update on the radio at 12.30 pm but there was nothing. I couldn't get FM at Trants. I knew Wednesday was test day so I assumed they were testing earlier than usual. I lit the stove and put the water on to boil ... There was no sound outside, just the siren ... That morning I'd prayed to God to save me from pyroclastic flows and surges - and he did.' Eyewitness 5. 'I heard a siren but paid no attention because it usually goes off on a Wednesday.' Eyewitness 3. 'I expected the flows to reach further in stages, not all the way to the airport at once. The reports did say it could come far. I had my own ''buffer zone" and was planning to move out once the flows reached Farm's bridge. What I was thinking was far-fetched from what it [the volcano] was really capable of. I had my own contingency plan - the vehicles were parked facing outwards. On the 25th when I saw the flow I had time to let my four beloved dogs in the yard loose, but in the back of my mind I thought it wouldn't go that far. They died. I was more naive than anything else.' Eyewitness 6. 'They [the scientists] said the truth but some people have to see to believe.' Eyewitness 12. 'Me never knew this thing would be so desperate until that very day; me never knew it so serious. I used to see it from where I live every day, but I never knew it were so dangerous. Got me house, clothes, sheep, everything. From the beginning me never move ... me say "Jesus Christ know me nah move" ... me swear to God me never move ... no matter what the MVO say, me would not move, not even if we saw pictures and people said all they could.' Eyewitness 13. 'Me got no time with radio ... me no listen to no radio. Me know it was dangerous to go to Streatham, but me

227

go there every day to look after the cattle. Me know the volcano; the volcano could kill me, but me had to look after me cattle ... if me had some other place to put cattle, me no been go back to Streatham.' Eyewitness 22. 'Some of these people who were working on the flanks of the volcano ... stopped working up there because of the level of danger ... based on the scientific information that was available to everyone ... Just as I thought it was my duty for my country to be at the airport, they thought it was their duty ... It was said with the airport, if an alternative mode of transport was found we would not have operated that airport; if farming areas had been provided for these farmers they would not have continued to farm on the flanks of the volcano.' Eyewitness 23. 'But I did not know the activity was so high that day. If I had known I would have stayed at the airport ... not everybody is going to listen [to the radio updates]. Most of the people normally listen to the siren and normally they check the siren during Wednesday time. So the siren went on and people going to think now it a normal check.' Perceptions regarding the scientists The MVO scientists expended considerable effort in public education and communication. They presented information and updates to the authorities and public on a daily basis that was generally appreciated and undoubtedly contributed to understanding and compliance with official advice (Clay et al. 1999b; Aspinall et al. 2002; Kokelaar 2002). However, the inability of scientists to predict the time and scale of specific phenomena frustrated some people, possibly causing them to make, and act upon, their own assessments. Eyewitness 25. 'Scientists were on the radio giving warnings on the 24 June and 25 June. They did their job. But their mistake was they never committed themselves to a time, so they seemed to downplay things because they couldn't predict. If they had made a prediction and got it right, it would have made so much difference.' Eyewitness 7. 'At one point I thought that there was too much guesswork, but then I heard the start of a radio programme [involving MVO staff] and I understood that some things couldn't be understood and we have to monitor and analyse.' Eyewitness 14. 'Me used to listen to the radio report every day and read about volcanoes. I lived in Harris ... knew that volcanoes were bad ... I believed what the scientists were saying, but they don't know when it was going to blow.' Eyewitness 21. 'Once you could understand the information, it was something you had to take heed of and listen to.'

Future mitigation and volcanic emergency management The successive evacuations and relocation of nine-tenths of Montserrat's pre-crisis population to emergency accommodation, including crowded shelters, created severe social and economic hardship (Clay et al. \999a; Kokelaar 2002). This had a major influence on risk-taking by the public prior to 25 June 1997. The comments of survivors show that after two years of disruption and displacement, the pressures of day-to-day living seemed more pressing than volcanic hazards which, although forecast, had failed to materialize

228

S. C. LOUGHLIN ET AL.

on a life-threatening scale. An understanding of volcanic hazards together with a full appreciation of the difficulties of prediction and limitations of scientific monitoring is crucial for any individual considering whether or not to undertake risk. These topics were discussed regularly on radio, but not everyone will listen to the radio and, of those that do, not all will understand.

a high-risk but expedient route, and only when scientists stopped this practice did most other people also desist from using the road. On the morning of 25 June, the rapid retreat of two MVO scientists from their observation position at White's Yard (Zone B). back to the airport, caused several individuals to follow their lead, undoubtedly saving lives.

Communication

Audible warnings

As the crisis deepened, many different people attempted to explain the risks to those still living in the Exclusion Zone, including government information officials, police, scientists, religious leaders, teachers, families and friends (see Aspinall et al. (2002) for a discussion of the MVO's role at this time). This advice was often on a one-to-one basis (possible on a small island, but not in bigger populations), but, despite this, some inhabitants stayed where they were. Some people looked to community leaders for guidance and the importance of closely involving religious leaders and other influential local figures in education roles and decision-making cannot be overstated. In Montserrat one respected and influential village elder remained a sceptic until 25 June 1997, and his was one of the most highly populated villages in the Exclusion Zone. A proportion of those living in the Exclusion Zone seemed to have been amongst the more vulnerable in society, being elderly, infirm or for other reasons unable to comprehend the danger.

The sirens at Plymouth and the airport (by then, intended to warn authorized entrants to the Exclusion Zone) could not be heard by those who were at Streatham, Windy Hill, Bethel, Tuitt's or Harris. Officially these areas were evacuated so no sirens were provided there. Despite this, people still expected some kind of warning before a major escalation in activity. Most people expected the volcano to provide an audible warning, which it did not. Only those who saw the flows early had adequate time to escape or take evasive action. The flows and surges were apparently silent except for explosions as houses were engulfed and the sound of vegetation burning in the singe zone. Sounds are muffled by particles inside pyroclastic flows and surges, but strong easterly winds may also have affected the propagation of sound. During the Mount St Helens eruption in 1980, sound was heard many kilometres away but not everywhere close to the volcano (Dewey 1985), and similar effects were observed in the 1979 eruption of the Soufriere of St Vincent, when explosions were heard hundreds of kilometres away but not at the foot of the mountain (W. P. Aspinall. unpublished information). Some of those who could hear the airport siren ignored it, assuming it was an early test (siren testing took place at 15:00 every Wednesday). A few people inside the Exclusion Zone had unrealistic expectations that the MVO helicopter could provide them with a warning and rescue them if necessary. Many people simply entrusted their well-being to God.

Public pressure Pressure on officials to grant public access to an Exclusion Zone will always be strong, particularly if the volcanic activity appears to be low. On Montserrat, many people wanted to return to their homes or businesses to retrieve belongings and stock, and determined individuals could always find ways of entering the Exclusion Zone. At Mount St Helens in 1980, people also found numerous ways of avoiding roadblocks (e.g. Saarinen & Sell 1985). Fifty-eight persons died at Mount St Helens on Sunday 18 May 1980, but it could have been many more. As a result of public pressure on officials, people were to be allowed into the Exclusion Zone later that day to gather belongings from their homes. Business and private companies can also exert political pressure. Loggers would have been officially at work in the Exclusion Zone at Mount St Helens on a weekday and the death toll could have been in the hundreds (Saarinen & Sell 1985). Impressive volcanic activity also attracts people into the Exclusion Zone, especially the media. On 3 June 1991 at Unzen Volcano, Japan, a pyroclastic surge detached from a blockand-ash flow and killed 43 people in the Exclusion Zone. This group included mainly media people and volcanologists who were intent on taking photographs of the pyroclastic flow activity.

Actions of officials and scientists Crisis management officials should be aware that proximity to 'official workers' might be taken by some people to imply a reduction in risk. On 25 June 1997, sufficient warning was given to official workers in Zone C (e.g. the airport) to allow them to evacuate in time (Appendix 1), but this relied upon the presence of an MVO scientist in radio communication with the MVO, as well as rehearsed evacuation plans. Despite these preparations, staff and members of the public at the airport were still at considerable risk (e.g. Clay et al. 1999a, b; Kokelaar 2002). Some people at the airport drew comfort from the presence of a scientist, and several people visiting Trant's village (near the airport in Zone B) also apparently drew comfort from this proximity. In mid-June 1997, some people were clearly following the scientists' example rather than the official advice. MVO scientists with radio communication used the main road across the island (within the Exclusion Zone) on a regular basis until mid-June. It was

Secondary hazards Several people survived in buildings at the margins of surge zones, but subsequent fires then posed a serious hazard. Strong winds prevailing at the time caused fires to spread to adjacent buildings downwind. Limited access by search-and-rescue personnel to the eastern margins of the affected area was only possible owing to the easterly wind blowing the smoke and fine ash westwards. Casualties from the 3 June 1991 surges at Unzen could not be flown from the local hospital because of the residual ash in the air. Roads that were congested by anxious citizens on the move had to be used instead (Shimabara Onsen Hospital 1993). In densely populated areas, separate fires could rapidly coalesce and lead to superfires analogous to the worst-case scenarios in saturation incendiary bombing or a nuclear strike in modern warfare (Solomon & Marston 1986). The conflagration of St Pierre in the 1902 eruption of Mont Pelee was a comparable event (Anderson & Flett 1903: Baxter 1990). and this parallel with the thermal consequences of nuclear explosions should be considered along with comparisons of blast damage (see Valentine 1998).

Television and radio Educational videos (e.g. lAVCEI's Understanding Volcanic Hazards and Reducing Volcanic Risk) were played on local-access television early on in the crisis, but this could not be maintained due to problems with the television service. Repeated television coverage of the activity throughout the crisis may have curbed the curiosity that drove some people to enter the Exclusion Zone, and educational video footage may have helped people to understand hazards they had not yet seen (e.g. detached pyroclastic surges). Even where maintenance of local-access television would require considerable expenditure, it could constitute a valuable emergency management

EYEWITNESS ACCOUNTS OF 25 JUNE 1997

tool. Montserrat benefited during the emergency from the exceptionally good local radio broadcasts (Station ZJB), but visual images of the impacts of flows and surges might have persuaded a few more people not to take unnecessary risks.

Conclusions Despite the efforts of scientists and officials, the warnings and the police checkpoints, some people and groups still entered, or resided in, the Exclusion Zone, thus exposing themselves to considerable risk. This was mainly due to the severe social and economic difficulties caused by the protracted crisis. Advance planning by government and the community as a whole would have helped to alleviate these problems. The volcanic activity had increased slowly over two years and people working daily on the volcano's flanks had become familiar with the steadily increasing runout of topographically confined dome-collapse pyroclastic flows as well as the cyclic behaviour. This familiarity led people to use their own experience to make decisions rather than take official advice. Despite warnings, they could not envisage a sudden dome collapse far worse than any they had seen before. This tragedy could be used as an example at future eruptions where there is a danger of complacency as a result of a gradual build-up of activity. Despite advice to the contrary, some individuals believed that the volcano itself would provide sufficient warning (audible or visual) of a major dome collapse and that their ability to escape was somehow within their control. Public education at future volcanic crises should focus on countering these falsely held perceptions, perhaps using Montserrat's tragedy as an example. It is possible to survive at the margins of a pyroclastic surge, particularly in a sturdy building resistant to infiltration by the hot ash, but a survivor will still have additional, secondary hazards to overcome. Fire, heat, smoke, hot deposits and re-suspension of fine ash could all be problems. If burn victims at Montserrat had been wearing strong shoes and full body cover, some of the deep burn injuries could have been avoided. People who survived surges were typically indoors. The best chance for survival is to be in an interior room, to keep doors and windows shut and to be at the opposite side of the house to the impact. The family at Bethel was fortunate to be in a three-storey house built into the volcano slope such that only the upper floor was badly damaged. The lack of visibility and the abundance of ash in the air often prevent helicopters being used for rescue purposes in the immediate aftermath of an event, and hot deposits can impede the progress of search-and-rescue workers on foot. This point should be clearly made to the public. Strong prevailing winds caused fires to spread to adjacent buildings downwind, and singed zones were extensive on the downwind side of surge deposits. In densely built-up areas superfires are likely to occur and, where appropriate, this danger with urban areas should be taken into account by emergency planners concerned with future volcanic crises. The involvement of community leaders, religious leaders and other influential local figures in public education is essential. If emergency action is necessary, many people will look to them for assistance and guidance. On Montserrat, many of the people living in the Exclusion Zone were from vulnerable groups such as the elderly and infirm. At future crises such people would respond best to advice from people they know. Videos of volcano hazards and their impacts (e.g. IAVCEI; Lea 1996-1999; Lea & Sparks 2000) are extremely effective educational tools that should be used to full advantage in areas at risk from volcanic activity. Detachment of pyroclastic surges is a hazard that requires special attention and video records of this eruption (by the MVO) may be useful to illustrate their potential impact. Although public showings of videos are essential if there is no widely available television service, local-access television is undoubtedly the most

229

effective medium, especially for large populations. Aerial footage of damage in Exclusion Zones may prevent people entering out of curiosity during protracted crises. Crisis management officials should be aware that proximity to areas where 'official workers' and scientists operate in a crisis might be taken by some people to imply that a generally acceptable level of risk exists. Volcanologists should also be aware that 'actions speak louder than words': their routine presence within an Exclusion Zone may suggest to some people watching that it is relatively safe to be there. That there is usually special back-up available for this work (e.g. radio links, scanners, helicopter) should be stressed to the public so that people tempted to enter the Exclusion Zone illegally or without authorization do not have false perceptions about the true level of risk exposure. The Montserrat eruption constituted a protracted and dangerous emergency on a small island. In contrast to some volcanic crises elsewhere, it was unavoidable that many of the public would come into close and frequent contact with scientists monitoring the activity. In this case, as exemplified by the views expressed by those members of the public most directly involved in the events of 25 June 1997, it is clear that this interaction was simultaneously both beneficial and detrimental to total public safety, as the events and circumstances illustrate. For volcanologists, these experiences provide further valuable contributions to the debate about how to best meet the challenges and responsibilities of providing appropriate scientific advice in a dangerous eruption. We sincerely thank all the eyewitnesses for giving accounts of this tragic event. In addition we are grateful to those people who helped us to locate survivors and provided additional information, especially the staff at the Emergency Operations Centre, Montserrat, M. McEvoy, D. Greenaway and D. Lea. This manuscript benefited from the critical comments of R. B. Waitt, B. Dade and R. Watts. S.L. and A.D.M. publish with permission of the Director, British Geological Survey (NERC).

Appendix 1 Emergency actions on 25 June 1997 07:00 to 08:00 LT. Heightened concerns about possible pyroclastic flow activity. MVO observer at airport alerted. Police HQ contacted; advised to keep checkpoint to Plymouth closed for the time being (it later transpired that essential services were already working in Plymouth, without the knowledge of the duty scientist). MVO observers deployed at Windy Hill to observe small pyroclastic flows in Mosquito Ghaut. 08:20. Police Commissioner advised that checkpoints into Plymouth could open, and essential services allowed access. 12:30. Police Commissioner visited MVO, briefed by Chief Scientist (CS). Check points into Plymouth closed. 12:45. Deputy Chief Scientist (DCS) arrived at MVO. CS/DCS discussed situation. Duty scientist at airport put on alert. 12:45 to 12:55. Essential services in Plymouth advised to evacuate. Port Authority, Monlec, Montserrat Mills contacted successfully, but direct contact with Texaco fuel tanker not made (the tanker later cut the fuel lines to the port and successfully withdrew). Plymouth siren sounded. Live update of situation broadcast on Radio Montserrat (ZJB). Field team at White's Yard advised to speed up operation. Field team requests permission to enter central corridor area (cross-island road); permission refused by CS with advice to withdraw to airport. 12:55. Pyroclastic flow activity started on dome. c. 12:58. Phased evacuation of airport recommended by MVO. c. 13:00. Immediate evacuation of airport initiated. Airport and MVO staff, recently arrived passengers and taxi drivers clear of airport in less than 5 minutes; plane on ground took off. The

230

S. C. L O U G H L I N ET AL.

Governor of Montserrat, having just landed at airport, accompanied MVO scientists to safe viewing position north of the airport. c. 13:00. Acting Governor, Chief Minister, Minister for Education and Health, Minister for Agriculture, Trade and Environment arrived at MVO for briefing. c. 13:00. Two MVO staff members took off in helicopter from Vue Pointe Hotel, near Salem. c. 13:10. All MVO staff accounted for and confirmed safe. c. 13:15. Acting Governor, etc. left MVO for Emergency Operations Centre (EOC), accompanied by DCS and Dr P. Baxter. 13:16. Helicopter reconnaissance over the airport picked up survivors in the Trant's area. 13:20. Ash advisory issued by MVO. 14:00. Helicopter searched for survivors in Spanish Point area. 14:45. Survivors sighted in Harris; evacuation initiated by MVO helicopter. 15:06. Search-and-rescue team approached from west by road and reported hot ash deposits at Dyer's. 15:14. Hot ash reported by Police in Belham River valley, near Cork Hill. 16:00. CS briefed authorities at EOC on observations of activity and status of current situation.

References ANDERSON. T. & FLETT, J. S. 1903. Report on the eruption of the Soufriere in St Vincent 1902 and on a visit to Montagne Pele in Martinique. Philosophical Transactions of the Royal Society of London, Series A, 100. 353-553. ASPINALL, W. P., LOUGHLIN, S. C., MICHAEL, F. V. ET AL. 2002. The Montserrat Volcano Observatory: its evolution, organization, role and activities. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London, Memoirs, 21, 71-91. BAPTIE, B., LUCKETT, R. & NEUBERG. J. 2002. Observations of lowfrequency earthquakes and volcanic tremor at Soufriere Hills Volcano. Montserrat, In: Druitt. T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano. Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21. 611-620. BAXTER, P. J. 1990. Medical effects of volcanic eruptions. 1. Main causes of death and injury. Bulletin of Volcanology. 52. 532-544. BAXTER, P. J.. NERI, A. & TODESCO. M. 1998. Physical modelling and human survival in pyroclastic flows. Natural Hazards, 17, 163-176. BERNSTEIN, R. S., BAXTER, P. J., FALK, H., ING. R., FOSTER, L. & FROST. F. 1986. Immediate public health concerns and actions in volcanic eruptions: lessons from the Mount St Helens eruptions May 18-October 18 1980. In: BUIST, A. S. & BERNSTEIN, R. S. (eds) Health effects of volcanoes: an approach to evaluating the health effects of an environmental hazard. American Journal of Public Health (supplement). 76. 25-37. BONADONNA, C., MAYBERRY, G. C.. CALDER, E. S. ET AL. 2002. Tephra fallout in the eruption of Soufriere Hills Volcano. Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21. 483-516.

CLAY. E.. BARROW. C.. BENSON. C.. DEMPSTER. J.. KOKELAAR. P.. PILLAI. N. & SEAMAN. J. 1999a. An evaluation of HMG 's response to the Montserrat Volcanic Emergeney. Volume I. DfID Report. CLAY. E.. BARROW. C.. BENSON. C.. DEMPSTER. J.. KOKELAAR. P.. PILLAI. N. & SEAMAN. J. 1999h. An evaluation of HMG's response to the Montserrat Volcanic Emergency. Volume II. DfID Report. DEWEY. J. M. 1985. The Propagation of Sound from the eruption of Mount St Helens on 18 May 1980. Northwest Science. 59. 79-92. DRI ITT. T. H.. CALDER. E. S. COLE. P. D. ET AL. 2002. Small-volume, highly mobile pyroclastic flows formed by rapid sedimentation from pyroclastic surges at Soufriere Hills Volcano. Montserrat: an important volcanic hazard. In: Druitt. T. H. & K O K E L A A R . B. P. (eds) The Eruption of Soufriere Hills Volcano. Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21. 281-306. FISHER. R. V. & H E I K F N . G. 1982. Mt Pelee Martinique: May 8 and 20 1902 pyroclastic flows and surges. Journal of IVolcanologvand Geothermal Research. 13. 339-371. FUJII. T. & NAKADA. S. 1999. The 15 September 1991 pyroclastic flows at Unzen Volcano (Japan): a flow model for associated ash-cloud surges. Journal of Volcunology and Geothermal Research. 89. 159-172. IAVCEI. 1995. Understanding Volcanic Hazards. Videotape. IAVCEI UNESCO. IAVCEI. 1997. Reducing Volcanic Risk. Videotape. IAVCEI UNESCO. KOKELAAR. B. P. 2002. Setting, chronology and consequences of the eruption of Soufriere Hills Volcano. Montserrat (1995-1999). In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserra.t from 1995 to 1999. Geological Society. London. Memoirs. 21. 1-45. LEA. D. 1996 1999. Price of Paradise. Series of video cassettes. Living Letters Studios. Montserrat. LEA. D. & SPARKS. R. S. J. 2000. Montserrat's Andeessite Volcano. Videotape. Living Letters Studios. Montserrat. L O I G H L I N . S. C. CALDER. E. S.. CLARKE. A. ET AL. 2002. Pyroclastic flows and surges generated by the 25 June 1997 dome collapse. Soufriere Hills Volcano. Montserrat. In: Druitt. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21. 191-209. OPPENHEIM. A. N. 1992. Questionnaire Design. Interviewing and Attitude Measurement. Pinter. London and Washington. ROSENBAUM. J. G. & WAITT. R. B. 1981. Summary of eyewitness accounts of the May 18 eruption. In: LIPMAN. P. W. & M L L L I N E A L X . D. R. (eds) The 1980 Eruptions of Mount St Helens. Washington. United States Geological Survey. Professional Papers. 1250. 53 68. SAARINEN. T. F. & SELL. J. L. 1985. Warning and Response to the Mount St Helens Eruption. State University of New York Press. Albany. SHIMABARA ONSEN HOSPITAL. 1993. Voleanic disaster of Shimabara in 1991. Hospital Report (in Japanese). SOLOMON. F. & MARSTON. R. Q. (eds). 1986. The Medical Implications of Nuclear War. National Academy Press. Washington DC. VALENTINE. G. A. 1998. Damage to structures by pyroclastic flows and surges, inferred from nuclear weapons effects. Journal of \ 'olcanologv and Geothermal Research. 87, 117-140. VOIGHT. B.. HOBLITT. R. P.. C L A R K E . A. B.. LOCKHART. A. B.. M l L L E R .

A. D.. LYNCH. L. & McMAHON. J. 1998. Remarkable cyclic ground deformation monitored in real-time on Montserrat. and its use in eruption forecasting. Geophysical Research Letters. 25(18). 3405-3408. VOIGHT. B.. SPARKS. R. S. J.. M I L L E R . A. D. ET AL. 1999. Magma flow instability and cyclic activity at Soufriere Hills Volcano. Montserrat. British West Indies. Science. 283. 1138-1142. WATTS. R. B.. HERD. R. A.. SPARKS. R. S. J. & YOLNG. S. R. 2002. Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano. Montserrat. In: Druitt. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano. Montserrat. from 1995 to 1999.

Geological society, London, Memoirs. 21. 115 152.

Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano, Montserrat P. D. COLE 1 , E. S. CALDER 2 , R. S. J. SPARKS 2 , A. B. CLARKE 3 , T. H. DRUITT4, S. R. YOUNG5, R. A. HERD6, C. L. HARFORD 2 & G. E. NORTON6 1

4

Centre for Volcanic Studies, University of Luton, Park Square, Luton, LU1 3JU, UK (e-mail: [email protected]) 2 Department of Earth Sciences, University of Bristol, Queens Road, Bristol, BS8 1 RJ, UK 3 Department of Geosciences, Penn State University, 503 Deike Building, University Park, PA 16802-2714, USA Department des Sciences de la Terre (UMR 6524 et CNRS), Universite Blaise Pascal, 63038 Clermont Ferrand, France 5 British Geological Survey, Murchison House, West Mains Road, Edinburgh, EH9 3LA, UK 6 British Geological Survey, Keyworth, Nottingham, NG12 5GG, UK

Abstract: Pyroclastic flows were formed at Soufriere Hills Volcano by lava-dome collapse and by fountain collapse associated with Vulcanian explosions. Major episodes of dome collapse, lasting tens of minutes to a few hours, followed escalating patterns of progressively larger flows with longer runouts. Block-and-ash flow deposit volumes range from <0.1 to 25 x 10 6 m 3 with runouts of 1-7 km. The flows formed coarse-grained block-and-ash flow deposits, with associated fine-grained pyroclastic surge deposits and ashfall deposits. Small flows commonly formed lobate channelized deposits. Large block-and-ash flows in unconfined areas produced sheet-like deposits with tapering margins. The development of pyroclastic surges was variable depending on topography and dome pore pressure. Pyroclastic surge deposits commonly had a lower layer poor in fine ash that was formed at the current front and an upper layer rich in fine ash. Block-and-ash flows were erosive, producing striated and scoured bedrock surfaces and forming channels, many metres deep, in earlier deposits. Abundant accidental material was incorporated. Pyroclastic flow deposits formed by fountain collapse were pumiceous, with narrow sinuous, lobate morphologies and well developed levees and snouts. Two coastal fans formed where pyroclastic flows entered the sea. Their seaward extent was limited by a submarine slope break.

Pyroclastic flows were a major feature at Soufriere Hills Volcano, Montserrat, between 1995 and 1999, and were generated by two principal mechanisms. First, collapse of the andesitic lava dome generated numerous block-and-ash flows. Second, two series of Vulcanian explosions in August 1997 and September to October 1997 generated pumice-and-ash flows by fountain collapse (Druitt et al. 2002b). In most cases, the pyroclastic flows were divisible into three components: a dense, coarse-grained basal part that was substantially confined to topographic depressions; an overlying, dilute, fine-grained and turbulent pyroclastic surge; and a buoyant ash plume. The relative importance of the three components varied considerably from one flow to another. In addition, a new phenomenon was documented in which a pyroclastic surge transformed by rapid sedimentation into a dense pyroclastic flow, termed a surgederived pyroclastic flow (Druitt et al. 2002a), which travelled substantially beyond the parent surge cloud. In this paper the term pyroclastic flow is used in a general sense for any pyroclastic density current without reference to particle concentration or transport mechanism. The high concentration basal parts of pyroclastic flows are termed either block-and-ash flows where they formed by dome collapse, or pumice-and-ash flows where they formed by fountain collapse. We describe the main features of the pyroclastic flow deposits formed in the 1995-1999 period. The paper develops the preliminary account of the pyroclastic flows from Soufriere Hills Volcano by Cole et al. (1998), and is complementary to other papers in this Memoir that document and discuss specific events (Loughlin et al. 2002a,b; Sparks et al. 2002; Ritchie et al. 2002). Table 1 lists all the major pyroclastic flows in the 1996-1999 period. We discuss mechanisms of pyroclastic flow formation, transport and emplacement, based on observations of the flows and their deposits.

Temporal development of pyroclastic flows Growth of the lava dome began on 15 November 1995 within English's Crater, a sector-collapse depression open to the east. In late March 1996 rockfalls and then block-and-ash flows began to spill

out of the open, eastern side of English's Crater into the Tar River valley. Between March 1996 and February 1997 rockfalls and block-and-ash flows were confined to the Tar River valley (Figs 1 and 2). The first substantial pyroclastic flows occurred on 27 and 31 March 1996, with runouts of 1 and 1.5km respectively, and another on 3 April (Fig. 2a) travelled 1.5 km from the dome. Blockand-ash flows first reached the sea on 12 May 1996, 2.7 km from the dome (Fig. 2b). Substantial sustained dome collapses occurred at the end of July, August and September 1996 (Table 1 and Fig. 2c-e), in association with rapid increases in the rate of extrusion of the lava dome (Sparks et al. 1998). On 17 September 1996, semi-continuous collapse over a period of nine hours formed block-and-ash flows, the largest of which travelled over 3 km and entered the sea (Fig. 2e). This collapse was followed within 2.5 hours by the first magmatic explosive activity of sub-Plinian style (Robertson et al. 1998). The combination of large-volume sustained dome collapse (see Table 2 for volumes) and explosive activity formed a 200-m-deep scar in the dome (Watts et al. 2002). Following three months of regrowth of the dome, block-and-ash flows caused by dome collapse formed again in the Tar River valley, in late December 1996 and January 1997. During 1997 the lava dome progressively filled English's Crater and started to spill out over low points on the crater rim. Blockand-ash flows then began to inundate other flanks, first travelling over the southern rim into the vicinity of Galway's Soufriere on 10 February 1997 (Fig. 3a), where in March numerous small flows built a talus apron. These initial flows travelled about 1 km and formed narrow, lobate deposits. Block-and-ash flows formed during a sustained dome collapse on 30 and 31 March travelled 3.6km down the White River valley (Fig. 3b), and a further sustained collapse on 11 April formed flows that travelled 500 m further down the valley, with a total runout of 4.1 km (Fig. 3c). These later flows were less topographically confined due to progressive infilling of the valley by previous deposits. Similar temporal development was illustrated by the movement of block-and-ash flows down Mosquito and Tuitt's Ghauts in May and June 1997 (Figs 4a, b). Initial rockfalls and small flows first moved 200m down Tuitt's Ghaut on 27 May 1997. Over the next few days the flows had runouts of up to 0.5 km, and on 5 June 1997 block-and-ash flows travelled 3.1 km down Tuitt's Ghaut (Fig. 4a).

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 231-262. 0435-4052/02/$15 © The Geological Society of London 2002.

231

Table 1. Major pyroclastic-flow-forming

events during the 1995-1999 phase of eruption of Soufriere Hills Volcano

7

March

Va T w T M W R R GGG G

Major pyroclastic flows and associated phenomena 31/3/96

- First dome-collapse pfs 1 .5 km down Tar River valley.

12/5/96

- First dome-collapse pfs travel to the sea 2.9 km down Tar River valley.

April May

—

June July

1996

9

-

29&31/7/96 - First major dome collapse. Pyroclastic surges travel 0.5 km across the sea. 12/8/96 - Major dome-collapse with multiple pfs to the sea.

Sept

-

4/9/98 17/9/96

Au

- Dome collapse with multiple pfs to the sea. - 9 hour dome collapse followed by sub-Plinian explosion plume to 13 km altitude.

Oct Nov Dec

1 9/1 2/96

Jan

8.16& 20/1/97

- Major dome collapses down Tar River valley. Many flows reach the sea.

Feb

10/2/97

- First rockfalls and small pfs down Galways Wall to the south.

March

1997

- First pfs down Tar River valley to the sea since 1 7/9/96.

30&31/3/97 • First major dome collapse down Galways Wall. Pfs down White River valley to 3.5 km.

April

11/4/97

- Major dome-collapse pfs down White River valley to 4.1 km.

May

15/5/97 27/5/97

- Dome-collapse pfs to the sea in Tar River valley, - First rockfalls and pfs down Tuitt's Ghaut.

5/6/97 16/6/97 25/6/97

- Dome-collapse pfs to 2.8 km in Tuitt's Ghaut. - Dome-collapse pfs to 2 km down Fort Ghaut. 17/6/97 - Dome collapse pfs to 4 km in Mosquito Ghaut. - Dome-collapse pfs 6.8 km down Mosquito Ghaut. 19 people killed. 30/6/97 D-c pf 3.5 km in Fort Ghaut. Late June - early July cyclical pf production every 8-12 hours in Mosquito & Fort Ghaut.

June

-

July Aug

'/

'/

Sept

,

Oct

1

Nov Dec

3/8/97 - Major dome collapse down Fort Ghaut; Plymouth largely destroyed. Triggers explosions. 4 to 12/8/97- 13 vulcanian explosions with radial fountain-collapse pfs. Mosquito Ghaut largely filled with deposits of many small dome collapse pfs. 21/9/97 - Major dome collapse down Tuitt's Ghaut; airport destroyed. Triggers explosions. 22/9/97 to 21/10/97 - 75 Vulcanian explosions; 74 with radial fountain-collapse pfs. 4 & 6/1/97 - Major dome collapses to south down White River valley to the sea. 26/12/97

- Sector collapse, debris avalanche and extensive energetic pfs and surges in around the White River valley.

Jan

Key

Feb

moderate pfs 1 -3 km 10/03/98

March

- Dome stops growing.

—

April

multiple episode of Vulcanian explosions fountain collapse pfs in most valleys TR Tar River WR: White River1 TG Tuitt's Ghaut:MG = Mosquito Ghaut; FG=Fort Ghaut; WG = White's Ghaut

May

1998

dome collapse

June July

3/7/98

- 2.5 hour dome-collapse pf down Tar River valley, Long Ground impacted for first time.

12/1 1/98

- Large dome-collapse pf down Fort Ghaut; pfs reach the sea.

Aug

Sept Oct Nov Dec

—

Jan Feb

- Fountain collapse pfs confined to Fort Ghaut and Tar River valley.

March

- Sporadic ash venting and Vulcanian explosions.

April

1999 iyyy

May

/ 5/6/99

- Dome collapse event on northern face of Tar River valley.

June July Aug

Sept Oct

- Dome growth recommences for first time since 10/3/98 Nov

Fig. 1. Map of Soufriere Hills Volcano, Montserrat, showing principal block-and-ash flow deposits formed between 1996 and 1999.

234

P. D. COLE ET AL.

DEPOSITS FROM PYROCLASTIC FLOWS

235

Table 2. Characteristics of selected pyroclastic flows Date

Deposit volume (x10 6 m 3 )

Duration (mins)

No. of flow pulses

Runout (km)

% Area of surge

Comments/location

Discrete flows 3 Apr. 1996 12 May 1996

0.2 0.4

3 7

3 3

2 3

40 40

Duration relates to longest of three Duration relates to longest of three

31 Mar. 1997 5 June 1997 17 June 1997

0.3 0.4 0.8

4 5 35

1 1 4

2 2.8 4

90 5 10

Tar River valley Tuitt's Ghaut Mosquito Ghaut

Sustained collapses 29/31 July 1996 11 Aug. 1996 17 Sep. 1996

4 3.5 12.3

180(29/7/96) 180 and 120 540

>6 >6 >6

>3.5 >3.5 >3.5

85 75 70

Tar River valley Tar River valley Tar River valley

30 Mar. 1997 11 Apr. 1997 25 June 1997 3 Aug. 1997 21 Sep. 1997 4/6 Nov. 1997

2.6 3 6.4 9.1 14.3 8

c.45 180 25 120 20 15-30 and 35

>6 >6 3 >6 2 >6

3.5 4 6.8 >4.5 >6.9 >4.5

45 35 80 20 40 N/C

White River valley White River valley Mosquito Ghaut Fort Ghaut Tuitt's Ghaut White River valley

3 July 1998

20-25

150

>6

>3.5

N/C

Tar River valley

Deposit volumes are non-DRE and include the associated pyroclastic surge deposits. Duration measured from seismic records. N/C, not calculated.

Mosquito Ghaut was inundated in a similar fashion as small rockfalls first overtopped English's Crater above the valley on 13 June 1997. These were followed by progressively longer flows (1.5km runout) over the next few days. On 17 June a block-and-ash flow travelled 4 km down Mosquito Ghaut (Fig. 4a). Eight days later, on 25 June, a flow with three block-and-ash flow pulses travelled down Mosquito Ghaut to within 50m of the sea, 6.8km from the dome (Fig. 4b) (Loughlin et al 2002a,b). The first minor rockfalls occurred down Fort Ghaut in June 1996. A year later, dome growth concentrated on the western side formed rockfalls in Fort Ghaut on 14 June 1997, and by 16 June pyroclastic flows had reached Gages Lower Soufriere, 2km west of the dome. Dome-collapse flows on 31 June 1997 had runouts of 3.5km and reached the eastern margins of Plymouth for the first time (Fig. 1). A sustained dome collapse on 3 August 1997 formed pyroclastic flows that travelled 4 km to the west, impacting large parts of Plymouth (Fig. 1). This was followed, between 4 and 12 August, by a series of 13 Vulcanian explosions (Druitt et al. 2002b), each involving generation of fountain-collapse pumice-andash flows (Fig. 5). Dome growth after 12 August filled the crater formed by these explosions and produced numerous block-and-ash flows onto Farrell's Plain (around Farrell's Yard; Fig. 1) to the north. On 21 September 1997, large-volume block-and-ash flows travelled down Tuitt's Ghaut impacting the airport 6km to the NE (Fig. 6). This was followed by a second series of 75 Vulcanian explosions from 22 September to 21 October. All but one of the explosions generated pumiceous pyroclastic flows by fountain collapse. The flows travelled simultaneously down several valleys, including Tuitt's Ghaut, Tar River valley, Fort Ghaut, White River valley and White's Ghaut. A map of pyroclastic flow deposits formed by a typical Vulcanian explosion on 18 October 1997 is shown in Figure 7. The runouts of fountain-collapse flows were generally between 3 and 6 km. Rapid dome growth at the end of October 1997 once again filled the Vulcanian explosion crater. Growth switched to the southern margin above Galway's Wall at the end of October. Extensive dome collapses occurred on 4 and 6 November 1997, forming

block-and-ash flows in the White River valley and extending the fan on the southern coast (Fig. 3d). In November and December 1997, dome growth was once again focused on the southern side and on 26 December a sector collapse involving both the lava dome and the flanks of the old volcanic edifice formed a debris avalanche and associated high-velocity pyroclastic density current (Ritchie et al. 2002; Sparks et al. 2002; Voight et al 2002). Dome growth ceased 2.5 months later, in March 1998. Block-and-ash flows occurred sporadically following cessation of dome growth (Norton et al. 2002). These formed by both dome collapse and explosions. Flows formed during the post-domegrowth period were quite varied in both volume and nature. One of the largest dome collapses took place on 3 July 1998 and produced block-and-ash flows that moved down the Tar River valley and reached the sea, as well as pyroclastic surges that spilled northeastwards 300 m out of the valley where they impacted the village of Long Ground (Fig. 1). Collapses on 12 November 1998 down Fort Ghaut were characterized by relatively fine-grained, valleyconfined pyroclastic flows with few blocks >2m in size. Pyroclastic surges developed only within the initial 1 km, and blocks in the distal regions, 5km from the dome, were typically <2m in diameter. Flows associated with post-dome-growth explosive events, e.g. 10 May 1999, were generated by fountain collapse and generally produced deposits with a greater proportion of fine-grained surge deposits than previous block-and-ash and pumice-and-ash flows. Clasts within these flows were semi-vesicular (Norton et al. 2002). Rockfalls and block-and-ash flows Observations Rockfalls and block-and-ash flows resulting from gravitational instability of the lava dome, both during and after growth (postMarch 1998), occurred almost daily from early 1996 through to November 1999, when a new period of dome growth started. Rockfalls consist of debris that does not readily fragment, such that

Fig. 2. Maps showing the development of pyroclastic flow deposits in the Tar River valley during 1996. (a) 3 April, (b) 12 May, (c) 29-31 July, (d) 12 August, (e) 17 September. TREH = Tar River Estate House. CP = Castle Peak dome. D = dome. Block-and-ash flow deposits not mapped for 17 September. Roman numerals relate to measured sections shown in Figures 11 and 18.

236

P. D. COLE ET AL.

Fig. 3. Maps showing the development of pyroclastic flow deposits over Galway's Soufriere and White River valley on the south side of the volcano in 1997. (a) 10 February and March, (b) 30-31 March, (c) 11 April, (d) 6 November.

they form short lobes with runouts of < 1 km. They generally lack vigorously convecting ash clouds (Fig. 8a). Block-and-ash flows, on the other hand, are composed of material that readily fragments and generates large amounts of fine-grained material. They are characterized by vigorously convecting ash clouds (Fig. 8b & c). For more detail on this distinction see Calder el al. (2002). Rockfalls and block-and-ash flows are considered as end-members

of a gradational spectrum of flow types. Dome-collapse block-andash flows travelled from < l k m to 7km. Both single-pulse and multiple-pulse flows generated by sustained collapses have been recognized. Durations of dome collapses varied between 1 minute for small flows to several hours for large events (Table 2). Measured block-and-ash flow-front velocities were up to 6 0 m s - 1 , although velocities of 5-20 m s - 1 were more typical. The pyroclastic flows

DEPOSITS FROM PYROCLASTIC FLOWS

237

Fig. 4. Maps showing the distribution of pyroclastic flow deposits in Tuitt's and Mosquito Ghauts during June 1997. (a) Pre-25 June 1997. Dates with arrows point to termini of successive pyroclastic flows, (b) 25 June 1997.

238

P. D. COLE ET AL.

Fig. 5. Map showing the distribution of pumice-and-ash flow deposits formed during Vulcanian explosions in August 1997 and between 22 September and 21 October 1997, adapted from Druitt el al. (2002b).

had a wedge-shaped, lobe-and-cleft frontal morphology (Fig. 8b), which was intermittently disrupted by poorly sorted mixtures of blocks and ash surging forwards from within the flow. Sustained block-and-ash flows had distinct pulses, where successive highervelocity pulses overtook earlier ones. During the course of the eruption, seven valleys (ghauts) that drain the volcano were progressively inundated by block-and-ash flows formed by dome collapse, in chronological order: Tar River valley, White River valley, Tuitt's Ghaut, Mosquito Ghaut, Fort Ghaut, Tyre's Ghaut and White's Ghaut (Fig. 1). During a typical episode of dome collapse, the largest-volume, and most extensive flows, tended to occur towards the later stages and were marked by short-lived pulses of intense seismic energy. Patterns varied in the build-up to a major episode of dome collapse, and could be linked to the behaviour of the dome (Watts et al. 2002). Major collapses often followed either a pulse in dome growth or a switch in domegrowth direction. In some cases (e.g. the dome collapse of 25 June 1997), collapses of increasing volume and extent occurred over the weeks preceding the major collapse (Fig. 4a, b). In other cases (e.g. the dome collapse of 21 September 1997, Fig. 6) the major collapse occurred within a few days of a distinct switch in dome-growth direction from north to NE.

Impact effects of block-and-ash flows In several instances houses, mainly in distal regions, were impacted by block-and-ash flows (e.g. 25 June 1997 and 3 August 1997). Houses impacted by blocks > 0 . 5 m in diameter suffered significant or total destruction of walls. In the distal 0.5km of runout there was minimal structural damage to vertical walls, and destruction was restricted to burning of all flammable material, usually causing roof collapse. Many houses were simply buried by block-and-ash flow deposits. Locally, block-and-ash flow deposits banked up against house walls that faced towards the volcano. In areas inundated by the block-and-ash flows most trees were felled. However, some trees that were engulfed by deposits remained standing in distal areas (i.e. the last 0.5 km of flow runout). Burning caused the trees that initially remained standing to fall down in the days following engulfment.

Morphology of block-and-ash flow deposits Deposits of small-volume block-and-ash flows and rockfalls formed extensive (composite) debris aprons around the base of the dome.

DEPOSITS FROM PYROCLASTIC FLOWS

239

Fig 6. Pyroclastic flow deposits formed on 21 September 1997, from one of the largest dome collapses of the current eruption. Roman numerals relate to sections in Figure 11. Inset: histograms showing the variation in density of coarse clasts (>3cm) in a distal lobe of the 21 September 1997 pyroclastic flow deposit.

Some were sheet-like, whereas many others formed overlapping lobate deposits (Fig. 9a). Such individual lobes were narrow, <20m wide, up to several hundred metres long, and some had well developed levees and bulbous snouts (Fig. 9a). These small blockand-ash flow deposits were similar to those formed at Colima volcano, Mexico in 1991 (Rodriguez-Elizarraras et al. 1991). Many of the small-to-moderate volume (<1 x 10 6 m 3 ) blockand-ash flow deposits were confined to the bases of valleys and had rather flat upper surfaces (Fig. 9b). Zones of singed vegetation at the edges of the deposits were narrow, typically <5m, and pyroclastic surge deposits were not apparent. Exceptions to this occured at bends where pyroclastic surges had detached, ridden up and deposited on valley flanks. Distal terminations were generally steep, characterized by accumulations of coarse boulders > 1 m in size and also, in places, piles of logs oriented perpendicular to the valley axis (Fig. 9b & c). Only rarely did pyroclastic surge deposits extend beyond the main block-and-ash flow deposits (e.g. 5 June 1997, Tuitt's Ghaut flow, Fig. 4a). Some of the most voluminous valley-confined block-and-ash flow deposits were up to 40m thick (e.g. 40m in the White River valley on 31 March 1997, and 20m thick in Pea Ghaut on 25 June

1997). Many of the greatest thicknesses were probably the result of several block-and-ash flow pulses that occurred during a single dome collapse or several rapidly successive flows from a discontinuous collapse. Where the larger-volume block-and-ash flows disgorged from the valleys and spread unconfined across plains (e.g. 25 June and 21 September 1997, Figs 4b and 6; Table 2) the associated deposits formed extensive sheets that ranged from >3m to locally
240

P. D. COLE ET AL.

Fig. 7. Map of pumice-and-ash flow deposits formed by fountain collapse during a Vulcanian explosion on 18 October 1997. Inset: Details of some distal lobes of deposits. The four Soufrieres 1-4, are named on Figure 1. patterns (Fig. 9d). The ridges had heights of around a decimetre and spacings of several metres. Ridges of deposit extended both upstream and downstream of large blocks. Similar ridges and furrows were observed on the surfaces of the Tar River (Fig. 2e) and White River fans following flows from large collapses. These

features were observed immediately following emplacement, and so were not generated by reworking. Numerous large blocks several metres in size protruded well above the deposit surface of sheets of block-and-ash flow deposits. Many blocks were capped by massive flow deposit, up to 30cm

DEPOSITS FROM PYROCLASTIC FLOWS

Fig. 8. (a) Rockfall extending to the base of the talus slope in the Tar River valley on 8 June 1997. Note the diffuse ash plume developing. Runout is approximately 1 km. (b) Dome-collapse pyroclastic flow on 6 November 1997 at the mouth of the White River, 4.5 km from the dome on the south side of the volcano. Note the pyroclastic surge travelling over the sea. Sediment-rich seawater is from block-and-ash flows that entered the sea minutes earlier. Buildings at bottom right give scale, (c) Broad, small pyroclastic flow near the base of the dome in the Tar River valley.

241

242

P. D. COLE ET AL.

Fig. 9. (a) Deposits of small block-and-ash flows and rockfalls forming a talus apron at the base of the south side of the dome near the position of Galway's Soufriere, May 1997. (b) Block-and-ash deposits in the White River valley immediately following the 11 April 1997 dome collapse. Note the filling of topography by flows, absence of surge deposits at valley margins and the trees remaining standing in the base of the valley, (c) Margin of the 21 September 1997 flow deposits at Trant's near WH Bramble Airport. Note concentrations of coarse pumice clasts and charcoaled wood. Donkeys for scale (mid-right), (d) Dendritic, ridge-and-furrow surface morphology on the 25 June 1997 block-and-ash flow deposit. Photo taken on 26 June 1997. Roads and houses indicate scale, (e) Deposits stranded on the top of a boulder in the 21 September 1997 block-and-ash flow deposit in the Spanish Point region. Hammer for scale.

thick (Fig. 9e). Such caps typically occurred >1 m above the upper surface of the surrounding deposit.

Lithology of block-and-ash flow deposits In the available sections through them, block-and-ash flow deposits were generally massive and ungraded, although grading was observed in some exposures (Figs 10 and 11). At one locality in the 21 September 1997 block-and-ash flow deposit, the lower

half of the 3 m thick deposit was reverse graded and the upper half was normally graded. This resulted in a distinctive coarse central zone (Figs lOa and 11c), and similar grading was observed in other sheet-like deposits from large dome collapses (Fig. 11b; e.g. 20 Jan 1997). Gas-escape structures were abundant within the deposits (Fig. 11b & c). The block-and-ash flow deposits were poorly sorted ( o > 2), containing fragments that ranged from fine ash to dense blocks up to 15m in size. Grain-size analyses of the matrix fraction (<32mm only) showed that median diameters ranged between — 1 . 5 and 0.9o and fine ash constituted <10wt 0 / o (Fig. 12).

DEPOSITS FROM PYROCLASTIC FLOWS

243

Fig. 10. (a) Section through 21 September 1997 block-and-ash flow deposits in Spanish Point. Flow direction was from left to right (see Fig. 6 for location). The pre-eruption surface is the dark horizon above the person's head, (b) The base of the 21 September 1997 block-and-ash flow deposit near the coast at Spanish Point. Flow direction was from left to right. The original surface is the dark horizon in the centre against which the spade rests. Note the appearance of the reverse graded lens at the base of the flow deposit that occurs only in a depression in the palaeotopography. Spade for scale. Lenses of fines-poor deposit, up to 20cm thick, underlie some block-and-ash flow deposits (Figs 10 and 11). These lenses are thickest and best developed in hollows and where the block-and-ash flows encountered topographic roughness (Fig. lOb). Where the flow travelled over flatter surfaces, these fines-poor deposits are thinner and/or absent. The lenses have sharp contacts with the

overlying block-and-ash flow deposit and reverse grading was locally observed within them (Fig. l1e). Locally the fines-poor deposits have penetrated small cracks in the pre-eruption surface. The clasts in the flows were porphyritic andesite of uniform composition with prominent hornblende and plagioclase phenocrysts (Devine et al. 1998; Murphy et al. 2000). Mafic inclusions

Fig. 11. Selected sections through block-and-ash flow deposits, (a) 12 May 1996 Tar River valley; (b) January 1997. Tar River fan: (c) 21 September 1997. north of Spanish Point; (d) pumice-and-ash flow deposits in Plymouth. Histograms show weight % versus phi size. For location of sections see Figures 2 and 6.

245

DEPOSITS FROM PYROCLASTIC FLOWS

were common (c. 1%) and blocks varied from homogeneous and structureless to strongly flow banded. Coarse clasts (>3cm) ranged in density from 1900 to 2700 kg m-3 (Fig. 13a, b). Pumiceous clasts were rare and were probably accidental, related to earlier Vulcanian explosions (e.g. concentrated in the margins of the 21 September 1997 block-and-ash flow deposit). Blocks varied in shape. Many were subangular with rounded edges and corners. Large blocks commonly displayed friction marks with polished slickensided surfaces overlying thin layers of cataclasite (Grunewald et al. 2000). The friction marks were elongate, typically 1 to 5 cm wide and 10cm to 1m long. Some of the largest friction marks had pseudotachylite surface layers c. 50 to 100um thick. These marks formed from sliding collisions with hard substrate and other large blocks during transport. Some blocks of dense, poorly vesicular andesite were rounded by concentric fractures where rock easily broke away from the surface. In contrast to block-and-ash flow deposits from some other volcanoes (Francis et al. 1974; Miyabuchi 1999), prismatically jointed blocks were rare. Temperatures of several block-and-ash flow deposits were measured using a thermocouple and ranged from 365 to 640°C (Table 3).

Synopsis Pyroclastic flows generated by dome collapse ranged from rockfalls with < 1 km in runout, through to block-and-ash flows that travelled up to 7km. Deposit volumes ranged from <10 3 m 3 to over 10 7 m 3 , with runout distance being correlated with flow volume (Calder et al. 1999). Many dome-collapse episodes involved multiple collapses over tens of minutes to hours, so that the eventual deposits commonly represented amalgamations from numerous individual block-and-ash flows. Measured or calculated block-and-ash flow velocities generally ranged from over 3 0 m s - 1 in proximal regions to < 1 0 m s - 1 in distal regions (Loughlin et al. 2002a, b). Burial of standing houses with intact walls, and trees left upright in block-and-ash flow deposits in distal regions, reflect the low-velocity emplacement. Three kinds of block-and-ash flow deposit geometry have been identified. Small rockfall and block-and-ash flow deposits ranged

Fig. 12. Median diameter (Md ) versus (a) sorting ( ) and (b) wt% fine ash (<63 um) for all pyroclastic deposits sampled from the eruption. Table 3. Measured temperatures Date

Block and ash flow deposits 10 Jan. 1997 25 June 1997 30 June 1997 21 Sep. 1997 Pyroclastic surges 9 Jan. 1997 20 Jan. 1997 31 Mar. 1997 3 July 1998

Temperature measured (°C)*

410 640 400 365 >200 99-121 99-149 260-288

Days after event

Comments

1 20 5 <1

Measured Measured Measured Measured

-

Direct temperature measurements using temperature patches fixed at or near the Tar River Estate House c. 1.7km from the dome on the north side of the Tar River valley

Pyroclastic surge deposits 17 Sep. 1996

326

Surge-derived pf deposits 25 June 1997

410 and 350

5

138

4

220 180

<1 <1

26 Dec. 1997 Pumice-and-ash flow deposits 25 Sep. 1997 17 Oct. 1997

15

3.2km from dome on Tar River fan 6 km from dome at Trant's 4km from dome at Plymouth 6km from dome at Trant's

Measured near Tar River Estate House using a thermocouple Measured 2.5km (Dyer's river) and 6km from dome (below Cork Hill). Measurement made 3.5km from dome Measured from deposits on edge of fan at mouth of Tar River valley approx 3.2km from source a few hours after emplacement

* Both deposit temperatures measured using a thermocouple after the event and direct pyroclastic surge temperatures measured using temperature patches are given.

246

P. D. COLE ET AL,

Fig 13. Histograms showing the density of coarse clasts (>3 cm) for block-and-ash flow deposits formed by (a. b) dome-collapse and (c-f ) pumice-and-ash flow deposits.

from narrow, channelized and lobate to thin sheets. Many deposits were valley-confined, with flat upper surfaces and abrupt frontal terminations. The largest-volume block-and-ash flow deposits formed thin sheets with diffuse tapering margins, and had a pattern of narrow ridges and furrows on their upper surface oriented approximately in the flow direction. Visual observations of block-and-ash flows, rounding of the edges and corners of blocks, and slickensided surfaces of large blocks (Grunewald et al. 2000) indicate that many of the blocks in the flows were tumbling. Particle interactions were thus important during large parts of the runout. The larger dome collapses formed block-and-ash flow deposits of quite variable thickness, and many deposits were massive in section. The local reverse-to-normal grading of some block-and-ash flow deposits might have related to amalgamated block-and-ash flow deposits formed during a single collapse. Low-density accidental material was concentrated at the margins of block-and-ash flow deposits.

Pyroclastic flows associated with Vulcanian explosions Observations Pyroclastic flows resulted from fountain collapse associated with 13 Vulcanian explosions between 4 and 12 August 1997 and 74 Vulcanian explosions between 22 September and 21 October 1997 (Table 1). Each explosion lasted a few tens of seconds and the pyroclastic flows generally formed from a single collapse pulse (Druitt et al. 2002/7). Collapse occurred from 300 to 650m above the crater rim at 950m a.s.l. In each case, an initial surge cloud travelled at initial velocities of 6 0 m s - ' near the collapsing fountain and rapidly decelerated before lofting (Druitt et al. 2002b). Following this, dense pumice-and-ash flows emerged from beneath the decelerating cloud, where they were channelled along the valleys and then spread out onto unconfined plains at about 1 0 m s - 1 (Figs 7 and 14a). Ash plumes that developed above the pumice-andash flows were weak in comparison (Fig. 14b) to those formed above dome-collapse flows of the same runout length.

Pyroclastic flows formed by Vulcanian explosions in August 1997 formed thin veneers of coarse pumice and lithics around the upper regions of the valleys draining the volcano (Fig. 5). Finegrained pyroclastic surge deposits were extensive in the first 2 km around the site of fountain collapse. At distances greater than 2km. pyroclastic surge deposits were poorly developed or absent (Figs 5 and 7).

Effects

of pumice-and-ash flows

Pumice-and-ash flows showed similar effects to block-and-ash flows where, in the final 0.5 km of runout, some trees and telegraph poles that were engulfed remained standing (Fig. 14c). Such a feature was observed in a number of locations, e.g. Fort Ghaut. Spanish Point and White River valley. Pumice-and-ash flows did not burn through the trunks of buried trees, although small branches were partly charred on the outer surfaces and other flammable material, e.g. paper and foam mattresses, was also partially charred. Bark removal on engulfed trees occurred up to 40cm above the upper surface of the deposit (Fig. 14c). Pumice-and-ash flows that impacted houses caused little structural damage. They flowed in through open doors and windows, without causing any structural damage to walls. In some cases, wire mesh fencing contained or deflected the pumice-and-ash flows within the last few hundred metres before the distal terminations.

Morphology of pumice-and-ash flow deposits Pumice-and-ash flow deposits were sinuous and lobate both within valleys and on relatively flat unconfined plains. Individual lobes were up to 300m long and up to 50m wide (Figs 7 and 14a.d). The thicknesses of individual flow-deposit lobes were typically
DEPOSITS FROM PYROCLASTIC FLOWS

247

Fig. 14. (a) Lobate fountain-collapse pyroclastic flow deposits in the White River valley. Photo taken 10 October 1997. Note houses for scale, (b) Fountain collapse pyroclastic flow moving down Tuitt's Ghaut following an explosion at 15:13 hours on 20 October 1997. (c) Tree engulfed by fountain-collapse pyroclastic flow deposits near Plymouth. Note bark of tree has been removed 40cm above the top of the deposit, (d) Pumice-and-ash flow deposit lobes on eastern edge of Plymouth formed on 20 October 1997. Note coarse, fines-poor levees and snouts. Central flow lobe is c. 10m wide, (e) Coarse massive pumice-and-ash flow deposits (PAF) overlying 21 September 1997 block-and-ash flow deposits (BAF). Spade rests on original ground surface, (f) Section through pumice-and-ash flow deposit in Plymouth on the west flank of the volcano. Flow direction was from right to left. The base of the flow deposit is defined by a region poor in fine ash (FP) (broken line shows boundary). Note diffuse gas-escape structures, poor in fine ash higher up in the deposit (G). FR = fines-rich upper layer; FU = flow unit. Scale is in 1 cm intervals.

deposits locally occurred stacked on top of each other. Levees and snouts were enriched in low-density pumice clasts compared to the interior of the deposit lobes (Fig. 13c-f). The surface of deposit margins, including the levees and snouts, were notably matrix-poor, with the uppermost 20 cm locally being completely matrix-free and clast-supported. Lithology of pumice-and-ash flow deposits The pumice-and-ash flow deposits typically contained clasts up to 30cm in diameter, with those larger than 3cm in size ranging in density from 800 to 2000 kg m-3 (Fig. 13c,d). They were distinct

from block-and-ash flow deposits in containing predominantly vesicular pumiceous clasts (compare Fig. 13a, b with 13c, d). Blocks with bread-crust-like surfaces, glassy exteriors and more vesicular interiors formed around 1 % of coarse clasts larger than 3 cm. Grain-size analyses of matrix samples (<32mm) show that they were slightly finer grained and more poorly sorted ( = 3.5-4.5) than many block-and-ash flow deposits (Fig. 12). The pumice-andash flow deposits that were analysed contained between 10 and 23wt% fine ash (Fig. 12). Numerous sections through pumice-and-ash flow deposits indicate that most were massive with reverse coarse-tail grading of pumice clasts (Figs 11d and 14e, f). Ungraded, massive deposits also occur (Fig. l1d). Approximately 50% of pumice-and-ash flow

248

P. D. COLE ET AL.

Fig. 14. (continued)

deposits studied comprise a <2 cm thick basal layer, poor in fine ash (Fig. 11d). The contacts between the basal layers and the overlying pumice-and-ash flow deposits varied from sharp to gradational. Locally, contacts between the two layers were intimate with broad fingers of fine-ash-poor deposit extending up into more ash-rich material above (Figs l1d and 14f). Gas-escape structures and local crude imbrication of elongate clasts were typical within the lower parts of the deposits (Fig. 14f). Temperatures of distal lobes of pumice-and-ash flow deposits formed by two different Vulcanian explosions were measured a few hours after emplacement by thermocouple and were 180 and 220 C.

Synopsis Observations indicate that the high-concentration, valley-confined pumice-and-ash flows formed by segregation from collapsing

fountains. Initially, the collapsing fountains spread away from the source as expanded turbulent currents (pyroclastic surges). The surge components rapidly dissipated, within 2 km of the source, and from these the valley-confined, dense pumice-and-ash flows emerged. Evidence for their dense, highly concentrated character includes lack of extensive surge deposits, concentration of lowdensity pumice in the upper and marginal parts of pumice-and-ash flow deposits and the levee and channel morphology. Velocities of pumice-and-ash flow fronts measured from videos were c. 1 0 m s - 1 . The lack of destructive effects on buildings also indicates gentle, low-velocity emplacement in distal regions. The lack of debarking of trees only 40cm above the top of the deposit indicates that the flows were not expanded by more than 50% of the final deposit thickness in this distal region. Lack of carbonization of wood indicates that the deposits were emplaced at temperatures less than 300 C, which agrees well with temperatures measured a few hours after emplacement.

DEPOSITS FROM PYROCLASTIC FLOWS

Similar lobate pumiceous pyroclastic flow deposits were formed from the 12 June, 22 July, 7 August and 16-18 October 1980 explosions of Mount St Helens (Rowley et al. 1981). Temperatures from the Mount St Helens deposits were markedly hotter (ranging from 350°C to 850°C; Banks & Hoblitt 1981) than the pumice-andash flow deposits described here. Pyroclastic surges Observations Pyroclastic surges were associated with both dome-collapse and fountain-collapse pyroclastic flows. Pyroclastic surges generated by fountain collapse were limited and are not considered further here. Pyroclastic surges typically developed from the upper parts of block-and-ash flows and their deposits were common overlying block-and-ash flow deposits (Fig. 11 a). Some pyroclastic surges were able to detach from, and move independently of, the parent block-and-ash flows (e.g. Figs 2, 4b and 6). For example, extensive pyroclastic surges were formed from block-and-ash flows on 25 June 1997 (Fig. 4b) and these ran up c. 70 m onto Windy Hill, 2.5 km SSE of the dome (Fig. 4b). Velocities in the order of 35 ms-1 are inferred (Loughlin et al. 2002a, b). The development of fine-grained pyroclastic surges from the block-and-ash flows was variable. To investigate the extent of surge detachment, maps were used to calculate the percentage area covered by pyroclastic surge deposits beyond the limits of blockand-ash flow deposits (area covered by pyroclastic surge deposits divided by area covered by block-and-ash flow deposits and pyroclastic surge deposits). There is no systematic relationship between the percentage area covered by the deposit of the detached surge and the volume of the block-and-ash flow deposit (Fig. 15). Large block-and-ash flows do not necessarily produce a greater percentage area impacted by the detached surge than smaller flows. Two small-volume flows shown in Figure 16 illustrate the variability in the area of detached surge. The 5 June 1997 Tuitt's Ghaut flow had only 5% by area of detached surge deposit (Fig. 16b), whereas the 31 March 1997 Tar River flow had 90% of the area covered by the surge alone (Fig. 16d). Effects

of pyroclastic surges

Pyroclastic surges associated with block-and-ash flows impacted buildings on a number of occasions, generally within more distal

Fig. 15. Plot of block-and-ash flow volume versus percentage area of pyroclastic surge detached (i.e. proportion of the area of the pyroclastic surge outside the area covered by the block-and-ash flow deposit).

249

regions. They impacted Streatham on 25 June 1997, Plymouth on 3 August 1997, and Tuitt's village on 21 September 1997, 2.5km NNW, 3.5km west and 3km NE of the dome, respectively (Fig. 1). These pyroclastic surges generally caused little structural damage to masonry walls of buildings (Fig. 17a). Damage was limited to the effects of the high temperature (burning), such that roofs, windows and doors were either totally or partly destroyed. In some cases surges did not penetrate rooms or parts of houses and damage was limited to stripping of bitumen roof tiles (Fig. 17b). Pyroclastic surges formed on 17 September 1996 impacted the Tar River Estate House, 1.7km ENE of the dome (Fig. 2). Some structural damage resulted, such as partial destruction of the balustrade and removal of the roof. Pyroclastic density currents formed on 26 December, 1997, however, caused extensive structural damage to buildings (Sparks et al. 2002). Felling of trees by pyroclastic surges was variable. The surges associated with the 25 June 1997 dome collapse felled trees 0.51.5km north of the dome, along the interfluves on either side of Mosquito Ghaut (Fig. 4b). In large areas at distances of more than 2km the surges did not fell trees or telegraph poles (Fig. 4b). Pyroclastic surges formed during the 21 September 1997 dome collapse left trees upright in large parts of distal impact regions (Fig. 6). Only in the more proximal region, <2km from the dome, were trees abundantly felled. Temperatures of some pyroclastic surges were measured directly using industrial temperature patches attached to walls of buildings and trees. These patches had variable temperature brackets of between 20 and 50°C. Temperatures of pyroclastic surges, determined using temperature patches, ranged from 100 to 290°C (Table 3).

General features of pyroclastic surge deposits Pyroclastic surge deposits formed following nine of the largest dome collapses were studied. The main features of the pyroclastic surge deposits are summarized in Table 4 and representative sections are shown in Figure 18. In a number of cases, pyroclastic surge deposits were studied at similar locations on the northern margin of the Tar River valley (12 May 1996, 17 September 1996, 16 and 20 January 1997, 3 July 1998). This section describes features common to the pyroclastic surge deposits formed following large dome collapses. A prominent feature of seven of the nine deposits examined was that they were massive, without any stratification (Figs 17c and 18). Normal grading was observed in four deposits (e.g. 25 June 1997, 3 August 1997; Table 4), although in five of the deposits grading was not observed. Only two of the nine deposits studied showed well developed stratification and cross-stratification (Table 4). Single units (i.e. derived from the passage of a discrete pyroclastic surge) studied were generally less than 50cm thick, although 10-20 cm was more typical. Locally, in depressions, they thickened up to 1 m. Most of the surge deposits examined show a bipartite layering, with a lower layer poor in fine ash (<20 wt% fine ash) overlain by an upper layer rich in fine ash (>20wt% fine ash). Contacts between the lower and the overlying upper layers were generally sharp. The lower layer decreased in thickness and locally pinched out on topographic highs, whereas the upper one was more continuous across small changes in topography in a number of examples (e.g. 25 June and 3 August 1997). Gas-escape structures, poor in fine ash, were abundant within the upper-layer deposits (Fig. 18) and typically extended from the layer base, or from coarse juvenile clasts. Locally the gas-escape structures were so numerous in the upper layer that they formed the majority of the deposit (e.g. 12 May 1996, 20 January 1997, 3 August 1997). Seared zones, where trees had been lightly burned or singed, occurred locally beyond the outer limit of the pyroclastic surge deposit. They were variable in width, ranging between 10 and 50m beyond the surge deposit, and were wider where winds blew the ash plume away from the block-and-ash flows (e.g. 25 June 1997).

250

P. D. COLE ET AL.

Fig. 16. Two small-volume pyroclastic flows and their deposits, (a) 5 June 1997, Tuitt's Ghaut block-and-ash flow, (b) 5 June 1997. Tuitt's Ghaut block-andash flow deposits (see Fig. 4a for map of deposits), (c) 31 March 1997. Tar River block-and-ash flow, (d) 31 March 1997. Tar River block-and-ash flow (f) and pyroclastic surge (s) deposits. Note the absence of pyroclastic surge deposits associated with the 5 June 1997 flow and abundant surge deposits associated with the 31 March 1997 flow.

We now describe two specific examples of pyroclastic surge deposits formed by dome-collapse pyroclastic flows: first, from pyroclastic surges on 17 September 1996 that produced massive deposits, and second from surges on 3 July 1998 that produced well developed cross-stratification.

Pyroclastic surge deposits of 17 September 1996

Grain-size analyses of the pyroclastic surge deposits (e.g. 17 September 1996, 25 June 1997, 3 August 1997, 21 September 1997, 3 July 1998) show that the lower-layer deposits range in median diameter (Md ) from —0.5 to +3, and the upper-layer deposits range from +2 to +4 (Fig. 12). The 12 May 1996 deposits were coarser grained (Md = —0.8 to +2.6) more poorly sorted than those of the larger volume dome collapses (Fig. 12). A conspicuous feature of the 17 September 1996 and 25 June 1997 pyroclastic surge deposits was that they locally contained coarse clasts up to 50 cm diameter. In the 25 June 1997 deposit, coarse blocks were concentrated at the base of up-current-facing sides of houses up to 1 km from the parent block-and-ash flow (Fig. 17b).

Extensive pyroclastic surges were associated with the block-and-ash flows formed during the nine-hour, semi-continuous dome collapse into the Tar River valley on 17 September 1996. Pyroclastic surges associated with at least one of these block-and-ash flows spilled out of the lower part of the valley on both its northern and southern margins (Fig. 2e). The pyroclastic surge deposits were examined at a number of locations between 1.4 and 1.7km NE of the dome on the northern side of the Tar River valley (see Fig. 2e for locations). Three units were identified (Fig. 18b), each of which was composed of a lower layer poor in fine ash (<64um) and an upper layer rich in fine ash (Fig. 17c). Contacts between these two layers were generally sharp, but contacts between the lower layer and pyroclastic surge deposits formed only minutes or hours earlier were gradational. The lower layers were non-stratified. They were up to 10cm thick, but quite variable, and locally absent. Clasts consisted of dome andesite and hydrothermally altered lithologies. Each upper layer was massive, although rare, faint stratification was observed in one section (Fig. 18b. locality iv and Table 4). The upper layers varied locally between 10 and 40cm thick. They contained abundant gas-escape structures, some of which stemmed from coarse clasts while others originated from the top of the lower layer. Coarse 'outsized* clasts were notable; for example, nine isolated blocks between 10 and 20cm in size occurred in an area of 1 m 2 within 40m of the northern margin of the pyroclastic surge deposit (locality i Fig. 2e). At least one clast. 12cm in diameter, had impacted 15 cm into the deposits formed hours earlier. Grain-size analyses of pyroclastic surge deposits formed on 17 September 1996 show that the lower layers contain < 20 wt% fine ash ( < 6 3 u m ) , although most lower layers contain < 1]0wt% fine ash (Fig. 19). The upper layers always contain >20% fine ash. The lower layers were coarser (Mdo = —0.2 to 2.7) than the upper layers (Mdo = 1.8 to 3.5). Sorting coefficients ( o) of both the upper and lower layers have similar ranges of between 1 and 2.3 (Fig. 19). Weight percentage histograms demonstrate that the lower layers, poor in fine ash, generally contain coarser fragments than the overlying deposits (Fig. 18b).

DEPOSITS FROM PYROCLASTIC FLOWS

251

Fig. 16. (continued)

Towards the northern limits of the pyroclastic surge deposit, away from the centre of the valley, the lower layers thinned, became intermittent and richer in fine ash (from 2 to 18wt%), whereas the upper layers showed no systematic grain-size variation (Fig. 19c). The temperature of these pyroclastic surge deposits measured by thermocouple 15 days after emplacement was 326°C.

Deposits 1.7km NNE of the dome (adjacent to the Tar River Estate House) varied between 15 and 30cm in total thickness. The lowermost deposit was a lens, up to 6cm thick, poor in fine ash. Three separate cross-stratified layers above this varied from 4 to 20cm thick and each contained climbing dune structures with wavelengths of up to 30cm. Mantle-bedded fine ash layers up to 4cm thick containing accretionary lapilli were interbedded between the cross-stratified layers (Figs 17d and 18e).

Pyroclastic surge deposits of 3 July 1998 Pyroclastic surges formed during the dome collapse on 3 July 1998 (four months after the dome had stopped growing) impacted the whole of the Tar River valley. Seismic signals related to this dome collapse were sustained as it waxed and waned over at least 2.5 hours. Pyroclastic flows travelled initially in a southeasterly direction, impacting Roche's Mountain, and were then deflected obliquely across the Tar River valley to the NE, spilling out of the valley on its northern margin. The village of Long Ground was impacted by pyroclastic surges for the first time during the eruption (Fig. 1). Pumiceous ballistic blocks were observed on Roche's Mountain and sandblasting and erosion of trees was common 1.7 km east of the dome.

Synopsis Pyroclastic surges of variable extent were associated with many dome-collapse pyroclastic flows. They represent the part of the ash cloud overlying the parent block-and-ash flow that is denser than air. They appear to originate by expansion of gases and entrained air from the block-and-ash flows and were thus analogous to 'ash-cloud surges' described from other volcanoes (Boudon et al. 1993; Abdurachman et al 2000; Kelfoun et aL 2000). Tree felling and building damage associated with flows from the largest dome collapses were extensive in the first 2km of pyroclastic

252

P. D. COLE ET AL.

Fig. 17. (a) View of Tuitt's village following impact by the pyroclastic surges of the 21 September 1997 flows. Note the absence of structural damage to walls of houses and abundance of upstanding trees. Several houses retain partial roofs. (b) House in Streatham village impacted by pyroclastic surge on 25 June 1997. showing the accumulation of boulders <0.5m in size on the volcano-facing side of house. Note damage is restricted to burning only and stripping of bitumen tiles from the roof. The surge impacted from the lower right-hand side. (c) Section through 17 September 1996 pyroclastic surge deposits on the northern side of the Tar River valley, with massive upper layer rich in fine ash. and coarser lower layer poor in fine ash (location i on Fig. 2e). (d) 3 July 1998 surge deposits immediately adjacent to the Tar River Estate House. Note presence of cross-stratified layers separated by accretionary lapilli-bearing fine ashfall layer (A.L.).

surge runout. At distances greater than 2km, tree felling and building damage were minimal or absent. Most trees had diameters <20cm and calculated velocities of pyroclastic surges were 35 m s - 1 (Loughlin et al. 2002b). Based on the assumptions of Clarke & Voight (2000), dynamic pressures of pyroclastic surges where most trees remained standing would have been <3 kPa. Using data presented by Valentine (1998), the beginning of structural damage to buildings would occur at dynamic pressures of >7kPa. The massive pyroclastic surge deposits are interpreted as deposited directly from turbulent suspension without significant traction. Normal grading is considered to occur during sedimentation from a turbulent medium, where coarser grains became concentrated at the current base and so they are deposited first. Cross-stratification and climbing dune structures represent tractional sedimentation. The typical bipartite layering, with an upper layer rich in fine ash and a lower layer poor in fine ash, occurs in most pyroclastic surge deposits on Montserrat and its origin is considered further in the 'Discussion' section. Grain-size analyses of the 17 September 1996 pyroclastic surge deposits show that the lower layers record sedimentation of relatively coarse material, poor in fine ash, which is interpreted to have occurred by winnowing from the front of the current. The

increase in the ash content of lower layers away from the valley axes (Fig. 19c) suggests that towards the lateral margins of the pyroclastic surge, the deposition of fine ash was less inhibited than at the front of the current. The sharp contacts between the lower and upper layers suggest that deposition was discontinuous. Some sparse outsized blocks >10cm in diameter within the pyroclastic surge deposits could have been picked up during passage of the surges, but others were derived from the dome. Some of the larger blocks in block-and-ash flows were observed on video to have considerable freedom of movement, sometimes moving beyond the flow front. Thus some blocks in the pyroclastic surge deposits may represent blocks derived from the parent block-andash flow. The lower temperatures of the pyroclastic surges (directly measured with patches) compared to the block-and-ash flow deposits are attributed to the larger amounts of cold air admixed into them. In addition, cooling will be more rapid in pyroclastic surges than in block-and-ash flows, owing to their finer grain size and hence more efficient heat transfer. However, the apparent lower temperatures of pyroclastic surges might also reflect a sampling bias, because the temperature patches obtained data from flows related to relatively small dome collapses.

DEPOSITS FROM PYROCLASTIC FLOWS

253

Fig. 17. (continued)

Ash plume deposits

Observations All pyroclastic flows developed buoyant ash plumes and these ascended to altitudes of up to 15km (Bonadonna et al. 2002). Ash plumes that developed above dome-collapse block-and-ash flows were generally more extensive and ascended to greater heights than those formed above fountain-collapse pumice-and-ash flows. The

development of ash plumes during runout of pyroclastic flows was variable, and strong pulses of plume production occurred where pyroclastic flows encountered changes in topography, such as breaks in slope or sharp bends.

Deposits Extensive fallout occurred from all ash plumes, as described by Bonadonna et al. (2002). Thin, fine-grained ash layers typically

P. D. COLE ET AL.

254 Table 4. Major features of pyroclastic surge deposits studied Collapse date

No. of units

Bipartite layering

Grading

Internal lamination

Localities comments

12 May 1996

1 or 2

Yes

None

Massive (rare weak lamination)

Eight sections examined at edge and centre of Tar River valley

17 Sep. 1996

3 or 4

Yes

Weak, normal

Massive (rare weak lamination)

Ten sections examined across north edge of Tar River valley

16 Jan. 1997

1

Yes

None

Massive

Five sections examined across north edge of Tar River valley

20 Jan. 1997

••)

Yes

None

Massive

Five sections examined across north edge of Tar River valley

25 June 1997

2

Yes

Normal

Massive (upper)

Five sections examined in Farrell's Streatham region

3 Aug. 1997

1

Yes

Normal

Massive

Seven sections sections examined in Plymouth on north side of Fort Ghaut

21 Sep. 1997

1

Yes*

None

Massive

Four sections examined in Tuitt's village region

26 Dec. 1997

2

Yes

Normal

Local crosslamination

Numerous sections examined across whole of region (see Ritchie et al. 2002)

3 July 1998

3

Yes*

None

Well developed cross-lamination

Four sections examined around the Tar River Estate House

* Except for one section.

formed downwind of the area affected by the pyroclastic flows, sometimes with their maximum thicknesses well away from the areas inundated by the source flows. These layers were up to 5cm thick, but were typically < l c m ; they were equivalent to coignimbrite ashfall associated with pyroclastic flows generated by explosive eruptions (Sparks & Walker 1977; Bonadonna et al. 2002). Grain-size analyses of ash plume deposits show them to be finer grained (larger Mdo) than, but overlapping slightly with, the ash-rich pyroclastic surge deposits (Fig. 12). Accretionary lapilli, as large as 11 mm in diameter, were typical in ash plume fallout deposits from pyroclastic flows (Ritchie et al. 2002). They were particularly abundant in deposits of ash plumes that formed where pyroclastic flows entered the sea. Although all ash plumes were to some extent affected by wind, particularly strong winds had an unusual effect on an ash plume that was derived from a block-and-ash flow that travelled 4km down Mosquito Ghaut on 17 June 1997. Although the flow was retained within the confines of the ghaut, a large area of scorched ground, 200m wide and 300m long, was formed on the interfluve outside the confines of the valley, on FarrelFs Plain immediately west of the bend in Mosquito Ghaut (Fig. 4a). Examination of vegetation in the scorched area showed that there was no flattening of grass. We therefore interpret this as being related to fallout from ash plumes with material hot enough to scorch the vegetation.

Erosion by pyroclastic flows and entrainment of lithic clasts Many of the dome-collapse block-and-ash flows eroded into the prehistoric lithified deposits, forming furrowed and striated surfaces with grooves and flute marks (Fig. 20a). Similar erosional effects caused by pyroclastic surges have been described from tuff rings by Fisher (1977), and from Mount St Helens by Kieffer & Sturtevant (1988). At Lascar Volcano, Chile, they were formed by pyroclastic flows (Sparks et al. 1997). At Montserrat, one of the most spectacular examples of erosion was that of a chute, 60m deep and 80m wide, cut into Galway's Wall on the south side of English's Crater by a sequence of dome collapses on 31 March 1997 (Fig. 20b). The 17 September 1996 pyroclastic flows eroded a 30-m-deep gorge 50-100 m wide down the whole length of the Tar River valley, in block-and-ash flow

deposits formed earlier in the eruption (see Fig. 2e). A common observation in the Tar River valley was that small- and mediumvolume flows would deposit in the valley and the deposits would then be eroded by larger-volume and more energetic flows. In several cases pyroclastic flows travelled over ground previously unaffected by pyroclastic flows. In these cases they eroded and incorporated distinctive red-orange, hydrothermally altered lithologies from prehistoric deposits. Such hydrothermally altered fragments form up to 9wt% of the clasts >1 mm in the 12 May 1996 pyroclastic surge deposits and nearly 100wt% of clasts > l m m within lower layers of the 25 June 1997 pyroclastic surge deposits. In other cases, material eroded from prehistoric deposits or those formed earlier during the eruption was petrologically indistinguishable from the juvenile dome rock. The 21 September 1997 dome-collapse pyroclastic flow overran and entrained pumice from pumice-and-ash flow deposits formed during the August 1997 Vulcanian explosions. Accidental pumice typically occurred only in the margins of deposits, within 20m of the edge of the flow lobes. Coarse clasts (>3cm) within the central part of flow lobes were composed predominantly of dense dome rock (Fig. 6 inset). We observed that there was a systematic relationship between the volume of pyroclastic flows and the maximum slope angle on which the related deposits were emplaced. Rockfalls and small pyroclastic flows formed a 33 talus slope around the dome. Generally, as the flows increased in size and runout, the slope at which transition occurred between erosion and deposition decreased. Flows reaching between 1 and 2km were erosive on slopes > 10-15 and flows reaching the sea were erosive on slopes >6-82. In the Tar River valley two prominent benches showed no deposition on their steep, seaward-facing scarps (see Fig. 2). In the Tar River valley the largest flows often eroded substantial amounts of earlier deposits and thus the valley contains only a limited amount of preserved deposit. Development of coastal fans Two main coastal fans of pyroclastic flow deposits were formed during the 1995-1999 phase of the eruption. A fan was built first at the mouth of the Hot River (within the Tar River valley) on the east side of the volcano (Figs 1. 2 and 21a.b). An extensive coastal

Fig. 18. Selected measured sections through pyroclastic surge deposits for a number of different dome collapses between 1996 and 1999. See Figure 2 for location of samples from 12 May 1996 and 17 September 1996. TRV = Tar River valley. TREH = Tar River Estate House.

256 P- D. COLE ET AL.

^SB&E^^^^' S£SSES^SS MS3SH??wS for d js^sif zr^r^t r ™ """fr * depoS,lira,s,opesd^/;™S(«- *«ing ' ""»»« °f

October 1997 and H *!

?

™

Vulca

™an explosions in

valleys by acoustic sounding in July 1998. In the Tar River case e Pyr ClaStIC fan WaS traCed ab Ut vef9S900m OOaP water «m ofoffshore °,depth, with an estimated° volume t ooove, at least M J x I O m- confirming observations that substantial parts of pyroclastic flows entered the sea. The submarine deposits were up to 50m thick and deposition took place on slopes <15 : For the White River fan, thickness changes of up to 50m occurred in submarine valleys, but the survey did not extend far enough offshore to locate the main deposits. In both the Tar River and White River fans, submarine bathymetry determined the deposit distribution.

Discussion

Distance from main valley axis (m) Fig. 19. (a) Median diameter (Mdo) versus sorting ( ) and (b) wt% fine ash for 17 September 1996 pyroclastic surge deposits from the northern edge of the Tar River valley. Tie lines in (a) and (b) connect samples from the same locality, (c) Wt% fine ash versus distance from the centre of the valley. For sample locations see Figure 2e.

Block-and-ash flow deposits formed from dome collapses at Montserrat between 1996 and 1999 were similar in many respects to those associated with other dome-forming eruptions such as at Mont Pelee, Martinique (Fisher & Heiken 1982; Boudon & Lajoie 1989), Merapi in Indonesia (Boudon et al. 1993; Abdurachman el al. 2000), Mount Unzen, Japan, between 1991 and 1995 (Fuji & Nakada 1999; Miyabuchi 1999) and Colima. Mexico, in 1991 (Rodriguez-Elizarraras et al. 1991). The other eruptions mostly formed both coarse block-and-ash flow deposits and fine-grained pyroclastic surge deposits, similar to those described here. Pumiceand-ash flow deposits were similar to those formed at Mount St Helens at several periods in 1980 (Rowley et al. 1981) and at Lascar in 1993 (Sparks et al. 1997). In the following section we discuss the origin of features that are either new and have not been previously described before or features that we consider important and require particular attention.

Deposit margins and surface morphology fan also formed at the mouth of the White River on the southern flank (Figs 3 and 21c). In addition, two small fans were formed by deposits of the 21 September 1997 pyroclastic flows, at the mouth of White's Ghaut and also near Spanish Point (Fig. 6) where flows entered the sea. Pyroclastic flows that moved through Plymouth on 3 August 1997 reached the sea and also formed a small fan. The fan at the mouth of the Tar River valley began to form on 12 May 1996. Numerous pyroclastic flows subsequently added to it (Fig. 21). Surveying undertaken periodically throughout the course of the eruption, including near-coast bathymetric surveys, allowed growth of the fan to be monitored (Fig. 21). Block-and-ash flows from four major dome collapses, on 29-31 July, 11 August, 4 and 17 September 1996. added 2.2, 1.2. 1.4 and 4.2 x 10 6 m 3 of material respectively (Table 5). During each of these collapses, the seaward extent of the Tar River fan increased and reached its maximum easterly limit on 17 September 1996. Since that time subsequent pyroclastic flows have thickened and extended the fan along the coast, in a north-south direction. In addition they have backfilled

Well developed lobes and levees were observed in the pumice-andash flow deposits, and in many of the small-volume dome-collapse rockfalls and block-and-ash flow deposits. The levees are considered to develop where clasts were segregated laterally from the flow. Low-density pumice clasts were concentrated in the margins and upper parts of pumice-and-ash flow deposits, indicating that effective segregation of clasts of different density occurred within the flows during runout. The fact that the surfaces of deposit margins were notably poor in fine-grained matrix may indicate that clasts were actually segregated from the main body of the flow. Larger-volume and more extensive block-and-ash flows that travelled across unconfined surfaces have a ridge-and-furrow 7 surface morphology, quite distinct from the lobes, levees and channel morphology of most smaller-volume deposits. However, the tapering margins of these unconfined sheet-like deposits were enriched in low-density accidental pumice clasts, tree trunks and the occasional gas canister. We envisage that as large-volume unconfined

DEPOSITS FROM PYROCLASTIC FLOWS

Fig. 20. (a) Scouring of the southern side of Paradise Ghaut just south of Harris village caused by the 25 June 1997 block-and-ash flows (see Fig. 4b for location), (b) Chute eroded into Galway's Wall formed during a fourhour dome collapse on 31 March 1997.

257

258

P. D. COLE ET AL.

Fig. 21. (a) Map showing growth of the Tar River fan: (b) profiles through the Tar River fan (location of profiles shown in a): (c) map showing growth of the White River fan. Contours in feet and isobaths in metres.

block-and-ash flows decelerated in the last few hundred metres, low-density objects and clasts that were picked up during runout were retained in the upper and leading part of the block-and-ash flow (Fig. 22a). We infer that the lower part of the block-andash flow became emplaced together with the large boulders, whereas the upper part continued moving (Fig. 22b). This resulted in the upper part of the block-and-ash flow moving further than the lower part, and the low-density fragments that were retained in the upper part of the flow becoming concentrated in the deposit margins (Fig. 22c). We infer that the ridge-and-furrow surface morphology forms where the upper part of the flow has detached from the lower part of the flow. Remnants of the upper part of the block-and-ash flow were left on the top of boulders, indicating that the moving pyroclastic flow was significantly thicker than the final deposit

Table 5. Volume additions to the Tar River fan Date

Volume addition (m3 x 10 6 )

29 31 July 1996

2.2

11 Aug. 1996

1.2

3 Sep. 1996 17 Sep. 1996

1.4 4.2

16 Jan. 1997 12 Feb. 1997

2.0 1.5 6.0

July 1998

DEPOSITS FROM PYROCLASTIC FLOWS

259

Fig. 22. Diagram summarizing the final stages of transport (c. 0.5 km) and emplacement of large-volume unconfined block-and-ash flow deposit, (a) Flow front entrains low-density material, which is concentrated towards the upper surface and front of the flow, (b) Lower part of the block-and-ash flow decelerates and comes to rest, whereas the upper part, containing the majority of low-density material, continues to move, (c) Final deposit is emplaced. Margins and front of the flow are rich in low-density material; upper surface of flow deposit has ridge-and-furrow morphology caused by motion of upper part of block-and-ash flow. Note deposit left stranded on top of boulders.

Pyroclastic surge development Our studies indicate that development and detachment of pyroclastic surges was unrelated to the volume of the block-and-ash flow deposit, but can be related to several independent factors including topography, temperature and dome pore pressure. Topographic effects such as the depth of valleys can play an important role. For example, deep valleys may contain small-volume block-andash flows and inhibit surge detachment. Such a situation may have

occurred in the White River valley (Fig. 3b, c) and in Tuitt's Ghaut on 5 June 1997 (Fig. 16a). Other small-volume flows, such as the one in the Tar River valley shown in Figure 16c, have a large proportion of detached surge. The Tar River valley has a steeper gradient than Tuitt's Ghaut, which may have promoted development of pyroclastic surges by enhanced entrainment and heating of air at the flow front. Volumes of some pyroclastic flows were such that the whole valley was filled with debris during passage of the flow, and pyroclastic surge detachment could readily

260

P. D. COLE ET AL.

occur unhindered by topography. The dome collapses of 25 June and 21 September 1997 both generated large-volume block-and-ash flows that completely filled the valley during motion. The pyroclastic surge associated with the 25 June 1997 collapse formed 80% by area of the deposits of this event, whereas that of 21 September formed only 40% (see Fig. 15). High pore pressures have been inferred as being important in the development of pyroclastic surges from dome-collapse pyroclastic flows (Sato et al. 1992; Fink & KiefTer 1993; Woods et al. 2002). Collapse of a region of the dome with high internal pore pressure is expected to result in rapid expansion of the gases as the rock mass disintegrates in the collapse. The expanding gas can entrain fine particles, but leaves coarse particles behind due to their inertia (Woods et al. 2002). Thus the resulting flow can become strongly stratified in grain size and concentration, with an upper, expanded, low-concentration part (the pyroclastic surge) and a lower, relatively high-concentration, coarsegrained part (the block-and-ash flow). The efficacy of this process depends strongly on pore pressure, as discussed in more detail by Woods et al. (2002). The gas pressure within the dome will depend on a number of factors, of which the permeability and extrusion rate are the most important. Models estimating gas pressures in the upper part of the conduit (Melnik & Sparks 2002) and in the dome (Woods et al. 2002) indicate typical gas pressure values of a few megapascals at the base of the dome. Woods et al. (2002) also calculated that pressure falls off rapidly only in the outer layers of the dome, within 10-20m of the surface. Larger-volume pyroclastic flows associated with sustained dome collapses involve material from deeper levels within the dome that are likely to have higher pore pressures. Such pore pressures will facilitate rapid expansion and development of surge clouds associated with larger flows (Calder et al. 1999; Woods et al. 2002). In contrast, regions of the dome that have been extruded for sufficient time for the pore pressure to relax to the low ambient pressure are not expected to generate such substantial surge clouds. The absence of a simple relationship between surge development and collapse volume (Fig. 15) indicates that regions of high pore pressure are heterogeneous and not evenly distributed within the dome.

Pyroclastic surge deposition Traction during pyroclastic surge sedimentation is controlled by four factors: velocity, current density, grain size and sedimentation rate. The abundance of massive pyroclastic surge deposits in the examples studied indicates that traction sedimentation was mostly suppressed during deposition. Cross-stratification has been observed only in the 3 July 1998 pyroclastic surge deposits and locally in the pyroclastic density current deposits of 26 December 1997. Sparks et al. (2002) estimate velocities of up to 90 ms- for the 26 December 1997 pyroclastic density current. Three features indicate that pyroclastic surges formed by the 3 July 1998 collapse were more energetic than from other dome collapses. First, the surges nearly overtopped Roche's Mountain, east of the dome, and were the most widespread of any of the pyroclastic surges in the Tar River valley. Second, pumiceous ballistics indicate that some explosive activity was associated with this collapse. Third, sandblasting of trees indicates that these flows had a velocity higher than most other pyroclastic surges, possibly >50ms - 1 . Thus, at Montserrat, cross-stratification was associated with deposits from high-velocity pyroclastic surges. The high velocities would have exerted strong shear stresses on the sedimenting particles. At lower velocities the shear stresses on particles would be less and a high sedimentation rate would bury grains before any lateral movement so as to form massive deposits (Arnott & Hand 1989). Control of massive-bed deposition by variations in grain size and grain density is excluded because of the relative constancy of these parameters in the different deposits. Bipartite pyroclastic surge deposits like those at Montserrat have been widely described from other volcanoes with lava domes, for example Mount St Helens (Hoblitt et al. 1981; Druitt 1992) and Mont Pelee (Boudon & Lajoie 1989; Charland & Lajoie 1989).

At Mont Pelee the origin of a similar lower layer poor in fine ash was attributed partly to flash burning of vegetation (Boudon & Lajoie 1989; Charland & Lajoie 1989). However, formation of identical layered deposits on the surfaces of deposits formed onlyhours earlier at Montserrat (Figs 17c and 18). as well as in areas where vegetation had already been destroyed, precludes burning of vegetation as a general explanation. Similar bipartite layering also occurs in turbidites and is attributed (Gladstone & Sparks 2002) to hydrodynamic effects in the parent gravity currents. A hydrodynamic model related to the fundamental structure of dilutesuspension gravity currents can explain the bipartite layering. Druitt (1992) attributed the paucity of fines in the lower layer Al of the Mount St Helens lateral blast deposit to a combination of residual turbulence and gas escape as particles sedimented. We favour a similar model in which the advancing front of the pyroclastic surge intermittently expands and moves ahead of the main part of the pyroclastic flow and. together with the upward turbulent motion that exists in a gravity current head, flushes out fines. This segregation of fines due to upward flow of turbulent eddies need not be exclusively related to the ingestion and expansion of cold air. but might be an intrinsic property of the head of the current itself, enhanced by factors such as sudden steepening of topography.

Erosion Dome-collapse block-and-ash flows have been described as characteristically monolithologic (e.g. Cas & Wright 1987. p. 111). Pyroclastic flows at Montserrat, however, show evidence of pronounced erosion. Significant quantities of lithic material were picked up and consequently the resulting block-and-ash flow deposits were commonly heterolithic. which was particularly evident when the incorporated lithic materials were hydrothermally altered rocks. On Montserrat seemingly monolithologic deposits may still contain significant quantities of eroded material as the prehistoric deposits are petrologically similar to those formed in the current eruption. Erosion and the incorporation of lithic fragments have imporant implications, as apparently juvenile clasts may be from older deposits or those formed earlier during the same eruption. Incorporation of cold clasts by erosion, together with air entrainment. may help explain some low emplacement temperatures of the Soufriere Hills deposits that were substantially lower than the dome temperature (850 C).

Conclusions (1)

Two mechanisms formed pyroclastic flows between 1996 and 1999 at Soufriere Hills Volcano: (1) dome collapse, and (2) fountain collapse during Vulcanian explosions. (2) Dome collapses produced block-and-ash flows that varied between 1 and 7km in runout. Three block-and-ash flow deposit geometries were identified: small lobate deposits at the base of the dome, valley-filling block-and-ash flow deposits, and sheet-like deposits across a wide range of volumes. Temperatures of block-and-ash flow deposits ranged from 365 to 640 C. (3) Fountain-collapse pumice-and-ash flows had runouts typically between 3 and 6 km and formed distinctly lobate deposits. The high-concentration pumice-and-ash flows segregated from turbulent, collapsing mixtures within 2 km of the dome. Efficient density segregation of clasts occurred within the pumice-and-ash flows such that deposit margins were enriched in low-density pumice. Pumice-and-ash flows were emplaced at around 200 C. (4) The extent of detachment of pyroclastic surges from blockand-ash flows was not simply related to volume. Crucial in surge detachment are topographic confinement and the slope gradient across which the surse travels. Another factor is the

DEPOSITS FROM PYROCLASTIC FLOWS

pore pressure of the collapsed dome material, which varies as a function of depth of origin in the dome, extrusion rate, dome permeability and residence time in the dome. Collapse of regions of the dome with high pore pressure can generate substantial surge clouds. (5) Pyroclastic surge deposits studied that formed from the larger dome collapses were typically massive in cross-section. Rapid sedimentation from turbulent suspension may have supressed traction. Most pyroclastic surge deposits show bipartite layering, with a lower, coarser layer poor in fine ash and an upper, ash-rich layer. The lower layer is considered to have been deposited from the front of the current where turbulence caused selective sedimentation mainly of coarse material. (6) Erosion by block-and-ash flows, both of prehistoric lithified deposits and of deposits formed earlier during the eruption, was common. The eroded material was capable of significantly cooling block-and-ash flows and gave rise to deposits that are not monolithologic. This work was supported by the Department for International Development. We thank helicopter pilots J. MacMahon and A. Grouchy of St Lucia Helicopters and the pilots of Bajan Helicopters whose skill allowed many of the observations described in this paper to be made. We express thanks to the staff of the MVO who assisted this study in many ways. R.S.J.S. acknowledges a NERC Professorship. Thorough and detailed reviews by A. Freundt, M. Branney and P. Kokelaar improved this paper considerably. This work is published with the permission of the Director of the British Geological Survey.

References ABDURACHMAN, E. K. BOURDIER, J-L. & VOIGHT, B. 2000. Nuees ardentes of 22 November 1994 at Merapi volcano, Java, Indonesia. Journal of Volcanology and Geothermal Research, 100, 345-361. ARNOTT, R. W. C. & HAND, B. M. 1989. Bedforms, primary structures and grain fabric in the presence of suspended sediment rain. Journal of Sedimentary Petrology, 59, 1062-1069. BANKS, N. & HOBLITT, R. 1981. Summary of temperature studies of 1980 deposits. In LIPMAN, P. W. & MULLINEUX, D. R. (eds) The 1980 Eruptions of Mount St Helens, Washington. US Geological Society, Professional Papers, 1250, 295-314. BONADONNA, C., MAYBERRY, G. C., CALDER, E. S. ETAL. 2002. Tephra fallout in the eruption of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 483-516. BOUDON, G. & LAJOIE, J. 1989. The 1902 Peleean deposits in the Fort Cemetery of St. Pierre, Martinique: a model for the accumulation of turbulent nuees ardentes. Journal of Volcanology and Geothermal Research, 38, 113-129. BOUDON, G., CAMUS, G., GOURGAUD, A. & LAJOIE, J. 1993. The 1984 nueeardente deposits of Merapi Volcano, Central Java, Indonesia: stratigraphy, textural characteristics and transport mechanisms. Bulletin of Volcanology, 55, 327-343. CALDER, E. S., COLE, P. D., DADE, W. B. ET AL. 1999. Mobility of pyroclastic flows and surges at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 26, 537-540. CALDER, E. S., LUCKETT, R., SPARKS, R. S. J. & VOIGHT, B. 2002. Mechanisms of lava dome instability and generation of rockfalls and pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 173-190. CAS, R. A. F. & WRIGHT, J. V. 1987. Volcanic Successions: Modern and Ancient. Allen and Unwin, London. CHARLAND, A. & LAJOIE, J. 1989. Characteristics of pyroclastic deposits at the margin of Fond Canonville, Martinique, and its implications for the transport of the 1902 nuees ardentes of Mt. Pelee. Journal of Volcanology and Geothermal Research, 38, 97-112. CLARKE, A. B. & VOIGHT. B. 2000. Pyroclastic current dynamic pressure from aerodynamics of tree or pole blow-down. Journal of Volcanology and Geothermal Research, 100, 395-412

261

COLE, P. D., CALDER, E. S., DRUITT, T. H., HOBLITT, R., ROBERTSON, R., SPARKS, R. S. J. & YOUNG, S. R. 1998. Pyroclastic flows generated by gravitational instability of the 1996-97 lava dome of Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3425-3428. DEVINE, J. D., MURPHY, M. D., RUTHERFORD, M. J., BARCLAY, J., SPARKS, R. S. J., CARROLL, M. R., YOUNG, S. R. & GARDNER, J. E. 1998. Petrologic evidence for pre-eruptive pressure and temperature conditions and recent reheating of andesitic magma erupting at the Soufriere Hills Volcano Montserrat. Geophysical Research Letters, 25, 3669-3672. DRUITT, T. H. 1992. Emplacement of the 18 May 1980 lateral blast deposit ENE of Mount St Helens, Washington. Bulletin of Volcanology, 54, 554-572. DRUITT, T. H., CALDER, E. S., COLE, P. D. ET AL. 2002a. Small-volume, highly mobile pyroclastic flows formed by rapid sedimentation from pyroclastic surges at Soufriere Hills Volcano, Montserrat: an important volcanic hazard. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 263-279. DRUITT, T. H., YOUNG, S. R., BAPTIE, B. ETAL. 20026. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306. FINK, J. H. & KIEFFER, S. W.1993. Estimates of pyroclastic flow velocities from explosive decompression of lava domes. Nature, 363, 612-615. FISHER, R. V. 1977. Erosion by volcanic base-surge density currents: Ushaped channels. Geological Society of America Bulletin, 88, 1287-1297. FISHER, R. V. & HEIKEN, G. 1982. Mt Pelee Martinique: May 8 and 20 1902 pyroclastic flows and surges. Journal of Volcanology and Geothermal Research, 13, 339-371. FRANCIS, P. W., ROOBOL, M. J., COWARD, M. P., COBBOLD, P. R. & WALKER, G. P. L. 1974. San Pedro and San Pablo volcanoes and their hot avalanche deposits. Geologiche Rundchau, 63, 357-388. FUJI, T. & NAKADA, S. 1999. The 15 September 1991 pyroclastic flows at Unzen Volcano (Japan): a flow model for associated ash-cloud surges. Journal of Volcanology and Geothermal Research, 89, 159-172. GLADSTONE, C. & SPARKS, R. S. J. 2002. The significance of grain size breaks in turbidite beds. Journal of Sedimentary Research, in press. GRUNEWALD, U., SPARKS, R. S. J., KEARNS, S. & KOMOROWSKI, J. C. 1999. Friction marks on blocks from pyroclastic flows at the Soufriere Hills Volcano, Montserrat: implications for flow mechanisms. Geology, 28, 827-830. HOBLITT, R. P., MILLER, C. D. & VALLANCE, J. W. 1981. Origin and stratigraphy of the deposits produced by the May 18 directed blast. In: LIPMAN, P. W. & MULLINEUX, D. R. (eds) The 1980 Eruptions of Mount St Helens, Washington. US Geological Survey, Professional Papers, 1250,401-419. KEIFFER, S. W. & STURTEVANT, B. 1988. Erosional furrows formed during the lateral blast at Mount St Helens, May 18 1980. Journal of Geophysical Research, 93, 14793-14816. KELFOUN, K., LEGROS, F. & GOURGAUD, A. 2000. A statistical study of trees damaged by the 22 November 1994 eruption of Merapi volcano (Java, Indonesia): relationships between ash-cloud surges and blockand-ash flows. Journal of Volcanology and Geothermal Research, 100, 379-393. LOUGHLIN, S. C., BAXTER, P. J., ASPINALL, W. P., DARROUX, B., HARFORD, C. L. & MILLER, A. D. 2002a. Eyewitness accounts of the 25 June 1997 pyroclastic flows and surges at Soufriere Hills Volcano, Montserrat. and implications for disaster mitigation. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London. Memoirs, 21, 211-230. LOUGHLIN, S. C., CALDER, E. S., CLARKE, A. B. ETAL. 2002b. Pyroclastic flows and surges generated by the 25 June 1997 dome collapse, Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 231-262. MELNIK, O. & SPARKS, R. S. J. 2002. Dynamics of magma ascent and lava extrusion at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 153-171. MIYABUCHI, Y. 1999. Deposits associated with the 1990-1995 eruption of Unzen Volcano, Japan. Journal of Volcanology and Geothermal Research, 89, 139-158.

262

P. D. COLE ET AL.

MURPHY, M. D., SPARKS, R. S. J., BARCLAY, J., CARROLL, M. R. & BREWER, T. S. 2000. Remobilization origin for andesite magma by intrusion of mafic magma at the Soufriere Hills volcano, Montserrat, WI. Journal of Petrology, 41, 21-42. NORTON, G. E., WATTS, R. B., VOIGHT, B. ETAL. Pyroclastic flow and explosive activity at Soufriere Hills Volcano, Montserrat, during a period of virtually no magma extrusion (March 1998 to November 1999). In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 467-481. RITCHIE, L. J., COLE, P. D. & SPARKS, R. S. J. 2002. Sedimentology of deposits from the pyroclastic density current of 26 December 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 435-456. ROBERTSON, R., COLE, P. D., SPARKS, R. S. J. ET AL. 1998. The explosive eruption of Soufriere Hills Volcano, Montserrat, 17 September 1996. Geophysical Research Letters, 25, 3429-3432. RODRIGUEZ-ELIZARRARAS, S., SIEBE, C, KOMOROWSKI, J-C., ESPINDOLA, JM. & SAUCEDO, R. 1991. Field observations of pristine block-and ashflow deposits emplaced April 16-17, 1991 at Volcan Colima, Mexico. Journal of Volcanology and Geothermal Research, 48, 399-412. ROWLEY, P. D., KUNTZ, M. A. & MACLEOD, N. S. 1981. Pyroclasticflow deposits. In: LIPMAN, P. W. & MULLINEUX, D. R. (eds) The 1980 Eruptions of Mount St. Helens, Washington. US Geological Society, Professional Papers, 1250, 489-512. SATO, H., FUJI, T. & NAKADA, S. 1992. Crumbling dacite dome lava and generation of pyroclastic flows at Unzen volcano. Nature, 360, 664-666. SPARKS, R. S. J. & WALKER, G. P. L. 1977. The significance of vitric enriched airfall ashes associated with crystal enriched ignimbrites. Journal of Volcanology and Geothermal Research, 2, 329-341.

SPARKS, R. S. J., GARDEWEG, M., CALDER, E. S. & MATTHEWS, S. 1997. Erosion by pyroclastic flows at Lascar volcano. Chile. Bulletin of Volcanology, 58, 557-565. SPARKS, R. S. J., YOUNG, S. R.. BARCLAY. J. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills Volcano. Montserrat: November 1995 to May 1997. Geophysical Research Letters, 25, 3421-3424. SPARKS, R. S. J., BARCLAY, J., CALDER, E. S. ETAL. 2002. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21, 409-434. VALENTINE, G. A. 1998. Damage to structures by pyroclastic flows and surges. Journal of Volcanology and Geothermal Research, 87, 117-140. VOIGHT, B., KOMOROWSKI, J.-C. NORTON. G. E. ETAL. 2002. The 26 December (Boxing Day) 1997 sector collapse and debris avalanche at Soufriere Hills Volcano, Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London. Memoirs, 21. 363-407. WATTS, R. B., HERD. R. A., SPARKS, R. S. J. & YOUNG, S. R. 2002. Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano, Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21. 115-152. WOODS, A. W., SPARKS, R. S. J., RITCHIE. L. J.. BATEY, J., GLADSTONE, C. & BURSIK, M. 2002. The explosive decompression of a pressurized lava dome: the 26 December (Boxing Day) 1997 collapse and explosion of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21. 457-466.

Small-volume, highly mobile pyroclastic flows formed by rapid sedimentation from pyroclastic surges at Soufriere Hills Volcano, Montserrat: an important volcanic hazard T. H. DRUITT1, E. S. CALDER2, P. D. COLE3, R. P. HOBLITT4, S. C. LOUGHLIN 5 , G. E. NORTON5, L. J. RITCHIE3, R. S. J. SPARKS2 & B. VOIGHT7 1 Laboratoire Magmas et Volcans (UMR 6524 & CNRS), Universite Blaise Pascal, 5 rue Kessler, 63038 Clermont-Ferrand, France (e-mail: [email protected]) 2 Department of Earth Sciences, University of Bristol, Queens Road, Bristol BS8 1RJ, UK 3 Centre for Volcanic Studies, University of Luton, Park Square, Luton LU1 3JU, UK 4 David A. Johnston Cascades Volcano Observatory, US Geological Survey, 5400 Mac Arthur Boulevard, Vancouver, WA 98661, USA 5 British Geological Survey, Murchison House, West Mains Road, Edinburgh EH9 3LE, UK 6 British Geological Survey, Keyworth, Nottingham NG12 5GG, UK 1 Department of Geosciences, Penn State University, University Park, PA 16802, USA

Abstract: Gravitational collapses of the lava dome at Soufriere Hills Volcano on 25 June and 26 December 1997 generated pyroclastic surges that spread out over broad sectors of the landscape and laid down thin, bipartite deposits. In each case, part of the settling material continued to move upon reaching the ground and drained into valleys as high-concentration granular flows of hot (120-410°C) ash and lapilli. These surge-derived pyroclastic flows travelled at no more than 1 0 m s - 1 but extended significantly beyond the limits of the parent surge clouds (by 3 km on 25 June and by 1 km on 26 December). The front of the 25 June flow terminated in a valley about 50m below a small town that was occupied at the time. Despite their small deposit volumes (5-9 x 10 4 m 3 ), the surge-derived pyroclastic flows travelled as far as many of the Soufriere Hills block-and-ash flows on slopes as low as a few degrees, reflecting a high degree of mobility. An analysis of the deposits from 26 December suggests that sediment accumulation rates of at least several millimetres per second were sufficient to generate pyroclastic flows by suspended-load fallout from pyroclastic surges on Montserrat. Surge-derived pyroclastic flows are an important, and hitherto underestimated, hazard around active lava domes. At Montserrat they formed by sedimentation over large catchment areas and drained into valleys different from those affected by the primary block-and-ash flows and pyroclastic surges, thereby impacting areas not anticipated to be vulnerable in prior hazards analyses. The deposits are finer-grained than those of other types of pyroclastic flow at Soufriere Hills Volcano; this may aid their recognition in ancient volcanic successions but, along with valleybottom confinement, reduces the preservation potential.

Pyroclastic surges are turbulent density currents of volcanic particles and gas that can travel at high velocities. They are commonly associated with block-and-ash pyroclastic flows formed by gravitational collapses of lava domes. Pyroclastic surges are hazardous due to their ability to traverse topographic barriers and interfluves, and are a serious concern during all lava dome eruptions (Yamamoto et al. 1993; Fisher 1995; Fujii & Nakada 1999). The high speeds and commonly high temperatures of pyroclastic surges render them devastating to life, vegetation and buildings up to several kilometres from the dome (Valentine 1998; Kelfoun et al. 2000; Voight & Davis 2000). In 1902, pyroclastic surges at Mont Pelee (Martinique) swept over the town of St Pierre, killing 28 000 people. More recently, they claimed 43 lives at Mount Unzen (Japan) in 1991 and 65 at Merapi Volcano (Java) in 1994 (Yamamoto et al. 1993; Abdurachman et al. 2000). Nineteen people were killed, and seven seriously injured, by block-and-ash flows and associated pyroclastic surges on Montserrat on 25 June 1997. The eruption of Soufriere Hills Volcano has highlighted an additional hazard from pyroclastic surges. Hundreds of dome collapses have occurred during the eruption due to retrogressive gravitational failures involving 104 to 108 m3 of andesite and lasting for periods of several minutes to hours (Calder et al. 1999, 2002; Cole et al. 1998, 2002). These collapses generated block-and-ash flows with accompanying pyroclastic surges. The most voluminous and fastest-moving surges formed during large collapses, involving young, pressurized parts of the dome and rapid decompression of entrapped gases, and spread out over broad sectors of southern Montserrat. Collapses of relatively old, degassed dome andesite, on the other hand, generated block-and-ash flows with relatively weak surge components. During two of the largest dome collapses (25 June and 26 December 1997) suspended-load fallout from pyroclastic surges generated high-concentration, granular flows of ash and lapilli, which in the former case travelled 3km beyond the limit of the parent surge. Another example occurred during a

collapse on 12 May 1996 but was much smaller, travelling only a few hundred metres (Cole et al. 2002). An important feature of these surge-derived pyroclastic flows is that they formed by sedimentation over large catchment areas and were able to drain into different valleys from those impacted by the main block-andash flows and pyroclastic surges. They thus affected areas not anticipated to be vulnerable in prior hazards analyses, and in one case terminated close to an occupied town. Given their small volumes compared with typical block-and-ash flows of Soufriere Hills Volcano, the hot surge-derived flows were notable for their long runout distances and for their ability to travel along drainages with gradients of only a few degrees. This combination of properties makes surge-derived pyroclastic flows a potentially important hazard around active lava domes. The aim of this paper is to describe the deposits and effects of the surge-derived pyroclastic flows, to use field measurements and calculations to constrain the processes of formation and transport, and to discuss hazard implications. Special attention is paid to one of the surge-derived flows generated on 26 December 1997, SE of the lava dome. This example was studied in most detail because the area was not threatened by later activity and fieldwork could be carried out in relative safety. In contrast, the example of 25 June 1997 was less well studied, as the deposits were rapidly buried and/or eroded by subsequent pyroclastic flows and lahars, and fieldwork was more hazardous.

Summary of the eruption during 1995-1999 The eruption of Soufriere Hills Volcano began with a phreatic phase on 18 July 1995 (Robertson et al. 2000; Gardner & White 2002). The lava dome first appeared on 15 November 1995, situated on an old dome in an ancient sector-collapse scar 1 km across called

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 263-279. 0435-4052/02/S15 © The Geological Society of London 2002.

263

264

T. H. DRUITT ET AL.

Fig. 1. Map of southern Montserrat showing place names and the total area of impact by pyroclastic flows and pyroclastic surges during the 1995-1999 eruptive period.

English's Crater (Fig. 1). Throughout 1996 and the first part of 1997 the dome was contained within the collapse scar, and domecollapse block-and-ash flows were channelled down the Tar River valley to the sea, where they built a delta. On the night of 17 September 1996, a large dome collapse triggered an explosive eruption (Robertson el al. 1998). In late March 1997, block-and-ash flows began to be shed southwards into the White River valley, and then in June westwards into Gages valley and northwards into Mosquito and Tuitfs Ghauts (Fig. 1). On 25 June, part of the dome collapsed down the northern flank of the volcano, and the resulting pyroclastic flows and surges killed 19 people and generated one of the surge-derived pyroclastic flows described below. Between 4 and 12 August there occurred 13 Vulcanian explosions and, following a large dome collapse to the north on 21 September 1997, another 75 explosions between 22 September and 21 October (Druitt el al. 2002). Large collapses occurred down the White River valley on 4 and 6 November. On 26 December 1997, the southern wall of English's Crater failed gravitationally, forming a debris avalanche (Sparks el al. 2002; Voight el al. 2002). At the same time a pyroclastic surge swept out across a 70 sector, devastating 10km 2 of southern Montserrat (Ritchie el al. 2002). Sedimentation from this surge generated surge-derived pyroclastic flows in several drainages.

Collapse of 25 June 1997

Chronology and events The dome collapse of 25 June 1997 generated block-and-ash flow and pyroclastic surge deposits with a total volume of 6.4 x 10 6 m 3 , or about 4.9 x 10 6 m 3 dense-rock equivalent (DRE; Calder el al. 2002). The collapse and ensuing events are described in detail by Loughlin el al. (2002 a,b), and the deposits and effects are described by Cole el al. (1998. 2002). Prior to collapse, the dome had a volume of 68 x 10 6 m 3 (Calder el al. 2002). Strong cyclic swarm seismicity and edifice deformation began on 22 June and increased in duration and intensity until, on 24 and 25 June, they merged into almost continuous tremor at peak edifice inflations (Voight el al. 1999). The main collapse and associated seismic signal on 25 June commenced during rapid deflation of the dome at about 12:55 local time (LT, four hours behind universal time). The seismic signal lasted about 25 minutes, with three particularly intense pulses starting at 12:57, 13:00 and 13:08 LT (Loughlin el al. 2002b). These three seismic pulses are believed to have coincided with the generation of three block-and-ash flows observed on time-lapse video and by eyewitnesses. The flows travelled down the deeply entrenched

SMALL PYROCLASTIC FLOWS: A HAZARD

265

Fig. 2. Map of the area impacted by pyroclastic flows and surges generated by the dome collapse of 25 June 1997.

Mosquito Ghaut and the Paradise River valley, almost reaching the sea and the third spilling into Spanish Point. Pyroclastic surges associated with flows 2 and 3 swept out successively across Farrell's plain and that of flow 3 passed over the village of Streatham and climbed a vertical height of 70m up Windy Hill (Fig. 2). Surge movement was mainly to the north and NW, as shown by treeblowdown directions, the orientation of charred surfaces, shadow zones behind houses, and the transport of a water tank (Fig. 2). Nineteen people died from the pyroclastic flows and surges, and seven others were seriously injured. The ash plume formed by lofting of the pyroclastic surges and by elutriation from the pyroclastic flows rose to a height of 10km within a few minutes. Surge-derived pyroclastic flow and its parent pyroclastic surge The 25 June 1997 surge-derived pyroclastic flow was generated by suspended-load fallout from the pyroclastic surges associated with pyroclastic flows 2 and 3 (Fig. 2). Surge-derived pyroclastic flow deposits formed in all the main distributory valleys feeding into the upper part of Dyer's River valley, indicating that the flow was fed from a wide area. High-concentration granular flows of ash and lapilli are inferred to have drained in a fluid-like manner off Farrell's plain westwards into Tyre's Ghaut and northwestwards into the headwaters of Dyer's River valley. Eyewitnesses on the road a few hundred metres east of Streatham saw ground-hugging flows of ash travelling downhill to the west along the main road almost at 90° to the original flow direction of the surge. They commented on how the ash was confined to the road, was hugging the ground, was 'boiling' and how it 'moved round the bends on the road like a vehicle' (Loughlin et al. 2002a). Once in Dyer's River valley, the multiple flows of ash and lapilli merged into a single surge-derived pyroclastic

flow, which then travelled 3km down Dyer's River and Belham River valleys, around a series of tight meanders, before terminating about 50m below the small town of Cork Hill (Figs 2 and 3b). The volume of the surge-derived flow deposit was estimated as 90000m 3 (non-DRE), or about 1.5% of the entire 25 June deposit volume of 6.4 x 106 m3 (Calder et al. 2002). The pyroclastic surge deposit, typically 5-20 cm thick, was examined on Farrell's plain one month after emplacement. A representative section (location 1, Fig. 2) is shown in Figure 3A. Two layers of similar thickness were present. The lower one (layer 1) consisted of grey, friable, fines-poor ash and lapilli that thickened into depressions and thinned onto highs, maintaining an even upper surface. Where emplaced onto furrowed fields, layer 1 thickened into the furrows and was commonly absent on the ridges. Clots of sheared soil and vegetation were common in layer 1 and rare blocks up to 10cm were present. Layer 2 consisted of brown ash richer in fine-grained components than layer 1 and tinged orange at the top by in situ oxidation. The thickness of layer 2 was approximately constant in any section, and there were segregation pipes rooted on fragments of vegetation. The whole deposit thickened and coarsened into drainages and topographic depressions. Near location 1 the deposit on the floor of one drainage several metres deep had a lower, fines-poor layer of blocks and lapilli and an upper layer of lapilli and ash. Trenching at this site suggested that the two layers of the surge deposit were continuous between the plain and drainage facies. It is inferred that both layers were laid down by a single pyroclastic surge, probably that associated with pyroclastic flow 3. Decriptions of the 25 June 1997 surge deposits at other locations are given by Loughlin et al. (2002b). The deposit of the surge-derived pyroclastic flow was examined in July at two locations: Dyer's bridge (location 2, Fig. 2) and the terminus below Cork Hill (location 3, Fig. 2). At Dyer's bridge

266

T. H. DRUITT ET AL.

Fig. 3. (a) The 25 June 1997 pyroclastic surge deposit on Farrell's plain at location 1 (Fig. 2). The deposit has a lower fines-poor layer (layer 1), and an upper layer (layer 2) richer in fine components. Knife for scale, (b) Aerial view of the Belham River valley, looking east, showing the 25 June 1997 surge-derived pyroclastic flow deposit (f) and associated singe zone (s). The deposit had already been partially eroded when this photo was taken. Gages Mountain (g) and the village of Dyers (d) are also marked.

(Figs 4 and 5) the surge-derived flow deposit was generally 0.5-1 m thick and draped the valley to a vertical height of 6m (Fig. 5). Only 10cm of flow deposit occurred on the bridge top surface. The deposit was pink to pale grey and had little internal structure (Fig. 4a). It consisted mainly of ash with subordinate lapilli of dense, juvenile andesite, and was noticeably finer grained than typical block-andash flow deposits of Soufriere Hills Volcano (Cole et al. 1998, 2002).

Blocks up to 10cm diameter made up a few per cent of the deposit, but these were heterolithologic with many altered and weathered rocks, suggesting that they were picked up by the flow further upstream. Concentrations of outsized blocks occurred at the side of the valley upstream of Dyer's bridge (Fig. 4b). The surface of the surge-derived flow deposit was uneven, reflecting the underlying topography. There were fresh exposures of

SMALL PYROCLASTIC FLOWS: A HAZARD

267

Fig. 4. (a) Surge-derived pyroclastic flow deposit (f) of 25 June 1997 near Dyers bridge (location 2, Fig. 2), showing the characteristically finegrained nature with sparse lapilli. The deposit is overlain by a thin lahar deposit (1). The scale is in inches (1 inch = 2.54 cm), (b) Concentration of blocks at the margin of the 25 June 1997 surge-derived pyroclastic flow deposit (f) at Dyers bridge. Person for scale.

the deposit in scar faces with dips of 30-60° towards the drainage axis. The scars were interpreted as the headwalls of breakaway avalanches formed immediately after deposition, as reported from pyroclastic flow deposits at Mount St Helens (Rowley et al. 1985) and Mount Pinatubo (Torres et al. 1997). Corrugated iron roofing material was wrapped around parts of the bridge and around trees at the valley sides, showing that the active flow had occupied the whole valley to a depth of 5-6 m (Fig. 5). The surge-derived flow at Dyer's bridge did not break branches off trees more than 1 cm in diameter. Many burnt logs and branches in the flow must have been derived from further upstream where the flow was more energetic. The associated pyroclastic surge deposit at this location reached at least 15m higher up the valley sides than the parent surgederived flow deposit and was 5-15 cm thick. It was composed of fine ash and contained carbonized twigs. The surge had singed the vegetation to heights of 10-15m above the valley floor in this area, but quite delicate twigs and leaves remained on the trees, suggesting that the surge cloud was travelling very slowly. There was a sharp grain-size break between the surge-derived flow deposit and the finer-grained pyroclastic surge deposit at 5-6 m above the valley

floor, showing that the surge-derived flow had a well defined upper surface. The trees were strongly burnt up to the level of the grainsize break. Leaves had been stripped from the trees to only 1 m above the top surface of the surge-derived flow deposit. At the distal limit of the surge-derived flow, in the valley below Cork Hill (location 3, Fig. 2), the deposit was flat-topped and 0.5-1 m thick, smoothing over the boulder-strewn valley bottom. Vegetation and trees were not singed above the top of the flow and leaves were not removed. The deposit at the terminus was finegrained, with sparse lapilli up to 1 cm in diameter. The singe zone from the associated surge cloud extended only about 3 m above the surface of the flow deposit. The terminus of the surge-derived flow was marked by a log jam oriented perpendicular to the valley axis. The temperature measured using a thermocouple at 30-3 5 cm depth in the surge-derived flow deposit was 410°C at Dyer's bridge and 350°C at the terminus, both five days after emplacement. The observations at Dyer's bridge (location 2) and at Cork Hill (location 3) imply the existence of a hot, high-concentration pyroclastic flow with a well defined upper surface and a weak overriding pyroclastic surge cloud. The surge at these locations is

268

T. H. DRUITT ET AL.

Fig. 5. Schematic cross-section through the surge-derived pyroclastic flow deposit of 25 June at Dyers bridge (location 2. Fig. 2).

attributed mainly to ash elutriation from the surge-derived flow, but may also have been in part a remnant of the more powerful parental surges further upstream. Both the surge-derived flow and accompanying surge cloud are inferred to have travelled slowly, as there were no significant superelevation effects on bends. Most trees remained standing, even where their bases were immersed in the deposit, and the flow left two concrete bridges intact. At peak discharge, the flow moved down the valley as a wave up to 6m deep, as shown from the relationships at Dyer's bridge, leaving behind it a deposit 0.5 to 1.0m thick draping the valley floor and sides as it spread out downstream. As the flow drained the valley axis, immediate remobilization of the flow deposit generated small breakaway scars. The fine-grained character of the surge-derived

pyroclastic flow deposit is attributed to its origin by suspended-load fallout from pyroclastic surges. Related volcanic hazards Pyroclastic flows in the Belham River valley were not expected from hazards analyses carried out prior to the 25 June 1997 collapse. Since the dome had most recently been growing on its northern side, and since block-and-ash flows throughout June had travelled northeastwards, towards the airport, the areas to the north and NE of the volcano were considered to be those at greatest risk. The town of Cork Hill had been considered relatively safe from impact and was

Fig. 6. Map of the area impacted by the pyroclastic density currents generated following the sector collapse of 26 December 1997.

SMALL PYROCLASTIC FLOWS: A HAZARD

still inhabited at the time of the 25 June collapse. As it turned out, the surge-derived pyroclastic flow terminated in the Belham River valley about 50m below Cork Hill and within about 400m of the village school with an estimated 200 children in it. A child was burnt while approaching the new deposits (Fig. 2). This incident highlights the fact that surge-derived pyroclastic flows can move in unexpected directions, well outside the area of the parent surge cloud and any associated block-and-ash flows.

Collapse of 26 December 1997 Chronology and events The collapse of 26 December began at 03:01 LT and the peak seismic signal lasted 11.6 minutes (Sparks et al. 2002; Voight et ai 2002). Failure of the southern flanks of the volcano generated a debris avalanche that travelled down the White River valley to the south coast (Fig. 6). The collapse was followed by a lateral blast and pyroclastic density current, which swept out across a 70°, 10km 2 sector south of the volcano and into the sea. The pyroclastic density current occurred in two pulses: a first, very powerful pulse and a second, less powerful one, forming a deposit with two normally graded layers (Ritchie et al. 2002). In total, the collapse and lateral blast involved about 46 x 10 6 m 3 of hydrothermally altered flank rocks and dome talus and about 35-45 x 106 m3 of the lava dome (Sparks et al. 2002; Voight et al. 2002). In many respects the 26 December pyroclastic density current resembled those generated by other lateral blasts, such as Mont Pelee in 1902 (Boudon & Lajoie 1989), Mount St Helens in 1980 (Hoblitt et al. 1981; Fisher 1990; Druitt 1992) and Bezymianny in 1956 (Belousov 1996). The general term pyroclastic density current is used to describe the flow generated by the 26 December 1997 lateral blast in order to emphasize the existence within it of strong vertical and flow-transverse gradients in grain size and particle concentration (Sparks et al. 2002; Ritchie et al. 2002). It differed from the pyroclastic surges from other collapses such as 25 June 1997 in that it transported large quantities of decimetre- or even metre-sized debris, particularly in the higher-concentration lower part. The events on 26 December 1997 have been reconstructed as follows. Gravitational collapse of the southern wall of English's Crater, with failure in hydrothermally altered rocks associated with Galway's Soufriere, generated a debris avalanche that travelled 4km to the south coast. Decompression of the dome interior then caused explosive expansion of high-pressure gases and production of a fast-moving, density-stratified pyroclastic current (Woods et al. 2002). Much of the basal, higher-concentration part of the current

Fig. 7. Map of Dry Ghaut showing the deposits and effects of the pyroclastic surge and surge-derived pyroclastic flow of 26 December 1997. Locations 1 to 3 are referred to in the text.

269

was channelled down the White River valley to the sea, while the upper, more dilute part spread out across the landscape. In this paper we refer to this upper, more dilute part as a pyroclastic surge. Surge-derived pyroclastic flows formed in all drainages inundated by the upper, more dilute part of the density current, including substantial ones down Germans and Gingoes Ghauts (Fig. 6). The deposits are referred to as block-poor pyroclastic density current facies by Ritchie et al. (2002). The surge-derived flow we consider below formed where part of the expanding surge swept over the saddle separating the White River valley from Dry Ghaut (Fig. 6). This isolated it from the main impact zone, simplifying field relationships and making the deposit and flow effects amenable to study. The Dry Ghaut surge-derived pyroclastic flow and its parent pyroclastic surge As the pyroclastic surge descended Dry Ghaut, rapid suspendedload fallout formed a high-concentration granular flow of ash and lapilli. Figure 7 shows the area of surge impact and the surgederived pyroclastic flow deposit. The volume (non-DRE) of the surge deposit in Dry Ghaut was estimated to be about 1.5 x 105 m3 and that of the surge-derived flow about 0.5 x 105 m3. The map also distinguishes a zone of tree blowdown in which many trees were toppled and/or removed by the surge, and a singe zone in which most trees were killed but remained standing. The height of the top of the singe zone above the valley floor decreased progressively downstream (Fig. 8). For the purposes of discussion the ghaut is divided into three regions: upper (<0.6km from the head of the ghaut), middle (0.6 to 2km) and lower (>2km). Upper region of Dry Ghaut (0-0.6km). The upper region of Dry Ghaut consists of a broad valley that becomes progressively constricted downstream (Figs 7 and 9). The top of the singe zone in this region lay about 100m above the ghaut floor, reaching a higher elevation on the south side than on the north, probably due to centrifugal effects as the pyroclastic surge swept from the north over the saddle separating the White River valley from Dry Ghaut. The pyroclastic surge deposit draped the ghaut sides to a thickness of a few centimetres to a few tens of centimetres. Grain size and thickness were greatest at the head of the ghaut and decreased rapidly downstream. At most locations the deposit overlay, in erosive contact, a few centimetres of buff-coloured ash with scattered pumice lapilli produced by the Vulcanian explosions of August, September and October 1997. Like the deposit of the

270

T. H. DRUITT ET AL.

Fig. 8. Profile down Dry Ghaut showing how the pyroclastic surge singe line on the northern and southern ghaut walls fall downstream due to a combination of surge deceleration, decrease in surge bulk density through sedimentation, buoyant lofting and inflow of ambient air. The pyroclastic surge cloud was very weak beyond about 2 km downstream of the ghaut entrance. Superelevation effects are observed where the pyroclastic surge turned bends in the ghaut. The sketches depict schematically an interpretation of the development of the pyroclastic surge (grey) and surge-derived pyroclastic flow (black) down the valley. Rapid suspended-load fallout in the upper ghaut caused the surge to become density stratified. By the middle ghaut, segregation of the lower, more concentrated part of the pyroclastic surge had generated a welldefined surge-derived pyroclastic flow.

25 June 1997 pyroclastic surge, the surge deposit in Dry Ghaut had two principal layers of comparable thickness. Layer 1 consisted of coarse, fines-poor lapilli and blocks (the latter being most abundant near the head of the ghaut, where blocks up to 15 cm occurred), and was laterally discontinuous, being thickest in depressions and in the lee of obstacles. The base of layer 1 was commonly marked by a zone of intense shearing, with surge deposit, clots of soil and Vulcanian ash, and charred vegetation mixed together. Overall, layer 1 was normally or inverse-to-normally graded. Layer 2 consisted of brown ash with scattered lapilli. The contact between the two layers was typically very sharp, although locally a more gradational contact showed that both layers were the product of a single depositional event. A third layer of thin fallout ash up to few millimetres thick capped the surge deposit. At some sites in the upper ghaut, two discrete surge units were observed, each with bipartite layering. Two surge units were also preserved widely across the main impact zone south of the dome (termed Units I and II by Ritchie el al 2002). Unit I was thicker, coarser-grained, and more extensive than Unit II, from which we infer that the main surge unit in Dry Ghaut was probably Unit I. Some repetition of layering in Dry Ghaut may also record local splitting of the surge around topographic irregularities, as occurred at Mount St Helens (Kieffer 1981; Druitt 1992). The surge deposit in upper Dry Ghaut thickened to 1 m or more in pre-existing drainage channels on the ghaut floor. The same twolayer stratigraphy was present in the channel facies, but both layers were thicker and coarser-grained than outside the channels. A typical section through channel fills consisted of a grey to mauve, finespoor layer of blocks and lapilli a few decimetres thick overlain by a few decimetres of brown ash and lapilli with matrix-supported blocks. Trenching confirmed that these layers passed without break into layers 1 and 2 respectively of the thin surge deposit outside the drainage channels. Locally, clast-supported accumulations of blocks and lapilli were observed at the bases of the drainage fills. Although we have mapped the drainage facies in the upper ghaut as surgederived flow deposit (Fig. 7), it is essentially overthickened surge deposit, which passes downstream into the true surge-derived flow deposit in a manner that could not be studied in detail during our fieldwork, owing to practical reasons of safety and limited helicopter access. The field relationships suggest that suspended-load fallout in the upper ghaut concentrated coarse debris towards the base of the pyroclastic surge, and that this accumulated preferentially in preexisting gullies and topographic lows. Abundant segregation pipes were present in layer 2 near the top of the drain-age fills. The pyroclastic surge deposit, traced up the valley sides from the ghaut floor to the top of the singe zone, showed essentially continuous decreases in grain size and thickness, indicating that the pyroclastic surge was vertically stratified in grain size, and probably

concentration, in the upper ghaut. No deposits resembling those of block-and-ash flows were observed in the upper region of Dry Ghaut, showing that the surge-derived pyroclastic flow further downstream formed by sedimentation from the pyroclastic surge, and not from any primary, high-concentration material swashed into the ghaut from the White River valley. Many small trees on the floor of the upper ghaut were felled and/or removed by the surge (Fig. 9). The width of the blowdown zone decreased rapidly down the ghaut, probably because the surge was decelerating and or becoming less dense as it went forward (Fig. 7). Felled trees were oriented obliquely inwards towards the ghaut axis, and in some small tributaries were reoriented directly downslope due to local formation and drainback of dense underflows, as also observed at Mount St Helens (Hoblitt & Miller 1984). At a given site, mature trees were felled more readily than small, younger trees, probably because the latter were more supple and, being smaller, offered less resistance to the surge. Many young trees remained standing in the blowdown zone, bent over in the direction of flow. Some young trees remained standing on the ghaut floor even where partly immersed in metre-thick surge deposits. Bark was abraded or removed on the upstream sides of trees, and that remaining was commonly charred and blistered. The char on young trees and bushes was commonly very thin (1 mm or less) and the wood inside unaffected, showing that the burn was intense, but short-lived. The boundary between the charred, upstreamfacing sides of branches and their uncharred downstream sides was very sharp (Fig. lOa), as also observed at Mount Lamington (Taylor 1958). The degree of tree damage decreased gradually from the blowdown zone upwards to the top of the singe zone. The evidence from the upper ghaut implies that, during this part of its passage, the pyroclastic surge was vertically stratified in grain size and density and was actively segregating into an upper, more dilute part and a lower, more concentrated part that was being concentrated into channels and depressions. As the surge swept through the upper ghaut, it decelerated and or decreased in density rapidly, as shown by the rapid downstream fall-off of grain size and extent of tree blowdown. Middle region of Dry Ghaut (0.6-2.0 km). The middle region of the ghaut is dominated by a steep-sided canyon. When the surge entered the middle ghaut it was travelling too slowly, and or was too dilute, to fell many trees. The valley sides here were covered by a surge deposit, commonly consisting of just a single layer less than a decimetre thick. The second surge unit (probably Unit II of Ritchie el al. 2002) was not observed outside the upper ghaut. On the ghaut floor a massive surge-derived pyroclastic flow deposit formed a continuous drape up to several decimetres thick over stream

Fig. 9. Panorama looking up (west) into the upper reaches of Dry Ghaut. The pyroclastic surge overtopped the saddle on the skyline and travelled towards the observer before being deflected towards the lower right of the field of view. The surge-derived pyroclastic flow deposit (f) occurs in the pre-existing drainages. The surrounding slopes are covered by the deposit from the pyroclastic surge. Areas in which trees are mostly blown down and/or removed (b) and those in which many trees remain standing (s) are distinguished. The field of view of the skyline is about 1 km wide (see Fig. 7).

272

T. H. DRUITT ET AL.

Fig. 10. Effects of the 26 December 1997 pyroclastic surge and surge-derived pyroclastic flow in Dry Ghaut, (a) Branch (held vertical) charred by the pyroclastic surge as it passed through the upper reaches of the ghaut. The extremely sharp limit to the char suggests that the heat was very intense and shortlived. Location 1 on Figure 7. (b) Log-jam generated by the surge-derived pyroclastic flow 7 as it passed through the lower reaches of the ghaut. The boulder was picked up upstream and rammed into the accumulated logs. Location 2 on Figure 7. boulders. In the old stream channels and in narrow gorges the flow deposit thickened to 1 m or more of massive ash with matrixsupported blocks and lapilli and lacking the bipartite layering of the surge deposit. Most trees in the middle ghaut remained standing right down to the ghaut floor and only where the progressively forming surgederived pyroclastic flow swept up and over the inside corners of bends were small trees on the valley sides systematically felled. The surge-derived flow had a maximum thickness no more than that of the resulting deposit. On the ghaut floor there was a clear boundary between where small trees had been pushed over by the highconcentration flow, and where trees remained standing; this boundary corresponded with the upper surface of the flow deposit. A particularly convincing example occurred 1.5km down the ghaut, where the flow travelled through a narrow gorge. In this gorge young trees with delicate branches remained upright and intact immediately above the top of the flow deposit, which was nearly 2m thick. The observations suggest that by the time the pyroclastic surge entered the middle ghaut, a high-concentration granular underflow (the surge-derived pyroclastic flow) was already developed, and that the two were separated vertically by a sharp concentration and rheological interface.

Lower region of Dry Ghaut (>2km). By the time the pyroclastic surge reached 2 km down the ghaut it had become very weak and is inferred to have mostly lofted; most of the moving mass was by then

concentrated in the surge-derived pyroclastic flow. The singe zone in the lower ghaut was less than 10m high, and only 2-3 m high at the flow7 terminus. Figure 11 is a map of one stretch of the ghaut where the surge-derived flow travelled round two bends. The surge-derived flow deposit on the ghaut floor was typically flat-topped and consisted of massive ash and lapilli with abundant charred wood. Wood fragments were commonly concentrated at the top of the deposit due to flotation in the high-concentration flow. Traced to the sides of the ghaut, the flow deposit had a very sharp limit, beyond which it passed abruptly into a few centimetres of fine-grained grey ash from the overriding ash cloud (Fig. 11). showing that the moving flow had a very well defined upper surface. Where the flow travelled around bends there were clear superelevation effects. w ? ith a rising and falling strandline of massive flow deposit where the flow had swashed up on the outer side of the bend. On the outside curve of one particularly sharp bend, part of the flow ran up out of the stream channel, dumped a pile of charred logs, and spread out over the gently sloping forest floor as a sheet about 40cm thick (location A. Fig. 11). Most trees remained standing in the lower ghaut, and many retained their leaves at heights greater than a few metres above the ghaut floor. This showed that the pyroclastic surge component was very weak by the time it reached this part of the ghaut and was probably sustained largely by elutriation of ash from the highconcentration surge-derived pyroclastic flow. Trees impacted by the surge-derived flow were commonly felled, and the bark stripped off their upstream sides. Older trees snapped, but younger ones were simply flattened down parallel to the flow direction. In several

SMALL PYROCLASTIC FLOWS: A HAZARD

273

Fig. 11. Interpretative map of part of the lower region of Dry Ghaut, where the surge-derived pyroclastic flow encountered two bends in the stream valley. The location is marked on Figure 7.

places the surge-derived flow left jams of partially charred logs piled up against trees or boulders (Fig. 10B). Some large trees remained standing in the flow deposit, presumably because the flow was too shallow and/or too slow to fell them. The pyroclastic flow terminated behind a pile of boulders 200 m from the sea. Near the terminus the deposit had an almost flat surface (Figs 12A and 13), with just 1-2 cm of fine ash on the adjacent valley sides. The singe zone here was only 2-3 m high. Trees in the singe zone retained very delicate twigs and many still had charred leaves on them, so the surge cloud was very weak near the flow terminus. Many trees remained standing in the flow and had suffered only minor bark loss and charring to a few tens of centimetres above the top of the deposit. There were no noticeable superelevation effects on bends near the terminus, showing that the speed of the surge-derived flow was low. The distal flow deposit was typically massive or (coarse-tail) normally graded, with a crude upstream-inclined fabric of coarser clasts, and was about 1 m thick (Figs 12B and 13). The gradient of the stream bed where the flow front came to rest was 7°. Temperatures of 120-140°C were measured in the flow deposit at depths of 25-35 cm four days after the eruption, accounting for the weak thermal effects on vegetation. Another 26 December 1997 surge-derived pyroclastic flow deposit produced near St Patrick's was 190°C nine days after emplacement.

Ghaut are distinguished on these graphs. In each case samples from different heights in a single section are connected by tielines. The plots show two important features of the surge-derived flow deposits. First, they are systematically finer grained than typical block-and-ash flow deposits of Soufriere Hills Volcano, but have grain-size compositions within the range of those of the associated surge deposits. This is consistent with derivation of the flows by sedimentation from the pyroclastic surges, as deduced from field relationships. Second, the grain-size compositions of the surgederived flow deposits are rather uniform. Those from the lower and middle regions of Dry Ghaut are very similar granulometrically, and in turn resemble the surge deposit in the upper region of Dry Ghaut. The observations suggest: (1) that the Dry Ghaut surgederived flow formed mainly by suspended-load fallout in the upper region of the ghaut (top few hundred metres); and (2) that the flow travelled through the middle reaches of the ghaut without significantly changing its grain-size composition. Physical parameters We have used field measurements and calculations to estimate the physical properties of the surge-derived pyroclastic flows and parent pyroclastic surges. Emphasis is placed on Dry Ghaut, since it is where our most detailed fieldwork was carried out.

Granulometry The grain-size compositions of deposits from the Belham River valley and Dry Ghaut pyroclastic surges and surge-derived pyroclastic flows are shown in Figure 14 on a standard r Md plot. Samples from the upper, middle and lower regions of Dry

The parent pyroclastic surges The surge associated with block-and-ash flow pulse 3 of 25 June 1997 ramped 70m up the side of Windy Hill. Simple conversion of

274

T. H. DRUITT ET AL.

Fig. 12. Surge-derived pyroclastic flow deposit from 26 December 1997 in Dry Ghaut. Location 3 on Figure 7. (a) About 200m from the terminus, showing the flow deposit (f) (cut by a later gully) and associated singe zone (s). (b) Close-up of the surge-derived flow deposit (0 showing its fine-grained nature with scattered, matrix-supported lapilli. Note also the crude upstream (right) inclined fabric of the coarser clasts. The topmost layer (1) is a lahar deposit of unknown later date. The light and dark patches on the left are surface effects.

kinetic to potential energy implies a speed of about 35ms l (see also Loughlin et al. 2002&). The frontal velocity of the 26 December 1997 pyroclastic surge across the main impact area south of the dome has been constrained by to have been in the range 5585ms - 1 , corresponding to an inferred internal velocity of perhaps 80-120ms-1 (Sparks et al. 2002).

We now consider the upper, dilute part of the 26 December 1997 pyroclastic density current (surge) that entered Dry Ghaut. The velocity was estimated crudely using measured superelevation effects on bends. This exploits the observation that at several places down the ghaut the singe line of the surge in Dry Ghaut rises on the outsides of bends and falls on the insides (Fig. 8), and this is

SMALL PYROCLASTIC FLOWS: A HAZARD

275

Fig. 13. Section through the surge-derived pyroclastic flow deposit about 200 m from its terminus in Dry Ghaut (reconstructed, pre-gullying). The boulders on the bed of the stream channel followed by the surgederived flow pre-dated the flow. The location is marked on Figure 7.

attributed to centrifugal effects. The balance of centrifugal and gravity forces on a bend generates a slope 9 on the surface of a current, given by:

where R is the mean radius of curvature of the bend, U is the current velocity, and 0 is the channel slope parallel to flow. Tan 0 was estimated at five bends of different radii by measuring the elevation difference AH of the upper limit of the singe zone across the valley, along with the valley width W (tan 0 = AH/W}. In all

five cases a significant superelevation effect was present. The calculated velocities range from 13 to 2 2 m s - 1 (Fig. 7, Table 1). The velocities can in turn be used to constrain the mean particle concentration in the surge. Britter & Linden (1980) found experimentally that the frontal velocity Uf of a turbulent density current on a slope in the range 5-90° is insensitive to the slope angle (due to balance of buoyancy and entrainment-related drag forces) and is given approximately by:

where Q is the internal volumetric flux per unit width of the current, p is the density of the ambient fluid, and Ap is the excess density of the current. Q is given by Q = U i h, where Ui is the internal current velocity and h is the current thickness. In experimental density currents, Uf is found to be about 60% of Ui Assuming that the velocities derived above from superelevation effects are estimates of the internal velocity of the surge, substituting Uf — 0.6Ui, Ui= 13-22ms -1 , p = 1.3kgm - 3 (density of air near sea level) and h = 50-100 m shows that the average bulk density of the surge cannot have exceeded about 1.4kgm - 3 , otherwise its velocity would have been greater. Similarly, the particle concentration in the surge cannot have exceeded about 0.1 vol% irrespective of what gas temperature (<850°C), gas density and solids density are assumed. This crude calculation ignores sedimentation and unsteadiness effects, but probably provides the right order of magnitude. Note that this reasoning applies only to the middle region of the ghaut, where it was possible to quantify the superelevation effects. In all likelihood the surge had a higher mean density and particle concentration in the upper ghaut. Moreover, the surge would have been density-stratified, so that density and concentration near the ghaut floor were higher than these mean values (Valentine 1987). Table 1. Velocity estimates for the Dry Ghaut surge and flow components from superelevation effects on bends Distance (km)

A// (m)

Pyroclastic surge 0.75 0 30 1.05 0 25 1.20 0 10 0 18 42 0 Surge-derived pyroclastic flow 5 3.5 2.33 5 1.5 2.60 5 3.6 Fig. 14. Plot of median grain size Mdo versus sorting (Inman 1952), showing data for the 25 June 1997 and 26 December 1997 pyroclastic surge deposits (a) and surge-derived pyroclastic flows (b) in the Belham River valley and Dry Ghaut. The fields of Montserrat dome-collapse block-andash flows are shown for comparison.

U (ms-1)

W (m)

R (m)

350 210 250 150 150

305 405 540 150 90

12 10 10 8 8

16 22 15 13 16

10 10 10

30 25 10

7 7 6

10 6 6

o

(°)

AH is the height difference of the top of the pyroclastic surge or flow deposit across the valley, width W; R is the mean radius of curvature of the bend; o is the mean valley gradient; and U is the estimated pyroclastic surge or flow velocity.

276

T. H. DRUITT ET AL.

The ranges of velocity and density of the pyroclastic surge appear to be consistent with the limited extent of tree blowdown in Dry Ghaut. Once the surge left the upper region, it was incapable of felling significant numbers of trees (Fig. 7). The ability of a surge to fell trees depends on the dynamic pressure exerted. This scales as P 3Uf, where 3 is the bulk density and Ui the internal velocity. In a review of nuclear weapon effects, Valentine (1998) gives a critical dynamic pressure of about 500-800 Pa for light (<30%) tree blowdown. This will depend on tree size and species, but is consistent with measured wind speeds of up to c. 40ms - 1 (P = c. 800 Pa) during Hurricane Hugo, which felled a considerable number of trees on Montserrat in 1989. For a surge of density 1.4kgm - 3 travelling down the middle region of Dry Ghaut at 2 0 m s - 1 , we have P = c. 280 Pa, which is indeed below the threshold for blowdown. The velocity and density of the surge at the top of the ghaut cannot be constrained, but it is likely that both were higher than further downstream. We estimate that on average about 30%, and locally as much as 90%, of trees were felled in the upper ghaut. This may require dynamic pressures as high as 2500 Pa (Valentine 1998). For a flow density of 2 k g m - 3 the calculated internal speed required to generate a dynamic pressure of 2500 Pa is 50m s - 1 , and for 5 k g m - 3 it is 30m s - 1 . These examples give the correct order of magnitude for a mean current density and average internal velocity over the height of a felled object. The internal velocities calculated for the pyroclastic surge in Dry Ghaut (<50 ms-1) are lower than those estimated for the pyroclastic density current across the main impact zone south of the lava dome (80-120ms- 1 ; Sparks et al. 2002). This is probably because: (1) Dry Ghaut is oriented perpendicular to the main current axis, so that the current impacted the head of the valley obliquely and the initial downstream velocity component was not very high; (2) only the upper, most dilute part of the current (pyroclastic surge) spilled over into Dry Ghaut, having a lower density (and thus lower velocity) than that along the main axis; (3) the velocities estimated for main impact zone were enhanced by gravitational acceleration down the flanks of the volcano south and SW of the lava dome. We can estimate very approximately the rate of sediment accumulation from the surge as it flowed down the ghaut. The rate of sediment accumulation from a turbulent suspension current is (Bursik & Woods 1996):

where {3 is the bulk density of the surge, n is the mass fraction of solids (c. 0.5, for a particle concentration of c. 0.1 vol%), C is the deposit density (c. 1000 kg m - 3 ), and w is the mean particle settling velocity . The settling velocity W of a spherical particle in gas can be expressed approximately as (Kunii & Levenspiel 1991):

material up to 6m high, which spread out longitudinally and thinned as it went. The speed of the flow must have been very low, probably a few metres per second, since many trees remained standing, even where partially immersed in the flow deposit, and the flow was unable to destroy a bridge over which it passed. The flow in Dry Ghaut was thinner during transport (
where u is the fluid viscosity, d is the particle diameter, p is the fluid density, a (>>p) is the particle density, and g is gravitational acceleration. The mean particle size in the surge was about 300500um (Fig. 14), so w is a few metres per second. Taking a surge density of 1.4kgm - 3 , appropriate for the middle stretch of the ghaut, yields an average sediment accumulation rate of a few millimetres per second. This implies a sedimentation duration of a few tens of seconds to generate the observed deposit, typically a few decimetres in thickness. Higher accumulation rates were likely along the valley axis due to effects of density stratification (Valentine 1987), and in the headwaters of the ghaut where the surge density and particle concentration were probably higher. The surge-derived pyroclastic flows Once formed, the surge-derived pyroclastic flows of 25 June and 26 December 1997 travelled slowly down their respective drainages under gravity. The Belham River valley flow moved as a wave of

Pyroclastic flows formed by three different mechanisms during the 1995-1999 eruptive period of Soufriere Hills Volcano: lavadome collapse, fountain collapse during Vulcanian explosions, and suspended-load fallout from pyroclastic surges generated during dome-collapse events (Calder et al. 1999: Cole et al. 2002; Druitt et al. 2002). The first two mechanisms have been described from many volcanoes and are relatively well understood (e.g. Fisher & Schmincke 1984; Cas & Wright 1997; Carey 1991; Druitt 1998). In this paper we have described the evidence for the third mechanism in some detail and have constrained approximately some of the relevant physical parameters. The fact that surge-derived pyroclastic flows were generated on at least two occasions, and at several locations, on Montserrat suggests that they may be a common threat around lava domes. Although we have stressed the Dry Ghaut example in this paper, surge-derived flows also formed in several other drainages along the south coast of Montserrat on 26 December 1997, with particularly large ones in Gingoes and Germans Ghauts (Fig. 6). Another, smaller, example also occurred during a collapse on 12 May 1996 (Cole et al. 2002).

277

SMALL PYROCLASTIC FLOWS: A HAZARD

Small-volume, high-concentration flows were also produced locally by suspended-load fallout from the 18 May 1980 lateral blast pyroclastic surge at Mount St Helens (Hoblitt et al. 1981; Hoblitt & Miller 1984; Brantley & Waitt 1988; Fisher 1990). These travelled down local slopes at low speeds, sometimes at right angles to the main surge transport direction, but were largely trapped by local topography. During the 1902 eruption of Mont Pelee, settling of particles from energetic surges generated pyroclastic flows that were finer grained than the associated primary block-and-ash flows of the Riviere Blanche (Fisher et al. 1980; Fisher & Heiken 1982). The term 'secondary' has been used to describe these flows (Hoblitt & Miller 1984; Brantley & Waitt 1988; Fisher et al. 1987); however, it is important to distinguish them from secondary pyroclastic flows formed by the remobilization of ignimbrite, such as followed the 1991 eruption at Mount Pinatubo (Torres et al. 1997). Moreover the term 'secondary' implies a time gap between surge emplacement and remobilization to form concentrated flows. This was not the case at Montserrat, where the emplacement of the pyroclastic surge and its derivative pyroclastic flow were essentially simultaneous. It is for this reason that we propose the term surgederived pyroclastic flow for this phenomenon. The main features of the Belham River valley and Dry Ghaut flows are summarized in Table 2. Both formed by suspended-load fallout from pyroclastic surges, which generated granular flows of ash and lapilli that then drained gravitationally into neighbouring valleys and topographic depressions. The flows were highly concentrated with well defined upper surfaces, they travelled at a few metres per second and the associated overriding surge clouds near the distal limits were very weak (in contrast to the parental surges that generated the surge-derived flows in the first place). The Belham River valley flow had a broader catchment area and a larger volume than that in Dry Ghaut. The former consequently travelled as a wave of debris up to 6 m high, reaching 3 km from the limit of the parent surge, whereas the Dry Ghaut flow nowhere exceeded a couple of metres in thickness and outran its parent surge by only 1 km. The Dry Ghaut example has been studied in some detail. As the dome collapsed, a powerful pyroclastic density current swept southwards, and part of the upper, dilute levels of it (pyroclastic surge) spilled over into Dry Ghaut. As it entered the ghaut, this surge was travelling at a few tens of metres per second and was capable of felling trees. However, it decelerated rapidly and began to segregate into a more concentrated lower part and an upper more dilute part. Evidence for deceleration is provided by the rapid decrease in grain size of the surge deposit, and narrowing of the tree-blowdown zone, over the first few hundred metres from the top

of the ghaut. As the surge decelerated, suspended-load fallout generated a high-concentration granular flow that cascaded downstream through the steep-sided gorge of the middle section of the ghaut. As it did so, it was fed by further fallout from the surge, which by now was travelling at no more than 20 m s-1 and was very dilute (0.1% or less of solids), although this was not sufficient to greatly modify the grain-size composition of the flow. The downstream reduction in height above the valley floor of the surge singe line probably records a combination of progressive surge deceleration, decrease in bulk density through sedimentation, drawing in of the sides of the surge cloud by advective indrafts of air, and buoyant lofting (Fig. 8). By the time the surge-derived pyroclastic flow reached 2km down the ghaut it was travelling at only a few metres per second and the accompanying pyroclastic surge was very weak. The front of the flow then continued a further kilometre before coming to rest 200m from the sea. In total, the surge-derived flow accounts for about a quarter of the material that entered Dry Ghaut (Table 2). Given their small volumes relative to many block-and-ash flows, an important feature of the surge-derived pyroclastic flows was their large runout distances compared to the height drop (i.e. high mobility in the sense of Calder et al. 1999). This was not a temperature effect, since the emplacement temperatures (120-410°C) were lower than those of less 'mobile' block-and-ash flows (up to 650°C). If the flows were partially fluidized by escaping gas, the gas cannot have been derived by exsolution because the diffusion rate of gas in volcanic glass is negligible at temperatures as low as 410°C or less. Nor can the gas have contained a significant component of externally derived steam since, to our knowledge, both the Belham River valley and Dry Ghaut were largely free of water prior to the 25 June and 26 December 1997 events. Two possible gas sources that cannot be ruled out are juvenile clast attrition and rupture of gas-filled vesicles during transport, and incorporation and combustion of vegetation. Calder et al. (1999) tentatively attributed the high mobility of the surge-derived flows to rapid sedimentation of material from turbulent suspension and the formation of a poorly sorted granular flow with low frictional resistance, perhaps due to the generation of high transient pore pressures. Recent studies have recognized different regimes of sedimentation from turbulent pyroclastic suspensions (Druitt 1998 and references therein). In some cases, particles settle directly to form a deposit. The deposit can be planar or cross-stratified if particles undergo late-stage traction, or massive if deposition occurs directly from suspension. The deposits of pyroclastic surges are laid down principally in this manner. In other cases, the settling particles continue to move upon reaching the ground, forming

Table 2. Comparison of the surge-derived pyroclastic flows of 25 June and 26 December 1997 Feature Similarities Origin of pyroclastic flow Nature of flow Flow velocity Nature of flow deposit

Differences Vol. of surge deposit (m3) Vol. of surge-derived flow deposit (m 3 ) Flow thickness in transport (m) Runout beyond surge (km) Ground slope at terminus (°) Mobility ratio, L/H Temperature of flow deposit (°C)

26 December (Dry Ghaut)

25 June (Belham valley)

Rapid suspended-load fallout from pyroclastic surge Concentrated granular flow with well-defined upper surface 1 0 m s - 1 or less, with weak accompanying ash cloud Hummocky veneer (upstream), flat-topped drainage fill (downstream); massive, poorly sorted, weak normal grading; unusually fine-grained for pyroclastic flow deposit 150000* 50000
* Volume of surge deposit within Dry Ghaut, not DRE. Volume of the whole pyroclastic surge deposit from that event, not DRE.

SOOOOOf 90000 <6 3 2 5 310-400

278

T. H. DRUITT ET AL.

high-concentration pyroclastic flows. These decouple from the pyroclastic surge and drain gravitationally into lows, sometimes outrunning the surge (Fisher 1990, 1995). The present study offers new insight into the formation of pyroclastic flows beneath turbulent pyroclastic suspensions. The 25 June and 26 December 1997 pyroclastic surges at Montserrat laid down thin, landscape-draping deposits, typically of two layers - a finesdepleted lower layer and an upper one richer in fines - as has been described from deposits of other pyroclastic surges at Mont Pelee (Boudon & Lajoie 1989), Mount St Helens (Hoblitt et al 1981; Fisher 1990; Druitt 1992) and Bezymianny (Belousov 1996). At Mount St Helens and Bezymianny, high-concentration flows of small volume formed by remobilization of rapidly accumulated and loosely packed slope facies (Fisher et al. 1987; Hoblitt & Miller 1984; Belousov 1996). Ponds of flow deposits at Mount St Helens commonly overlie the blast surge deposit, showing that remobilization took place following initial sediment accumulation. At Montserrat we have not observed a surge deposit below that of its surgederived flow, but this could be due to erosion or to the necessarily cursory nature of our field studies. It is possible that flow formation at Montserrat took place both simultaneously with surge deposition (i.e. without the material coming to rest) as well as by remobilization shortly thereafter. The presence of breakaway scars on the surface of the Belham River valley deposit shows that parts of the flows themselves came to rest and then remobilized. We envisage a complicated combination of suspended-load fallout, advection of high-concentration flows to the valley floor, and some remobilization, in the production of the Montserrat surge-derived pyroclastic flows. It has been speculated that the formation of high-concentration pyroclastic flows from turbulent suspensions is favoured by high rates of suspended-load fallout (Druitt 1992, 1998). Rapid sedimentation of granular materials can, under some conditions, generate excess pore pressures in the accumulating sediment layer, reducing the intergranular friction and allowing the sedimented layer to flow away (Miller 1990; Freundt 1999). The average sediment accumulation rate in the middle reaches of Dry Ghaut was probably a few millimetres per second. Higher values were likely in the upper ghaut and along the valley axis due to density stratification effects (Valentine 1987). Bursik et al. (1998) estimated an average bulk density of c. 1.5kgm - 3 for the Mount St Helens pyroclastic surge 12km from source, which also generated highconcentration pyroclastic flows of local extent, and this corresponds to an accumulation rate of a few millimetres per second (Equation 3). We tentatively infer that sediment accumulation rates of a few millimetres per second may be required for the formation of pyroclastic flows beneath turbulent pyroclastic suspensions. Values of this order are predicted by theoretical models of ignimbrite emplacement, which is known to involve widespread formation of highly concentrated granular flows from turbulent suspensions (Bursik & Woods 1996; Druitt 1998; Freundt 1999). The deposits of surge-derived pyroclastic flows are massive or normally graded and are characteristically finer grained than other types of pyroclastic flow deposits. They consist principally of ash with subordinate lapilli, blocks comprising only a small percentage of the deposit. The relatively fine grain size of these flows is attributed to their origin by sedimentation from turbulent suspensions in which the transport of abundant coarse blocks was not possible. The fine grain size may be a distinguishing feature of surge-derived pyroclastic flows at other volcanoes, and may aid recognition of their deposits in ancient successions. The deposits are, however, relatively thin and, owing to their deposition along valley axes, readily eroded, giving them poor preservation potential. The surge-derived flow deposits in the Belham River valley and Dry Ghaut had both been largely removed by erosion a year after their formation. The recognition of surge-derived pyroclastic flows has important implications for volcanic hazards assessments. The ability of pyroclastic surges to generate highly mobile, high-concentration pyroclastic flows poses an important, but hitherto underestimated, threat around lava domes. A particularly important feature of these

flows is their ability to develop in drainages different from those affected by the main block-and-ash flows and associated pyroclastic surges. They therefore have the potential to impact upon areas not previously anticipated in hazards analyses. The Department for International Development is thanked for financial support in monitoring Soufriere Hills Volcano. E.S.C. acknowledges a NERC studentship, L.J.R. a studentship from the University of Luton, and R.S.J.S. a NERC Professorship. Helicopter pilots J. McMahon and A. Grouchy skillfully transported the authors in the field. M. Branney, A. Freundt and P. Kokelaar kindly reviewed the manuscript. This paper is published with the permission of the Director of the British Geological Survey.

References ABDURACHMAN, E. K., BOURDIER, J.-L. & VOIGHT, B. 2000. Nuees ardentes of 22 November 1994 at Merapi volcano, Java, Indonesia. Journal of Volcanology and Geothermal Research, 100, 345-362. BELOUSOV, A. 1996. Deposits of the 30 March 1956 directed blast at Bezymianny Volcano, Russia. Bulletin of Volcanology, 57, 649-662. BOUDON, G. & LAJOIE, J. 1989. The 1902 Peleean deposits of the Fort Cemetery of St. Pierre, Martinique a model for the accumulation of turbulent nuee-ardentes. Journal of Volcanology and Geothermal Research, 38, 113-129. BRANTLEY, S. & WAITT, R. B. 1988. Interrelations among pyroclastic surge, pyroclastic flow, and lahars in Smith Creek valley during first minutes of 18 May 1980 eruption of Mount St. Helens. USA. Bulletin of Volcanology, 50, 304-326. BRITTER, R. E. & LINDEN, P. F. 1980. The motion of the front of a gravity current travelling down an incline. Journal of Fluid Mechanics, 99, 531-543. BURSIK, M. I. & WOODS, A. W. 1996. The dynamics and thermodynamics of large ash flows. Bulletin of Volcanology, 58, 175-193. BURSIK, M. I., KURBATOV, A. V., SHERIDAN, M. F. & WOODS, A. W. 1998. Transport and deposition in the May 18 1980, Mount St. Helens blast flow. Geology, 26, 155-158. CALDER, E. S., COLE, P. D., DADE, W. B. ET AL. 1999. Mobility of pyroclastic flows and surges at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 26, 537-540. CALDER, E. S., LUCKETT, R., SPARKS, R. S. J. & VOIGHT, B. 2002. Mechanisms of lava dome instability and generation of rockfalls and pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21. 173-190. CAREY, S. N. 1991. Transport and deposition of tephra by pyroclastic flows and surges. In: FISHER, R. V. & SMITH. G. A. (eds) Sedimentation in Volcanic Settings. SEPM Special Publication, 45, 39-57. CAS, R. A. F. & WRIGHT, J. V. 1987. Volcanic Successions. Modern and Ancient. Allen and Unwin, London. COLE, P. D., CALDER, E. S., DRUITT. T. H. ET AL. 1998. Pyroclastic flows generated by gravitational instability of the 1996-97 lava dome of Soufriere Hills Volcano, Montserrat. Journal of Geophysical Research Letters, 25, 3425-3428. COLE, P. D.. CALDER, E. S., SPARKS, R. S. J. ET AL. 2002. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano. Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London. Memoirs, 21, 231-262. DADE, W. B. & HUPPERT, H. E. 1998. Long runout rockfalls. Geology, 26, 803-806. DRUITT, T. H. 1992. Emplacement of the 18 May 1980 lateral blast deposit east-northeast of Mount St. Helens. Washington. Bulletin of Volcanology, 54, 554-572. DRUITT, T. H. 1998. Pyroclastic density currents. In: GILBERT, J. S. & SPARKS, R. S. J. (eds) The Physics of Explosive Volcanic Eruptions. Geological Society, London, Special Publications, 145, 145-182. DRUITT, T. H., YOUNG, S. R., BAPTIE, B. ET AL. 2002. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306.

SMALL PYROCLASTIC FLOWS: A HAZARD FISHER, R. V. 1990. Transport and deposition of a pyroclastic surge across an area of high relief: the 18 May 1980 eruption of Mount St. Helens, Washington. Geological Society of America Bulletin, 102, 1038-1054. FISHER, R. V. 1995. Decoupling of pyroclastic currents: hazards assessment. Journal of Volcanology and Geothermal Research, 66, 257-263. FISHER, R. V. & HEIKEN, G. 1982. Mt. Pelee: May 8 and 20 1902, pyroclastic flows and surges. Journal of Volcanology and Geothermal Research, 13, 339-371. FISHER, R. V. & SCHMINCKE, H.-U. 1984. Pyroclastic Rocks. SpringerVerlag, Berlin. FISHER, R. V., SMITH, A. L. & ROOBOL, M. J. 1980. Destruction of St. Pierre, Martinique, by ash-cloud surges, May 8 and 20 1902. Geology, 8, 472-476. FISHER, R. V., GLICKEN, H. X. & HOBLITT, R. P. 1987. May 18 1980, Mount St. Helens deposits in South Coldwater Creek, Washington. Journal of Geophysical Research, 92, 10267-10283. FREUNDT, A. 1999. Formation of high-grade ignimbrites. Part II. A pyroclastic suspension current model with implications also for low-grade ignimbrites. Bulletin of Volcanology, 60, 545-567. FUJII, T. & NAKADA, S. 1999. The 15 September 1991 pyroclastic flows at Unzen Volcano (Japan): a flow model for associated ash-cloud surges. Journal of Volcanology and Geothermal Research, 89, 159-172. GARDNER, C. A. & WHITE, R. A. 2002. Seismicity, gas emission and deformation from 18 July to 25 September 1995 during the initial phreatic phase of the eruption of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 567-581. HAYASHI, J. N. & SELF, S. 1992. A comparison of pyroclastic flow and debris avalanche mobility. Journal of Geophysical Research, 97, 9063-9071. HOBLITT, R. P. & MILLER, C. D. 1984. Comment on 'Mount St. Helens 1980 and Mount Pelee 1902 - flow or surge?' Geology, 12, 692-693. HOBLITT, R. P., MILLER, D. C. & VALLANCE, J. E. 1981. Origin and stratigraphy of the deposits produced by the May 18 directed blast. In: LIPMAN, P. W. & MULLINEAUX, D. R. (eds) The 1980 Eruptions of Mount St. Helens, Washington. US Geological Survey, Professional Papers 1250, 379-400. INMAN, D. L. 1952. Measures of describing the size distribution of sediments. Journal of Sedimentary Petrology, 22, 125-145. KELFOUN, K., LEGROS, F. & GOURGAUD, A. 2000. Statistical study of damaged trees related to the pyroclastic flows of November 22 1994, at Merapi Volcano (Central Java, Indonesia): Relation between ash-cloud surge and block-and-ash flow. Journal of Volcanology and Geothermal Research, 100, 379-394. KIEFFER, S. W. 1981. Fluid dynamics of the May 18 blast at Mount St. Helens. In: LIPMAN, P. W. & MULLINEAUX, D. R. (eds) The 1980 Eruptions of Mount St. Helens, Washington. US Geological Survey, Professional Papers, 1250, 379-400. KUNII, D. & LEVENSPIEL, O. 1991. Fluidization Engineering (second edition). Butterworth Heinemann, Oxford. LOUGHLIN, S. C., BAXTER, P. J., ASPINALL, W. P., DARROUX, B., HARFORD, C. L., MILLER, A. D. 20020. Eyewitness accounts of the 25 June 1997 pyroclastic flows and surges at Soufriere Hills Volcano, Montserrat, and implications for disaster mitigation. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 211-230. LOUGHLIN, S. C., CALDER, E. S., CLARKE, A. B. ETAL. 2002. Pyroclastic flows and surges generated by the 25 June 1997 dome collapse, Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 191-209.

279

MILLER, T. F. 1990. A numerical model of volatile behaviour in non-welded cooling pyroclastic deposits. Journal of Geophysical Research, 95 19349-19364. PARISEAU, W. G. & VOIGHT, B. 1979. Rockslides and avalanches: basic principles and perspectives in the realm of civil and mining operations. In: VOIGHT, B. (ed.) Rockslides and Avalanches, 2: Engineering Sites. Elsevier, Amsterdam, pp 1-92. RITCHIE, L. J., COLE, P. D. & SPARKS, R. S. J. 2002. Sedimentology of deposits from the pyroclastic density current of 26 December 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 435-456. ROBERTSON, R., COLE, P., SPARKS, R. S. J. ET AL. 1998. The explosive eruption of Soufriere Hills Volcano, Montserrat, 17 September, 1996. Geophysical Research Letters, 25, 3429-3432. ROBERTSON, R. E. A., ASPINALL, W. P., HERD, R. A., NORTON, G. E., SPARKS, R. S. J. & YOUNG, S. R. 2000. The 1995-1998 eruption of the Soufriere Hills volcano, Montserrat, WI. In: FRANCIS, P., NEUBERG, J. & SPARKS, R. S. J. (eds) Causes and Consequences of Eruptions of Andesite Volcanoes. Philosophical Transactions of the Royal Society, Mathematical, Physical and Engineering Sciences, 358, 1619-1638. ROWLEY, P. D., MACLEOD, N. S., KUNTZ, M. A. & KARPLAN, A. M. 1985. Proximal bedded deposits related to pyroclastic flows of May 18 1980, Mount St. Helens, Washington. Geological Society of America Bulletin, 96, 1373-1383. SPARKS, R. S. J., BARCLAY, J., CALDER, E. S. ETAL. 2002. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 409-434. TAYLOR, G. A. 1958. The 1951 eruption of Mount Lamington, Papua. Australian. Bureau of Mineral Resources, Geology and Geophysics Bulletin, 38, 1-117. TORRES, R. C., SELF, S. & MARTINEZ, M. L. 1997. Secondary pyroclastic flows from the June 15 1991, ignimbrite of Mount Pinatubo. In: NEWHALL, C. G. & PUNONGBAYAN, R. S. (eds) Fire and Mud: Eruptions and Lahars of Mount Pinatubo, Philippines. University of Washington Press, 665-678. VALENTINE, G. A. 1987. Stratified flow in pyroclastic surges. Bulletin of Volcanology, 49, 616-630. VALENTINE, G. A. 1998. Damage to structures by pyroclastic flows and surges inferred from nuclear weapons tests. Journal of Volcanology and Geothermal Research, 87, 117-140. VOIGHT, B. & DAVIS, M. J. 2000. Emplacement temperatures of the November 22 1994 nuee ardente deposits, Merapi volcano, Java. Journal of Volcanology and Geothermal Research, 100, 371-378. VOIGHT, B., SPARKS, R. S. J., MILLER, R. C. ET AL. 1999. Magma flow instability and cyclic activity at Soufriere Hills Volcano, Montserrat, British West Indies. Science, 283, 1138-1142. VOIGHT, B., KOMOROWSKI, J.-C., NORTON, G. E. ETAL. 2002. The 26 December (Boxing Day) 1997 sector collapse and debris avalanche at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 363-407. WOODS, A. W., SPARKS, R. S. J., RITCHIE, L. J., BATEY, J., GLADSTONE, C. & BURSIK, M. I. 2002. The explosive decompression of a pressurized volcanic dome: the 26 December 1997 collapse and explosion of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 457-466. YAMAMOTO, T., TAKARADA, S. & SUTO, S. 1993. Pyroclastic flows from the 1991 eruption of Unzen volcano, Japan. Bulletin of Volcanology, 55, 166-175.

Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat T. H. DRUITT 1 , S. R. YOUNG2, B. BAPTIE2, C. BONADONNA3, E. S. CALDER 3 , A. B. CLARKE 4 , P. D. COLE5, C. L. HARFORD 3 , R. A, HERD 6 , R. LUCKETT2, G. RYAN 7 & B. VOIGHT 4 1 Laboratoire Magmas et Volcans (UMR 6524 & CNRS), Universite Blaise Pascal, 5 rue Kessler, 63038 Clermont-Ferrand, France (e-mail: [email protected]) 2 British Geological Survey, Murchison House, Edinburgh EH9 3LA, UK ^Department of Earth Sciences, University of Bristol, Queens Road, Bristol BS8 1RJ, UK 4 Department of Geosciences, Pennsylvania State University, University Park, PA 16802, USA 6 Centre for Volcanic Studies, University of Luton, Park Square, Luton LU1 3JU, UK 6 British Geological Survey, Keyworth, Nottingham NG12 5GG, UK 7 Environmental Science Department, Institute of Environmental and Natural Sciences, University of Lancaster, Lancaster LAI 4YQ, UK

Abstract: In 1997 Soufriere Hills Volcano on Montserrat produced 88 Vulcanian explosions: 13 between 4 and 12 August and 75 between 22 September and 21 October. Each episode was preceded by a large dome collapse that decompressed the conduit and led to the conditions for explosive fragmentation. The explosions, which occurred at intervals of 2.5 to 63 hours, with a mean of 10 hours, were transient events, with an initial high-intensity phase lasting a few tens of seconds and a lower-intensity, waning phase lasting 1 to 3 hours. In all but one explosion, fountain collapse during the first 10-20 seconds generated pyroclastic surges that swept out to 1-2 km before lofting, as well as high-concentration pumiceous pyroclastic flows that travelled up to 6 km down all major drainages around the dome. Buoyant plumes ascended 3-15 km into the atmosphere, where they spread out as umbrella clouds. Most umbrella clouds were blown to the north or NW by high-level (8-18 km) winds, whereas the lower, waning plumes were dispersed to the west or NW by low-level (<5 km) winds. Exit velocities measured from videos ranged from 40 to 140ms - 1 and ballistic blocks were thrown as far as 1.7 km from the dome. Each explosion discharged on average 3 x 10 5 m 3 of magma, about one-third forming fallout and two-thirds forming pyroclastic flows and surges, and emptied the conduit to a depth of 0.5-2 km or more. Two overlapping components were distinguished in the explosion seismic signals: a low-frequency (c. 1 Hz) one due to the explosion itself, and a high-frequency (>2Hz) one due to fountain collapse, ballistic impact and pyroclastic flow. In many explosions a delay between the explosion onset and start of the pyroclastic flow signal (typically 10-20 seconds) recorded the time necessary for ballistics and the collapsing fountain to hit the ground. The explosions in August were accompanied by cyclic patterns of seismicity and edifice deformation due to repeated pressurization of the upper conduit. The angular, tabular forms of many fallout pumices show that they preserve vesicularities and shapes acquired upon fragmentation, and suggest that the explosions were driven by brittle fragmentation of overpressured magmatic foam with at least 55vol% bubbles present in the upper conduit prior to each event.

Vulcanian explosions are a common feature of andesitic volcanoes (Morrissey & Mastin 2000). Examples include historic eruptions of Arenal (Costa Rica), Ngauruhoe (New Zealand), Fuego (Guatamala) and Augustine (Alaska) volcanoes (Melson & Saenz 1973; Martin & Rose 1981; Nairn & Self 1978; Kienle & Shaw 1979). Recent examples include explosions at Lascar (Chile) in the period 1986-1996 (Matthews et al 1997), Pinatubo (Philippines) in 1991 (Hoblitt et al 1996) and Galeras (Columbia) in 1992 (Stix et al 1997). Individual Vulcanian explosions typically discharge between 102 and 10 6 m 3 of magma and comminuted accidental debris in cannon-like detonations, generating buoyant plumes mostly between 5 and 20km high. Most of the erupted mass is discharged on a time scale of 10 to 103 seconds. Exit velocities ranging from about 50 to at least 300 m s - 1 have been observed and blocks up to 2m or more can be thrown ballistically up to several kilometres during the initial vent-clearing phase (Self et al 1979; Fagents & Wilson 1993). Fallback of part of the discharging material (fountain collapse) generates pyroclastic flows. Successions of powerful explosions occur from some volcanoes, with intervals ranging from 102 to 107 seconds (Self et al 1979). Vulcanian explosions are attributed to the interaction of magma with external water or to the sudden release of highly pressurized, vesicular magma beneath a cooled or degassed cap (Self et al 1979; Fagents & Wilson 1993; Woods 1995; Stix et al 1997; Sparks 1997). Involvement of external water is invoked when there is direct field evidence, or when the gas content implied by models exceeds likely maximum magmatic values of a few per cent (Self et al 1979; Fagents & Wilson 1993). This paper concerns two episodes of Vulcanian explosions that took place in the second half of 1997 at the lava dome of Soufriere

Hills Volcano, Montserrat. Thirteen of these occurred in August and 75 in September and October. A remarkable feature was the repeated and regular nature of the explosions, intervals ranging from 2.5 to 63 hours with a strong mode at c.10 hours. The activity in August was accompanied by cyclic patterns of edifice deformation and seismic energy release (Voight et al 1998, 1999). The explosions generated plumes up to 15km and, in all but one, fountain collapse formed pumice-and-ash pyroclastic flows that travelled up to 6 km from the vent. The cyclicity of the explosions permitted accurate short-term forecasting and hazard assessment over more than a month of intense activity. It also allowed unusually detailed study by a variety of techniques. Explosions were filmed both during the day and at night, and when possible from multiple locations. Maximum plume heights were estimated, fallout and pyroclastic flow deposits were mapped and sampled, and measurements were made of the sizes and distributions of ballistic blocks. Seismicity before, during and after explosions was recorded by the Montserrat Volcano Observatory (MVO) broadband system (Neuberg et al 1998), and edifice deformation was measured by a combination of electronic distance measurement, global positioning system (GPS) and tiltmeter (Voight et al 1998, 1999). We describe the 1997 explosions and their products and make estimates of erupted masses, exit velocities, fragmentation pressures and conduit drawdowns. In some cases only ranges and averages of the parameters are presented, as practical considerations of time and safety limited data acquisition for most individual events. We also describe the eruption seismic signals and recognize two components: one related to the explosion itself and the other to fountain collapse, ballistic impact and pyroclastic flow transport. The data and analysis in this paper are complemented by numerical

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306. 0435-4052/02/S15 © The Geological Society of London 2002.

281

282

T. H. DRUITT ET AL.

modelling of the explosion plumes by Clarke et al. (2002) and of the associated conduit flow by Melnik & Sparks (2002b). Bonadonna el al. (2002a, b) describe the fallout from the explosions and develop mathematical models of tephra dispersal. The pyroclastic flows and their deposits are described by Cole et al. (2002). Two other periods of explosive eruption at Soufriere Hills Volcano have been described by Robertson et al. (1998) and Norton et al. (2002). During the night of 17/18 September 1996 there occurred a 50-minute sub-Plinian eruption that formed a plume at least 11.3 km high (above sea level), but no fountain-collapse pyroclastic flows (Robertson et al. 1998). Multiple short-lived (Vulcanian) explosions occurred in late 1998 and in 1999, during the period of virtually no magma extrusion (Norton et al. 2002). These relatively weak explosions generated plumes that typically rose to heights of 3-6 km above sea level and fountain-collapse pyroclastic flows that travelled up to 3 km from the lava dome. We begin by summarizing the main features of the 1995-1999 eruptive period and, in particular, events of the period from July to October 1997 during which the explosions described in this paper occurred. Montserrat local time (four hours behind universal time) is used throughout the paper unless noted. Place names are given on Figure 1. All plume heights are given above sea level.

The eruption of Soufriere Hills Volcano from 1995 to 1999 Soufriere Hills Volcano is an andesitic lava dome complex situated in southern Montserrat, in the Lesser Antilles island arc. Detailed overviews of the 1995-1999 eruptive period have been given by Young et al. (1997, 1998), Kokelaar (2002) and Sparks & Young (2002). The eruptive vent was situated in an ancient sector-collapse scar (English's Crater) about 1 km across and open to the east (Fig. 1). The western rim of English's Crater is called Gages Wall and the southern rim is called Galway's Wall (Fig. 1). The flanks of Soufriere Hills are scarred by radial valleys (locally called ghauts), which served to channel pyroclastic flows. Initial phreatic explosions began in July 1995. Magma reached the surface in November 1995, and a lava dome began to form. The first dome-collapse pyroclastic flows occurred in March 1996, and flows first reached the sea down the Tar River valley in May of the same year. Major dome collapses occurred in July and August, 1996, and on 17 September 1996 a major collapse of the dome was followed by an explosive eruption (Robertson et al. 1998). Dome collapses and associated pyroclastic flows continued throughout 1997, with particularly large ones occurring on 25 June, 3 August, 21 September, 4 November and 6 November (Cole et al.

Fig. 1. Map of southern Montserrat, showing the dome location inside English's Crater, the principal drainages (ghauts) around the dome, and the area of impact from pyroclastic surges and flows of the entire 1995-1999 phase of the eruption.

EPISODES OF CYCLIC EXPLOSIVE ACTIVITY

2002). The two episodes of Vulcanian explosions reported in this paper followed the collapses of 3 August and 21 September 1997. On 26 December 1997, Galway's Wall of English's Crater failed, sending 80-90 x 106 m3 of the wall, lava dome and dome talus down the White River valley as a debris avalanche and high-energy blockand-ash flows (Sparks et al. 2002). A pyroclastic density current generated by decompression of gases trapped in the dome interior devastated 10 km2 of southern Montserrat. Magma extrusion ceased in March 1998, then resumed again in November 1999. The total volume of magma discharged over the 28 months of dome formation was 0.3 km3 dense-rock equivalent (DRE). Overall magma flux and the sizes of gravitational collapses increased during the period, but with some significant fluctuations. Discharge rates during the first year of extrusion were mostly less than 2 m3 s - 1 , but by June 1997 had risen to more than 7-8 m3 s-1 (Sparks et al. 1998; Sparks & Young 2002). The 1995-1999 magma is a crystal-rich andesite containing phenocrysts of plagioclase, hornblende, orthopyroxene, quartz, and Fe-Ti oxides set in a groundmass of microphenocrysts, microlites and rhyolitic glass (Murphy et al. 2000). It is believed to have formed by heating and remobilization of a pre-existing crystal-rich (60-65 vol%) mush. The pre-eruptive liquid phase of the magma was saturated with % water, as determined from glass inclusion analysis and phase equilibria studies (Barclay et al. 1998; Devine et al. 1998#). This corresponds to a water-saturated magma reservoir depth of 5 to 6 km beneath the vent, which is consistent with seismic evidence (Aspinall et al. 1998). Dome growth was accompanied by cyclic patterns of ground deformation and seismicity with periodicities of 3 to 30 hours and attributed to non-linear processes of gas exsolution, crystallization, rheological stiffening and pressurization in the conduit beneath the lava dome (Voight et al. 1998, 1999). Ground deformation was measured by tiltmeters installed near the rim of English's Crater. During a typical tilt cycle there was a slow inflation of the edifice (5-30urad), followed by abrupt deflation and associated gas emission, ash-venting, dome collapse or explosion. Associated seismic swarms built up during inflation, then declined during deflation. Seismicity included volcanotectonic and long-period earthquakes, rockfall, pyroclastic flow and explosion signals, and tremor. Most seismic signals forming the swarms were of hybrid type, which combined the high frequencies of volcanotectonic earthquakes with low-frequency components (Miller et al. 1998). Two observatory stations were active during the explosive periods of 1997. Prior to early September, the MVO was sited 6km NW of the dome (MVO South, Fig. 1), whereas from then onwards it was located in northern Montserrat (MVO North, not shown on Fig. 1).

Eruptive chronology from July to October 1997

Buildup to the August explosions The scar left by the 4.9 x 106m3 (DRE) 25 June dome collapse (Loughlin et al. 2002) began to fill in rapidly during the last week of June, about 65% of the void having been filled by 1 July. Between 28 June and 5 July there was intense pyroclastic flow activity. Multiple block-and-ash flows were shed up to 3.5km down Mosquito and Fort Ghauts, up to 1.1 km down Tuitt's Ghaut, and up to 0.5km down the White River valley. Many of these started with a resounding boom, and a vertical ash column ascended to an altitude of more than 10km. A period of intense ash emissions began on 8 July and persisted through 13 July. Peaks in ash emission often coincided with the peaks of tilt cycles. Some preceded small block-and-ash flows into Mosquito Ghaut and Gages valley; small ash columns reached heights of no more than 3km before dissipating. By 17 July the highest point on the new dome growth nested in the 25 June scar had reached 957m above sea level. The total volume of the dome on 17 July was estimated to be 75 x 106 m3. The level of seismic and eruptive activity in late July was generally low. The 25 June scar

283

had filled in, and the dome had a more or less flat summit 960m above sea level. Activity was characterized by low-amplitude broadband tremor associated with multiple rockfalls and small block-and-ash flows, particularly over Gages Wall.

The August explosions In retrospect, tilt and hybrid cycles related to the impending explosive activity began on 31 July. Pale, weakly convecting plumes of ash rose almost continuously to heights of between 4.5 and 6km, and small block-and-ash flows travelled as far as 2 km down Gages valley and Tuitt's Ghaut. High levels of long-period and hybrid seismicity continued on 1 August, peaking at one event per minute at the top of the tilt cycle, and a number of detonations were heard coming from the volcano. At least one correlated with an impulsive signal recorded on the broadband seismometers and explosive activity was surmised. A view of the dome on the same day revealed a small horseshoe-shaped depression in its west side above Gages Wall. Activity over the next three days was characterized by further tilt cycles and hybrid swarms every 9-12 hours. Associated venting became increasingly explosive and the block-and-ash flows more voluminous. The dome was observed at 14:00 on 3 August, when an active face of large blocks and spines was present high above Gages Wall. From 18:00 to 20:30, at the peak of the afternoon tilt cycle on 3 August, a succession of block-and-ash flows travelled down Gages valley. A boom heard at 16:28 at MVO South may have been a small explosion of the dome or of a gas tank ignited by the flows. Then at 18:10 a 7.0 x 10 6 m 3 (DRE) block-and-ash flow descended the length of Fort Ghaut to the sea, causing extensive damage in Plymouth. The first clear explosive activity occurred on 4 August. A blockand-ash flow at 06:30 was accompanied by a loud rumbling, followed by fallout of lithic and crystal lapilli up to 5mm in diameter at MVO South. A second explosion at 16:43, following a hybrid swarm, sent a dark grey jet inclined at about 60° to the horizontal northwards from the dome. Large blocks were observed to follow ballistic trajectories. Moments later, block-and-ash flows swept 3.5km down Tuitt's Ghaut, 3.5km down the Tar River valley to the sea, 4km down Fort Ghaut to the sea, and an unknown distance down the White River valley. The resulting ash plume rose to 4.5km, and fragments of dome rock and dense pumice as large as 15 mm fell at MVO South. Between the morning of 5 August and the morning of 12 August another 11 explosions occurred. They are listed in Table 1. Each generated pumice-and-ash pyroclastic flows by fountain collapse and showered the island with pumice-rich fallout. The first nine explosions occurred regularly every 10 to 12 hours during or immediately after hybrid earthquake swarms. The last two occurred on 11 and 12 August, and were slightly weaker than the others. The August explosions occurred from a circular crater excavated in the summit of the dome. Observations of crater development were hampered by cloud cover, but the following sequence is deduced. Rockfalls over Gages Wall at the end of July and on 1 and 2 August generated a summit crater with a low lip to the west. Significant enlargement took place on 3 August during the 18:10 collapse. Further enlargement of the crater took place during the explosions of 4 August, which discharged large quantities of dense dome rock as well as pumice. By midday on 5 August (after the explosion of 04:45, but before that of 16:57; Table 1), there existed a circular, funnel-shaped crater at the top of the dome with a rim diameter of 300 20 m. The crater was seen again clearly at 07:30 on 7 August, when the highest point on the northern rim lay at 940 m above sea level, and that on the southern rim at 980m. There was also a low lip in the crater wall (870 m) above Gages valley, showing that the crater was at least 110m deep. The crater persisted through the explosion of the morning of 8 August, and was seen from MVO South at 17:00 on 9 August. When seen again at midday on 10 August, a small new lobe of lava nested within the crater was visible above the western crater lip. Following the final explosion on 12 August, lava continued to be extruded within the crater.

284

T. H. DRUITT ET AL.

Table 1. Characteristics of the Vulcanian explosions of August, September and October, 1997 Date

4 Aug. 4 Aug. 5 Aug. 5 Aug. 6 Aug. 6 Aug. 7 Aug. 7 Aug. 7 Aug. 8 Aug. 8 Aug. 1 1 Aug. 12 Aug. 22 22 22 23 24 24 24 25 25 25 26 26 27 27 27 28 28 28 29 29 29 29 30 30 1 1 1 2 2 2 4

Sep. Sep. Sep. Sep. Sep. Sep. Sep. Sep. Sep. Sep. Sep. Sep. Sep. Sep. Sep. Sep. Sep. Sep. Sep. Sep. Sep. Sep. Sep. Sep. Oct. Oct. Oct. Oct. Oct. Oct. Oct.

Local time*

06:30 16:43 04:45 16:57 04:02 14:36 00:34 12:05 21:55 10:32 20:51 11:38 10:12 00:57 10:45 20:42 07:23 00:34 10:54 17:16 03:54 11:09 20:05 04:25 14:56 00:01 09:46 17:15 04:28 10:34 23:03 06:26 11:23 16:48 21:57 04:43 17:44 05:00 11:34 17:40 01:05 12:53 22:50 08:33

Intervalf (hrmin)

10:13 12:02 12:12 11:05 10:34 9:58 11:31 9:50 12:37 10:19 62:47 22:35

9:48 9:57 10:41 17:11 10:20 6:22 10:38 7:15 8:56 8:20 10:31 9:05 9:45 7:29 11:13 6:06 12:29 7:23 4:57 5:25 5:09 6:46 13:01 11:16 6:34 6:06 7:25 11:48 9:57 33:43

Plume height! (km)

3.0 4.6 6.1-10.7 9.1 9.1-12.2 11.0-12.2 12.2-13.7§ 9.8 9.1-10.7 12.2 11.0

>9.1 10.7-12.2 9.1-10.7 10.7-12.2 ~7.6 12.2

Pyroclastic flows

Date

Ta W Tu M Ty G Wr

X

X

X

O

X

X

0

X

X

X

0

X

X

0

X

X

X

X

X

X

X

X

X

X

X

X

X

X

0

X

X

X

X

X

X

X

0

X

X

X

X

X

X

X

X

X

X

X

X

0

0

0

X

X

0

X

X

0

X

X

0

X

X

0

X

X

0

X

X

0

X

X

O

X

0

12.2 12.2 7.6 12.2§ 12.2 <15.2§ 13.7 13.7

X

X

X

X

X

X

X

X

X

0 0

X

0

X

X

X

X

x

X

0

O O

X

X

X

4.6-7.6 10.7-12.2 >12.2 12.2-13.7 9.1-13.7 <12.2 <<10.7 ~4.6

O

X

X

X

0

X

X

X

X

X

X

X

X

X

X

X

X

X

X

X

X X

X

12.2 9.1 11.0

Local time*

X

X

X

X

X

X

X

X

X

X

X

X

4 Oct. 5 Oct. 5 Oct. 5 Oct. 6 Oct. 6 Oct. 6 Oct. 7 Oct. 7 Oct. 8 Oct. 8 Oct. 9 Oct. 9 Oct. 10 Oct. 10 Oct. 1 1 Oct. 12 Oct. 12 Oct. 13 Oct. 13 Oct. 14 Oct. 14 Oct. 14 Oct. 15 Oct. 15 Oct. 15 Oct. 15 Oct. 16 Oct. 16 Oct. 16 Oct. 16 Oct. 16 Oct. 17 Oct. 17 Oct. 17 Oct. 17 Oct. 18 Oct. 18 Oct. 19 Oct. 19 Oct. 20 Oct. 20 Oct. 21 Oct. 21 Oct.

18:27 02:53 10:41 18:41 02:44 10:42 17:50 04:06 16:02 03:47 15:10 03:03 12:32 04:13 18:40 17:57 07:55 22:24 09:32 15:24 01:36 13:48 23:16 05:47 08:33 14:50 22:20 02:51 06:35 09:44 14:20 18:48 04:01 12:35 16:05 23:18 06:48 15:17 05:13 21:27 05:04 15:13 11:39 19:02

Interval! (hr:min) (km)

Plume height;

9:54 8:26 7:48 8:00 8:03 7:58 7:08 10:16 11:56 11:45 11:23 11:53 9:29 15:41 14:27 23:17 13:58 14:29 11:08 5:52 10:12 12:12 9:28 6:31 2:46 6:17 7:30 4:31 3:44 3:09 4:36 4:28 9:13 8:34 3:30 7:13 7:30 8:29 13:56 16:14 7:37 10:09 20:26 7:23

12.2-13.7 7.6-9.1 7.6-9.1 5.3 6.9 7.6 <12.2 >9.8 13.7 <9.1-10.7 ~12.2 10.1 12.2 12.5 9.4 6.1-7.6

Pyroclastic flows

Ta W Tu M Ty G Wr x

x

x X

X

X

X

X

X

X

X

X

X

X X

X

x

x

X

X

X

O

X

x

X

X

X

X

X

X

X

X

X

X

X

X

X

X

X

X

X

X

x x x x

X

X

X

X

X

X

x x x x

>4.6

3.1-7.6 4.6-6.1 >4.6 >3.7 3.7-4.6 4.6 3.1-4.6 6.1 7.6 7.6-9.1 6.7 9.1 4.6-7.6 12.2 <9.1 5.5-9.1 7.6 9.1-10.7 >9.1 10.7

x

x

X

X

X

X

X

X

O

O

O

0

0

O

O

O

O

O

O

O

x

X

X

0

O

O

x

X

X X

O

O

O

0

x x x X X

X X

X

x x

X

X

X

X

X

X

X

X X

X

X

X

x

* Local time is 4 hours behind universal time. fTime interval since the previous explosion. | Best estimates of maximum plume heights (above sea level) based on National Oceanic and Atmospheric Administration (NOAA) data, supplemented by Montserrat and Trinidad airport reports and by Abney-level measurements of the plume top made from the MVO. A blank denotes lack of data. § Additional height estimates were obtained using plume-top temperatures on GOES satellite images (Bonadonna el al. 2002b) for three explosions: 12:05, 7 Aug. (9.3km); 14.56, 26 Sep. (11.3km); 09:46, 27 Sep. (10.8km). f Presence (x) or absence (o) of fountain-collapse pumice-and-ash flows. Ta, Tar River valley; W, White's Ghaut; Tu. Tuitfs Ghaut; M. Mosquito Ghaut; Ty, Tyre's Ghaut; G, Gages valley; Wr, White River valley. A blank denotes lack of observations.

Following the initiation of Vulcanian activity in early August, major changes were made to the volcanic hazards map of the island, resulting in an enlargement of the evacuated zone and northward displacement of inhabitants under threat from this new style of activity (Kokelaar 2002). Renewed dome growth after the August explosions The new dome lobe in the summit crater developed a large spine, the elevation of which was 950m by 13 August. By 14 August the crater was nearly filled in, and by 19 August the entire dome had

reached its pre-3-August volume, growth being concentrated in an area about 150m wide above Gages valley. Initially during this period the seismic activity continued to be characterized by intense hybrid earthquake swarms occurring at approximately 8-hour intervals. The intense hybrid activity that characterized the August explosive period continued until 19 August, with block-and-ash flows occurring mainly in the upper Gages valley from the growing dome, immediately after peaks in earthquake activity. Throughout late August and early September, rockfall and pyroclastic flow signals dominated the seismic records, with activity confined to the western and northern flanks of the dome. A small dome collapse (about 1 x 10 6 m 3 ) took place on 30 August. A slow

EPISODES OF CYCLIC EXPLOSIVE ACTIVITY

increase in block-and-ash flow activity occurred through midSeptember, with material being shed across Farrell's Plain and into Tuitt's Ghaut. Rapid dome growth was occurring at this time, with the unstable active face located above the northern wall of the crater. Earthquake activity remained at a low level, with occasional swarms of hybrid earthquakes, some of which included the largest individual events recorded by the broadband network up to that time. Heightened long-period earthquake activity was also notable, with events often occurring immediately before pyroclastic flow generation. At least one long-period event, on 16 September, correlated with an audible detonation from the dome. By 28 August, the dome volume was more than 80 x 106 m3 and the total volume of magma erupted to form both lava and pyroclastic deposits was 160 x 106m3 DRE.

Dome collapse of 21 September Hybrid earthquake swarms recommenced 24 hours prior to the dome collapse of 03:54 on 21 September and continued after the collapse at the same level. The swarms were neither long nor intense compared with previous ones; individual events were not large and neither were there visual signs of imminent collapse. The onset of increased seismic amplitude related to this collapse is timed at 03:54, although the first 8 minutes of activity was not at a high level, probably registering small dome collapses with pyroclastic flows in the upper and middle parts of Tuitt's Ghaut. A sharp increase in signal amplitude at 04:02 marked the start of sustained high-amplitude signals on all stations, typical of those generated during major dome collapse. Minor pulsing occurred throughout this phase, which lasted until 04:17, a total of 15 minutes. Within this phase, the highest amplitude signals were recorded between 04:11:1 and 04:13:7; the signal amplitude at this time was markedly higher than at any other time, peaking at 04:12:1. The signal had dropped to background level by 04:24, giving a total duration for the dome collapse of 30 minutes. During the collapse, which involved 11.0 x 106m3 (DRE) of the dome, block-and-ash flows moved down Tuitt's Ghaut and White's Ghaut to the ocean, spreading out over the area of Spanish Point (Cole et al. 2002). Associated pyroclastic surges covered interfluves between the ghauts, causing the burning of Tuitt's and parts of other villages not overrun by the flows. The ash plume associated with the block-and-ash flows reached an altitude of 9-12 km, causing ash fall over much of Montserrat. The post-collapse dome had a deep scallop-shaped scar on its northern flank, extending back about 300m. There was a

Fig. 2. The crater formed early during the explosion sequence of September and October 1997. The photo is taken from the NE. The crater has a rim diameter of about 300 m and opens into Tuitt's Ghaut.

285

prominent opening and chute above the head of Tuitt's Ghaut, eroded by pyroclastic flows as dome collapse proceeded.

The September and October explosions The first explosion of this episode occurred about 20 hours later at 00:55 on 22 September and the last at 19:02 on 21 October. A total of 75 explosions occurred over a 30-day period (Table 1). Intervals varied between 2.5 and 33.5 hours, with an average of 9.5 hours. No systematic variation of plume height with time occurred from 22 September until 11 October, when there occurred a series of at least ten relatively weak explosions with plume heights of 7 km or less over a period of five days (Table 1). Gaps in the plume-height record at this time mean, however, that it cannot be excluded that larger explosions also occurred. After 16 October, plume heights increased again until the end of the explosive episode on 21 October. The first explosion produced a crater at the southern edge of the 21 September collapse scar. The shape and size of this crater changed only gradually thereafter, becoming larger in diameter and probably deeper. The volume of the dome at this point was 68 x 10 6 m 3 . An estimate of the volume of the crater excavated during the first few explosions was 2 x 106m3 (Fig. 2). Successive explosions deposited a tephra rampart on the northern side of the crater, effectively completing the near-circular crater wall across the 21 September collapse scar. Minor reaming of the crater walls occurred throughout the explosion sequence, but good views into the crater were scarce so that accurate estimates of the volume increases were impossible to obtain. Within a day of the cessation of explosions on 21 October, a lava lobe was seen growing within the crater. Within a few days, lava had filled the crater and overspilled the tephra rampart, and by 3 November had largely filled the entire 21 September collapse scar. Observations also suggest that a small lobe started to grow in the base of the explosion crater during the longest break between explosions (2 to 4 October; Table 1).

The Vulcanian explosions The characteristics of the explosions are listed in Table 1. Of the 88 explosions, 37 occurred during hours of darkness and about 30 under poor weather conditions. About 20 explosions were well observed and documented, five of which were at night. The following descriptions are based on field observations and on ongoing

286

T. H. DRUITT ET AL.

Fig. 3. Sequence of events during a typical Vulcanian explosion in 1997.

analysis of video footage. To a first approximation, the explosions and their products were similar and they are described together. The main features of a typical explosion are shown schematically in Figure 3. Sequences of photographs of three typical explosions are shown in Figures 4, 5 and 6.

General description of the explosions Visual precursors to the explosions were rare, although on several occasions increases in fumarolic activity around the dome were noted in the preceding seconds to minutes. Each explosion consisted of two fairly well defined phases: (1) an initial, high-intensity phase lasting about 10 minutes and including the main explosion and peak magma discharge rates (a few tens of seconds), fountain collapse, formation of a buoyant eruption plume, and the ascent of the plume to its neutral buoyancy level in the atmosphere; (2) a drawn-out, much lower-intensity waning phase lasting a few tens of minutes (typically 1 to 3 hours) and characterized by relatively weak, pulsatory venting of gas and ash. Fountain collapse was limited to the first 10-20 seconds of each explosion. Each explosion began with the rapid rise of numerous dark-grey finger jets of ash and debris (Fig. 3a). Condensation of atmospheric moisture ahead of the jets, indicative of shock waves, was observed

in some explosions (Clarke el al. 2002). There then followed a loud and steady roaring, like an aircraft, punctuated by further detonations as subsequent jets were discharged. The initial jets were relatively weak, but subsequent ones became progressively more vigorous with time over the first few seconds of the explosion. As the jets rose, they decelerated rapidly. Decimetre- to metre-sized ballistic blocks detached from the leading edges of many jets and were thrown outwards in high, curving arcs as far as 1.7km from the dome. Impact of the blocks with the ground kicked up clouds of ash visible from a distance (Fig. 6). The finger jets, having reached their maximum elevation, then collapsed back towards the ground. Simultaneously, part of the material that had ingested enough air began to rise as one or more buoyant plumes, which then merged into a single large plume. Viewed from a distance, the collapsing fountain took the form of a hemispherical cap, through which the buoyant plume then pierced (Fig. 3b). The fountain-collapse height was typically between 300 and 650 m above the crater rim, although, in some explosions, later jets remained momentum-driven up to more than 1000 m. Impact of the collapsing material with the ground generated highly expanded pyroclastic surges, which swept out from the volcano with frontal velocities of 30-60 m s - 1 (Figs 3c, 4, 5 and 6). On some occasions, the surges moved almost as quickly horizontally as the central plume was ascending vertically, so that the explosion cloud appeared to expand equally in all directions. The surges decelerated rapidly 1-2 km from the dome, then lifted

EPISODES OF CYCLIC EXPLOSIVE ACTIVITY

287

Fig. 4. Photographs of the explosion of 12:05 on 7 August 1997. The plume from this explosion ultimately rose to about 13 km. Times after the onset of the seismic signal: (a) 27 s, (b) 36 s, (c) 42 s, (d) 48s, (e) 58s, (f) 76s. The collapsing fountain is visible to the left in (a) and to the right in (b) and (c). Fountain collapse generated pyroclastic surges and flows in all the major drainages around the volcano. The three individual plumes shown in Fig. 15 and discussed in the text are numbered 1 to 3; g, Gages Mountain; sg, St George's Hill. For scale, the top of the plume in (e) is about 3 km above sea level and 2 km above the lava dome. Photographs taken from NW of the volcano by K. West.

off the ground to form buoyant ash plumes (Fig. 3d). Shortly thereafter, highly concentrated pumice-and-ash pyroclastic flows were observed advancing at about 10ms - 1 down one or more valleys around the dome as thin (0.5-1 m), granular avalanches with associated billowing clouds of elutriated ash. It is surmised that the pyroclastic flows formed by rapid fallout of debris from the collapsing fountain and initial pyroclastic surges. In most of the explosions, pyroclastic flows occurred in all major ghauts around the dome. Some explosions in September and October were angled to the north, probably because the horseshoe-shaped dome summit crater opened in that direction (Fig. 2). Runout distances varied between explosions, but were typically 3-6 km, with the flows taking up to a few hundred seconds to reach their distal limits. Only one explo-

sion (06:35 on 16 October) did not generate pyroclastic flows, and another generated only very small ones (05:47 on 15 October). Fountain collapse was clearly visible at night. First, a brightly incandescent cloud was seen rising over the dome. Moments later, a ring of coarse, incandescent debris fell back from height along steep, outwardly inclined trajectories onto the slopes surrounding the dome. Fountain collapse was short-lived, no more than 10 to 20 seconds. Incandescence in the initial pyroclastic surges disappeared rapidly over a few seconds as the surges entrained air and cooled. The central explosion plumes had fast-rising bulbous heads and narrow central stems (Figs 4, 5 and 6) and reached heights of 3-15 km, with an average of about 10 km (Figs 7 and 8). The height

288

T. H. DRUITT ET AL.

Fig. 4.

estimates (probably ) were supplied by the US National Oceanic and Atmospheric Administration (NOAA) Satellite Analysis Branch in Washington DC, based on ash-cloud movements tied to radiosonde wind data from Puerto Rico and Guadaloupe. Additional ground-based height estimates made on Montserrat using an Abney level were in broad agreement with the NOAA heights. For three explosions, height estimates were obtained using plume-top temperatures from GOES satellite images (Bonadonna et al. 2002b; Table 1). Upon attaining their maximum altitude, which typically took about 10 minutes, the plume heads spread out to form umbrella clouds (Fig. 9), which then detached from their stems and were carried to the north or NW by high-level (8-18 km) winds. Satellite images of one such cloud (14:56 on 26 September) are presented by

(continued)

Bonadonna et al (2002/?). Some plumes were richer in steam and thus paler in colour than others and tended not to rise as high. The weaker plumes also had more poorly developed umbrellas, and were more liable to be blown off-centre by the wind. Ash clouds generated by the lofting of the pyroclastic surges and by elutriation from pyroclastic flows were gradually drawn up and incorporated into the central plume by inward-moving currents of air (Fig. 3d). After several minutes, each explosion settled into a phase of waning, relatively low-intensity discharge, generating a low, bentover plume transported mainly to the west or NW on low-level trade winds (below about 5 km). This decoupling of the high umbrella (north or NE) and low, waning plume (west or NW) was characteristic of many of the explosions. The waning plume then

EPISODES OF CYCLIC EXPLOSIVE ACTIVITY

289

Fig. 4. (continued)

decreased slowly in height over a period of 1-3 hours. Venting during the waning phase was often pulsatory on a time-scale of 1-2 minutes. Lightning occurred during many of the explosions, and was particularly evident at night. It appeared in the eruption column a few seconds after the initial explosion and continued for up to 10 minutes. Cloud-to-cloud strikes accompanied by thunderclaps were most common. Despite the presence of condensed steam in some of the plumes, it is not believed that external water played any significant role in the explosions. The groundwater system of Soufriere Hills Volcano had essentially dried out by mid-1997 after 20 months of lava extrusion, and there was no evidence of a significant source of groundwater.

The explosion of 12:05 on 7 August, 1997 This explosion was studied in particular detail from video footage. The events and their timing are listed in Table 2, which is exploited later to compare with calculations. The explosion was filmed by a time-lapse video camera mounted at MVO South (frame interval 1.4 s), two ordinary video cameras at MVO South and at Fleming, and an ordinary video camera aboard the MVO helicopter. Still photographs were taken roughly every 2 s from MVO South until 45 s into the explosion. Another set of images taken by K. West is shown in Figure 4. Correlation of all image sets permitted detailed reconstruction of the explosion. The timer on the Fleming video s camera had been previously correlated with GPS time to

Fig. 5. Explosion at 14.02 on 6 August 1997, showing the typical development of the buoyant plume, which rose to between 9 and 12 km above sea level. Partial fountain collapse during the initial stages of the explosion sent pyroclastic flows and surges down valleys to the north (left) and west (right). either side of Gages Mountain (g). Photographs taken from MVO South by T. H. Druitt.

Fig. 6. Explosion at 15:13 on 20 October 1997. Fountain collapse generated pyroclastic surges and flows visible to the west (right) and north (left) of Gages Mountain (g). Ash was thrown up by the ground impact of ballistic blocks (b). The buoyant plume ultimately rose to about 10km. Note the buildings for scale in the foreground. Photographs taken from the NW by P. Cole.

292

T. H. DRUITT ET AL.

Fig. 7. Histograms of plume height and explosion repeat interval for the 1997 Vulcanian explosions.

Fig. 8. Variations of (a) plume height and (b) explosion repeat interval with time during the episode from 22 September to 21 October 1997. The repeat interval is that which preceded the explosion concerned, which is why none is reported for the first explosion. The horizontal axis shows the explosion number during this period (see Table 1). An absence of bars indicates a lack of data. In mid-October there occurred a series of relatively weak explosions with short repeat intervals.

allowing correlation of images with the broadband seismic signal measured by the Galway's Estate seismometer (Fig. 10). Time zero in Table 2 is taken as the onset of the seismic signal (as received at the Galway's Estate seismometer) at 12:04:44. Emer-

gence of the first jet above the crater rim, as estimated by backextrapolation of height-time curves (see below), followed 1 s later. Given that the velocity of long-period seismic waves at Montserrat is 1800-1900ms-1 (Neuberg et al. 1998). the travel time for the

EPISODES OF CYCLIC EXPLOSIVE ACTIVITY

293

Fig. 9. (a, b) Development of the umbrella cloud during a typical explosion (09:46 on 27 September 1997). The plume height in (b) is about 10.8 km above sea level. Photograph taken from a boat NE of Montserrat by B. Poyer. The boat was retreating from Montserrat, so that (a) was taken from closer than (b).

seismic signal from the dome to the seismometer was about 1.5 s. About 2.5 s therefore separated the onset of the seismic signal and emergence of the first jet above the crater rim. Given the vertical exit velocity of this jet (80 10ms - 1 , see below), this suggests a depth of about 200m for the summit crater at the time of the explosion, which is compatible with the observed crater diameter (300 20 m) and a typical angle of rock stability. The explosion generated at least three separate plumes (numbered 1 to 3) visible from MVO South and Fleming that merged, after about 100s, into a single, large plume (Fig. 4). There may have been other plumes not visible from these observation points. Plume 1 emerged first over the southern or southwestern flank of the dome, rapidly followed by plume 2 to the north (Fig. 4a). Each decelerated as it rose, part collapsing back to the ground, and part ascending buoyantly. The fallback height of the two plumes was estimated visually as a few hundred metres above the crater rim and the collapsing material first hit the ground behind Gages Mountain about 18s into the explosion (Table 2). The resulting pyroclastic surge was first visible behind Gages Mountain at 22.8 s. As this surge travelled out from the dome, a thin veil of ash was thrown up all over Gages Mountain, either due to seismic shaking or to a blast of preceding air. The surge then ramped over the north face of Gages Mountain before decelerating abruptly, ceasing forward motion, and lifting buoyantly off the ground at about 45 s (Fig. 4b, c and d). Collapse over the northern flanks lagged a few seconds behind that over Gages. The northern fallback curtain was first observed at 19.1 s and it hit the ground three seconds later (Fig. 4a). The resulting surge was first visible at 27.8 s advancing at about 45 m s-1 down the headwaters of Mosquito Ghaut.

Large blocks thrown northwards ahead of plume 2 followed ballistic trajectories. The first blocks were seen to hit the ground at 21.6s, throwing up ash. Impacts then migrated northwards away from the dome, reaching the maximum range of 1.6 km about 5 s later (26.9 s). At 27.9 s, as plumes 1 and 2 became buoyant, plume 3 broke out at high speed at the top of the column (Fig. 4b). It rapidly decelerated and began to rise buoyantly, before separating into two (Fig. 4c and d). Plume 3 remained momentum-driven up to 1200m above the crater rim. After a few tens of seconds, thin pumice-and-ash pyroclastic flows were observed travelling slowly down Tuitt's Ghaut, Mosquito Ghaut and the Tar River valley, reaching their maximum limits of 3-6 km about 200 s after the onset of the explosion. The central plume rose to between 12.2 and 13.7km according to NOAA data, forming a large umbrella. The waning phase of the explosion lasted an hour.

Explosion products Fallout tephra from the explosions had three sources (Bonadonna et al. 2002b). (1) Pumiceous (and minor lithic) blocks, lapilli and ash from the main central plumes and umbrella clouds. Pumice clasts as large as 10cm in mean diameter fell on St George's Hill, 6.5cm on the South Soufriere Hills, 6.5cm on northern Plymouth, 4.5cm on Windy Hill and at Cork Hill, 4cm at MVO South and 2 cm in northern Montserrat. (2) Ash from plumes generated by the lofting of pyroclastic surges and by elutriation from pyroclastic flows (termed co-pyroclastic-flow plumes). This was mostly drawn

294

T. H. DRUITT ET AL.

Table 2. The 12:05 explosion of 7 August 1997 Time (s)

Event

0 1 7 17 17.4 18

Start of explosion seismic signal (phase 1)* Emergence of explosion jet 1 at 95 1 0 m s - 1 Emergence of explosion jet 2 at 95 10 m s - 1 Emergence of explosion jet 3 at > 130ms - 1 Fallout visible behind Gages Mountain from MVO South Start of seismic signal from fountain collapse and pyroclastic flows (phase 2) Fallout curtain descending over the north flank First ballistics hit Farrell's Plain, 1.2km north of the vent First ballistics hit Paradise Plain, 1.2km north of the vent Collapsing fountain hits the north flank Pyroclastic surge visible behind Gages Mountain Ballistics reach maximum range on Paradise Plain, 1.6km north of the vent Pyroclastic surge in Mosquito Ghaut, 1.7km from source, travelling at c. 4 5 m s - 1 Jet 3 arrives at the top of the plume Pyroclastic surge passes Gages soufriere on the west flank Pyroclastic surge ramps over Gage's Mountain and lofts Pyroclastic surge reaches maximum runout on the Farrell's Plain and begins to loft Drop in intensity of the phase 2 seismic signal Pyroclastic flows reach the foot of St George's Hill on the west flank Pyroclastic flow reaches the Paradise River, 3.5km from source, at 10ms - 1 Pyroclastic flow level with Harris, 3.4km from source, at 9ms- 1 Pyroclastic flow reaches sea on Tar River delta, 3.3 km from source, at 13-25ms-1 End of pyroclastic flow seismic signal; continuing tremor (phase 3) End of the explosive eruption

19.1 21.6 22.0 22.2 22.8 26.9 27.8 27.9 34.3 45 58 70 108 167 187 202 300 c. 3600

*The seismic signal was measured at the Galway's Estate station (Fig. 10). The time for seismic waves to reach this station from the dome was about 1.5s, so emergence of jet 1 occurred about 2.5s after the onset of the explosion seismic activity.

up and mixed into the central plumes; however, where observed separately, fallout from this source was often sporadic due to generation from surges and flows in different sectors around the volcano. (3) Ash from the low, waning plume. Except within a couple of kilometres from the lava dome, the total fallout from individual explosions seldom exceeded just a scattering of clasts or a layer of ash no more than a few millimetres thick. Fallout distributions were complicated by mixing of material from the three sources and by variable wind directions, wind intensities and plume heights. Western Montserrat was affected by fallout from all three sources, whereas fallout from the umbrellas dominated in the north. A map of the areas impacted by the fountain-collapse pyroclastic flows and surges is given in Figure 10. Pyroclastic flows travelled down the Tar River and White River valleys as far as the sea, and down Fort Ghaut to within a few hundred metres of the shoreline. Pyroclastic flows discharged to the north by the August explosions travelled down Tuitt's and Mosquito Ghauts, then on as far as the village of Farm (6 km). Those in September and October did the same, but also entered White's Ghaut and spread out over the surface of the 21 September dome-collapse block-and-ash flow deposit (Cole et al. 2002), locally reaching the sea. Some pyroclastic flows in September and October were channelled preferentially down Tuitt's Ghaut because the explosions were angled to the north. Many explosions sent flows down Tyre's Ghaut, then into Dyer's River valley. The deposits from the pyroclastic flows are described by Cole et al. (2002). They consisted of numerous anastomosing lobes with multiple breakouts (Fig. 11). The distal ends of individual lobes were typically 10-50m wide and 0.5-1 m thick (Fig. 11c,d), with

well defined levees and snouts rich in relatively low-density pumice boulders. Temperature measurements made approximately 2 hours after flow emplacement ranged from 180 to 220°C. This is consistent with the night video footage of the explosions, which shows that the discharging material lost heat very rapidly during fountain collapse, presumably by entrainment of air, as also seen in numerical models of the explosions (Clarke et al. 2002). Vesicularities of clasts (>3cm) from the explosion deposits were calculated from density measurements made by the waterimmersion technique. Vesicularities of pumice clasts from both fallout and pyroclastic flows (from all but the first three explosions in August) ranged from 55 to 75vol% (26 clasts). In contrast, ballistic blocks discharged during the initial vent-clearing phase of each explosion were mostly dense, with Vesicularities much lower than 55%. Dense clasts also occurred in the pyroclastic flows, but only up to 5% by volume (Cole et al. 2002; Clarke et al 2002). These are interpreted as fragments of crater rubble, degassed magma plugging the conduit prior to each explosion, or material picked up from the ground. No systematic differences in pumice vesicularity were observed between different explosions, with the exception of the first three of the August series, in which the Vesicularities of fallout clasts ranged from 0 to 75 vol%. The ejection of important quantities of dense andesite as fallout could be due to reaming out of the August summit crater by these early explosions. It is not known if a similar process took place in September. A notable feature of the fallout pumices was the high abundance of tabular clasts with planar or curviplanar surfaces and sharp edges (Fig. 12). Many pumices were lenticular in cross-section, with one flat face and a convex face on the other side. Between 20 and 50% of fallout pumices a few centimetres in diameter had tabular shapes. Many had been slightly deformed and rounded by minor post-fragmentation inflation, showing that their angular shape was not due to ground impact. Neither were they associated on the ground with other pieces of the same block, as expected if impact breakage had occurred. The shapes of the clasts are attributed to brittle fragmentation of an already vesicular magmatic foam with at least 55 vol% bubbles. In contrast, pumice blocks in pyroclastic flows from the explosions were typically rounded and subequant due to abrasion during fountain collapse and transport.

Cyclic patterns of edifice deformation, seismicity and explosion Cycles of edifice deformation and hybrid seismicity were closely associated with the August explosive activity (Voight et al. 1998, 1999). A typical cycle consisted of slow inflation of the dome followed by rapid deflation and explosion, giving a saw-tooth pattern on tiltmeter records (Fig. 13a). Seismic swarms began up to several hours before each explosion and culminated near the tilt peak or during the deflation phase at between one and four events per minute (Fig. 13b, c). Deformation records ceased during the explosion of 16:57 on 5 August, when the only operating tiltmeter (Fig. 10) was destroyed, but the cycles continued to be evident from the seismic record. All but one of the August explosions occurred shortly before or shortly after the peak in seismicity. The close correlation between seismicity and eruption onset permitted accurate prediction of the explosions during this period. Repeat intervals between the August explosions ranged from 10 to 14 hours, except for the last two of the series, which were preceded by intervals of 63 and 23 hours respectively (Table 1). In contrast, there was much less precursory seismicity before each of the September and October explosions. Precursor seismic swarms occurred on 14 occasions, although only a few included periods of intense seismicity such as those preceding most of the August explosions. When a swarm did occur, it tended to continue for a short time after the explosion. On no occasion did the precursor seismicity develop in such a way as to enable accurate forecasting of explosions as in August. There was also no tiltmeter functioning at the time, so there are no tilt data from this period.

EPISODES OF CYCLIC EXPLOSIVE ACTIVITY

295

Fig. 10. Map of the pyroclastic flow and pyroclastic surge deposits from the 1997 explosions. The locations of the seismometers and tiltmeter operational at the time are also shown. The tiltmeter was destroyed at 16:57 on 5 August and was not replaced until after the explosions had finished.

Explosion forecasting in September and October relied on the observed periodicity of the explosions themselves, which ranged from 2.5 to 33.5 hours, with an average of 9.5 hours (Figs 8 and 9). Seismic signals from the explosions Each explosion generated a seismic signal captured by the MVO broadband system. A typical example (16:57 explosion on 5 August 1997) is shown in Figure 14a, with a blow-up of the first 1.4 min in Figure 14b. Each signal began abruptly and remained at a high level for several minutes, then decayed over a period of a few tens of minutes to 3 hours. Selected signal parameters are listed in Table 3. There were three discrete phases to each signal: (1) a long-period part, typically of 10-20 s duration; (2) a higher amplitude pyroclastic flow signal lasting a few minutes; and (3) harmonic tremor lasting 1-3 hours (average 80 minutes) during the long period of waning discharge that terminated each explosion. Phases 1 and 3 had similar frequency spectra, with the main energy between 0.6 and 1.7 Hz (Fig. 14c). Phase 2 contained much more energy distributed over a range of higher frequencies, principally 2 to 20 Hz. Filtering of the raw signals enabled us to separate the signals into low-frequency and high-frequency components (Fig. 14a and b).

We applied a time-domain recursive Butterworth filter between 0.5 and 1 Hz and a high-pass filter at 2 Hz. The low-frequency component was present throughout all three phases of the seismic signal, whereas the high-frequency component occurred only during phase 2 when it was superimposed on the low-frequency one. The low-frequency component is interpreted as the vibrational response of the magmatic conduit to the explosion itself (Neuberg & O'Gorman 2002). During most explosions, it remained at a high level for about 45-70 s before falling to a much lower level, which agrees with visual observations for the duration of peak discharge. The low-frequency component fluctuated in intensity with time, as visible on Figure 14b for the explosion at 16:57 on 5 August. This is typical of the amplitude modulation of a longperiod seismic signal (Neuberg & O'Gorman 2002), and is not thought to be due to eruption unsteadiness. Video analysis of the 12:05 explosion of 7 August revealed no obvious correlation between eruptive intensity and the amplitude of the low-frequency seismic component. The high-frequency component of phase 2 was due to a combination of ballistic impact, fountain collapse and pyroclastic flow, and had a frequency spectrum typical of pyroclastic flow signals, for example at Montserrat (Miller et al. 1998) and Mount Unzen, Japan (Uhira et al. 1994). Its onset coincided with the first impact of the

296

T. H. DRUITT ET AL.

Fig. 11. Pyroclastic flow deposits from the 1997 Vulcanian explosion., (a) Buff-coloured pumice-and-ash pyroclastic flow deposits from the August explosive period overlying grey block-and-ash flow deposits of the 25 June 1997 dome collapse. View looking NNE down Tuitts Ghaut and Pea Ghaut. Paradise River valley joins Tuitt's Ghaut from the left. (b) Buff-coloured pumice-and-ash pyroclastic flow deposits from August 1997 in White River valley. The dome lies hidden in cloud to the right. Road for scale. (c) Pumice-rich snout of a pyroclastic flow lobe from the September/October 1997 explosions, near Spanish Point. The lobe is about 1 m high and about 12m across. It overlies block-and-ash flow deposits of the 21 September 1997 dome collapse. (d) Pyroclastic flow lobes from the September/October 1997 explosions overlying block-and-ash flow deposits of the 21 September 1997 dome collapse. Houses of the community of Spanish Point on the left. The lobes have well defined levees and snouts rich in coarse pumice boulders.

collapsing fountain and ballistic blocks with the ground. This is confirmed by analysis of the 12:05 explosion of 7 August, in which phase 1 lasted 18s, and the end of phase 1 corresponded approximately with the collapsing fountain hitting the flank of the volcano behind Gages Mountain (Table 2). Ballistics were first observed to hit the north flank of the volcano 21.6 s into this explosion and the collapsing fountain touched down 0.6 s later. The 10-20 s duration of seismic phase 1 is therefore the transit time for fountain collapse, i.e. the time for the momentum-dominated

eruption jets and ballistic blocks to reach their maximum height, then fall to the ground. The duration of the high-frequency signal (i.e. of phase 2) corresponded with the time necessary for the pyroclastic flows to reach their distal limits (200-500 s). The abrupt drop in intensity of the phase 2 signal about 55 s after the explosion onset (arrow, Fig. 14a) is believed to be due to pyroclastic flows nearest the seismometer concerned ceasing movement, while those in other valleys further away were still in motion, resulting in a drop in apparent seismic energy production.

EPISODES OF CYCLIC EXPLOSIVE ACTIVITY

297

Fig. 12. Fallout pumices from the August explosions. The fragments are tabular with angular edges and curviplanar surfaces, and were formed by brittle fragmentation of a pressurized magmatic foam resident in the conduit prior to each explosion. The scale is in centimetres.

Fig. 13. (a) Tiltmeter and (b, c) seismic data for the period 30 July to 14 August 1997. The tiltmeter (Fig. 10) was destroyed by the explosion at 16:57 on 5 August. Cycles of slow inflation of the dome, followed by rapid deflation, are evident from the tilt data. Seismic data are from the St George's Hill seismometer (Fig. 10) and show cyclic variations in total seismic amplitude (RSAM) and number of triggers per 10 minutes that are in phase with the tilt cycles. Explosions are marked by the letter E and the dome collapse of 18:10 on 3 August by the letter C. The events marked P were the two relatively weak initial explosions of 4 August (Table 1).

Eruptive volumes We now estimate the volumes of magma discharged during the explosions and the partitioning of material between fallout tephra and pyroclastic flows. The fallout includes that from (1) the main, central plume and umbrella, (2) the co-pyroclastic-flow plumes, and (3) the low, waning plume. We use the following mean densities based on field and laboratory measurements: airfall ash 1100 kgm - 3 , pumice-and-ash flow deposit 1350 k g m - 3 , and dense juvenile andesite2600kgm - 3 .

The uncompacted volume of fallout tephra from the months of August, September and October has been estimated by Bonadonna et al (2002b) at 22 x10 6 m 3 . All but 106m3 of this (21 x 106m3 or 8.9 x 10 6 m 3 DRE) is attributable to the 88 Vulcanian explosions. The estimate is based on extrapolating isopach data to infinity using an exponential decay model. It is thought to be a minimum estimate, at least 10% too low, because study of GOES satellite images shows that considerable quantities of very fine (2-20 um) ash in fact travel further than predicted by the exponential model (Bonadonna et al. 2002b). Raising the figure by 10% and dividing

298

T. H. DRUITT ET AL.

by 88 gives an average of at least 1.1 x 10 5 m 3 DRE of fallout tephra per explosion. Another estimate of fallout volumes can be obtained from plume height data (Table 1). The maximum ascent heights of the

explosion plumes ranged from 3 to 15km. Since the duration of peak discharge (c. 50 s) was an order of magnitude shorter than the time for the plumes to reach their maximum altitude (c. 500 s), the plumes can be treated as discrete thermals of a given initial mass. This agrees with visual observations that the plumes developed the bulbous heads characteristic of thermals (Figs 4, 5 and 6). The ascent height of a volcanic thermal in the atmosphere can be expressed approximately by: H = 1 . 8 9 ( Mc[T-

To]) 0.25

(1)

where M is the mass of solids, T is magma temperature (about 1100K; Barclay et al. 1998), T0 is the atmospheric temperature at vent level (about 300 K), c is the specific heat of the solids (typically 1 1 0 0 J k g - 1 K - 1 ) , and is the fraction of particles contributing to the thermal mass of the plume (Morton et al. 1956; Woods & Kienle 1994). This relationship has been shown to be appropriate for fine-grained ash clouds with a major part of their ascent below the tropopause (Woods & Kienle 1994). On the timescale (t) of plume ascent in the atmosphere, particles of radius r can attain thermal equilibrium with the gas phase only if: r

<

(2)

(Woods 1995) where K is the thermal diffusivity of the particles (c. 10 -6 m2 s - 1 ). The Montserrat plumes took on the order of 500s to reach their maximum heights, so only particles smaller than 1 cm or so attained thermal equilibrium with the gas during plume ascent. These typically constitute about 80% of particles discharged during explosive eruptions (e.g. Druitt 1992; Woods & Bursik 1991), so 0 is taken as 0.8. Equation 1 yields individual plume masses ranging over three orders of magnitude, equivalent to 0.01-17.5 x 10 5 m 3 DRE of magma, reflecting the wide range of plume heights. The heights used were the mean values of the ranges given in Table 1. They were corrected by first subtracting the height of the dome during the explosions (c. 1000m), since the buoyant thermals formed above the dome, not at sea level. The average DRE volume calculated in this way was 3.8 x 10 5 m 3 , which is over three times that estimated from the field measurements given above (1.1 x 10 5 m 3 ). One reason for the discrepancy may be the parameter 0, which is poorly constrained. Another reason may lie in the plume height estimates. The calculated volumes are a strong function of plume height due to the one-quarter-power dependence in Equation 1. Independent estimates of plume heights for three explosions made from GOES images (Bonadonna et al. 2002b) yield lower values than those provided by NOAA (Table 1), the differences ranging from 10 to 30%. If we reduce all the plume height estimates by 20% and reapply equation 1 while retaining 0.8, we obtain a range

Fig. 14. (a) Seismic signal from the 16:57 explosion of 5 August 1997, recorded on the seismograph at Galway's Estate (Fig. 10). The unfiltered signal shows phase 1, phase 2 and the first minute of phase 3. The abrupt drop in intensity of the phase 2 signal (arrow) may be due to pyroclastic flows nearest the seismometer coming to rest. The low-frequency (0.5-1 Hz) filtered component is attributed largely to fragmentation and conduit flow. The high-frequency (>2 Hz) component is due to fountain collapse, ballistic impact, and pyroclastic flow. (b) Enlargement of the first 1.4 minutes of the same signal. Phase 1 is interpreted as the time for the ballistic blocks and collapsing fountain to first hit the ground. Pulsing of the 0.5-1 Hz component is attributed to resonance of seismic waves in the conduit. (c) Frequency spectra for the different phases of the explosion signal, as well as for the background seismicity. Two spectra are shown for phase 2: one prior to the intensity drop (arrow in (a)) and one after it. The lowfrequency component of the explosion signal is present throughout the explosion, whereas the high-frequency component occurs only in phase 2.

299

EPISODES OF CYCLIC EXPLOSIVE ACTIVITY Table 3. Characteristics of the explosion seismic signals Date

Time (local)

4 Aug. 4 Aug. 5 Aug. 5 Aug. 6 Aug. 6 Aug. 7 Aug. 7 Aug. 7 Aug. 8 Aug. 8 Aug. 1 1 Aug. 12 Aug.

06:30 16:43 04:45 16:57 04:02 14:36 00:34 12:05 21:55 10:32 20:51 11:38 10:12

22 Sep. 22 Sep. 22 Sep. 23 Sep. 24 Sep. 24 Sep. 24 Sep. 25 Sep. 25 Sep. 25 Sep. 26 Sep. 26 Sep. 27 Sep. 27 Sep. 27 Sep. 28 Sep. 28 Sep. 28 Sep. 29 Sep. 29 Sep. 29 Sep. 29 Sep. 30 Sep. 30 Sep. 1 Oct. 1 Oct. 1 Oct. 2 Oct. 2 Oct. 2 Oct. 4 Oct.

00:57 10:45 20:42 07:23 00:34 10:54 17:16 03:54 11:09 20:05 04:25 14:56 00:01 09:46 17:15 04:28 10:34 23:03 06:26 11:23 16:48 21:57 04:43 17:44 05:00 11:34 17:40 01:05 12:53 22:50 08:33

Duration phase 1 (s)

Max. vertical velocity explosion (nms - 1 )*

Duration phase 2 (s)

10

43960

360

18 19 17 19 18

101270 65560 89160 50410 84950 34500

300

Max vertical velocity py flows (nms-')t

330 270 300 240 300

12 10.6 10 8 10 9.6 6 14.5 8.5 14 14 17 17 17 20 22.2 10.4 17.5 12.9 23.5 18 7 34 16 4 14 18 7 17 15

31950 56125 24113 16932 25965 20763 8925 53240 17256 27772 13522 62372 38362 24208 14339 139868 13179 133280 10142 40622 29554 21468 20005 189880 1950 5442 41329 47944 29 551 30904

480 360 270 300

330 240 300 300 300 240 300 270 360 270 240 300 270 240 270 270 270 210 240 300 240 450

123235 155352 107300 40525 42514 64639 19651 103013 42964 65515 35585 50483 57430 45939 32940 78465 17927 115986 30784 29695 82990 54007 21610 75990 35951 32284 38498 47944 12119 161781

Date

Time (local)

Duration phase 1 (s)

Max. vertical velocity explosion (nms- 1 )*

Duration phase 2 (s)

4 Oct. 5 Oct. 5 Oct. 5 Oct. 6 Oct. 6 Oct. 6 Oct. 7 Oct. 7 Oct. 8 Oct. 8 Oct. 9 Oct. 9 Oct. 10 Oct. 10 Oct. 1 1 Oct. 12 Oct. 12 Oct. 13 Oct. 13 Oct. 14 Oct. 14 Oct. 14 Oct. 15 Oct. 15 Oct. 15 Oct. 15 Oct. 16 Oct. 16 Oct. 16 Oct. 16 Oct. 16 Oct. 17 Oct. 17 Oct. 17 Oct. 17 Oct. 18 Oct. 18 Oct. 19 Oct. 19 Oct. 20 Oct. 20 Oct. 21 Oct. 21 Oct.

18:27 02:53 10:41 18:41 02:44 10:42 17:50 04:06 16:02 03:47 15:10 03:03 12:32 04:13 18:40 17:57 07:55 22:24 09:32 15:24 01:36 13:48 23:16 05:47 08:33 14:50 22:20 02:51 06:35 09:44 14:20 18:48 04:01 12:35 16:05 23:18 06:48 15:17 05:13 21:27 05:04 15:13 11:39 19:02

19 13 4 14 21 14 5

5250 42141 5644 31789 41720 28892 9210

270 240 210 300 240 270 300

24 22 15 13 20 16 17 15 14

96213 17342 31663 34058 21303 42172 23 915 39798 15306

300 300 300 300

49257 36828 54754 74980 38356 81507 41035 74243 25726

39 18 13

28399 20828 14414

270 210 240 300 240

15813 30044 48731 25354

10 34 31 19

240

0

44 12 6 26

16538 48463 3252 18674 7061 32847 39357 22517 35000

25185 34041 45914 15536 0 11275 15117 39790 52444

15 26 8 10 22 55 17 18 21

33870 27381 14923 7655 46554 30973 15588 66336 20120

300 240 360 450 480 270 270 480 240 480

52215 32620 45512 42377 1 1 1 028 44074 23411 78478 33765 22438

300 300

Max. vertical velocity py flows (nms- 1 )t 28641 40914 19594 56612 39381 37664 55690

* Maximum vertical component of the velocity spectrum for the explosion component of the seismogram. Values for August were measured from the seismometer at Galway's Estate. Those for September and October were measured on the Windy Hill seismometer. The two data sets are therefore not directly comparable. t Maximum vertical component of the velocity spectrum for the pyroclastic flow component of the seismogram, measured at Windy Hill.

of DRE volumes from 0.01 to 6.6 x 10 5 m 3 , with an average of 1.4 x 105 m3, which is more consistent with the field estimate. The total volume of pumice-and-ash pyroclastic flow deposits generated during the two episodes of Vulcanian explosions was 32 x 10 6 m 3 , equivalent to an average of 1.9 x 105m3 DRE per explosion. This was calculated from volume surveys of the main ghauts carried out throughout the Soufriere Hills eruption. The method involved surveying of the ground surface by helicopter using rangefinder binoculars and GPS (Sparks et al. 1998). Only for one explosion (15:17 on 18 October) was a detailed survey carried out of the pyroclastic flow deposits from a single event in enough detail to calculate a volume. The resulting map is given in Cole et al. (2002). This explosion was average in magnitude (NOAAestimated plume height 9.1 km), and the calculated pyroclastic flow

volume (2.3 x 105 m3 DRE) agrees broadly with the overall average given above. The variation of pyroclastic flow volume with explosion magnitude cannot be determined. There are indications from the data in Table 1 that flows from larger explosions (as indicated by higher plumes) entered more ghauts around the dome, and thus were perhaps more voluminous, although this is not possible to quantify owing to the incompleteness of the observations. On the other hand, there is no obvious correlation between pyroclastic flow runout and plume height. For example, flows from the relatively small explosions at 11:34 on 1 October (plume height 4.6 km) and at 17:57 on 11 October (plume height 6.9 km) had runouts down Tuitt's Ghaut of 4.5 and 5.5 km respectively, which is comparable to, or greater than, those from explosions with higher plumes. Neither is there any

300

T. H. DRUITT ET AL.

correlation between plume height and the amplitude or duration of the high-frequency (pyroclastic flow) component of the explosion seismic signals that might suggest systematic variation of pyroclastic flow volume with explosion magnitude. We conclude that the average Vulcanian explosion during August, September and October 1997 discharged a total of about 3.0 x 10 5 m 3 DRE of magma - 1.1 x 10 5 m 3 as fallout and 1.9 x 10 5 m 3 as pyroclastic flows - although there was a large variation from the smallest to the largest explosions. Since the volume of ash in the co-pyroclastic-flow plumes was c. 10% of the total fallout (Bonaonna et al. 2002b), the partitioning of magma during an average explosion is estimated to have been as follows: central plume, umbrella cloud and waning plume 1.0 x 10 6 m 3 , pyroclastic flows 1.9 x 10 6 m 3 , and co-pyroclastic-flow ash plumes 0.1 x 10 6 m 3 . About two-thirds of the material ejected during an average explosion underwent fountain collapse to form pyroclastic flows. Exit velocities during the explosions Exit velocities during the explosions (the velocity at which the material left the dome summit crater) were estimated from analysis of video footage, distributions of ballistic blocks and, more crudely, from the observed fountain-collapse heights and durations of phase 1 of the explosion seismic signals.

Observed fountain-collapse height and duration of seismic phase 1 A crude estimate of exit velocity (u) at the start of each explosion can be made from the estimated fountain collapse heights (h = 300-650 m above the summit crater rim) during the initial 10-20s of the explosions. Numerical modelling has shown that fountain-collapse height can be approximated by h u2/2g (Dobran et al. 1993; Clarke et al. 2002), which implies a vertical exit velocity at crater-rim level of 80-115m s - 1 . For such velocities, the transit time (t) of the collapsing fountain from initiation to ground impact would be approximately t 2u/g, or 16-23 s, which is consistent with the observed durations of phase 1 of the explosion seismic signals (10-20 s).

Analysis of video footage More accurate estimates of vertical ascent velocities were made for the three main plumes of the explosion at 12:05 on 7 August, using the northern crater rim (940m) as reference level. Video footage from two sites (MVO South and Fleming) was analysed in order to construct height-time curves and thus to estimate ascent velocities. The height of each of the three plumes was measured as a function of time and corrected for perspective effects using the equations of Sparks & Wilson (1982). Distortions were small because the plumes were filmed from distances of several kilometres and scaling corrections were approximately linear. The vertical and horizontal fields of view for the two cameras were determined by filming a known landscape using the same settings as during the explosion. Tracings of the three plumes are shown in Figure 15, and the height-time data are shown in Figure 16a. Plume ascent velocities were calculated using best-fit polynomials to smooth the heighttime curves, then differentiating the polynomials (Fig. 16b). For all three plumes the velocity first decreased, reached a minimum in the range 15 to 3 0 m s - 1 , then increased again. The velocity minimum is interpreted as the transition from momentum-driven to buoyancydriven behaviour and occurred at heights of 450-650 m (plumes 1 and 2) and 1200m (plume 3) above the crater rim. Exit velocities for the three plumes were estimated by backextrapolation of the velocity curves to crater-rim level. Plume 2 left the crater 7 s into the explosion with a vertical velocity of 80 1 0 m s - 1 ; but, since it was initially inclined at about 60° to the horizontal, the absolute exit velocity was about 95 1 0 m s - 1 .

Fig. 15. Tracings of plumes 1 to 3 of the explosion of 12:05 on 7 August (Fig. 4). The data were measured from video footage taken from Fleming and corrected for perspective effects using the equations of Sparks & Wilson (1982). Numbers are the time in seconds after the onset of the explosion and refer to the overlying plume front. The tracing intervals are either 2 or 3s. The star marks the vent location.

The data for plume 1 do not allow accurate extrapolation, but the velocity and emergence angle were about the same as for plume 2. Plume 3 left the vent vertically 17 s into the explosion at a velocity of at least 1 3 0 m s - 1 . Clarke et al. (2002) have analysed the same video footage of the 12:05 explosion on 7 August using a similar method, but without distinguishing the three individual plumes recognized here (i.e. by tracing the progress of the front of the entire plume). Their data are -1 consisted with an initial exit velocity of and a velocity-minimum height of about 450m above the crater rim. They also analysed footage of the 14:36 explosion on 6 August, for which the exit velocity and velocity-minimum height were estimated to be -1 and 650m respectively. A more detailed analysis of video footage of three explosions in October 1997 (17:50 on 6 October, 16:02 on 7 October and 12:32 on 9 October) has been carried out by Formenti & Druitt (in prep.). The analysis involved plotting height-time curves for individual finger jets and extrapolating velocities back to crater-rim level using a fluid dynamic model for a jet. The calculated exit velocities range from 40 to 140m s-1 and in all three explosions increased with time as fragmentation progressed to deeper levels in the conduit. Thus the slowest jets emerged at the start of the explosion and the fastest ones about 10s later. Higher exit velocities may have occurred subsequently, but any such jets were hidden by the billowing clouds of ash.

Analysis of ballistic trajectories The locations and sizes of ballistic blocks also served to constrain exit velocities. Ballistic crater fields were mapped by helicopter following the August explosions using onboard GPS, and block diameters were estimated to within 0 cm (Fig. 17). No accurate measurements were made in September and October due to safety considerations, but observations indicate that they were similar. The August explosions threw blocks out to 1.7 km from the crater centre. The distribution was approximately symmetrical around the dome. Blocks with the largest ranges had diameters of 0.7m (to the north)

EPISODES OF CYCLIC EXPLOSIVE ACTIVITY

301

Fig. 16. (a) Heights and (b) velocities as a function of time for the three plumes of the 12:05 explosion on 7 August (Fig. 15). The lines simply connect the data points.

and 1.2m (to the south). Blocks smaller than 0.4m are not shown in Figure 17, but became increasingly abundant nearer the dome, reflecting the stronger air drag experienced by small blocks (Bower & Woods 1996). Two models (Self et al 1980; Waitt et al 1995) have been used to estimate launch velocities for the ballistic blocks. In the Self et al. (1980) model the maximum range for a block several decimetres in size is achieved for an optimum launch angle of about 35°. Launch angles as low as 30° are observed on video footage, so the optimum angle was assumed in the calculations. Selected launch velocities and travel times calculated using the model are shown on Figure 17. These take into account the elevation difference between the crater rim and landing site. Blocks 0.7m large on Farrell's Plain require launch speeds of about 160ms - 1 to reach their range of 1.6km,

540m below the crater rim. The 1.2m block launched over Galway's Wall requires a velocity of 135ms - 1 . The highest velocities (250 m s - 1 ) are required by 0.4m blocks that landed 1.7km from the vent. However, it was unclear in the field whether these were fragments of larger blocks that had broken on impact, so the result is ambiguous. Launch velocities up to 160m s-1 are therefore required by the Self et al. (1980) model to explain the ballistic data, which is higher than plume exit velocities measured from video footage (up to 140ms - 1 ). Calculated flight times for the ballistics, which range from about 17 to 27 s (Fig. 17), are, however, broadly consistent with observations of the 12:05 explosion of 7 August, in which the first ballistic impacts on Farrell's Plain were observed at 21.6 s (Table 2). The same ballistic data have been modelled by Clarke et al. (2002) using the numerical scheme of Waitt et al. (1995). This

302

T. H. DRUITT ET AL.

Fig. 17. Sizes and ranges of ballistic blocks from the 13 explosions in August 1997. The numbers are the launch velocities and flight times calculated using the model of Self el al. (1980), taking into account air drag and the elevation differences between the impact sites and the crater rim. The calculations assume an optimum eruption angle of 352 to the horizontal. Locations shown by squares are south of the dome, dots between Tar River valley and White's Ghaut, upwardpointing triangles between White's Ghaut and Tuitt's Ghaut, and downward-pointing triangles between Tuitt's Ghaut and Mosquito Ghaut.

Fig. 18. Summary of a single explosive cycle in 1997: (a) immediately prior to explosion onset; (b) during the explosion; (c) during the interval between explosions. See the text for discussion.

requires lower initial velocities for the same 35° launch angle as used above (< 1 2 5 m s - 1 as opposed to < 160ms - 1 ), since the drag coefficients assumed in this model result in lower form drag than in the model of Self et al. (1980). The result is in better agreement with the exit velocities observed. Discussion Explosion cyclicity Activity of Soufriere Hills Volcano in 1997 involved a regime of cyclic explosive behaviour, with an average interval of about 10 hours. Repetitive Vulcanian explosions have been reported from other volcanoes. For example, six explosions occurred regularly at Mount Ngauruhoe, New Zealand, on 19 February 1975 at intervals of between 0.5 and 1 hour (Nairn & Self 1978). Over the period 12 to 14

June 1991, a sequence of four explosions occurred at Mount Pinatubo, Philippines, at intervals ranging from 10 to 28 hours, culminating in the climatic eruption (Hoblitt et al 1996). At Volcan Galeras, Colombia, six explosions took place at intervals of between 8 and 181 days in 1992-1993 (Stix et al. 1997). Twenty-three explosions occurred at Tokachi-dake, Hokkaido, between 16 December 1988 and 5 March 1989, an average of three to four days apart (Katsui et al. 1990). Repetitive explosive behaviour at Montserrat is attributed to the cyclic build-up and release of magmatic pressure beneath a rheologically stiffened plug of degassed magma at shallow levels in the conduit below the dome (Voight et al. 1998, 1999). A type of stickslip effect has been invoked to explain cyclic conduit pressurization at Montserrat, resulting in cyclic deformation of the dome and surrounding terrain (Voight et al. 1999; Denlinger & Hoblitt 1999; Wylie et al. 1999). Hybrid earthquakes are attributed to hydrofracturing and associated gas flow in rock or crystal-rich magma at

EPISODES OF CYCLIC EXPLOSIVE ACTIVITY

303

Table 4. Estimates of physical parameters for the 1997 explosions Parameter

Value*

Plume height (km) Ascent duration of plume (s) Total magma volume discharged (m3 DRE) Duration of peak discharge (s) Exit velocity (ms - 1 ) Fountain-collapse height (m) Mass fraction entering collapse fountain Initial velocity of pyroclastic surges ( m s - 1 ) Runout of pyroclastic flows (km) Runout duration of pyroclastic flows (s) Typical velocity of pyroclastic flows ( m s - 1 ) Fragmentation pressure (MPa) Velocity of fragmentation wave ( m s - 1 ) Conduit withdrawal depth (km) Duration of waning phase of explosion (min) Explosion interval (min) Magma ascent velocity between explosions ( m s - 1 )

c. 10(3 to 15) c.500 3x 10 5 45-70 40-140 300-650 c.2/3 30-60 3-6 c.500 10 5-15 10-50 0.5 to >2 c. 60 (20-190) c. 600 (100-3 700) >0.02

* Average values, with ranges in brackets.

the peak of each pressurization cycle (Neuberg et al. 1998; Voight et al. 1999). Synchronized tilt cycles and hybrid swarms during the August explosive episode provided accurate indicators of the pressurization state of the system, enabling MVO volcanologists to anticipate many of the explosions successfully and to reduce the threat to the population. They also facilitated study of the explosions and their products. Subsequently, in September and October, when there was no tiltmeter and hybrid swarms were weak or absent, the strong periodicity of the explosions themselves played this role.

Explosion mechanisms Figure 18 summarizes schematically the events throughout one cycle of 1997 explosive activity at Soufriere Hills Volcano. Our best estimates of the physical parameters are given in Table 4. Explosive eruption commenced when the conduit overpressure exceeded the strength of the cap of degassed crystal-rich magma. During the initial few seconds, crater rubble and fragments of disrupted, degassed plug were thrown out, forming ballistic showers. A fragmentation wave then descended the conduit into the region of pressurized magma, resulting in a rapid escalation of exit velocities from about 40 to 140ms - 1 . Eruption was highly unsteady, peak discharge lasting just a few tens of seconds with the highest intensity over the first 10-20 s. Each explosion discharged on average about 3 x 105 m3 DRE of magma. Since the conduit diameter during the 1995-1999 phase of the eruption is estimated at 25 to 30m from spine dimensions and the widths of early vents (Watts et al. 2002), and is not believed to have varied greatly with time (Voight et al. 1999), the conduit drawdown during an average explosion was about 500 m below the crater floor (which itself was about 800m above sea level). This is a DRE drawdown; the actual average drawdown of vesicular magma would have been greater, but not more than 1 km. The largest explosions must have emptied the conduit to depths of 2km or more. Since peak discharge lasted a few tens of seconds, the velocity of the fragmentation wave down the conduit is constrained to have been of the order of l0-50ms -1 , although the initial value could have been greater. High-intensity eruption probably ceased once the wave reached a level in which the magma pressure was not sufficiently large to drive fragmentation. Ash then continued to be discharged for 1-3 hours, but at a greatly reduced rate. Numerical models of the plume dynamics (Clarke et al. 2002) and conduit flow (Melnik & Sparks 2002b) reproduce several key features of the explosions, including their highly transient nature, peak exit velocities in the range 80-140 m s - 1 , and discharge durations and conduit drawdowns comparable to those observed. The models are based on the rapid decompression and expansion of

Fig. 19. Explosion magnitude (as measured by plume height) as a function of the time interval between explosions: (a) interval preceding the explosion, and (b) interval following the explosion. See the text for discussion.

gas-rich, pressurized magma beneath a degassed plug, thus supporting this interpretation of the eruption dynamics. The eruption columns were partially unstable in all but one explosion. On average about two-thirds of the erupted material collapsed back to form pyroclastic surges and flows, while the other third, including probably a large proportion of smaller particles, was carried up into the plume. However, these proportions may have varied considerably between individual explosions. Fountain collapse occurred in the first 10-20s of each explosion from a few hundred metres above the crater rim. Vertical velocity profiles in the plumes reveal velocity minima corresponding to the transition from momentum-driven to buoyancy-driven behaviour. This is analogous to the superbuoyant regime of sustained eruption columns, which is intermediate between fully stable (convective) and fully unstable (collapsed) regimes (Bursik & Woods 1991) and is seen in the explosion simulation of Clarke et al. (2002). Once each explosion was over, magma rose in the conduit by viscous flow at a couple of centimetres per second or more (1-2 km in 10 hours). This exceeds the critical ascent speed of about l ^ c m s - 1 for amphibole breakdown and explains the presence of hornblende phenocrysts lacking breakdown rims in the explosion pumices (Devine et al. 1998b). In at least some cases the conduit was totally refilled prior to the next explosion and a small dome appeared in the crater. Repressurization of the conduit then occurred until conditions were right for the next explosion, although the exact mechanism is not well understood. Figure 19 shows that a weak correlation exists between plume height and the intervals between explosions for both the August and September-October episodes. Positive correlations exist between (1) plume height and the interval prior to a given explosion (Fig. 19a) and (2) plume height and the interval following a given explosion (Fig. 19b). Correlation 1 would suggest that large explosions result from long preceding intervals, perhaps allowing the build-up of larger magma pressures in the conduit. Correlation 2 is consistent with a scenario in which large explosions drain the conduit to deeper levels, so that longer

304

T. H. DRUITT ET AL.

intervals are then required to refill the conduit prior to the next explosion. The data do not distinguish between these two mechanisms, although correlation 2 appears visually to be slightly better than correlation 1.

pascals) relative to the overlying plug and surrounding conduit walls. The presence of pressurized, gas-charged magma at very high levels in the conduit immediately prior to each explosion is consistent with the observation that exit velocities in excess of 100 m s-1 or more were achieved only a few seconds after each explosion began (Clarke et al 2002; Melnik & Sparks 2002b).

Conduit pressurization and explosive fragmentation Pressurization of the magmatic conduit at Montserrat is attributed to non-linear vertical pressure gradients caused by large viscosity variations that accompany exsolution of water from magma (Sparks 1997; Massol & Jaupart 1999). Magma viscosity is a strong function of water content, particularly at low pressures (Hess & Dingwell 1996). The estimated viscosity of non-degassed magma at Montserrat is about 106 Pas and that of completely degassed magma about 10 14 Pas (Voight et al. 1999). This is believed to have generated large magma overpressures (magma pressure minus lithostatic pressure) at shallow levels in the conduit. Another effect is the development of high gas pore pressures in the ascending magma due to (1) viscous resistance to vesicle expansion, which increases as the liquid exsolves gas (Massol & Jaupart 1999), and (2) growth of microlites in the degassed, undercooled liquid, which forces further gas into vesicles (Stix et al 1997; Sparks 1997). Tilt amplitudes and far-field deformation measurements at Montserrat are consistent with maximum magma overpressures of about ten to a few tens of megapascals a few hundred metres below the base of the dome (Shepherd et al 1998; Voight et al. 1999). The conduit flow modelling of Melnik & Sparks (2002a) predicts steep pressure gradients and overpressures up to l0MPa in the upper conduit. The angular, platy shapes of many of the 1997 fallout pumices with 55-75 vol% vesicles are consistent with brittle fragmentation of a pressurized magmatic foam present in the upper conduit prior to each explosion. Brittle fragmentation of magma requires steep pressure gradients and fast decompression rates in order to drive the magma through the glass transition limit (Dingwell 1996). This has been observed experimentally by Alidibirov & Dingwell (1996), who showed that pressure differentials across the fragmentation interface of a few megapascals can be sufficient to drive brittle failure, generating platy fragments with shapes very much like those at Montserrat. Recent experimental work has shown that the tensile strength of crystal-rich magma like that at Montserrat may be of the order of 20MPa or more (Martel et al 2001). As the fragmentation wave descended the conduit during each explosion, the pressurized foam broke up into tabular fragments that were then accelerated to the surface. Magma fragments erupting from the vent apparently had sufficiently high viscosities due to gas exsolution to suppress post-fragmentation expansion, enabling pumices to retain vesicularities and angular shapes close to those acquired at fragmentation (the viscosity quench effect; Thomas et al 1994). Pumice incorporated into pyroclastic flows were subsequently rounded by abrasion during transport. The presence of magmatic foam with at least 55% bubbles in the upper conduit can be used to provide an independent estimate of magma pressure prior to each explosion. The bulk water content of the magma prior to ascent was about , based on the water content of glass inclusions ; Barclay et al 1998; Devine et al 1998a) and the estimated crystal content in the magma reservoir at 5-6km depth (60-65vol%; Murphy et al 2000). In the Appendix we show that the total confining pressures required for magmatic foam with 1.6 3 wt% water to have vesicularities of 55-75% are 5-15MPa. One key feature of the explosions is that the products are dominantly pumiceous, with dense clasts making up no more than 5% of those erupted (Clarke et al 2002). Given that an average explosion emptied the conduit to about 500 m (DRE) depth, the plug of degassed magma present in the conduit prior to each explosion can have been no more than about 25m thick, corresponding to an overburden of less than 1 MPa. Given that pressures of 5-15 MPa are required in the magmatic foam just below this cap, this suggests that the foam must have been very significantly overpressured (by at least a few mega-

Initiation of episodes of explosive activity on Montserrat Each of the two episodes of explosive activity in 1997 was triggered by a major dome collapse (3 August and 21 September), as was the explosive eruption on 17 September 1996 (Robertson et al 1998). In each case, sudden removal of part of the dome led to the conditions for explosive fragmentation. This was not immediate, the delays being 2.5 hours (17 September 1996), 10 hours (3 August 1997) and 20 hours (22 September 1997), showing perhaps that time was necessary for the build-up of sufficient conduit pressure for this to occur. Many dome collapses occurred during the 1995-1999 period, but only three are known to have triggered major vertical explosions. We exclude here the relatively weak explosions of late 1998 and 1999, which may have been triggered by slow pressure build-up in the slowly crystallizing lava dome and conduit during the period of virtually no magma extrusion (Norton et al 2002). One important factor was probably that the 17 September 1996 and 21 September 1997 collapses resulted in two of the largest height reductions of the active dome-growth area during the 1995-1999 period (130m and 230m respectively), causing large decompressions of the conduit (at least 3.5 and 6 MPa). The height reduction from the 3 August 1997 collapse was not observed clearly, but it is inferred to have been at least 110m (3 MPa) from the form of the crater observed four days later. Sudden decompression of at least 3 MPa therefore appears necessary to trigger explosive fragmentation at Montserrat. Conduit flow beneath lava domes involves complex feedback effects and sudden decompressions can force systems from effusive to explosive behaviour (Jaupart & Allegre 1991; Woods & Koyaguchi 1994). An additional effect in 1997 may have been the high magma discharge rate. The time-averaged magma discharge rate increased throughout the 1995-1999 period, and by August 1997 had reached 7-8 m3 s-1 (Sparks et al 1998; Sparks & Young 2002). High discharge rates favour explosive fragmentation by limiting the time available for magma degassing during ascent (Jaupart & Allegre 1991; Melnik & Sparks 2002a). High magma flux during August, September and October of 1997 may have helped to prime the conduit for explosive activity once a suitably large dome collapse occurred. Strangely, the largest dome collapse of the 1995-1999 period (26 December 1997; Sparks et al 2002) decompressed the conduit by at least 8 MPa. giving rise to a violent lateral blast and pyroclastic density current, but triggered no vertical explosion from the conduit and produced little pumice. This highlights the complexity of the system and the existence of important effects not considered here. Conclusions Two episodes of cyclic explosive activity occurred at Soufriere Hills Volcano in 1997. Thirteen explosions took place in August and another 75 in September and October. The activity had a major impact on southern Montserrat and triggered northward enlargement of the evacuation zone in mid-August. Like the explosive eruption of 17 September 1996. both episodes in 1997 were preceded by major dome collapses that decompressed the conduit by 3 MPa or more. Delays of 3 to 20 hours then followed before explosive activity commenced. Large gravitational collapses are a prerequisite for vertical explosive eruption at Montserrat. The explosions were highly unsteady, with the most intense phase lasting only a few tens of seconds. Peak discharge was accompanied by ballistic showers, exit velocities up to 140 m s - l , and (in all but one event) fountain collapse from a few hundred metres above the crater rim over the first 10-20s of each explosion. Pyroclastic flows travelled up to 6km down all major drainages around the

EPISODES OF CYCLIC EXPLOSIVE ACTIVITY dome and entered the sea on the south and east coasts. Buoyant eruption plumes with large, bulbous heads rose to 3-15 km in the atmosphere, then spread out as umbrella clouds. After 10 minutes or so, each explosion settled into a waning phase that typically lasted an hour and generated a low, bent-over ash plume. Fallout and pyroclastic flows/surges from the explosions accounted on average for one-third and two-thirds of the magma discharged, respectively. The explosions emptied the conduit to a depth of 0.5-2 km, perhaps more in some cases. Filtering of explosion seismic signals permitted distinction of a low-frequency (c. 1 Hz) component due to the explosion itself and a high-frequency (>2 Hz) component due to ballistic impact, fountain collapse and pyroclastic flow. Relative timing of the onsets of the two components provided information on the flight durations of ballistic blocks and on the transit time for fountain collapse, from inception to first ground impact. Explosions in August were accompanied by cyclic patterns of seismicity and edifice deformation. Repeated slow inflation, followed by rapid deflation, of the volcano recorded cycles of build-up, then release, of pressure beneath the dome. The explosions were driven by rapid decompression and brittle fragmentation of overpressured magmatic foam in the upper conduit and occurred at intervals of 2.5 to 63 hours, with a mean of 10 hours. Synchronized tilt cycles and hybrid earthquake swarms during the August explosions provided accurate indicators of the pressurization state of the system, enabling volcanologists to anticipate many of the explosions and reduce the threat to the population. In September and October, when there was no tiltmeter and hybrid swarms were weak or absent, the strong periodicity of the explosions themselves played this role. We thank the staff of the M VO for their very important contributions in the study of the 1997 explosions. D. Lea, M. Sagot and D. Williams kindly provided us with video footage of the explosions and allowed us to study it. D. Williams helped us in the analysis of video footage. B. Poyer kindly provided the photographs in Figure 7. Careful reviews by T. Koyaguchi, L. Wilson and P. Kokelaar are gratefully acknowledged.

Appendix

Estimation of fragmentation pressures during the explosions We estimate the pressure necessary for the magma with 60-65 vol% crystals (Murphy et al. 2000) and a bulk water content of % to have 55-75 vol% vesicularity. Consider a unit volume of crystal-bearing magmatic foam immediately prior to fragmentation. The volume fraction of bubbles is X and the volume fraction of crystals in the liquid phase is F. The masses of gas, liquid, and crystals are given by:

Mg = pgX

M 1 = A(1-A-)(1-F) Mc = Pc(l - X)F

(Al)

where M is mass, p is density and the subscripts g, 1 and c stand for gas, liquid and crystals respectively. Given the solubility law for water in magmatic liquid, n = P 1 / 2 , where n is a mass fraction and a is approximately 4.1 x 10-6 Pa 1/2 for rhyolite (the composition of interstitial glass in the pumices), we can write the mass balance equation for water in the foam Mg + M 1 aP 1/2 = N(Mg + M1 + Mc)

(A2)

where N is the bulk mass fraction of water in the magma. The density of the gas is given by:

(A3) where T is temperature (about 860°C or 1133 K; Barclay et al. 1998) and r is the gas constant (462 J k g - 1 K-1 for water). Given a bulk water content N, we can use Equation A2 to estimate the vesicularity X of the foam as a function of pressure P. For a bulk water content of , bubble contents of 55-75 vol% require pressures in the range 5-15MPa.

305

References ALIDIBIROV, M. A. & DINGWELL, D. B. 1996. Magma fragmentation by rapid decompression. Nature, 380, 146-148. ASPINALL, W. P., MILLER, A. D., LYNCH, L. L., LATCHMAN, J. L., STEWART, R. C., WHITE, R. A. & POWER, J. A. 1998. Soufriere Hills eruption, Montserrat 1995-1997: volcanic earthquake locations and fault plane solutions. Geophysical Research Letters, 25, 3397-3400. BARCLAY, J., RUTHERFORD, M. J., CARROLL, M. R., MURPHY, M. D., DEVINE, J. D., GARDNER, J. & SPARKS, R. S. J. 1998. Experimental phase equilibria constraints on pre-emptive storage conditions of the Soufriere Hills magma. Geophysical Research Letters, 25, 2437-3440. BONADONNA, C., MACEDONIO, G. & SPARKS, R. S. J. 2002a. Numerical modelling of tephra fallout associated with dome collapses and Vulcanian explosions: application to hazard assessment on Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 517-537. BONADONNA, C., MAYBERRY, G. C., CALDER, E. S., SPARKS, R. S. J., CHOUX, C., JACKSON, P., LEJEUNE, A. M., LOUGHLIN, S. C., NORTON, G. E., ROSE, W. L, RYAN, G. & YOUNG, S. R. 2002b. Tephra fallout in the eruption of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 483-516. BOWER, S. M. & WOODS, A. W. 1996. On the dispersal of clasts from volcanic craters during small explosive eruptions. Journal of Volcanology and Geothermal Research, 73 19-32. BURSIK, M. I. & WOODS, A. W. 1991. Buoyant, superbuoyant and collapsing eruption columns. Journal of Volcanology and Geothermal Research, 45, 347-350. CLARKE, A. B., NERI, A., VOIGHT, B., MACEDONIO, G. & DRUITT, T. H. 2002. Computational modelling of the transient dynamics of the August 1997 Vulcanian explosions at Soufriere Hills Volcano, Montserrat: influence of initial conduit conditions on near-vent pyroclastic dispersal. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 319-348. COLE, P. D., CALDER, E. S., SPARKS, R. S. J. ET AL. 2002. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoir. DENLINGER, R. P. & HOBLITT, R. P. 1999. Cyclic behavior of silicic volcanoes. Geology, 27, 459-462. DEVINE, J. D., MURPHY, M. D., RUTHERFORD, M. J. ET AL. 1998a. Petrologic evidence for pre-eruptive pressure-temperature conditions, and recent heating, of andesitic magma erupting at the Soufriere Hills Volcano, MONTSERRAT, W. I. Geophysical Research Letters, 25, 3669-3672. DEVINE, J. D., RUTHERFORD, M. J. & GARDNER, J. C. 1998b. Petrologic determination of ascent rates for the 1995-1997 Soufriere Hills Volcano andesite magma. Geophysical Research Letters, 25, 3673-3676. DINGWELL, D. B. 1996. Volcanic dilemma: Flow or blow? Science, 273, 1054-1055. DOBRAN, F., NERI, A. & MACEDONIO, G. 1993. Numerical simulations of collapsing eruption columns. Journal of Geophysical Research, 98, 4231-4259. DRUITT, T. H. 1992. Emplacement of the 18 May 1980 lateral blast deposit ENE of Mount St. Helens, Washington. Bulletin of Volcanology, 54, 554-572. FAGENTS, S. A. & WILSON, L. 1993. Explosive volcanic eruptions - VII. The ranges of pyroclasts ejected in transient volcanic explosions. Geophysical Journal International, 113, 359-370. HESS, K. U. & DINGWELL, D. B. 1996. Viscosities of hydrous leucogranite melts: a non-arrhenian model. American Mineralogist, 81, 1297-1300. HOBLITT, R. P., WOLFE, E. W., SCOTT, W. E., COUCHMAN, M. R., PALLISTER, J. S. & JAVIER, D. 1996. The pre-climactic eruptions of Mount Pinatubo, June 1991. In: NEWHALL, C. G. & PUNONGBAYAN, R. S. (eds) Fire and Mud: Eruptions and Lahars of Mount Pinatubo, Philippines. University of Washington Press, 457-511. JAUPART, C. & ALLEGRE, C. J. 1991. Gas content, eruption rate and instabilities of eruption regime in silicic volcanoes. Earth and Planetary Science Letters, 102, 413-429.

306

T. H. DRUITT ET AL.

KATSUI, Y., KAWACHI, S.. KONDO, Y. ET AL. 1990. The 1988-89 explosive eruption of Tokachi-dake, Central Hokkaido, its sequence and mode. Bulletin of the Volcanological Society of Japan, Series 2, 35, 111-129. KIENLE, J. & SHAW, G. E. 1979. Plume dynamics, thermal energy and long-distance transport of vulcanian eruption clouds from Augustine volcano, Alaska. Journal of Volcanology and Geothermal Research. 6. 139-164. KOKELAAR, B. P. 2002. Setting, chronology and consequences of the eruption of Soufriere Hills Volcano, Montserrat (1995-1999). In: DRUITT. T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs, 21, 1-43. LOUGHLIN, S. C, CALDER, E. S., CLARKE, A. B. ET AL. 2002. Pyroclastic flows and surges generated by the 25 June 1997 dome collapse, Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs. 21, 191-209. MARTEL, C., DINGWELL, D. B., SPIELER, O., PICHEVANT, M. & WILKE. M. 2001. Experimental fragmentation of crystal- and vesicle-bearing silicic melts. Bulletin of Volcanology, 63, 398-405. MARTIN, D. P. & ROSE, W. I., JR. 1981. Behavioural patterns of Fuego volcano, Guatemala. Journal of Volcanology and Geothermal Research, 10,67-81. MASSOL, H. & JAUPART, C. 1999. The generation of gas overpressure in volcanic eruptions. Earth and Planetary Science Letters, 166. 57-70. MATTHEWS, S. J., GARDEWEG, M. C. & SPARKS, R. S. J. 1997. The 19841996 cyclic activity of Lascar Volcano, Northern Chile: cycles of dome growth, dome subsidence, degassing and explosive eruptions. Bulletin of Volcanology, 59, 72-82. MELNIK, O. & SPARKS, R. S. J. 2002a. Dynamics of magma ascent and lava extrusion at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 153-171. MELNIK, O. & SPARKS, R. S. J. 2002b. Modelling of conduit flow dynamics during explosive activity at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs, 21, 173-190. MELSON, W. & SAENZ, R. 1973. Volume, energy and cyclicity of eruptions at Arenal volcano, Costa Rica. Bulletin of Volcanology, 37, 416-437. MILLER, A. D., STEWART, R. C., WHITE, R. A. ET AL. 1998. Seismicity associated with dome growth and collapse at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3401-3404. MORRISSEY, M. M. & MASTIN, L. G. 2000. Vulcanian eruptions. In: SIGURDSSON, H. (ed.) Encyclopedia of Volcanoes. Academic Press. San Diego, 463-476. MORTON, B., TAYLOR, G. I. & TURNER, J. S. 1956. Turbulent gravitational convection from maintained and instantaneous sources. Proceedings of the Royal Society, A234, 1-23. MURPHY, M. D., SPARKS, R. S. J., BARCLAY, J., CARROLL, M. R. & BREWER, T. S. 2000. Remobilization of andesite magma by intrusion of mafic magma at the Soufriere Hills Volcano, Montserrat, West Indies. Journal of Petrology, 41, 21-42. NAIRN, I. & SELF, S. 1978. Explosive avalanches and pyroclastic flows from Ngauruhoe 1975. Journal of Volcanology and Geothermal Research, 3, 39-60. NEUBERG, J. & O'GORMAN, C. 2002. A model of the seismic wavefield in gas-charged magma: application to Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 603-609. NEUBERG, J., BAPTIE, B., LUCKETT, R. & STEWART, R. 1998. Results from the broadband seismic network on Montserrat. Geophysical Research Letters, 25, 3661-3664. NORTON, G. E., WATTS, R. B., VOIGHT, B. ET AL. 2002. Pyroclastic flow and explosive activity at Soufriere Hills Volcano, Montserrat during a period of virtually no magma extrusion (March 1998 to November 1999). In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London, Memoirs, 21, 467-481. ROBERTSON, R., COLE, P., SPARKS, R. S. J. ET AL. 1998. The explosive eruption of Soufriere Hills Volcano, Montserrat, West Indies, 17 September, 1996. Geophysical Research Letters, 25, 3429-3432.

SELF. S.. WILSON. L. & N A I R N . I. A. 1979. Vulcanian eruption mechanisms. Nature. 277. 440-443. SELF, S., KEINLE. J. & HUOT. J. P. 1980. Ukinrek Maars. Alaska: II. Deposits and formation of the 1977 craters. Journal of Volcanology and Geothermal Research. 7. 39-65. SHEPHERD. J. B.. HERD. R. A.. JACKSON. P. & WATTS. R. 1998. Ground deformation measurements at the Soufriere Hills Volcano. Montserrat: II: Rapid static GPS measurements June 1996-June 1997. Geophysical Research Letters. 25. 3413-3416. SPARKS. R. S. J. 1997. Causes and consequences of pressurization in lava dome eruptions. Earth and Planetary Science Letters, 150. 177-189. SPARKS. R. S. J. & WILSON. L. 1982. Explosive volcanic eruptions - V. Observations of plume dynamics during the 1979 Soufriere eruption. St. Vincent. Geophysical Journal of the Royal Astronomical Society. 69. 551-570. SPARKS. R. S. J. & YOUNG. S. R. 2002. The eruption of Soufriere Hills Volcano. Montserrat (1995-1999): overview of scientific results. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21, 45-69. SPARKS, R. S. J., YOUNG. S. R.. BARCLAY. J. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills Volcano. Montserrat, West Indies: November 1995 to December 1997. Geophysical Research Letters. 25. 3421-3424. SPARKS. R. S. J.. BARCLAY. J.. CALDER. E. S. ET AL. 2002. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano. Montserrat. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21.409 434. STIX. J.. TORRES. R. C.. NARVAEZ. M. L.. CORTES. G. P.. RAIGOSA. J. A.. GOMEZ. D. M. & CASTONGUAY. R. 1997. A model of Vulcanian eruptions at Galeras Volcano. Columbia. Journal of Volcanology and Geothermal Research. 77. 285-304. THOMAS. N.. JAUPART. C. & VERGNIOLLE. S. 1994. On the vesicularity of pumice. Journal of Geophysical Research. 99. 15633-15644. U H I R A , K., YAMASATO. H & TAKEO. M. 1994. Source mechanism of seismic waves excited by pyroclastic flows observed at Unzen volcano. Japan. Journal of Geophysical Research. 99. 17757-17773. VOIGHT. B.. HOBLITT. R. P.. CLARKE. A. B.. LOCKHART. A. B.. MILLER. A. D., LYNCH. L. & MCMAHON. J. 1998. Remarkable cyclic ground deformation monitored in real-time on Montserrat. and its use in eruption forecasting. Geophysical Research Letters. 25. 3405-3408. VOIGHT. B.. SPARKS. R. S. J.. MILLER. R. C. ET AL. 1999. Magma flow instability and cyclic activity at Soufriere Hills Volcano. Montserrat. British West Indies. Science. 283. 1138-1142. WAITT, R. B.. MASTIN. L. G. & MILLER. T. P. 1995. Ballistic showers during Crater Peak eruptions of Mount Spurr Volcano, summer 1992. The 1992 Eruptions of Crater Peak Vent, Mount Spurr Volcano, Alaska. US Geological Survey. Bulletin 2139. WATTS. R. B.. HERD. R. A.. SPARKS. R. S. J. & YOUNG. S. R. 2002. Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21. 115-152. WOODS. A. W. 1995. A model of vulcanian explosions, Nuclear Engineering and Design. 155. 345-357. WOODS. A. W. & BURSIK. M. I. 1991. Particle fallout, thermal equilibrium and volcanic plumes. Bulletin of Volcanology. 53. 559-570. WOODS, A. W. & KIENLE. J. 1994. The dynamics and thermodynamics of volcanic clouds: theory and observations from the April 15 and April 21 1990 eruptions of Redoubt Volcano. Alaska. Journal of Volcanology and Geothermal Research. 62. 273-299. WOODS, A. W. & KOYAGUCHI. T. 1994. Transitions between explosive and effusive eruption of silicic magmas. Nature. 370, 641-645. WYLIE, J. J.. VOIGHT. B. & WHITEHEAD. J. A. 1999. Instability of magma flow from volatile-dependent viscosity. Science. 285. 1883-1885. YOUNG. S.. SPARKS. R. S. J.. ROBERTSON. R.. LYNCH. L. & ASPINALL. W. P. 1997. Eruption of Soufriere Hills Volcano in Montserrat continues. EOS, Transactions, American Geophysical Union. 78. 404. 408-409. YOUNG. S. R., SPARKS. R. S. J.. ASPINALL. W. P.. LYNCH. L. L.. MILLER. A. D.. ROBERTSON. R. E. A. & SHEPHERD. J. B. 1998. Overview of the eruption of Soufriere Hills volcano. Montserrat. 18 July 1995 to December 1997. Geophysical Research Letters. 25. 3389-3392.

Modelling of conduit flow dynamics during explosive activity at Soufriere Hills Volcano, Montserrat O. MELNIK 1,2 & R. S. J. SPARKS2 1 Department of Earth Sciences, University of Bristol, Bristol BS8 1RJ, UK (e-mail: [email protected]) 2 Institute of Mechanics, Moscow State University, 1 Michurinskii prosp., Moscow 117192, Russia

Abstract: Magmatic explosive activity at Soufriere Hills Volcano involved a sub-Plinian eruption, on 17 September 1996 and two series of repetitive short-lived (c. 1 min) Vulcanian explosions in 1997. Explosive activity followed major collapses of the dome. We have modelled unsteady conduit flow in explosive eruptions after unloading. Two cases are investigated: (i) equilibrium between gas dissolved in the melt and bubbles for sustained sub-Plinian eruption; and (ii) no mass transfer between pre-existing gas bubbles and melt for Vulcanian explosions. The models for Vulcanian explosions agree with observations of erupted volumes, eruption durations (tens of seconds), typical drawdown depths (a few hundred metres to c. 2 km), exit velocities and discharge rates. Explosive mixtures are predicted to have high densities consistent with the occurrence of fountain collapse. The models for sub-Plinian eruption show good agreement with observed erupted volumes and drawdown depths (c. 4km). Three fragmentation criteria were studied, namely fragmentation at fixed porosity, at a critical gas overpressure, and at a critical elongation strain rate. Results are similar for the three cases, but the critical overpressure and critical strain-rate criteria both predict strong pulsations, whereas the fixed-porosity criterion predicts continuous fragmentation. Pulsations are caused by feedback, with the threshold conditions for magma fragmentation being repeatedly crossed. Pulsations are indicated from seismic and video observations.

Explosive volcanism has been a prominent feature of the eruption of Soufriere Hills Volcano, Montserrat. Such activity has included early phreatic explosions (Young et al. 1998; Bonadonna et al. 2002), sub-Plinian magmatic explosive activity on 17 September 1996 (Robertson et al. 1998), two series of repetitive short-lived Vulcanian explosions from 3 to 12 August 1997 and from 22 September to 21 October 1997 (Druitt et al. 2002) and small sporadic Vulcanian explosions in the period following dome growth after March 1998 (Norton et al. 2002). This paper is concerned with magmatic explosive eruptions, all of which were preceded by major collapses of the dome, unloading pressurized magma in the conduit. This explosive activity has been documented elsewhere (Robertson et al. 1998; Druitt et al. 2002). Clarke et al. (2002) develop a numerical model for decompression and discharge of the explosive mixture into the atmosphere, using a simplified conduit model as an initial condition. This paper focuses on the unsteady magma flows in the conduit during these explosive eruptions and on the influence of magma fragmentation processes on dynamics. Explosive activity, caused either by a sudden decompression or when the gas overpressure in rising magma reaches a threshold value, are modelled as unsteady flows, taking account of vertical viscosity variations and fragmentation conditions. In the case of Soufriere Hills Volcano, all the episodes of explosive activity have occurred shortly after a major dome collapse in which the conduit has been rapidly decompressed. Repetitive individual Vulcanian explosions could be triggered when a critical gas overpressure in growing bubbles is exceeded in rapidly rising magma between the explosions. The Vulcanian explosions of 1997 were short-lived (typically a few tens of seconds). In contrast, the sub-Plinian explosive activity of 17 September 1996 reached a peak after 10 minutes, declining thereafter over a period of 30 minutes. Episodes of explosive activity did not develop into sustained steady eruptions and did not tap directly into the chamber. The models here therefore focus on transient flows, in contrast to published models which have been largely concerned with sustained explosive eruptions where changes with time are sufficiently slow that steady conditions can be assumed (Wilson et al. 1980; Dobran 1992; Barmin & Melnik 1993; Woods & Koyaguchi 1994). We compare different assumptions on fragmentation conditions to establish whether the results are sensitive to the exact mechanism of fragmentation. Three assumptions are compared: fragmentation at a fixed volume fraction of bubbles (VF) (Sparks 1978; Wilson et al. 1980), fragmentation at a fixed overpressure in the growing bubble (OP) which exceeds the tensile strength of the magma (Barmin & Melnik 1993; Melnik 2000), and fragmentation at a threshold where the elongation strain rate (SR) exceeds the magma strength (Papale 1999).

Magmatic explosive eruptions at Soufriere Hills Volcano This section describes the magmatic explosive eruptions at Soufriere Hills Volcano, largely based on the studies of Robertson etal. (1998) and Druitt et al. (2002). The sub-Plinian activity of 17 September 1996 followed a 9-hour (11:30 to 20:30 local time (LT)) period of continuous dome collapse. The course of the activity is depicted in the seismic record (Fig. 1). Explosive activity initiated at 23:42 LT after a relatively quiet period from 20:30 LT. The seismic energy reached a peak after 10 minutes and declined with prominent fluctuations over a 30-minute period thereafter. Various observations (Robertson et al. 1998) indicate that the eruption column reached about 14-15 km high, with a peak discharge rate of 3000m3 s-1 (9 x 106 kgs - 1 ). Ballistic clasts up to 1.5m diameter reached 2.1km from the dome. Robertson et al. (1998) estimated a launch velocity of 180m s-1 and explosion pressure of at least 24 MPa to account for the ballistic ejection. Various observations and inferences from ejecta volumes, conduit dimensions, post-eruption seismicity and re-establishment of dome growth two weeks later are consistent with draining of the conduit down to a depth of about 4 km. The ejecta include several components with contrasted densities and textures. The main pumice is moderately well vesiculated (7001200 kg m - 3 ) and is inferred to derive from the conduit by continuous discharge after the explosive disruption of more consolidated degassed rocks in the uppermost parts of the conduit. The ballistic blocks include dense, glassy non-vesicular blocks, a suite of poorly vesicular dense pumice (1200-2000 kg m - 3 ), and various lithified volcanic breccias. These rocks are interpreted as being derived form the uppermost few hundred metres of the conduit and represent vesiculated magma in the core of the upper conduit, chilled and fully degassed magma at the conduit walls (the glassy rocks) and conduit wallrock breccias. These blocks are inferred to have been discharged ballistically in the first several minutes of eruption when explosive intensity was at a maximum. It is estimated that about 11 x 106 m3 of dome rock collapsed, leaving a scar 200m deep over the conduit and decompressing the magma conduit by at least 4 MPa (Robertson et al. 1998; Calder et al. 2002). The two series of repetitive Vulcanian explosions followed major dome collapses on 3 August 1997 and 21 September 1997. They differed from the 17 September 1996 sub-Plinian explosive activity in a number of ways. They took place as a series of explosions with 13 explosions between 3 and 12 August and 75 explosions between 22 September and 21 October. Each series involved quasi-periodic behaviour, with rise of magma occurring between each explosion. The average interval between explosions

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 307-317. 0435-4052/02/$15 © The Geological Society of London 2002.

307

308

O. M E L N I K & R. S. J. SPARKS

Fig. 1. One-minute realtime seismic amplitude measurement (RSAM) seismic record over 8 hours during sub-Plinian explosive activity of 17 September 1996. Three peaks of seismic activity are marked (1-3).

from 22 September to 21 October was 9.5 hours, with a range of intervals from 4 to 33 hours. Repeated Vulcanian explosions were short-lived, with the most intense activity occurring over a few tens of seconds (Druitt et al. 2002), followed by periods of strong ashventing lasting typically 30 minutes to an hour. The explosions involved fountain collapse and generation of pumice-and-ash flows. Ballistic clasts (up to 1.2m diameter) were ejected distances up to 1.6km with observed velocities of 110-140m s - 1 . Eruption column heights ranged from 3km to a maximum of 15km, with inferred peak discharge rates of hundreds to a few thousand of cubic metres per second. Druitt et al. (2002) estimated a typical volume of ejecta as 3 x 10 5 m 3 (dense rock equivalent), although volumes of the largest individual explosions may have exceeded 10 6 m 3 . If the conduit has a diameter of 30m (see Melnik & Sparks 2002) then typical drawdown depths are 200-1000m. Pumice clasts in the Vulcanian explosions have angular and platy character (Druitt et al. 2002). Alidibirov & Dingwell (1996) have reproduced similar angular, platy shapes in laboratory experiments in which natural dome samples and pumice were decompressed from fluid pressures of a few megapascals. The observations on Vulcanian ejecta from Soufriere Hills Volcano support brittle fragmentation of vesicular magma under conditions where the overpressure exceeded the tensile strength of the magma. We infer that a fragmentation wave penetrated into already vesiculated magma.

Description of the physical model for explosive volcanic eruptions

Fig. 2. Schematic representation of flow regimes during explosive eruptions.

the conduit. At the outlet of the conduit the exit pressure is higher than atmospheric pressure due to the weight and viscous resistance of the overlying dome. At the moment t = 0 at the top of the conduit we assume a sudden pressure drop down to the atmospheric value caused by dome collapse, and we study the dynamics of the flow in the conduit caused by this pressure drop. We will examine the influence of fragmentation criteria, pre-eruption parameters and assumptions on the mechanism of mass transfer between magma and growing bubbles on the eruption behaviour. The unsteady model is based on the model developed by Melnik (2000) and is used to describe the motion of the multiphase mixture in the conduit, with some modifications to make the numerical code easier and faster, and to account for the presence of crystals in the ascending magma.

System of equations We consider the following mechanism of explosive eruption. As the overlying dome collapses, the pressure at the top of the conduit decreases rapidly. Conditions soon thereafter exceed a critical condition and explosive activity begins. A rarefaction wave propagates into vesiculated magma. In the rarefaction wave the mixture accelerates and fragmentation occurs when some critical condition, which will be discussed below, is reached. Fragmented magma forms a gasparticle dispersion, which propagates to the exit of the conduit to form a volcanic column in the atmosphere. Figure 2 shows schematically the processes which occur in the conduit. The problem of unsteady magma flow dynamics in a volcanic conduit is considered. Steady-state solutions for the flow of magma during lava-dome extrusion (Melnik & Sparks 1999) are used to define the initial porosity, velocity and pressure distribution along

The mechanical description of the conduit flow during explosive eruptions has been discussed in Barmin & Melnik (1993) and Melnik (2000). For unfragmented magma the main assumptions are: (i) the ascent velocity of bubbles and the gas velocity in permeable media (Melnik & Sparks 2002) are negligible in comparison with the velocity that the mixture attains between the rarefaction wave and fragmentation level (Fig. 2); (ii) temperature variations are small due to the high thermal capacity of magma; and (iii) no additional bubble nucleation is assumed as the rarefaction wave propagates into initially vesiculated magma. We also neglect changes in crystal content due to microlite crystallization (Melnik & Sparks 2002), as the time scale of this process is much longer than the duration of explosive eruptions.

MODELLING OF EXPLOSIVE ERUPTIONS

We now estimate the degree of non-equilibrium of mass transfer between the melt phase in the magma and the bubbles. In an explosive eruption two end-member situations can be considered. First, magma in the conduit has some distribution of vesicularity and gas pressure at the onset of explosive flow, and the propagation of the fragmentation front is too fast for any further diffusive mass transfer from the melt into vesicles. This end-member case is completely non-equilibrium with no mass transfer. Second, propagation of the fragmentation wave is sufficiently slow that diffusion has time to maintain equilibrium between the melt phase and the free-gas phase. This end-member is an equilibrium case. In principle the system can move from one end-member to the other during an explosive eruption, as discussed further below. At the initial stage of eruption, when the rarefaction wave is strong, fragmentation occurs just after the decompression and therefore mass transfer can be neglected. At later stages the fragmentation wave stops, or decreases its velocity significantly, and mass transfer can become important, with equilibrium attained if the flow conditions become sufficiently slow. A dimensionless criterion can be defined to evaluate the degree of equilibrium of mass transfer. Alidibirov & Dingwell (1996) estimated the velocity of a fragmentation wave, VF, to be in the range of tens to over one hundred metres per second. The speed of sound for the mixture, C = (dp/dp)1/2, is in the range of a hundred to several hundred metres per second, depending on the volume concentration of bubbles. Therefore, the length of the region between the front of the rarefaction wave and fragmentation wave (see Fig. 2), LF, is equal to (C - VF)t. The characteristic time of fragmentation is then tF = L F / (V F + V), where V is an average velocity of unfragmented magma. This time can be compared with the characteristic time of diffusion of dissolved gas tD = h s 2 /D (here hs is the thickness of the liquid shell surrounding the bubble and D is a diffusion coefficient) to establish whether mass transfer is significant. The parameter PeD = h s 2 (V F + V)/DLF determines the degree of nonequilibrium, with PeD > 1 being the case where diffusion can be neglected and PeD V and PeD = h s 2 (D(C/V F - 1)t) - 1 . The value of N is in the range of 1010 -1014 m-3 (Navon & Lyakhovski 1998) and the volume fraction of bubbles before the fragmentation is estimated to be between 0.3 and 0.6 (Melnik 2000). Even for an extreme set of parameters (N = 10 1 4 m - 3 ,D=10 - 1 1 m2s-1, = 0.6 and C/VF = 1.5) the condition of equilibrium mass transfer (PeD < 0.1) requires t > 40 s. During this time a fragmentation wave will travel several kilometres. Thus during magma fragmentation, mass transfer between melt and bubbles is negligible. In the second situation, after the fragmentation wave stops (VF = 0), we can rewrite PeD = hs2V/(DCt). Average velocity can be estimated by means of the Pousieulle law as V = pd 2 /32uL F where A/? is a pressure drop in the rarefaction wave, d is the conduit diameter and u is magma viscosity. Thus the final form of the criterion is: PeD = (3/4 N) 2/3 (l - 1/3)2 pd 2 /32uD(Ct) 2 . As magma degasses, the product uD is expected to remain approximately constant since melt viscosity increases and water diffusivity decreases as gas is lost. A typical value of uD is 10 - 4 N for u = 107Pa s and D= 1 0 - 1 1 m 2 s - 1 for Montserrat andesite at 850°C with 5wt% dissolved water in the rhyolitic melt (Barclay et al 1998). The upper estimate of PeD for N= 10 10 m- 3 and a = 0.3 gives PeD = 11.7/t 2 . Thus mass transfer reaches an equilibrium state after approximately 10s. The lower estimate for PeD (N = 10 1 4 m - 3 and a = 0.6) indicates equilibrium mass transfer when t > 0.23 s. Therefore the system can rapidly transform from negligible mass transfer to the equilibrium case when the fragmentation wave stops. There are several processes which have not yet been incorporated into models of conduit flow during explosive eruptions. These include continuous nucleation generating a size distribution of bubbles, interaction of bubbles at high concentrations, and a full

309

analysis of diffusion of gas in the melt shells surrounding growing bubbles. Incorporation of the diffusive bubble growth in particular requires elaborate numerical models (Proussevitch et al. 1993; Navon & Lyakhovski 1998). Here we consider only the two endmember cases: equilibrium mass transfer and total absence of mass transfer between melt and bubbles to bound the range of behaviours. For the gas-particle dispersion we assume that the particle temperature is constant and equal to the temperature of the gas phase. This is justified by a high heat capacity, large mass concentration of particles and small size of particles as in previous models (e.g. Wilson et al. 1980). We also neglect the variation of the gas temperature since there is rapid heat exchange between the gas and small particles. At the outlet of the volcanic conduit, where the particle concentration is low and the expansion rate of the gas is substantial, the gas temperature will decrease by no more than 3-5% (Barmin & Melnik 1993). We also consider that relative velocity between particles and the gas is small in comparison with the mixture velocity. With these assumptions the system of flow equations for both unfragmented magma and gas-particle dispersion (with u = 0) can be written in the following form:

(la) (1b) (1c) (1d) (1e) (1f)

Here the following notations are used: q are densities (subscript: g, gas; c, condensed phase, crystals plus melt; no subscript, mixture), p0g is a density of pure gas, V is the mixture velocity, pg and pc are pressures, n is number density of bubbles, a is the bubble radius, dis the conduit diameter, u is the mixture viscosity, T is magma temperature, R is gas constant, t is time and x is a vertical coordinate. The system consists of the continuity equations for gas and condensed components (Equations la and 1b) and number density of bubbles (Equation 1c), the momentum equation for the mixture as a whole (Equation 1d), as well as the Rayleigh-Lamb equation (see Scriven 1959) for bubble growth (Equation le). Equations 1f and lg are the perfect gas law and the volume fraction of bubbles. Gravity forces and the conduit resistance (in Pousieulle form) are taken into account in the momentum Equation 1d. We can introduce densities in a way to avoid mass transfer terms in the mass conservation Equations la and 1b. For the case of equilibrium mass transfer the densities of the gas and the condensed phase can be written as follows:

(2a) (2b) (2c) Here a is volume concentration of bubbles, (3 is volume concentration of crystals in condensed phase, pm and px are the densities of melt and crystals, c is mass fraction of dissolved gas, and Cf is a solubility coefficient. In Equations 2 the densities are defined in such a manner that Equation 2b incorporates both bubbles and dissolved gas, and Equation 2a incorporates only melt and crystals.

310

O. M E L N I K & R. S. J. SPARKS

In the case of no mass transfer, pg is the density of the gas in the bubble phase and pc includes the remaining dissolved gas in the melt phase:

(3a) (3b) We constrained magma rheological properties for the case of the Soufriere Hills eruption, as discussed by Melnik & Sparks (2002). Magma viscosity is a strong function of the volume concentration of dissolved gas and crystal content. In the case of equilibrium mass transfer, viscosity is calculated for equilibrium concentration of dissolved gas, and in the case of no mass transfer the conservation of viscosity in the melt phase is incorporated into Equations 1: (4)

The system of Equations 1 and 2 can be simplified due to the assumption of equal component velocities and equilibrium of mass transfer. From Equations la, 1b and 1c we can obtain the following integral relationship between the densities of components: (5)

which allows density and volume fraction of bubbles to be calculated as functions of pressure. The Rayleigh-Lamb Equation le can be developed in the form: (6)

In the case of no mass transfer = 0. Taking into account Equations 5 and 6, Equations 1 can be reduced to the Euler equations of gas dynamics with a complicated equation of state and an additional term due to pressure non-equilibrium:

Initial and boundary conditions We use the steady-state solution of the lava dome extrusion model (Melnik & Sparks 2002) to determine the initial distribution of parameters along the conduit. We assume that there is no further crystallization during the explosive eruption and that the crystal content (3) is constant. For the case of equilibrium mass transfer, volume concentration of bubbles is a function of pressure only, therefore gas loss due to the development of magma permeability is assumed to be negligible. This limitation is not assumed in the case of no mass transfer, where the volume fraction of bubbles is treated independently of pressure. We assume that the top 200m of the conduit is occupied by the gas-particle dispersion with pressure equal to atmospheric pressure: this avoids difficulties with boundary conditions at the initial stages of eruption before the flow structure is developed. Particular choices of the length of this zone make no difference to the outlet flow parameters a few seconds after the beginning of eruption. Values of parameters after Melnik & Sparks (2002) used for the calculations are listed in Table 1. Exit pressure at the top of the conduit is determined by the hydrostatic weight of the dome and its viscous resistance, and therefore depends on discharge rate and initial dome height before the collapse. We use a value of discharge rate (about 3 m3 s-1) that gives an exit pressure of 12MPa and a volume fraction of bubbles equal to 0.6 to compare the three fragmentation criteria. Figure 3 shows the pre-eruptive distributions of volume fraction of gas and pressure along the conduit for the two end-member cases. Only one boundary condition at the bottom of the conduit is needed in the case of equilibrium mass transfer and three boundary conditions in the case of no mass transfer. As the volume erupted in an individual explosion is much smaller than the chamber volume, it is reasonable to assume fixed pressure in the chamber throughout the eruption. The vesicularity and viscosity of the magma feeding into the base of the conduit are also assumed to be constant. If outflow from the conduit is supersonic or sonic, no boundary condition is needed because disturbances related to the atmospheric part of the flow cannot propagate into the conduit. In contrast, subsonic exit conditions require the conduit flow to be coupled with the flow in the atmosphere by assumption of continuity of all parameters at the top of the conduit. In this case, an artificial boundary condition is developed. We assume that, when the exit pressure is higher than atmospheric pressure and velocity is subsonic velocity remains sonic and

(7)

Table 1. Parameters for the Soufriere Hills andesite eruption used in modelling

Here A is a coefficient that indicates the flow regime: = 1 for bubbly liquid and A = 0 for gas-particle dispersion. In the case of no mass transfer, the volume fraction of bubbles remains an independent variable and pressure is a function of both density and a (Equation la). Because the conduit resistance differs strongly between the regimes of unfragmented magma and gas-particle dispersion (Fig. 2), magma discharge rate depends on the velocity of the fragmentation wave. Different criteria can be developed to characterize fragmentation. The earliest suggestion was that fragmentation occurs if the volume fraction of bubbles exceeds some critical value (Sparks 1978; Wilson et al. 1980). However, recent studies (Barmin & Melnik 1993; Melnik 2000; Alidibirov & Dingwell 1996; Papale 1999) suggest that fragmentation occurs when a critical overpressure (OP, defined as the difference between the pressure in a growing bubble and in the melt) or a critical elongation strain rate (SR) of magma is reached. In spite of different criteria for fragmentation, these latter two models suggest that the product of viscosity and elongation strain rate will reach a threshold value at fragmentation. We will study numerically the influence of different fragmentation conditions on the discharge rate.

Parameter

Symbol Value range

Magma chamber depth

L

5km

PC

l0MPa

Melt water content

C0

5%

Barclay et al. (1998)

Magma temperature

T

850 C

Barclay et al. (1998): Murphy et al. (2000)

Magma crystal content

3

0.6

Murphy et al. (2000)

Conduit diameter

D

30m

Dimensions of spines and early crater: hornblende reaction rims (Devine et al. 1998: Watts et al. 2002)

Density of melt

Pm

2300 kg m-3

Density of crystals

Px

2700 kg m-3

Solubility coefficient

cf

Magma chamber overpressure

4.1 x 10 - 6 Pa - 1

Information sources

Earthquakes (Aspinall el ill. 1998) and phase equilibria (Barclay el al. 1998)

2

Stolper (1982)

MODELLING OF EXPLOSIVE ERUPTIONS

311

Table 2. Calculation sets Run number

Fragmentation criteria

Mass transfer

1 2 3 4 5 6 7 8 9

VF (0.60) OP(l0MPa) SR(250MPa) VF (0.60) VF (0.70) VF (0.80) OP(l0MPa) OP(50MPa) SR (250 MPa)

Equilibrium Equilibrium Equilibrium No No No No No No

VF, fixed volume fraction of bubbles; OP, critical overpressure in growing bubble; SR, critical elongation strain rate. Critical values are given in parentheses.

Figure 4 shows profiles of (a) discharge rate, (b) volume fraction of bubbles and (c) pressure for run number 1 which adopts the volume fraction (VF) fragmentation criterion and equilibrium mass transfer, with initial conditions shown by curve 1 on Figure 3. A rarefaction wave propagates into the conduit, accelerating the mixture and decompressing it. The fragmentation wave, which follows the rarefaction wave, splits the flow domain into two different regions: bubbly magma and gas-particle dispersion. As the magma viscosity and volume fraction of bubbles are functions of pressure only, all the properties are fixed at the fragmentation level. High magma viscosity before fragmentation prevents rapid acceleration of the mixture in the rarefaction wave; therefore disturbances cannot propagate far into unfragmented magma. There is a zone of

Fig. 3. Distributions of (a) volume fraction of bubbles and (b) pressure in the conduit used as initial conditions in the calculations of unsteady explosive eruption. Curve 1 corresponds to conditions of impermeable magma; curve 2 takes into account gas loss due to the development of permeability in the ascending magma. Pressure at the top of the conduit is 12MPa.

pressure decreases in a way such that discharge rate at the outlet is equal to the discharge rate in the nearest mesh point in the conduit. This condition dampens oscillation of parameters during the transition from supersonic to subsonic regime and gives a monotonic decrease of exit pressure to the atmospheric value. This boundary condition can be overcome in a coupled model taking into account both conduit flow and atmospheric plume dynamics. This model is under development, but is not considered here. A detailed technical description of the numerical model is given in the Appendix.

Results of numerical modelling We present the results of simulations of the explosive eruptions generated by a sudden pressure drop at the top of the conduit. We focus on the role of fragmentation criteria and mass transfer between growing bubbles and the surrounding melt in the eruption dynamics. Parameters in the models are listed in Table 1. Table 2 summarizes the set of calculations for reference.

Fig. 4. Typical profiles of (a) discharge rate, (b) volume fraction of gas, and (c) pressure in the conduit during an explosive eruption. Fragmentation wave propagates down the conduit. Corresponding times (in seconds) are marked on the curves.

312

O. MELNIK & R. S. J. SPARKS

sharp gradients near the fragmentation level. As the velocity of the fragmentation wave decreases, gradients become smoother and the rarefaction wave separates from the fragmentation wave. Similar parameter distributions were observed for all other runs. Figure 5 shows (a) magma discharge rate at the outlet of the conduit and (b) position of fragmentation level versus time for runs 1 to 3 which investigate different assumptions for fragmentation with the same initial conditions and physical properties of magma as described in Figure 3. In case of the VF criterion, there is always a level in the conduit where the porosity reaches the fragmentation threshold, so fragmentation occurs continuously throughout the eruption. With time the system comes to a steady state, but this evolution is very slow, as the relaxation time in highly viscous magma is long. In the case of the OP and the SR criteria, fragmentation occurs as a series of discrete events. As the fragmentation wave propagates downwards, the pressure in the gas-particle dispersion after the fragmentation increases because its inertia prevents rapid evacuation of fragmented material. This leads to decrease in magma viscosity before fragmentation; consequently the critical fragmentation conditions cannot be reached and fragmentation stops. Between fragmentation events pressure above the fragmentation level decreases as material is evacuated and viscosity

Fig. 5. Variations of (a) discharge rate and (b) fragmentation depth with time for the case of equilibrium mass transfer between melt and bubble phase. Fragmentation criteria: VF, fixed volume fraction of bubbles; OP, critical overpressure in growing bubbles; SR, critical elongation strain rate.

increases again. Each fragmentation event leads to rapid inflow of newly fragmented material into the gas-particle dispersion zone, which causes sharp increases in flow rate. This increase propagates up the conduit and causes large changes in discharge rate at the surface. Thus the models incorporating more realistic fragmentation criteria produce strong fluctuations in discharge rate, although the broad evolution of the eruption is similar to all three cases. For run number 1 (VF fragmentation criterion. Fig. 5), the velocity at the top of the conduit is initially supersonic and reaches a maximum value of 2 5 0 m s - 1 after about 10s with a maximum Mach number value of 2.5. The maximum pressure at the top of the conduit is 2.7MPa. To estimate the characteristic timescale of discharge rate decrease we approximated the calculated values (after the maximum in discharge rate) by means of exponential functions. This approximation gives the characteristic timescale (when the current value is l/e times smaller then the initial value) of 140 s over the first 200 s of eruption, 1140 s from 200 s until the end of the explosion, and an average of 890 s for the whole discharge interval. The initial fragmentation wave velocity is 9 2 m s - 1 . but it decreases to 2 0 m s - 1 after 23s. After 10 minutes of eruption the velocity of the fragmentation wave is only 1 . 2 m s - 1 . The whole duration of eruption is about an hour (if explosive discharge is assumed to stop as exit velocity becomes less then 5 m s - 1 ) ; therefore only previously fragmented magma evacuates from the conduit during the last 50 min. Total erupted mass is 4.5 x 109 kg, which gives an erupted volume of approximately 1.7 x 10 6 m 3 dense rock equivalent of lava. The fragmentation penetrates to a depth of about 4km in run 1. In the case of OP and SR fragmentation criteria, the velocity of the fragmentation wave is much higher, reaching 260 and 2 0 0 m s - 1 respectively. As the influx of fragmented material is higher, the mixture accelerates up to velocities of 370 and 2 8 0 m s - 1 respectively (corresponding Mach numbers are 3.8 and 3.1). Fragmentation stops abruptly as critical values of OP or SR become unreachable. The duration of the first fragmentation event is 9s in the case of OP and 4.5 s in the case of SR. Then the boundary between the bubbly unfragmented magma and gas-particle dispersion starts ascending with a velocity of about O . l m s - 1 . As the pressure at the fragmentation surface is not fixed, the bubble fraction decreases to values of 0.25-0.3 as the fragmentation wave propagates downwards. After a series of several fragmentation events, the fragmentation level reaches a depth of 3500m for SR and arrives at the magma chamber for the OP criteria. Corresponding total erupted masses are 4.9 and 5.2 x 109 kg. In the initial stage of eruption the fragmentation wave almost follows the rarefaction wave and the assumption of equilibrium mass transfer is not applicable close to the fragmentation level. Thus, as discussed further below, the models are not likely to be realistic in the initial stages. We therefore examine the other endmember case of no mass transfer and the explosive eruption of a vesiculated magma column with a distribution of porosity and pressure given by the dome extrusion dynamics (Fig. 5; Melnik & Sparks 2002). Figure 6 shows the variation in (a) discharge rate and (b) the position of the fragmentation level with time for the case of no mass transfer and the VF fragmentation criterion for initial distribution of parameters represented by curve 2 on Figure 3. Runs 4-6 differ in the critical bubble fraction for the VF fragmentation criterion. For comparison, the results from run 1 are also plotted (dashed line). The velocity of the fragmentation wave is much smaller then for the case of run 1: 49, 27 and 1 4 m s - 1 for a = 0.6. 0.7 and 0.8 respectively. Corresponding maximum velocities at the top of the conduit are 130, 118 and 9 4 m s - 1 , with Mach numbers of just over unity. As we assume no mass transfer, the fragmented mixture is denser because it contains less gas (minimum gas fractions at the vent are 0.85, 0.89 and 0.92, respectively). Therefore the discharge rate (density x velocity x cross-sectional area of conduit) is similar to the case of equilibrium mass transfer (22.2 x l 0 6 k g s - 1 for run 1 and 19.2 x 1 0 6 k g s - 1 for run 4). Discharge rate decreases more rapidly than in the case of equilibrium mass transfer, as the rate of magma fragmentation is smaller and decreases faster as

313

MODELLING OF EXPLOSIVE ERUPTIONS

1 Fig. 6. Variations of (a) discharge rate and (b) fragmentation depth with time for the case of no mass transfer between melt and bubbles. VF fragmentation criterion: critical value of VF is labelled on curves. Dashed curve represents solution of equilibrium (e) mass transfer equations for comparison, with VF = 0.6. the fragmentation wave comes into regions of low porosity. Maximum depth of fragmentation is significantly smaller (1700m for run 4 instead of 4100m for run 1). Increase in the critical volume fraction value decreases the maximum value of the discharge rate by about two, but the discharge rate decreases asymptotically with the same speed, as it is controlled by the inertia of the fragmented gas-particle dispersion. Fragmentation depth also decreases as the threshold porosity for fragmentation increases. Figure 7 shows variations of (a) discharge rate and (b) depth of fragmentation level with time for the case of OP (10 and 50MPa) and SR (250 MPa) fragmentation criteria in the case of no mass transfer (runs 7, 8 and 9, Table 2). The use of a higher fragmentation threshold (50 MPa) in run 8 is justified by high volume content of crystals (about 60%) in Montserrat magma, which should increase the overall strength of magma. The curve for run 4 using the VF fragmentation criterion is shown for comparison (dashed lines). The initial velocity of the fragmentation wave is higher than in the VF case criteria (200 and 150m s-1 in the case of OP for runs 7 and 8, 1 1 0 m s - 1 in the case of SR). As the viscosity of magma in the conduit depends only on its preemptive distribution for the case of no mass transfer, its value decreases rapidly with depth and fragmentation stops quickly. As all

10 100 time (s)

1000

Fig. 7. Variations of (a) discharge rate and (b) fragmentation depth with time for the case of no mass transfer between the melt and growing bubbles. dissolved gas in this case remains in the melt, viscosity does not increase after the fragmentation stops. Maximum values of bubble overpressure or elongation strain rate decrease monotonically with time after the end of the initial fragmentation phase, as the intensity of the rarefaction wave decreases. Fragmentation processes cannot therefore start again. The assumption of no mass transfer must break down at later stages of the eruption and the model does not predict the asymptotic behaviour of the eruption correctly. There are two possible developments. First, diffusive mass transfer becomes sufficiently fast that gas-melt equilibrium is attained, so that much greater depths of the conduit are tapped, as in runs 1 to 3. Second, mass transfer may be too slow for explosive conditions so lava-dome extrusion resumes. In the case of the OP criterion, fragmentation stops after 6 and 4.5s and the depth of fragmentation is 870 and 580m for runs 7 and 9, respectively; in the case of SR criterion these values are 3 s and 250m respectively. When the fragmentation stops there is an outflow of previously fragmented material from the conduit. Discharge durations are 300, 130 and 80s, respectively. Figure 8 shows calculated (a) gas fraction, (b) exit velocity, and (c) bulk density of the eruption mixture for runs presented in Figure 7. Although the models differ in detail according to the fragmentation criteria, they all show common features. The maximum in velocity occurs in the first few seconds and declines thereafter. The maximum in density of the discharging mixture occurs several

314

O. M E L N I K & R. S. J. SPARKS density at later times will therefore tend to change conditions towards convective uprise. Summarizing the results of the calculations, assumptions on the intensity of mass transfer between melt and bubbles, and fragmentation criteria make significant differences to the eruption behaviour. In the case of no mass transfer, the velocity of the fragmentation wave and the duration of eruption both decrease compared to the case with mass transfer. Application of the VF criterion of fragmentation generates more moderate and longer-lived eruptions. In the case of equilibrium mass transfer, the velocity of the fragmentation wave reaches a few hundred metres per second, which is substantially higher than velocities observed in experiments (Alidibirov & Dingwell 1996). As discussed earlier, these high calculated velocities are unrealistic, because the timescales are too short for mass transfer to occur. Because of rapid influx of fragmented material into the gas-particle dispersion, the discharge rate at the top of the conduit is also overestimated. A totally non-equilibrium model produces fragmentation rates comparable with experimental data. A prominent feature of the equilibrium model is that fragmentation occurs in a series of separate pulses separated from each other by periods of magma ascent without fragmentation. Each fragmentation episode generates a sharp increase in eruption intensity at the outlet of the conduit.

Comparison of the results with field observation data

Fig. 8. (a) Volume fraction of bubbles, (b) exit velocity and (c) bulk density at the top of the conduit as a function of time for the case of no mass transfer. Abbreviations as in Figure 5. seconds later. The bulk densities and volume fraction of particles in discharging gas-particle dispersions are very high (100-250 kg m-3 and 0.1-0.3 respectively) and so the conditions for fountain collapse are strongly favoured (Clarke et al. 2002). Declining mixture

We have developed models for unsteady explosive eruptions using different fragmentation criteria, assumptions of the initial distribution of porosity, and assumptions of the importance of mass transfer between melt and gas phases. As summarized in Table 1, the models are illustrated with parameters thought to be relevant to conditions during the explosive activity of Soufriere Hills Volcano. We now compare features of the models with observations during this eruption. We suggest that the short-lived, repetitive Vulcanian explosions are examples close to the end-member case of no mass transfer. The timescales of the Vulcanian explosions were a few tens of seconds and are comparable to those predicted by the model. In both models and observed Vulcanian explosions, peak intensity occurs very early (order of 10 s). Peak discharges of several thousand cubic metres per second are predicted and are comparable to those estimated from column heights (Druitt et al. 2002) for some of the larger Vulcanian explosions. The natural explosions merged into a longer waning stage of vigorous degassing, which may represent the discharge of fragmented material and escape of gas from the top of the rising, but non-fragmenting magma below. The drawdown depths of Vulcanian explosions are estimated at a few hundred metres to c.2 km (Druitt et al. 2002) and are consistent with those estimated in the models. The models predict peak velocities at the early stage followed by a maximum in erupted mixture density after the order of 10s. Conditions for fountain collapse are strongly favoured in this first stage, and conditions favourable for convective column formation develop later as mixture density decreases, and also as mass transfer between melt and gas phases becomes more important. The broad sequence of fountain collapse, then vertical convective column development, is in accord with observations (Druitt et al. 2002). Figure 9 summarizes our interpretation of the two series of repetitive Vulcanian explosions in 1997. Magma rising in the conduit increases its internal pressure, resulting in inflation as recorded by tiltmeters and hybrid seismicity (Voight et al. 1999). The system reaches a condition where the internal pressure equals the tensile strength of the magma and an explosion is triggered. Melnik & Sparks (2002) have shown that steady solutions for lava extrusion show overpressures comparable to. or greater than, the tensile strength of crystal-rich magma (a fewmegapascals) at a fewhundred metres depth in the conduit. Thus the system will attempt to reach steady-state flow conditions, but will be interrupted when the overpressure exceeds the tensile strength. The system fails

MODELLING OF EXPLOSIVE ERUPTIONS

Fig. 9. Schematic view of the sequential (1-3) processes in the conduit during one cycle in the series of repetitive Vulcanian explosions. See text for explanation.

repeatedly in explosions. The field evidence of angular platy pumice supports brittle fragmentation and development of a fragmentation wave, as envisaged by Alidibirov & Dingwell (1996). These observations also indicate that the OP fragmentation criterion is to be preferred to the SR criterion, since there is no textural evidence for strain elongation in Montserrat pumice. An explosion ensues, building up quickly in peak intensity and then waning after about a minute. Most of the Vulcanian explosions resulted in fountain-collapse conditions. This is consistent with the no-masstransfer end-member, which releases much less gas, so that the erupted mixture densities are high. Fragmentation stops abruptly and unfragmented magma starts to rise. Although mass transfer must begin again, it is evidently too sluggish to re-establish or maintain conditions for explosive eruption. The downturn of the cycle involves deflation almost sufficient to recover all the strain stored during inflation (Voight et al. 1999). During this period there is vigorous ash-venting. We interpret this stage as due to gas being lost from the top surface of the rising magma column. Eventually the mass transfer and gas exsolution become sufficiently important that pressure once again starts to rise in the magma, with the system moving towards an overpressured state and towards the critical condition for another explosion several hours later. The sub-Plinian activity of 17 September 1996 differs from the Vulcanian explosions in several ways that suggest it is better described by the equilibrium case. The eruption was longer-lasting and more energetic, as shown by greater ranges of ballistics and by the total mass of ejecta being much greater than that of the largest Vulcanian explosion (Robertson et al. 1998). The eruption is estimated to have tapped down to 4km below the top of the conduit, and the subsequent rise of magma took two weeks with no development of repetitive explosive activity. There was also only a vertical convective column with no fountain collapse, suggesting low densities of the discharging mixture and substantial gas exsolution. The equilibrium mass transfer models show comparable features with larger exit velocities, drawdown to near the base of the conduit and comparable peak discharge rates of a few thousand cubic metres per second. A prominent feature of the models using the OP and SR fragmentation criteria is marked pulsations. We interpret the strong pulsations in the RSAM record of the seismic energy (Fig. 1) as recording the predicted pulsations. The equilibrium model is almost certainly unrealistic early in this eruption and so the prediction of the maximum intensity peak at about 10s is not thought to be realistic. Rather, we anticipate that the first few tens of seconds involved clearing of a rheologically

315

stiffened magma due to degassing and crystallization in the upper part of the conduit, before a discharge of deeper vesiculated magma could develop. This early stage is better modelled by the no-masstransfer model. Subsequently, each later fragmentation event would be better described by the no-mass-transfer model, as the timescale of these events is much shorter than that of diffusive bubble growth. The notion of an initial vent-clearing phase is consistent with strong ballistic ejection of denser lithologies from the uppermost conduit (Robertson et al 1998), followed by more vesiculated pumice. We interpret this eruption as evolving from the no-mass-transfer case to the equilibrium case, so a more comprehensive model which incorporates mass transfer will be required for complete description. An interesting feature of the 17 September 1996 explosive activity is the 2.5 hour gap between the end of dome collapse and the onset of the explosion (Robertson et al. 1998). This suggests that the unloading was not quite sufficient for the fragmentation conditions to be reached. We suggest that the unloading triggered gas exsolution in the conduit and this led to a build-up of critical conditions of overpressure in the uppermost parts of the conduit. In this sense the triggering mechanism is similar to that of the two series of Vulcanian explosions and is not explicitly unloading. Exsolution of gas in stiffened magma resulted in attainment of a critical overpressure threshold. Thus major dome collapses favour development of conditions for explosive eruption, but are not the specific trigger of the explosions themselves. Conclusions We have developed models of unsteady explosive eruptions following decompression of a magma column. The full problem would require incorporation of mass transfer as gas exsolves from the melt to the bubbles and would involve many complications and uncertainties. However, progress can be made by exploring two end-member cases. In the equilibrium case the distribution of gas between melt phase and bubbles is everywhere in equilibrium and so porosity is only a function of pressure. This case attains where the timescale for diffusion is small compared to the timescale of magma decompression and ascent. In the no-mass-transfer case only gas exsolved prior to onset of explosive discharge can expand and there is no further mass transfer. This case is attained when the timescale for diffusion is large compared to the timescale of explosive eruption. Our exploration of typical values of parameters suggests that both end-members can occur in nature. In particular we have interpreted the explosive activity at Soufriere Hills Volcano in terms of the two end-members. (1)

The Vulcanian explosions are envisaged as close to the nomass-transfer case where slowly rising magma develops a porosity and overpressure structure in the conduit that evolves to critical conditions for explosive fragmentation. The short durations, energetics and tendency for fountain collapse are consistent with the model, which predicts parameters broadly comparable to those observed. The associated patterns of repetitive explosions are also consistent with development of overpressures to trigger onset of explosive activity (Voight et al. 1999), followed by an abrupt halt of fragmentation and slower degassing. (2) The sub-Plinian activity of 17 September 1996 is interpreted as close to the equilibrium case, although the earliest stages and each discrete fragmentation event are likely to be closer to the no-mass-transfer case. Again the model predicts conditions consistent with observations such as discharge rates, eruption duration, volumes and inferred deep drawdown depth. (3) We have also studied the influence of fragmentation criteria on eruption dynamics, comparing the cases of constant-porosity criterion (VF), a critical overpressure criterion (OP) and a critical strain rate criterion (SR). Overall the differences in terms of averaged properties of mixture at the outlet of the conduit are small, as they are strongly controlled by the inertia of the gas-particle dispersion. The OP and SR criteria predict

316

O. MELNIK & R. S. J. SPARKS

strong pulsations for the equilibrium mass transfer conditions. Feedback processes cause fragmentation to pause and then resume as the erupted mixture is discharged. There are only subtle differences between the OP and SR results, although the former is favoured for Soufriere Hills Volcano on the basis of observations of shapes and textures of pumice clasts. The introduction of strong pulsations is in good agreement with seismic energy fluctuations (Fig. 1) and also observed pulsations in Vulcanian explosions. The authors acknowledge a grant from the Royal Society for the visit of O. Melnik to Bristol University, NERC Research Grants GR3/11683, GR3/11020 and GR3/10679. R.S.J.S. acknowledges support from the Leverhulme Trust (F/182/AL) and NERC through a Research Professorship. O.M. acknowledges support from the grant of Russian Foundation for Basic Research (99-01-01042). Comments by A. Barmin and A. Woods and reviews by A. Neri and T. Druitt were much appreciated.

Appendix

else end if

else where

k

— max(| -(u)|)

end if

Marquina's flux formula F M (u l . u r ) is then:

where r p (u l ), r p (u r ) are the right (normalized) eigenvectors of the Jacobian matrices A(u l ), A(u r ). The first-order scheme based on Marquina's flux formula is thus:

Numerical method Due to the presence of a moving boundary separating unfragmented magma and gas-particle dispersion with very different physical properties, computation of the unsteady problem for Equations 1 has some difficulties. For example, in the bubbly flow zone, the pressure gradient is governed by mixture weight and conduit resistance with the inertial term being small. In the gas-particle dispersion, conduit resistance is negligible and inertia is dominant. Near the fragmentation surface all terms in Equations 1 are of the same order. The fragmentation wave is a zone of very steep gradients of the main variables, and has a thickness much smaller than the conduit length (Barmin & Melnik 1993; Melnik 2000). Another difficulty arises from the differential form of fragmentation criteria (Equation 6). All complications listed above produce strong requirements on the numerical method in terms of accuracy and stability. We have used numerical methods of'large particles' (Davidov & Belotzerkovskii 1980) and the Lax-Wendroff method with additional smoothness of solution (Ramos 1995). Results were satisfactory only for weak explosive eruptions and the condition of fixed bubble fraction as a fragmentation criterion. Finally, we used the flux-splitting algorithm developed by Donat & Marquina (1996). The idea of this method can be explained if we rewrite the governing equations in vector form, keeping only the convective terms:

The values of unknowns u are calculated in the middle of the mesh cells; u l and ur represent the values on the left and right boundaries of the cell respectively. These values will be used for the calculation of fluxes (mass and momentum) between cells respectively. The algorithmic description of Marquina's flux formula is as follows. Given the left and right states, we compute the 'sided' local characteristic variables and fluxes:

for p= 1 , 2 , . . . , m . Here 1 p (ul), l p (u r ), are the (normalized) left eigenvectors of the Jacobian matrices A(u l ), A(u r ) (A(u) = f/ u). Let l (u l ),... , m (u/) and / ( u r ) , . . . , m (u r ) be their corresponding eigenvalues. We proceed as follows: For k = 1 , . . . , m, If (u) does not change sign in [ul ur], then If (u l ) > 0 then

As the magma viscosity is very high in the bubbly flow regime, viscous terms [4/3>( / x)] f( )( V x) and (32 d2) require high accuracy. They are treated implicitly. Therefore the momentum Equation 1d leads to a three-point implicit scheme which is solved by means of the Thomas method (Tannehill et al. 1997). Convective terms in this equation are taken in explicit form as given by Marquina's flux. The usual Courant-Friedriehs-Levy (CFL) condition and CFL number of 0.5 are used in all calculations to define the time step. The code was tested on the analytical solution of a shock tube problem for a perfect gas and gives an accuracy about 1 % for 500 nodes. As the fragmentation wave is an area of steep gradients we used finer grid (2500 points or 2m in space) to solve the flow accurately. A 500-point grid gives an error of 5% in the calculated discharge rate at the top of the conduit and 15% in evaluation of the position of the fragmentation level. The difference in solution using 2500 and 5000 grid points is negligibly small.

References ALIDIBIROV. M. A. & DINGWELL, D. B. 1996. Magma fragmentation by rapid decompression. Nature, 380, 146-148. ASPINALL, W. P.. MILLER, A. D., LYNCH, L. L.. LATCHMAN. J. L., STEWART, R. C. WHITE, R. A. & POWER. J. A. 1998. Soufriere Hills eruption. Montserrat: 1995-1997: volcanic earthquake locations and fault plane solutions. Geophysical Research Letters, 25, 3397-3400. BARCLAY. J., CARROLL, M. R., RUTHERFORD. M. J., MURPHY, M. D., DEVINE. J. D., GARDNER, J. C. & SPARKS. R. S. J. 1998. Experimental phase equilibria: constraints on pre-eruptive storage conditions of the Soufriere Hills magma. Geophysical Research Letters, 25. 3437-3440. BARMIN. A. A. & MELNIK. O. E. 1993. Features of eruption dynamics of high viscosity gas-saturated magmas. Fluid Dynamics, 28. 195-202. BONADONNA, C., MAYBERRY. G. C.. CALDER, E. S. ET AL. 2002. Tephra fallout in the eruption of Soufriere Hills Volcano, Montserrat. In. DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London, Memoirs, 21,483-516. CALDER, E. S., LUCKETT, R., SPARKS. R. S. J.. & VOIGHT, B. 2002. Mechanisms of lava dome instability and generation of rockfalls and pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London. Memoirs. 21, 173-190. CLARKE, A. B., NERI, A., VOIGHT. B.. MACEDONIO, G. & DRUITT, T. H. 2002. Computational modelling of the transient dynamics of August 1997 Vulcanian explosions at Soufriere Hills Volcano. Montserrat: influence of initial conduit conditions on near-vent pyroclastic dispersal. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs. 21. 319-348.

MODELLING OF EXPLOSIVE ERUPTIONS DAVIDOV, Y. M. & BELOTZERKOVSKII, O. M. 1980. Method of Large Particles in Gas Dynamics. Mir, Moscow. DEVINE, J. D., RUTHERFORD, M. J. & GARDNER, J. C. 1998. Petrologic determination of ascent rates for the 1995-1997 Soufriere Hills volcano andesite magma. Geophysical Research Letters, 25, 3673-3676. DOBRAN, F. 1992. Non-equilibrium flow in volcanic conduits and application to the eruption of Mt. St. Helens on May 18 1980 and Vesuvius in AD79. Journal of Volccanology and Geothermal Research, 49, 285-311. DONAT, R. & MARQUINA, A. 1996. Capturing shock reflections: an improved flux formula. Journal of Computational Physics, 125, 42-58. DRUITT, T. H., YOUNG, S. R., BAPTIE, B. ET AL. 2002. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306. MELNIK, O. E. 2000. Dynamics of two-phase conduit flow of high-viscosity gas-saturated magma: Large variations of Sustained Explosive eruption intensity. Bulletin of Volcanology, 62, 153-170. MELNIK, O. E. & SPARKS, R. S. J. 1999. Non-linear dynamics of lava dome extrusion. Nature, 402, 37-41. MELNIK, O. E. & SPARKS, R. S. J. 2002. Dynamics of magma ascent and lava extrusion at the Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 153-171. MURPHY, M. D., SPARKS, R. S. J., BARCLAY, J., CARROLL, M. R. & BREWER, T. S. 2000. Remobilization of andesite magma by intrusion of mafic magma at the Soufriere Hills volcano, Montserrat, West Indies Journal of Petrology, 41, 21-42. NAVON, O. & LYAKHOVSKI, V. 1998. Vesiculation processes in silicic magmas. In: GILBERT, J. S. & SPARKS, R. S. J. (eds) Physics of Explosive Eruptions. The Geological Society, London, Special Publications, 145, 27-50. NORTON, G. E., WATTS, R. B., VOIGHT, B. ET AL. 2002. Pyroclastic flow and explosive activity at Soufriere Hills Volcano, Montserrat, during a period of virtually no magma extrusion (March 1998 to November 1999). In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 467-481.

317

PAPALE, P. 1999. Strain-induced magma fragmentation in explosive eruptions. Nature, 397, 425-428. PROUSSEVITCH, A. A., SAHAGIAN, D. L. & ANDERSON, A. T. 1993. Dynamics of diffusive bubble growth in magmas: isothermal case. Journal of Geophysical Research, 98, 22283-22308. RAMOS, I. J. 1995. One-dimensional, time-dependent, homogeneous, 2-phase flow in volcanic conduits. International Journal for Numerical Methods in Fluids, 21, 253-278. ROBERTSON, R. E. A., COLE, P., SPARKS, R. S. J. ET AL. 1998. The explosive eruption of Soufriere Hills Volcano, Montserrat, West Indies, September 17, 996. Geophysical Research Letters, 25, 3429-3432. SCRIVEN, L. E. 1959. On the dynamics of phase growth. Chemical Engineering Science, 10, 1-13. SPARKS, R. S. J. 1978. The dynamics of bubble formation and growth in magmas - a review and analysis. Journal of Volcanology and Geothermal Research, 3, 1-37. STOLPER, E. 1982. Water in silicate glasses: an infrared spectroscopic study. Contributions to Mineralogy and Petrology, 81, 1-17. TANNEHILL, J. C., ANDERSON, D. A. & PLETCHER, R. H. 1997. Computational Fluid Mechanics and Heat Transfer. Taylor & Francis, London. VOIGHT, B., SPARKS, R. S. J., MILLER, A. D. ET AL. 1999. Magma flow instability and cyclic activity at Soufriere Hills Volcano, Montserrat. B.W.I. Science, 283, 1138-1142. WATTS, R. B., HERD, R. A., SPARKS, R. S. J. & YOUNG, S. R. 2002. Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 115-152. WILSON, L., SPARKS, R. S. J. & WALKER, G. P. L. 1980. Explosive volcanic eruptions - IV. The control of magma properties and conduit geometry on eruption column behaviour. Geophysical Journal of the Royal Astronomy Society, 63, 117-148. WOODS, A. W. & KOYAGUCHI, T. 1994. Transitions between explosive and effusive eruption of silicic magmas. Nature, 370, 641-645. YOUNG, S. R., SPARKS, R. S. J., ROBERTSON, R., LYNCH, L., MILLER, A. D., SHEPHERD, J. & ASPINALL, W. A. 1998. Overview of the Soufriere Hills Volcano and the eruption. Geophysical Research Letters, 25, 3389-3392.

Computational modelling of the transient dynamics of the August 1997 Vulcanian explosions at Soufriere Hills Volcano, Montserrat: influence of initial conduit conditions on near-vent pyroclastic dispersal A. B. CLARKE 1 , A. NERI 2 , B. VOIGHT1, G. MACEDONIO3 & T. H. DRUITT4 1 Department of Geosciences, Penn State University, University Park, PA 16802, USA (e-mail: [email protected]) 2 CNR-CSGSDA, Department of Earth Sciences, Pisa, Italy 3 Osservatorio Vesuviano, Napoli, Italy 4 Laboratoire Magmas et Volcans, Universite Blaise Pascal et CNRS, Clermont-Ferrand 63038, France

Abstract: This paper presents numerical models of the Vulcanian explosions that occurred in 1997 at Soufriere Hills Volcano. Plume evolution and velocities were calculated for the well-documented and typical explosions of 6 and 7 August 1997, and these data and other observations were compared to transient, axisymmetric, multiphase flow simulations of coupled conduit evacuation and pyroclastic dispersal. Pre-explosion conduit conditions were estimated from Montserrat data, using a simple gas solubility law and assuming that conduit magma flow had stagnated with a constant overpressure prior to the explosions. Reference simulation input parameters include conduit diameter of 30m, crater diameter of 300m, meltwater content of , grain sizes of 30, 2000 and 5000um, and conduit overpressure of l0MPa. The numerical simulations of the explosions resolved highly unsteady vent exit conditions such as velocity, pressure and mass flux, and the spatial and temporal dispersal of pyroclasts during the initial few minutes was investigated using one gas phase and two or three solid phases representing pyroclasts of different size. Our simulations produced transitional eruptive regime behaviour, dividing the erupted mass into a portion that generated a radial pyroclastic current fed by a collapsing column, and a convective portion that generated a buoyant plume. This behaviour generally mimicked the observed explosions. The movement of different particle sizes was tracked, with fine particles dominantly influencing the convective behaviour of the central plume and ash plume thermals generated above the pyroclastic currents. Simulated initial vent velocities ranged from 85 to 120 m s - 1 , collapse heights ranged from 450 to 1370m above the vent, initial pyroclastic current velocities ranged from 40 to 6 0 m s - 1 with surge runouts to 1.8km, drawdown depths in the conduit were a few hundred metres, and simulated pyroclastic current deposit temperatures ranged between 135 and 430°C. Subsets of these results are in reasonable agreement with observed and measured parameters of the 1997 explosions. The best match was intermediate between our reference simulation, which assumed no loss of volatiles from the conduit during rise from the magma reservoir and which appeared too energetic, and another simulation in which much volatile leakage was assumed. The results suggest that volatile depletion in the conduit was an important factor in influencing the dynamic behaviour of the Vulcanian explosions on Montserrat.

A total of 88 short-duration Vulcanian explosions, nearly all accompanied by radial fountain collapse, occurred at Soufriere Hills Volcano, Montserrat in 1997 (see Fig. 1). Thirteen occurred in August and 75 more occurred during September and October. These explosions provided an unprecedented opportunity for repeated observation and monitoring (Fig. 2; Druitt et al 2002), as only a few events of this type have been observed closely (Nairn & Self 1978; Sparks & Wilson 1982; Hoblitt 1986). Because of the intensive real-time monitoring of precursory seismicity and tilt, it was possible over the short term to forecast (on the timescale of several hours with an uncertainty of tens of minutes) the onset of many of these explosions (Voight et al 1998, 1999; Druitt et al 2002), thus facilitating scientific preparations for impending explosions. Therefore, many explosions were documented in unusual detail by video and repetitive still photography, by theodolite surveying of eruption plumes, by tephra sampling, and by monitoring of broadband seismicity and deformation. Numerical models of short-duration Vulcanian explosions with time-varying vent flux have not received much attention, and no eruption model to date has tried to combine highly unsteady vent dynamics with explosive dispersal of pyroclastics. Recent improvements in numerical code development provided new opportunities for our study. Therefore, we have used an axisymmetric, multipleparticle-size, numerical code (Neri 1998; Neri et al. 2001b) to calculate the unsteady and transient vent flux following vent cap failure and resulting pyroclast dispersal (explosion) for a number of assumed pre-explosion conduit profiles. The observational data fall into two categories: (1) those which constrained initial conduit conditions and input parameters for our numerical models of explosive pyroclast dispersal; and (2) those which describe the explosions and provide a basis for comparison with the results of the numerical models.

In this paper, we first present a general description of the 1997 Montserrat explosions, and follow this by a brief review of eruption modelling, focusing on what is known about fountain-collapse pyroclastic current generation and Vulcanian explosions. Next, we summarize the pyroclast dispersal model we used (Neri 1998; Neri et al. 2001b) and our methods for determining the pre-explosion conduit conditions. The solutions focus on the near-vent region (within a few kilometres) over the first 90 to 150 s of the explosions. Then we use several numerical simulations to investigate the influence of key input parameters on results, namely conduit overpressure, mass fraction of water, and fines fraction. Finally, we compare our results against observations and inferred parameters of the 1997 explosions at Soufriere Hills Volcano in order to test the pyroclast dispersal (explosion) model and its assumptions. Characteristics of the cyclic Vulcanian explosions An overview of the 1997 Vulcanian explosions and their physical parameters is given in Druitt et al. (2002). The main features relevant to our numerical simulations are summarized here. The Vulcanian explosions of 1997 occurred in two episodes, the first between 4 and 12 August, and the second between 22 September and 21 October. The explosions began with an initial phase of ballistic throwout followed by a phase of powerful fountaining of a pyroclastic mixture, which lasted 10-20s. Condensation of atmospheric moisture caused by shock waves was noted on video footage of some explosions. The condensation patterns propagated radially away from the vent ahead of the expanding fountain of gas and pyroclasts during the first few seconds of the explosions. The vertical jets collapsed to generate fountains and high-velocity radial pyroclastic surges that moved to distances of about 2 km (Fig. 3),

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 319-348. 0435-4052/02/$15 © The Geological Society of London 2002.

319

320

A. B. C L A R K E ET AL.

Fig. 1. Map of southern Montserrat and Soufriere Hills Volcano, showing the active lava dome (vent) location inside English's Crater, the principal drainages (ghauts) about the dome, and the area affected by pyroclastic surges and flows during 1995-1999 (after Druitt et al. 2002). Photograph and video observation points include Montserrat Volcano Observatory (MVO South) and Fleming.

and also high-concentration pumice-and-ash flows that were generally confined to channels and ran out to distances of 3 to 6 km. Subsequently, buoyant-convecting plumes developed from the fountains and rose to heights of 3 to 15 km, where they generally spread out as umbrella clouds. The explosions were dominated by the buoyant plumes for roughly 500 s. Buoyant ash plume thermals also formed above the surges and channelled flows and tended to rise and move inward to join the main plume. The explosions ended with approximately an hour of waning exhalations characterized by a bent-over plume. Collapse of the fountain generally occurred 10 to 20s after explosion initiation, from heights of 300-650 m above the crater rim (400-750 m above the vent). The individual explosions expelled on average 3.0 x 10 5 m 3 of magma 1.1 x 10 5 m 3 as fallout and 1.9 x 105 m3 as surges and pumice-and-ash flows, evacuating the conduit to depths of 500 to 2000 m. Vent exit velocities ranged from 40 to 140ms - 1 . The pyroclastic surges developed slope-parallel velocities of 30-60 m s - 1 , whereas the channelled pumice-and-ash flows typically had velocities of 10 m s-1. The emplacement temperatures of pumice-and-ash flows ranged from 180 to 2200C (Druitt et al. 2002; Cole et al. 2002), with air entrainment cooling the mixture with respect to its eruption temperature of about 8600C (Barclay et al. 1998). The explosions occurred from a conduit 30m ( 5m) in diameter (Voight et al. 1999) into a flared crater, approxi-

mately 300m 0 m) in diameter. Our paper focuses mainly on two well-documented events at 14:35 local time (LT) on 6 August and at 12:05 LT on 7 August 1997 (all times given are local times), with simulations attempting to reproduce the near-vent behaviour over the first 90 to 150s. Background on explosion modelling and overview of our model Fountain collapse occurs when an ejected pyroclast-gas jet does not have enough momentum to continue rising, and, having failed to become positively buoyant, falls back toward and across the ground surface as a dispersion of hot particles (fragmented magma and lithic clasts) and gas (Sparks & Wilson 1976; Sparks et al. 1978; Neri & Macedonio 1996a). Early understanding of plume behaviour was derived from numerical work on turbulent gravitational convection (Morton et al. 1956). Experimental and steady-state, pseudogas models Sparks et al. (1978) described the physics of column collapse and pyroclastic flows by applying steady-state conservation of mass

MODELLING DYNAMICS OF VULCANIAN EXPLOSIONS

321

Fig. 2. Photographs of the explosion at 14:35 LT on 6 August 1997. Times after the low-frequency onset of the explosion signal at 14:35:12 LT: (a) 22s (b) 37s (c) 55 s (d) 91 s (e) 96 s. In (a) the column was about 800m high and collapse had begun, but pyroclastic currents had not clearly penetrated the veil of fallout. Gages Mountain (Fig. 1) is in foreground below plume. In (b), pyroclastic currents were descending Mosquito Ghaut (left) and Gages valley (right), and other major drainages around the volcano. In (c), the pyroclastic currents had thickened by development of dilute ash plumes, and the bulbous central plume, depleted in coarse particles, was rising buoyantly. By (d), active tan-coloured ash-rich thermals that developed over the pyroclastic current, and rose upward and inward, were being incorporated into the stalk of the buoyant central plume. A full view of the central plume is shown in (e), which rose ultimately to about 12km above sea level, where it spread to form an umbrella cap that was subsequently sheared from the main column and transported NE by high-altitude winds. Photographs taken from 7 km NW at MVO South by B. Voight.

and momentum to a gas-particle mixture in which the particles and gas are in thermal and kinetic equilibrium. Bursik & Woods (1996) added the conservation of energy expression to the description of the ash flows but did not develop the fountain model further. In all these studies the gas-particle mixture, or pseudogas, is treated as a single-phase fluid with bulk properties determined by the volumetric proportion, size and temperature of the particles. The main parameters controlling fountain formation were found to be the vent radius, gas content and initial vent velocity (or a combination of exsolved gas content and vent velocity, i.e. vent mass flux), whereas flow runout is also controlled by air entrainment and particle sedimentation (Bursik & Woods 1996). Reasonable agreement has been claimed between theory and observations (Wilson et al 1978; Turner 1979; Sparks & Wilson 1982; Sparks 1986); however, as noted by the original authors, these relationships are valid only within the context of one-dimensional steady-state pseudogas theory, where pyroclasts are very fine-

grained, and they were generally developed for steady Plinian-type eruptions. They have been used to estimate first-order time-averaged column behaviour, but the required simplifications impede detailed comparison with the transient, unsteady, multidimensional, and multiphase nature of real eruptions. Shock-tube and gas-particle experiments also suggest that the pseudogas, steady-state, onedimensional approximation may not be a good assumption for highvelocity two-phase volcanic flows (Anilkumar et al. 1993; Sparks et al. 1997; Neri & Gidaspow 2000).

Steady discharge, multiphase models Computer codes that were originally designed for atomic explosion simulation or fluidization studies have been adapted and further developed for the study of volcanic eruptions (Harlow & Amsden

322

A. B. C L A R K E ET AL.

Fig. 2. (continued)

1971, 1975; Amsden & Harlow 1974; Valentine & Wohletz 1989; Valentine et al. 1992; Dobran et al 1993; Gidaspow 1994). These codes enabled solutions to time-dependent, two-phase (one solid, one gas) compressible Navier-Stokes equations. They addressed thermal and kinetic disequilibrium between the solid and gas phases and time-dependent behaviour of the plume and the flows, while holding vent conditions steady. In addition to the previous findings of Sparks et al (1978) and Woods (1988) these results showed that plume behaviour is sensitive to the ratio of conduit pressure to atmospheric pressure at the vent (Valentine & Wohletz 1989). The codes also led to recognition of important relationships between plume behaviour and simulation particle size. Under otherwise identical vent conditions, simulations with larger particle sizes (larger Rouse numbers) are more likely than those with smaller particle sizes to fall back and form pyroclastic currents (Valentine & Wohletz 1989). This is caused by two phenomena, namely slower heating of the surrounding gas by the larger particles and their higher settling velocity.

Neri & Macedonio (1996b) made a first attempt to account for the grain-size distribution of the eruptive mixture by adding a second particle phase to the numerical solution of Dobran et al. (1993). Different drag terms between gas and particles were included for different particle sizes, and a drag term was added to account for collisions between particles of different sizes. Their model results indicate that the particles of different sizes have considerably different dynamics and affect one another's behaviour, thus changing the resulting plume behaviour and runout dynamics of the pyroclastic currents. Although the aforementioned numerical models provide insight into many of the factors that control eruption column behaviour, like the pseudogas models described above, they specifically address Plinian-style eruptions where the vent conditions are quasi-steady. None directly treats the sudden decompression of a pressurized conduit or chamber, or resulting short-pulse explosions, and therefore they do not directly apply to the Vulcanian explosions witnessed at Soufriere Hills Volcano in 1997.

MODELLING DYNAMICS OF VULCANIAN EXPLOSIONS

Fig. 2. (continued)3

323

324

A. B. CLARKE ET AL.

Fig. 3. Schematic representation of selected features of Vulcanian explosions.

Vulcanian models Several studies have explored the physics of short-duration Vulcanian explosions, with important contributions by Self et al. (1979), Kieffer (1981), Turcotte et al. (1990), Fagents & Wilson (1993) and Woods (1995). Self et al. (1979) treated the mixture as a pseudogas, and compared results to Vulcanian explosions observed at Ngauruhoe, New Zealand. Kieffer (1981) developed equations to describe the sudden decompression of a high-pressure magma chamber due to the failure of the north flank of Mount St Helens on 18 May 1980. Fagents & Wilson (1993) examined ranges of pyroclasts ejected. Turcotte et al. (1990) developed a set of equations based on the one-dimensional shock-tube problem, using unsteady conservation laws for mass and momentum and treating the mixture as a pseudogas. Woods (1995) extended the Turcotte et al. model by allowing the mixture to cool adiabatically, and by including a factor to account for particles not in thermal equilibrium with the gas. In addition, Wohletz et al. (1984) used the KACHINA numerical code (Amsden & Harlow 1974) to simulate a caldera-forming eruption through a sudden decompression of a non-homogeneous magma chamber. The code treats one solid phase and one gas phase, and focuses on the initial unsteady blast phase, and on shock wave phenomena.

Our model The previous Vulcanian models addressed transient vent conditions, and related the initial conduit pressure to exit conditions. In this paper, we advance the work of our predecessors by (1) applying a two-dimensional model with multiple particle sizes, which solves for unsteady conduit flow and resulting pyroclast dispersal, (2) constraining pre-eruptive conduit conditions with analyses of observational data, (3) comparing explosion model simulations with well-documented explosions, and (4) varying some key conduit parameters in order to better understand their effects on explosion simulation results.

The pyroclast dispersal model, named PDAC2D (Neri 1998; Neri et al. 200la,b), is an extension of the three-phase flow code used by Neri & Macedonio (I996b) and solves a set of equations expressing the conservation of mass, momentum and energy for one gas phase, and a number of solid phases representative of particles of different sizes. The gas phase consists of a mixture of water vapour and atmospheric air. The fundamental transport equations are solved on an axisymmetric computational grid (5m to 40m grid spacing), as discussed in detail in the Appendix.

Input conditions for explosion model For our simulations of short-duration Vulcanian explosions, several input parameters were required. The most important of these were the conduit, crater and plugging cap geometry, the twodimensional (axisymmetric) topography of the region surrounding the vent, the initial conduit gas pressure, gas mass fraction and gas volume fraction as functions of depth in the conduit, and the sizes and densities of solid particles. Information acquired on Montserrat was used to constrain these input parameters, as described in the following sections.

Conduit and crater diameters, cap thickness and ground topography The conduit diameter of approximately 30m ) was constrained by spine dimensions, magma ascent rates and volume extrusion rates (Voight et al. 1999). The crater diameter of roughly 300m 0 m) was measured using dual-position theodolite measurements approximately one half hour prior to the 12:05 event on 7 August. The crater depth was approximately 100 m. The cap thickness was assumed to be 20m, which is a size roughly consistent with the estimated total volume of dense clasts per explosion (roughly 1-5% by volume of pumice-and-ash flow deposits).

Fig. 4. Topography used in the simulations. The grid is defined for an axisymmetric solution, thus the crater and conduit dimensions are radii.

MODELLING DYNAMICS OF VULCANIAN EXPLOSIONS

We used the ground topography representative of the north side of the volcano (Fig. 4). The north sector is characterized by a relatively smooth, low-slope fan with average slope approximately 22°, bordered by channels which, by mid-summer 1997, were nearly filled by pyroclastic debris. The model sloping-fan topography thus was more or less realistic for a broad sector of the near-vent region. The model assumption of axisymmetric flow is a reasonable approximation to reality, because the Vulcanian events in August of 1997 are described as axisymmetric (Druitt et al 1998, 2002), with fountain collapses and radial pyroclastic surges occurring simultaneously in all sectors and with pyroclastic current runouts similar in all directions.

Pressure and gas fraction profiles along the pre-explosion conduit For this study, the pre-explosion conduit was modelled quite simply, mostly on the basis of Sparks (1997). We assumed that magma flow in the conduit immediately prior to the explosions had stagnated due to viscous plugging at the vent. Specifically we assumed constant overpressure with depth, since for magma systems with large vertical viscosity gradients, almost all excess pressure drop from the chamber to the surface is concentrated near the top of the conduit (due to the higher viscosity of the magma near the surface), with excess pressure being roughly constant below the upper reaches of the conduit. There is some basis for assuming stagnated flow, because the Vulcanian explosions occurred in association with oscillatory flow (Voight et al. 1999), and modelled conduit flow rates were relatively small just prior to explosions (Wylie et al. 1999; Denlinger & Hoblitt 1999). The actual flow conditions and overpressure distributions were undoubtedly more complicated (Melnik & Sparks 2002a), but our assumptions were practical for a first approximation. We estimated gas mass and gas volume fractions as functions of depth in the conduit using the following solubility and hydrostatic laws. Symbols are summarized in Table 1. The mass fraction, ne, of exsolved water vapour in the bulk magma (melt + crystals + vesicles) at a given conduit depth was taken according to Henry's law as: (1)

where n0 is the mass fraction of water in the melt, z is depth in the conduit, Pg(z) is the total gas pressure in the conduit at depth z, s and (3 are experimentally determined constants for the appropriate melt composition, and 0 is the total volume fraction of crystals in the upper conduit magma (excluding vesicles). The water dissolved in the melt in the chamber, n0 was taken as 4.3 0.5 wt% (Devine et al. 1998; Barclay et al. 1998) and the total crystal volume fraction, was taken to a constant 0.65 (Murphy et al. 2000). The chemistry of the melt phase is rhyolitic (Murphy et al. 1998), for which s = 4.1 x l0 -6 N1/2m -1 and 0 = 0.5 (Wilson et al. 1980). The pressure distribution, Pg(z), is given by:

Jo

(2)

where AP is the overpressure at a given depth, which we assume to be constant with depth for each of the simulations presented in this paper, p is the bulk density of the overlying gas-magma mixture, and g is acceleration due to gravity. The water vapour is treated as an ideal gas with density, pg(z), as follows:

325

Table 1. Symbols and variables Symbol Description Ah aatm aj aw CD..k Cpgik Cs dk,j Dgk Dkj e Ev g G(Eg) hc hg Kk M Ms ne n0 Nuk Patm Pg Pg(z) P2 Pr Qk R R Rek s t Tg,k VDREA VDREE Vmax Vj vg vkj z a (3 P £g

Cross-sectional area of ballistic block Acoustic velocity in atmospheric air Acceleration of ballistic block in j direction Acoustic velocity in water vapour Drag coefficient for kth solid particle size Specific heats of gas and kth solid particle size Smagorinsky's constant (assumed to be 0.1) Particle diameter of kth and jth solid particle sizes Gas-solid drag coefficient Drag coefficient between kth andy'th solid particle sizes Restitution coefficient for a particle-particle collision Specific expansion energy of conduit water vapour Acceleration due to gravity Solid elastic modulus Fountain collapse height (above the vent) Gas enthalpy Thermal conductivity of kth solid particle size Mass of ballistic block Mach number of shock wave Mass fraction of exsolved water vapour in the bulk magma Mass fraction of water in the melt Nusselt number for kth particle size Atmospheric pressure Total gas pressure Total gas pressure in the conduit (at depth z) Pressure on high-pressure side of shock wave (see Fig. 5) Prandtl number Heat transfer coefficient between the gas and the kth particle size Ideal gas constant for water vapour Radius of conduit Reynolds number of kth particle size Solubility constant for Henry's law Time from onset of explosion Viscous stress tensor for the gas and kth solid particle size DRE volume available in the conduit DRE volume ejected from the conduit Maximum vent velocity for a given simulation Velocity of ballistic in j direction Velocity of water vapour Velocity of kth and j3th solid particle sizes Depth in conduit Restitution coefficient for non-head-on collisions Exponent for Henry's law Gas overpressure Gas volume fraction E Volume fraction of kth solid particle size k Ekj Maximum solids volume fraction for particles of size j and k ES Solid volume fraction in conduit Specific heat ratio for atmospheric air atm m Specific heat ratio for water vapour-magma mixture w Specific heat ratio for water vapour Gas molecular viscosity g Effective gas viscosity ge Effective gas turbulent viscosity gt k Viscosity of kth solid particle size Particle volume fraction of kth or jth size at maximum packing k.j p Bulk density of gas-magma mixture pa Atmospheric air density Pg Gas density Pkj Density of kth and jth solid particle sizes ps Density of melt-crystal mixture (solid) in the pre-explosion conduit Gas deformation tensor g Coulombic repulsive component among solid particles Solid stress tensor for kth particle size

(3)

where R = 4621kg -1 K - 1 i s the universal gas constant divided by the molecular weight of water, and T is the temperature of the water vapour, which we assume to be isothermal with the meltcrystal mixture (from here on referred to as 'solid') at 1133K

(860°C). If the density of this solid phase, ps, is constant with depth, then the bulk density of the gas-solid mixture is: (4)

326

The solid volume fraction

A. B. CLARKE ET AL. s

at a given depth is: (5)

And the gas volume fraction is simply: (6)

At the depth where the solid fraction s — 0.70 (30% vesicularity), which represents a bulk density slightly below 2000 kg m - 3 , a solid boundary is assumed in our pyroclastic dispersal model. Although this assumption is arbitrary, it is necessary to specify a solid boundary at the base of the conduit in order to effectively simulate the dispersal of the particles. This does not appear to be an unreasonable assumption because our pumice density measurements show that only 13% of randomly sampled clasts have vesicularities less than 30%. Because magma viscosity was very high, bubble expansion after fragmentation was probably minimal (Thomas et al. 1994; Druitt et al, 2002). Therefore, we assume that the vesicularity of pumice represents essentially the pre-explosion state of magma in the conduit, allowing us to claim that roughly 87% of material ejected had pre-explosion solid volume fraction, s < 0.70 (>30% vesicularity). The depth at which s = 0.70 (30% vesicularity) is referred to as effective conduit depth, DE. The dense rock equivalent (DRE) volume of solid material above this boundary is called DRE volume available, VDREA. The DRE volume of solid material ejected permanently from the conduit, that is, that which does not fall back into the conduit during the simulation, is termed DRE volume ejected, VDREE .

Pre-explosion specific expansion energy It should be noted that changing a single conduit parameter (e.g. overpressure or bulk H2O mass fraction), changes several characteristics of the modelled pre-explosion conduit, including VDREA and the effective depth of the simulation conduit. Because of this, it is difficult to attribute differences in simulated explosion behaviour to a single changed conventional conduit parameter. Therefore, to assist interpretation, we combined total pressure, exsolved gas mass fraction, DRE volume available, and conduit depth into a single parameter. This parameter represents the expansion energy due to gas overpressure per unit mass of solid material and is called specific expansion energy, Ev, which provides a convenient way to compare simulation results. Due to the very fast decompression of the system, we assumed adiabatic conditions (Woods 1995) and used the following equation to calculate this energy (Self et al. 1979; Neri et al. 1998):

(7)

where is the specific heat ratio for water vapour (the specific heat at constant pressure divided by the specific heat at constant volume: . = 1.25), Patm is atmospheric pressure, Pg(z) is the total gas pressure, DE is the effective depth of the conduit, r is the radius of the conduit, and is the mass of the material for a segment of the conduit of depth dz.

Ballistic analysis Numerous ballistics were observed during the initial seconds of the Vulcanian explosions of early August 1997. These ballistics were not affected by the motion of the plume because, in general, they were very large and dense, with fall velocities close to the initial explosion gas velocity, making them Type I ballistics according to Self et al. (1980). Documentation of the resulting crater size and position, and clast diameters, estimated from a helicopter, resulted

in a set of 24 ballistic range and size data (Druitt et al. 1998, 2002). We have used these data, along with basic dynamics equations and drag relationships, to estimate the initial velocities of the ballistics (Wilson 1972; Fagents & Wilson 1993; Waitt et al. 1995). The basic equation adopted for the ballistic analysis is dv j /dt = aj where aj is the acceleration (or deceleration) of the ballistic due to components of gravity and drag in the j direction. This acceleration is approximated by the methods of Wilson (1972) and Waitt et al. (1995) as follows: (8) and (9)

where vx and vv are the ballistic velocities in the horizontal and vertical directions respectively, pa is the density of the surrounding atmosphere, v is the total velocity of the ballistic, Ab is the ballistic cross-sectional area, CD is the drag coefficient for the ballistic, M is the ballistic mass, and g is acceleration due to gravity. Calculations for M and Ab assume a spherical clast and dome rock density of 2250 kg m-3 (Sparks et al. 1998). The coefficient of drag, CD, for smooth spheres was assumed. CD varies from 0.07 to 0.20 and is a function of Reynolds number, Re, such that CD = 0.11 log(Re) - 0.55 over a range 4 x 105 < Re < 6 x 106, estimated from Achenbach (1972) and Waitt et al. (1995). The drag coefficient is poorly constrained because the shapes of the ballistics vary greatly. Others, such as Wilson (1972), Fagents & Wilson (1993), Fudali & Melson (1972) and Self et al. (1979), have assumed that irregularly shaped ballistics have CD ranging from 0.7 to >1.0. However, in a more recent analysis, Waitt et al. (1995) suggest that these high values of drag coefficient result in unrealistically high launch velocities, particularly for smaller ballistics. As theoretical justification, they offer the evidence that a dimpled golf ball achieves a greater range than a smooth one, illustrating that the rough surface of ballistics can contribute to reduced drag. In general, although roughness increases skin-friction drag, it delays flow separation and reduces form drag, which is the largest component of total drag. Equations 8 and 9 were solved for initial velocities (vx.i and vy.i) by a fourth-order Runge-Kutta numerical scheme as done by Wilson (1972) and Waitt et al. (1995). We used an average ground slope of 22°. Drag was neglected in the first 1 0 - 2 s because the initial velocity and therefore the initial drag were not known. There was no significant difference in calculated initial velocity when drag was neglected during the first 1 0 - 2 s compared to when drag was neglected during the first 10 - 4 s, indicating that neglecting drag for the first 1 0 - 2 s did not significantly affect results. Total initial ballistic velocity is referred to as Vi. Launch angle is a significant factor. The angles for any specific ballistic clasts and associated craters are unknown. Therefore we executed the method above using angles at 5C intervals, from 30° to 70°. Values of initial velocity ranged from 93 to 1 1 6 m s - 1 for a 30C launch angle and from 131 to 1 6 9 m s - 1 for a 70 0 launch angle. The optimum launch angle was between 30 and 350.Druitt et al. (2002) carried out similar calculations using the ballistic model of Self et al. (1980), which adopts higher drag coefficients than those used here. For an optimum launch angle of about 350, they estimated launch velocities of up to 1 6 0 m s - 1 . Analysis of video footage gives exit velocities of between 40 and 1 4 0 m s - 1 (our work, and Druitt et al. 2002), suggesting that the Self et al. (1980) model overestimates exit velocity and that the values calculated in this paper are probably more realistic.

Estimation of overpressure Estimation of the gas overpressure beneath the conduit cap has been approached in several ways. Robertson et al. (1998) suggested explosion pressures of 10-27MPa for the sustained explosion of

MODELLING DYNAMICS OF VULCANIAN EXPLOSIONS

327

17 September 1996. Voight et al (1999) suggest that tilt data could be used to calculate pre-explosion overpressures, but simple halfspace elastic models resulted in only upper-bound estimates of several tens of megapascals. In the same paper, an extrusion model led to an estimate of overpressure approximately 11 to 25 MPa. The conduit flow models of Melnik & Sparks (2002a) predict overpressures up to 10 MPa. A lower-bound estimate of overpressure is the rock tensile strength, which is on the order of 4 MPa (Voight et al 1999). Druitt et al. (2002) use pumice vesicularities (55-75%) to place total confining pressures between 5 and 15 MPa. The highest of these vesicularities probably represent the upper parts of the conduit immediately beneath the cap. Our assumed 20m thick cap represents less than 0.5 MPa of overburden; therefore the calculated overpressure falls between 4.5 and 14.5 MPa, which is in reasonable agreement with the aforementioned values determined by independent methods. We have therefore chosen the middle of the range of estimates, roughly 10 MPa, as a reasonable estimate of overpressure for our reference simulation. In one of our variations, we used an overpressure of 7 MPa to illustrate the effect of reduced overpressure on results.

Fragmentation description Models of magma flow unsteadiness developed by Melnik & Sparks (2002b) test the significance of different fragmentation assumptions, such as fragmentation at a fixed volume concentration of bubbles (Sparks 1978; Wilson et al 1980), at a bubble overpressure threshold in excess of the magma tensile strength (Melnik 1999), or at a critical elongation strain rate (Papale 1999). The field evidence at Montserrat includes angular platy pumice that lacks bubble elongation texture (Druitt et al 2002). These observations suggest brittle fragmentation produced by a fragmentation wave (Alidibirov & Dingwell 1996) and support the overpressure threshold fragmentation criterion of Melnik (1999). At Soufriere Hills Volcano the sudden decompression of the conduit, due to fracture and disruption of the conduit cap, occurred when the strength of the cap rock was exceeded by the overpressure in the conduit. This disruption of the cap greatly increased the pressure difference between the vesicles near the top of the conduit and the surrounding environment, which was suddenly reduced to atmospheric pressure (Fig. 5). This new pressure state exceeded the fragmentation threshold, thus initiating the fragmentation wave. The following paragraphs describe how our model represents this condition and enumerate the associated assumptions. In our numerical model, the pre-explosion conduit at time t=0 was represented by a two-phase mixture whose properties (pressure, gas mass fraction and gas volume fraction) were described as a function of depth by the equations presented above. The 30m diameter conduit was discretized in a number of cells 10 m deep and 7.5m wide and no radial variation in the initial conditions was considered. At t = t1 > 0, the pressure disequilibrium between the overpressured conduit and the confining pressure produced a decompression wave which travelled down the conduit. As the decompression wave reached a particular depth, the mixture of gas and particles at that depth began to flow out of the conduit, whereas the undisturbed portion of the mixture below the decompression wave remained in magmastatic equilibrium. Such a representation of the conduit embodies three key assumptions. We first assumed that the melt phase at any depth quenched and fragmented nearly instantaneously when the decompression wave reached it, allowing the mixture of gas, crystals and melt to be represented as a gas-particle flow. This is supported by the results of Melnik & Sparks (2002b) which indicate that fragmentation occurs immediately after the decompression wave has reached a given depth. The second assumption is that the fragmentation process involved no energy loss, which in any case was probably relatively small (1-5% according to Alidibirov 1994). Third, we assumed that no further water exsolution occurred once

Fig. 5. Schematic representation of the numerical conduit, which defines the initial conditions for explosion simulations. The conduit is subdivided into grid cells, with input parameters that include solid volume fraction, S(Z), water vapour volume fraction, g(z), diameters of particles, and total pressure of water vapour, Pg(z}. Input values are discussed in the text and summarized in Table 2. The figure on the left also represents a shock tube at time zero, representing the pre-explosion, capped conduit, and on the right at time t1, representing the condition after disruption of the caprock.

disruption of the cap and decompression began. The propagation of a fragmentation front, and acceleration of the fragmented gasparticle mixture to the surface, is too rapid for significant diffusive mass transfer to occur during Vulcanian explosions (Gardner et al. 1999; Melnik & Sparks 2002b), and thus only gas that had exsolved before the cap was disrupted participated in the explosion.

Particle sizes The sizes of solid particles chosen for our simulations (up to three sizes) have been constrained by our grain-size analyses of the pumice-and-ash flow deposits, which considered the full range of particle sizes observed in the outcrops. Our estimates of the clasts ranged from a maximum size of 50cm to a minimum size of 0.9 m. The mean grain size of the pumice-and-ash flow deposits, accounting for a 30 wt% loss of fines (<2 mm diameter) to ash plumes associated with pyroclastic density currents is c. 2000 m. The mean plus 1.5 standard deviations (SD) is about 5000 m, and the mean minus 1.5 SD is 30 m. We used 2000 m and 30 m grains to represent 70 and 30wt%, respectively, of the pressurized solid magma in the lower conduit for the reference simulation. The 2000 m particles are representative of the mean grain size of the mixture, whereas the 30 m particles are representative of the fine portion. The density of all solid phases is 2600 kg m - 3 . The cap that plugged the conduit was represented as a region of closely packed grains (0.87 volume fraction of solids) of 5000 m diameter at the top of the conduit. Obviously this is a simplification, but a 0.87 solid volume fraction is broadly consistent with the

A. B. CLARKE ET AL.

328

density of dome rock, roughly 2250-2450 kg m- 3 (B. Voight, unpublished data). The gas separating the grains in the caprock was set at magmastatic pressure. This is consistent with the hypothesis that gas in the confining caprock had largely escaped, and therefore the cap was not overpressured. The gas permeability of the caprock was assumed to be zero with respect to underlying magma; therefore, the gas pressure in the conduit below was maintained until the explosion simulation initiated with cap disruption.

Simulations performed and main results We have conducted several simulations with this model for a number of sets of pre-explosion conduit conditions, using variations of initial overpressure, bulk magma water content, fines fraction, and exsolved gas volume fraction. Our reference simulation, Simulation 1 (Siml), is characterized by an overpressure of 10 MPa and meltwater content of 4.3 wt% (no = 0.043; bulk water content = 1.5wt%), by the presence of three particle sizes (two for the conduit and one for the caprock), effective conduit depth DE= 1020m, DRE volume available VDREA — 4.2 x 10 5 m 3 , and specific expansion energy Ev 14 600 J k g - 1 . The effects of the change of a single input parameter were then evaluated. In Simulation 2 (Sim2), we decreased the overpressure to 7 MPa, resulting in 15 000 J k g - 1 . Next, in Simulation 3 (Sim3), we increased the meltwater content to 4.8 wt% (n0 = 0.048;

bulk water content = 1.7wt%), with £ v = 17 200 J k g - 1 . In Simulation 4 (Sim4), we replaced the fines fraction (30 m) with 2000 m particles in the conduit, resulting in only one conduit particle size. EY for Sim4 is the same as that of Sim 1, £v = 14 600 J k g - 1 . Simulation 5 (Sim5) represents the initial conditions of Siml after 12 hours of leakage of exsolved gas from the conduit through the surrounding country rock, resulting in EY = 6600 J k g - 1 . The model for gas flow through porous media introduced by Jaupart & Allegre (1991) was used to determine the approximate gas volume fraction for Sim5, using permeability of 2.7 x 1 0 - 1 4 m 2 . Simulation input parameters are summarized in Table 2. Conduit profiles of total gas pressure and gas volume fraction for the reference simulation and four principal variations are shown in Figure 6. Siml - the reference simulation This section is devoted to presenting the output of the reference simulation in order to illustrate the type of results that our pyroclast dispersal model can produce using input parameters constrained by field observations. Overall development and timing of the plume. The evolution of the reference simulation, Siml. is shown in Figure 7a with contoured

Table 2. Initial conduit parameters for simulations Simulation

Overpressure (MPa)

Initial melt H 2 O (mass fraction)

Cap particle size ( m)

Particle no. 2 size ( m)

Particle no. 3 size ( m

Approx. depth* (m)

Specific expansion energy ( J k g - 1 )

DRE volume available (10 5 m 3 )

Siml Sim2 Sim3 Sim4 Sim5

10 7 10 10 10

0.043 0.043 0.048 0.043 0.043

5000 5000 5000 5000 5000

2000 (70%) 2000 (70%) 2000 (70%) 2000(100%) 2000 (70%)

30(30%) 30 (30%) 30 (30%) 30 (30%)

1020 1300 1300 1020 270

14600 15000 17200 14600 6600

4.2 4.8 5.1 4.2 1.2

Total conduit volatile energy was estimated assuming adiabatic conditions. * Depth is to point where solid volume fraction £s = 0.70, below which is a solid boundary in the simulation.

Fig. 6. Conduit parameter profiles as a function of depth for the main simulations. Siml and Sim4 have the same profiles: however. Sim4 lacked fine particles, (a) Total water vapour pressure: (b) gas volume fraction.

MODELLING DYNAMICS OF VULCANIAN EXPLOSIONS

329

Fig. 7. Simulation 1. (a) Distribution of the total particulate volumetric fraction in the atmosphere at 20, 45, 60, 90, 120 and 150s. The colour contour levels shown are the exponents to the base 10 and correspond to -8, -7, -6, -5, -4, -3, -2 and — 1. The outer two contours represent the limit of visibility. (b) Instantaneous velocity vectors for 2000 m particles (red) and 30 m particles (blue) at 20, 45, 60, 90, 120 and 150s. (c) Distribution of temperatures for 2000 m particles at 20, 45, 60, 90, 120 and 150s. Temperature increases from yellow to red.

colour zones of log10 of total volumetric particle concentration. The most dilute contour shown is log10 = - 8. The lightest (yellow) two to three zones mark the threshold of particle concentration visible to the naked eye. Figure 7a shows that the explosion developed as a nearly hemispherical expanding cap. By 4 s the erupted mixture expanded beyond the crater wall, with the top of the mixture approximately 300m above the vent, and the plume top contained a small indentation. By 10s pyroclastic material reached the base of the dome, about 300 m from the conduit, and the top of the mixture reached 700m above the vent. The column displayed an ellipsoidal bulge, and contained a conical indentation at its crest. This indentation had closed by 17s with the plume 500m wide and 850m high above the vent. At 20 s the plume crested at the axis and an

overhang had developed, which became more pronounced over the next 20 s. After 40 s the evolving separation between the collapsing material and the buoyant plume became evident, and by 45s the distinct nose of a pyroclastic density current reached 600 m radially but remained under the overhang. The plume crest rose to 1000m above the vent. By 60 s the rising central part of the plume reached 1500m above the vent (2.4km above sea level (a.s.l.)), and the pyroclastic current extended about 1100m and protruded beyond the overhang. The gravity separation of particle sizes linked the enhanced buoyancy of the axial plume with the increased density of the pyroclastic current. By 90s the buoyant low-density central plume showed a dramatic rise, and convoluted buoyant coignimbrite ash plume thermals developed above the pyroclastic density currents. By 120s these convoluted thermals joined the

330

A. B. CLARKE ET AL.

Fig. 7. (continued)

rising central plume. The radial progression of the density current stagnated after 150s at approximately 1850m. Figure 7b shows velocity vectors for the 2000 m particles in red and for the 30 m particles in blue for Siml. At 2 s the particles are exploded radially at high velocity (c. 180ms - 1 ) but by 20s maximum velocities had declined to 130ms - 1 , and a clockwise rotating vortex developed in the outer plume. The vortex is shown also by the swirl of hot ash-gas mixture in the temperature plot (Fig. 7c). The circulation pattern extended to the ground for fine particles, which feed back into the recycling fountain, but many of the heavier coarse particles broke out of the vortex and moved groundward outside the crater wall, similar to behaviour expected of incipient fountain collapse. By 45 s the ascending plume was restricted to a narrow axial zone, whereas particle velocity trends on the upper flank were downslope and parallel to topography, indicating full development of a pyroclastic current with runout beyond 500 m. The current was fed by the collapsing plume, with maximum downward velocities and temperatures existing just beyond the crater rim (Fig. 7b, c). A vortex remained in the overhanging plume, influencing particularly the fines, whereas coarse particles fell inward toward the pyroclastic current source. By 60s this pattern evolved into overhanging high hot zone and a hot axial core (Fig. 7c), with most coarse particles moving inwards and downwards, but with the fines

contributing to a buoyant mixture and leading to upward acceleration of the central plume (Fig. 7b). By 90 s the hot plume was dominated by energetic buoyant rise, with velocities accelerating with column height. A rising large buoyant vortex created a bulged plume 1 km above the vent. Differences were displayed by particles of different sizes, with the fines contributing to rise of a buoyant plume above the pyroclastic current, aided by a strong winds blowing radially inward. Pyroclastic current outward movement was restricted to a thin hot zone above the ground, and had ceased by 150s. By then, the lower 3 km of the main plume had stagnated apart from a hot vent and very narrow warm axial zone, and overall the pattern was dominated by vortices forced by the hot ash thermal rising inward from the stagnated pyroclastic current. Evolution of vent conditions: mass flux, velocity and pressure. Figure 8 displays the resultant vent parameters (top of the conduit) for the five main simulations, with key results also summarized in Table 3. For Siml, the maximum vent mass flux was 7 . 5 x l 0 7 k g s - 1 (2.9 x 10 4 m 3 s - 1 DRE), which occurred at 0.4s. The vent mass flux then fell sharply and stabilized at about 3 . 7 x l 0 7 k g s - 1 (1.4 x 10 4 m 3 s - 1 DRE) for 11 s before a further decline. Much of the

MODELLING DYNAMICS OF VULCANIAN EXPLOSIONS

331

Fig. 7. (continued)

mass ejection occurred in the first 15 s although a waning discharge of about 1 x 10 7 kgs- 1 (0.4 x 10 4 m 3 s- 1 DRE) continued beyond 30s. The maximum vent mixture velocity (115ms - 1 ) occurred at 2-12s, after which the velocity declined to slightly under 100ms - 1 by 30 s. The pressure at the vent peaked at roughly 0.2 s at 4.2 MPa, and declined to about one-half that value within a few seconds, and another one-half between 12 and 30 s. Vent Mach number exceeded 1.2 within a few seconds and maintained this value for over 30s.

rents. The simulations resulted in a vertical temperature gradient with maximum temperatures occurring near the middle of the thickness of the deposit, similar to patterns measured in deposits in the field at Montserrat. For deposit temperatures roughly 100m upstream of the stagnated pyroclastic current deposit terminus, Siml produced maximum emplacement temperatures of 135°C. Siml ejected a total DRE volume of 3.9 x 10 5 m 3 (1.0 x 10 9 kg), evacuating to a depth of roughly 960m.

Pyroclastic density current behaviour andDRE volume ejected. Sim 1 produced initial slope-parallel pyroclastic current velocities about 40-60 m s - 1 near their onset, an average frontal speed of 31 m s - 1 to 60s and about 1 0 m s - 1 after 60s on the shallower outer flank. Emplacement temperatures are analogous to the simulation particle temperatures after the stagnation of the pyroclastic cur-

Comparison of various simulations to Siml The main aim of this section is to compare simulations in order to illustrate the effects of key pre-explosion conduit parameters on pyroclast dispersal. In a later section, we compare the simulations to the observed explosions.

A. B. CLARKE ET AL.

332

Fig. 8. Parameters at the vent for the first 30s of the simulations, (a) Vent mass flux (kgs - 1 ). (b) Vertical velocity at vent for the gas-particle mixture (ms - 1 ). (c) Vent pressure (MPa). (d) Vent Mach number, for Siml only.

Siml, and by the time pyroclastic density currents were initiated after 42s no overhang existed (compare Siml and Sim2 at 45s). By 60 s the rapidly rising plume stretched 1900m above the vent (2.8km a.s.l.), 400m higher than Siml, and no wide central plume existed as in Siml. After 80s ash plume thermals rising above the pyroclastic current began to develop, and became more pronounced after 100-140s but did not merge with the narrow central plume. The cumulative changes in pre-explosion conduit conditions resulted in a vent velocity similar to Siml. with a 5-10% increased velocity after 12s, lower mass flux by c. 20% over the first 14s (Fig. 8), and an increase in DRE volume ejected by approximately 5%, to 4.1 x 10 5 m 3 (1.1 x 10 9 kg). evacuating to 1140m depth. Although it seems counterintuitive, the decrease in initial conduit pressure did not much change the maximum vent velocity because it resulted in a greater proportion of exsolved water vapour

Sim2 - decreased pre-explosion conduit overpressure. Sim2 started with a 30% reduction in pre-explosion conduit overpressure as compared to Siml. This reduction led to significant changes in other pre-explosion conduit parameters. Volume fraction of exsolved gas was larger, effective conduit depth was nearly 30% longer (1300 m), and DRE volume of available material (4.8 x 10 5 m 3 ) was roughly 15% greater, and consequently, total specific expansion energy was greater by 3% (Table 2). The overall development of Sim2 was generally similar to that of Siml (Fig. 9). However, Sim2 exhibited a more rapidly rising central plume and displayed less effective integration of pyroclastic current ash plumes with the rising central plume. The mixture expanded beyond the crater wall at 3 s with the top of the mixture 260 m above the conduit, similar to Siml at the same time. At 20s the plume showed little development of a plume overhang in comparison to Table 3. Selected simulation results Simulation

Max. vent velocity ( m s - 1 )

Maximum vent mass flux (107 k g s - 1 )

Collapse height (m above the vent)

Timing of collapse (s)

Pyroclastic current runout (m)

DRE volume ejected (10 5 m 3 )

Pyroclastic current deposit temperature ( C)*

Siml Sim2 Sim3 Sim4 Sim5

115 115 120 120 85

7.5 6.2 7.5 3.3 7

970 1000 1160 1370 450

28 26 35 22 15

1850 1850 1800 850 1300

3.9 4.1 4.6 3.6 1.1

135 210 100 430 290

* At 100m from terminus.

MODELLING DYNAMICS OF VULCANIAN EXPLOSIONS

333

Fig. 9. Simulation 2. Distribution of the total paniculate volumetric fraction in the atmosphere at 20, 45, 60 and 120 s. The contours are exponents to the base 10 as shown in Figure 7a.

334

A. B. CLARKE ET AL.

(due to the conduit model we used; see Equation 1), which resulted in a slightly greater specific expansion energy, Ev. The greater proportion of exsolved water vapour also explains the reduction in vent mass flux over the first 14s and the slightly higher fountain collapse. A more rapidly rising central plume was caused by a reduced bulk plume density, which also reflected the greater volume fraction of water vapour at the vent. The slight increase in DRE volume ejected can be attributed to an increased VDREA. Sim2 produced pyroclastic flows with velocities similar to Siml, but with higher emplacement temperatures of 210°C (compared with 135°C) which reflects less efficient entrainment of air into the current. Sim3 - increased bulk water content. Increasing chamber meltwater content by 12%, as in Sim3, increased the active simulation conduit depth, DE, by 30% (to 1300 m) and increased DRE solid volume available (5.1 x 10 5 m 3 ) by over 20%, and increased specific expansion energy (17 200 J kg - 1 ) by almost 20% as compared to Siml. The general evolution of Sim3 was similar to that of Siml (Fig. 10), with a more buoyant central plume and more obvious development of ash plumes above pyroclastic currents. The early plume evolution was similar to Siml, and by 20 s the plume top rose 1000m above the vent (100m higher than Siml). An overhang developed which was protruded at the toe by a pyroclastic current initiated just before 45s. By 60s pyroclastic currents had reached 1000m and both the central core plume and the overhang plume (in contrast to Siml) experienced vigorous buoyant rise (compare Figs 7 and 10). Compared with Siml, maximum vent velocity was 10-15% greater at 120ms - 1 (Fig. 8) and total mass erupted increased by nearly 20% to 4.6 x 10 5 m 3 DRE volume (1.2 x 10 9 kg) ejected, evacuating to 1200m depth. The greater vent velocity, greater collapse height and enhanced air mixing reduced pyroclastic current peak emplacement temperatures to 100°C. Sim4 - no fines fraction. Because the effective conduit depths, volume fractions of exsolved water vapour, and overpressures were identical for Sim4 and Siml, we can attribute all behavioural differences between these two simulations to the absence of 30 m particles from Sim4. Figure 11 shows the results of Sim4. At 20 s the plume height was almost 1200m above the vent (250m higher than Siml) and 7 s later there was a hint of pyroclastic current initiation and a growing overhang. The axial plume continued strong growth, exceeding 2000m by 37s and by 45s both axial and overhang plumes had risen twice as high as Siml. At 60-90s Sim4 showed continued plume development, but the thin pyroclastic current had moved about 200m at c. 5 m s - 1 , without an associated ash plume. Sim4 resulted in a 5-10% greater vent velocity than Siml, with a maximum about 120ms - 1 at 3-12 s (Fig. 8). Sim4 also resulted in a significantly lower, non-peaked mass flux during the first 2s, with a maximum of 3.3 x 10 7 kgs - l (1.3x 10 4 m 3 s - 1 DRE) and a 7% decrease in total DRE volume erupted to 3.6 x 10 5 m 3 (0.94 x 10 9 kg) with evacuation to 900m depth (Table 3). This reduced mass flux is mainly a result of the lower drag to mass ratio of 2000 m versus 30 m particles. The 30 m particles in the other simulations were more efficiently ejected from the conduit by the expanding water vapour. The emplacement temperature of pyroclastic currents was much greater than the three other simulations at 430°C. Whereas finegrained particles (30 m) that transfer heat more efficiently make up 30% of the conduit mixtures of Siml, Sim2 and Sim3, the 2000 m particles of Sim4 allow less efficient heat transfer, increasing emplacement temperatures. Sim 5 - volatile leakage from conduit. Sim5 started with a substantial reduction in pre-explosion conduit volatiles as compared to Siml. This reduction led to significant changes in other pre-

explosion conduit parameters. Effective conduit depth was less than one-third (c. 270 m), as was DRE volume of available material (1.2x 10 5 m 3 ), and consequently, total specific expansion energy was less than half (Table 2). The explosion developed as a nearly hemispherical expanding cap. By 5 s the erupted mixture expanded beyond the crater wall, with the indented top of the mixture approximately 300m above the vent. By 15s pyroclastic material reached the base of the dome, and the top of the mixture reached 450m above the vent. Also at 15-20 s the evolving separation between the collapsing material and the buoyant plume became evident, and radially directed pyroclastic currents were initiated (Fig. 12). No overhanging plume developed as in Siml. By 45s the nose of a pyroclastic density current was well formed and reached 600 m radially and the top of the plume was 1000m above the vent. By 60s the rising and still indented plume reached 1700m above the vent (2.6km a.s.l), the pyroclastic currents extended more than 1200m and early stages of convoluted ash thermal development was evident above the pyroclastic density currents. By 90s the near-vent sections of the pyroclastic density current, at 300 to 800 m radial distance, showed an energetic ash thermal that joined the rising central plume, and forward progression of the density current stagnated at 1300m. Velocity vectors show that rapid deceleration occurred during the first 25s from maximum velocities of 1 3 3 m s - 1 to 5 2 m s - 1 (figure not shown). Small, downward particle velocities occurred in the outer plume by 5s and by 15s the greatest downward particle velocities were occurring outside the crater wall, similar to behaviour expected of fountain collapse. By 25 s particle velocity trends on the upper flank were parallel to topography and pointed downslope, indicating full development of pyroclastic currents with runout beyond 500m. Simultaneously, inward-facing velocity vectors occurred along the outside surface of the plume and pyroclastic current, indicating that the motion of the 2000 m particles was being influenced by air entrainment and by settling effects. The maximum vent mass flux was 7 x l 0 7 k g s - 1 (2.7 x 10 4 m 3 s - 1 DRE), which occurred at 0.4s (Fig. 8). After 1s the vent mass flux fell sharply, then stabilized at about 3 . 7 x l 0 7 k g s - 1 (1.4 x 1 0 4 m 3 s - 1 DRE) for about 2s before a further rapid decrease. Most of the Sim5 mass ejection occurred in the first 10s, in contrast to the other simulations. Also in contrast to other similations, the Sim5 maximum vent mixture velocity of 8 5 m s - 1 occurred at about 2 s, after which the velocity declined steadily, to approximately 2 0 m s - 1 by 30s. The pressure at the vent peaked at roughly 0.2s at 3.9MPa, and declined to about one-fourth that value within 10s. Sim5 produced initial slope-parallel pyroclastic current velocities as high as 4 0 m s - 1 near their onset, slowing to 5 m s - 1 after 60s. Emplacement temperatures were about 2900C, compared with 1350C for Siml, reflecting the less energetic vent dynamics and less efficient air entrainment. DRE volume ejected by Sim5( 1.1 x 10 5 m 3 ; 0.29 x 10 9 kg), was less than one-third of that ejected by Siml because Sim5 started with one-third less available material VDREA.

Comparison of simulated behaviour with actual explosions In this section we compare our numerical model explosions with the 1997 Soufriere Hills Vulcanian explosions in order to illustrate the efficacy of our models in simulating real events, and their ability to improve our understanding of fundamental processes and eruption dynamics.

Evolution of plume geometry The explosion evolution exhibited by the simulations, and especially Siml, mimicked well the general characteristics of the observed explosions (Druitt et al. 2002). Like the observed explosions, Siml began with a spherical expanding cap, followed by expansion

MODELLING DYNAMICS OF VULCANIAN EXPLOSIONS

0.0

0.5

1.0

1.5

2.0

0.0

0.5

1.0

335

1.5

2.0

Fig. 10. Simulation 3. Distribution of the total particulate volumetric fraction in the atmosphere at 20, 45, 60 and 120s. Contours as in Figure 7a.

336

A. B. CLARKE ET AL.

Fig. 11. Simulation 4. Distribution of the total particulate volumetric fraction in the atmosphere at at 20. 45. 60 and 100 s. Contours as in Figure 7a.

MODELLING DYNAMICS OF VULCANIAN EXPLOSIONS

Fig. 12. Simulation 5. Distribution of the total particulate volumetric fraction in the atmosphere at 20, 45, 60 and 90 s. Contours as in Figure 7a.

337

338

A. B. CLARKE ET AL.

upward and outward, beyond the crater rim (compare Figs 7a and 2). Then, 10-30s into the simulation, the upward velocity of particles in the plume declined or reversed and a vortex-bearing overhang developed off centre of the plume axis (Fig. 7b). After 40s into the simulation, roughly 10s after the minimum axial plume rise velocity had been achieved, the overwhelmingly downward particle motion within the plume generated a fast-moving radial, basal pyroclastic current that soon passed beyond the overhang radius (Fig. 7a, b,c). Simultaneously, the upward velocity of the axial part of the plume increased because its density had been reduced due to settling of the coarser particles, reflecting the segregation of particles of different sizes. This segregation occurred during plume evolution, and continued during the lateral propagation of the pyroclastic current (Fig. 13). Later in the development of Siml, buoyant thermals developed above the pyroclastic currents, and subsequently joined the rapidly rising convective plume. These model results, illustrating a transitional eruptive regime of simultaneous convective plume behaviour and emplacement of pyroclastic density currents, are consistent with photographic and video observations of the Montserrat explosions (Fig. 2; Druitt et al. 2002). The series of photographs of the 6 and 7 August explosions (Figs 2 and 14) show clearly the typical development of an overhanging plume edge, which constrains the geometry of clast support surfaces (Carey & Sparks 1986). From the overhang, an opaque vertical curtain of coarse fallout developed early in the evolution of the explosion (Fig. 14a), obscuring one's view of the region under the overhang. The pyroclastic current did not develop from this fallout, but rather, as indicated by the simulations, originated from an annular funnel-shaped sector of a collapsing mixture of gas and pyroclasts within the plume interior region, which subsequently pierced through the opaque veil of fallout as a fast-moving, coherent, dense stream of gas and clasts (Fig. 14). Simultaneously with the emplacement of pyroclastic currents, an envigorated upward rise of the lightened, buoyant central plume occurred, reflecting the segregation of coarse and fine particles and the decreased density of the plume due to settling (Fig. 2; cf. Druitt et al. 2002, fig. 4). Buoyant thermals also developed above the pyroclastic currents (Druitt et al. 2002, fig. 4), and subsequently joined the rapidly rising convective plume. All these observations were captured by our model results, and thus the models illuminate internal plume processes which render the observed phenomena more understandable.

Of the simulations, both Siml and Sim3 displayed an overhanging plume when pyroclastic currents initiated. In Sim2 the decrease in initial conduit pressure resulted in a greater proportion of exsolved water vapour, and the combination of factors led to a narrower plume at the time of fountain collapse (compare Siml and Sim2 at 45 s), and no overhang when pyroclastic currents initiated. Although this geometry is inconsistent with the 'typical' plume at 12:05 on 7 August, broad behavioural variations were exhibited in the 88 explosions of 1997 (Druitt et al. 2002). A somewhat similar geometry was rendered by Sim5, with less volatile content due to leakage. As a result of leakage the plume was less energetic and no overhang developed, and pyroclastic currents were initiated within 15-20s from the onset of the explosion. This timing aspect of Sim5 is in closer correspondence with field observations than those of other simulations. However, one typical characteristic of the observed explosions was not duplicated by our simulations and is briefly discussed next. Druitt et al. (2002) describe how each explosion began with 'the rapid rise of numerous dark-grey finger jets of ash and debris', with later stages marked by the coalescence of the pyroclastic mixture into a single, complex, large plume. Our simulations did not reproduce this initial phase of numerous jets, because we assumed in our model homogeneous conduit material and instantaneous removal of the cap. Instead, all simulated explosions began with a single, expanding, bulbous, hemispherical plume. The multiple, initial jets of the observed explosions probably indicate unsteady or pulsating movement of the fragmentation wave into the conduit, and/or irregular break-up of the conduit plug. This pulsing, calypsonic nature of the explosions probably indicates an inhomogeneous distribution of pressurized bubbles in the conduit magma. Massol & Jaupart (1999) show that different parts of a single magma batch experience different degassing histories during ascent in a conduit, and in particular, the horizontal variation of bubble pressure across a conduit may be quite large at fragmentation. Because of such horizontal variations and inhomogeneities, the wave may have paused periodically on its way down the conduit, and the fragmentation wave velocity may have been influenced by such local property variations, resulting in propagation of a wave of irregular shape and velocity. Our simulations do not allow for such fragmentation wave complexities, nor for irregular vent geometries, because we assumed smoothly varying and horizontally

Fig. 13. Distribution of individual paniculate volumetric fraction in the plumes, for 5000 m particles (green), 2000 m particles (red) and 30 m particles (blue), at 45 and 90s of Siml. Contour levels as in Figure 7a. In all three-particlesize simulations, segregation between coarse and fine particles occurred during plume evolution as well as during emplacement of pyroclastic currents.

MODELLING DYNAMICS OF VULCANIAN EXPLOSIONS

339

Fig. 14. Photographs of the explosion at 12:05 LT on 7 August 1997. Images have been reversed to facilitate comparison with simulations. Times after the low-frequency onset of the explosion signal at 12:04:47: (a) 18s, (b) 25s, (c) 32s, (d) 35s. In (a) the column was about 550m high and collapse had begun internally, but pyroclastic currents had not pierced the opaque curtain of fallout. Gages Mountain in foreground left, Farrell's Plain and Mosquito Ghaut to right of fallout (see Fig. 1). In (b), pyroclastic currents make a first appearance on Farrell's Plain, and can also be seen behind Gages Mountain. Soon afterward in (c), a pyroclastic surge crosses Gages Mountain toward observer, and further pyroclastic current build-up is seen at the base of the fallout curtain facing the plain. In (d) the currents move outward on a broad front, and thicken by development of dilute ash clouds. The central plume exhibits more buoyancy as a consequence of loss of dense mixtures to the pyroclastic currents. Photographs taken from MVO South (Fig. 1), by R. Herd.

homogeneous conduit profiles. However, such variations could be considered in future modelling. The vent flux unsteadiness produced by our simulations generally reflects the systematic decrease in water vapour volume fraction with depth in the conduit, not the pulsating movement of the fragmentation wave.

Plume ascent rate Here we compare simulations with data from two 1997 Vulcanian explosions, obtained from timed video. We conducted photogrammetric work on digitized video images for the 6 August explosion at 14:35 and the 7 August explosion at 12:05, using a variation of the general methods of Schmid (1953), Rosenfield (1959; Doyle et al 1966) and Sparks & Wilson (1982). Selected ground points on the images were located in the field by a global positioning system (GPS), and simple computer codes were then developed to convert

the video image co-ordinates to real-space co-ordinates. The combination of co-ordinate position change and image timing enabled the determination of explosion front velocities. Druitt et al. (2002, fig. 16) also measured the plume velocities of the 12:05 7 August explosion by similar procedures. Our analysis followed only the absolute top of the visible plume through time, treating the complex plume as a single entity, whereas Druitt et al. (2002) measured the ascent rates of three separate lobes of the 12:05 7 August explosion plume. Our simplified analysis resulted in ascent rates and trends generally consistent with Druitt et a/.'s (2002) measurements of plumes 1 and 2; however, we measured a minimum velocity (of equal magnitude to theirs) which occurred c. 10s earlier. We have used our video measurements for general comparisons with simulations. In Figure 15, the variation of the simulated height of the central plume over time is compared against data for the two August 1997 Vulcanian explosions. In Figure 16, plume-front vertical velocities are shown over time. The zero reference time on these figures

A. B. CLARKE ET AL.

(d)

Fig. 14. (continued)

indicates the explosion onset, which we correlate with the initiation of the strong low-frequency signal on the seismic records. The explosion onset times assumed are 14:35:12 LT for 6 August, and 12:04:47 LT for 7 August. A few seconds of higher frequency signal preceded the low-frequency onset, and we interpret this as reflecting the destruction of the conduit plug prior to the explosion. Estimates of error in plume-front position from our video analysis, due to video frame measurement errors, have been made by allowing 2 mm error in locating the edge of the plume in the images (taken from images of dimensions 22.5cm by 16.9cm) in both the horizontal and vertical directions. The maximum error for the 6 August event is 5 m (resulting in maximum velocity error of 25ms"" 1 ) and that for the 7 August event is 5 m (resulting in velocity error of 23ms" 1 ), both when the plume was between 2000 and 2500m a.s.l. The error at altitudes nearer the vent for the 6 August event is 5 m (resulting in velocity error of 7ms - ) and that for the 7 August event is 0 m (resulting in velocity error of 9 m s - 1 ) . The video camera was set at a higher zoom level during

the 6 August event, explaining the difference in error between the two events. For both cases, early velocities exceeded 40m s-1 and decreased with time (Fig. 16). The 6 August plume front reached its minimum rise velocity of about 20 m s - 1 roughly 25s after event onset, at a plume height of roughly 750 m above the vent (1600 m a.s.l.). whereas for the 7 August eruption the velocity minimum of < 2 0 m s - 1 was achieved by c. 15 s at a height of approximately 550 m above the vent (1400m a.s.l.). The minimum velocity occurs at the boundary between momentum-dominated jet and buoyancy-dominated convective plume, marking approximately the fountain-collapse height and onset time. Based on seismic record interpretation, pyroclastic flows tended to begin 5-10s after the minimum plume rise velocity. In comparison, collapse initiated for Siml at a height of 970m above the vent at 28 s; for Sim2, 1000 mat 26s; for Sim3. 1160 mat 35s; for Sim4, 1370 mat 22s; and for Sim5, 450m at 15s (Table 3). The observational data sets indicate that both August plumes then underwent modest increases in speed, reaching secondary

MODELLING DYNAMICS OF VULCANIAN EXPLOSIONS

341

351

o • — — —

August 6,1997 explosion August 7,1997 explosion Siml Sim2 Sim3 Sim4 — Sim5

Fig. 15. Height of rising central plume for the first minute of the explosion, comparing observational and simulated data. Note that the 6 August explosion is slightly more energetic in the first 10s than that of 7 August. In general, the observed data fall between Siml and Sim5 over the first 30 s.

30

40

Time (s) 200

O August 6, 1997 • August 7, 1997

180

— Siml — Sim2 — Sim3 Sim4 — Sim5

160

Sim4,

Buoyancy dominated (acceleration of plume due to sedimentation)

Fig. 16. Plume-front vertical velocity for the first minute of the explosion, comparing observational and simulated data. The stippled area denotes the region of 7 August and 6 August minimum rise velocities, and marks approximately the transition from jet-dominated to buoyancydominated regimes. Vertical velocities for observed explosions were computed from differentiation of curve fits of the height-time values of Figure 15. In general the field data fall between Siml and Sim5, as discussed in text.

0

10

maximum velocities of 20-30 ms-1 about 28-40 s into the explosions. The increase in vertical velocity after the fountain collapse occurred because the bulk density of the central plume was reduced by sedimentation of large particles during fountain collapse, increasing buoyancy of the upper plume. The central plumes of Siml, Sim2 and Sim3 ascended at similar rates, with initial velocities between 80 and 9 0 m s - l , and declined to minima approaching zero at 28-35 s into the explosion. Then, the simulation plume ascent velocities climbed to 40-60 m s - 1 ,

20

30

40

50

60

Time (s) compared to <40 m s-1 for the observed plumes at 60 s (Druitt et al 2002, fig. 16b). Thus, vertical plume velocities of Siml, Sim2 and Sim3 followed qualitatively the general trend of the observed explosions, but their deceleration to a velocity minimum was slower, thus delaying pyroclastic current initiation. Sim4, lacking conduit fines, accelerated rapidly beyond 20 s and reached an ascent rate of 200m s-1 by 45 s. This vertical velocity far exceeds that of the observed events and represents a clear example of the unrealistic results that can be obtained by ignoring the fine

342

A. B. C L A R K E ET AL.

component of the eruptive mixture. The high ascent rate is explained by the rapid sedimentation of large particles comprising the erupted solid mass in the plume, thus more effectively reducing the bulk density of the heated central plume. The Sim5 plume, with conduit volatile leakage, most closely matched the 7 August event, although its plume-front velocity declined further than observed events (Fig. 16). The Sim5 plume reached a secondary maximum velocity of 40m s-1 at approximately 40 s into the simulation, whereas the secondary maxima of the two observed events were smaller. Although Sim5 reproduced the timing of the velocity minima most closely, it failed to reproduce the plume overhang beneath which the pyroclastic currents emerged. Siml did this most effectively as described above. The simulations all produced minimum plume-rise velocities less than those observed, implying that the real explosion plumes never quite stagnated (Figs 15 and 16). This may partly reflect the fact that the simulations did not duplicate in detail the multiple-jet character of the real explosions. Many of these jets occurred in rapid succession during the first 10s (Formenti & Druitt 2000), adding heat energy and turbulent air mixing to the complex plume. The simulated particle size distribution was also a factor (compare Siml and Sim4, Fig. 16).

Plume collapse and pyroclastic current initiation At 15-20s Sim5 produced an unstable fountain 450m high that collapsed to generate radial pyroclastic currents. In contrast, the Siml, Sim2 and Sim3 fountains collapsed between 28 and 35s into the simulations, producing pyroclastic currents at roughly 40 s. This delayed collapse relative to Sim5 occurred because Siml, Sim2 and Sim3 exhibited very high initial vent velocities, due to greater volatile volume fractions, which increased fountain collapse height to about 1000m or more and delayed pyroclastic current initiation. Sim4, with only coarse particles, produced a collapsing fountain earlier than Siml, Sim2, Sim3 and Sim5 because it lacked fine particles. As expected from conservation of energy, for all simulations, the collapse height is related to the peak vent velocity by, approximately, hc = V2max/2g, where hc is the fountain collapse height above the vent and Vmax is the maximum vent velocity. The simulations also illustrated a positive linear correlation between specific expansion energy, Ev, and fountain collapse height. The model results may also be compared to the seismic signals from the Vulcanian explosions. Druitt et al. (2002) recognized two components in the signals: a low-frequency component interpreted as the vibrational response of the magmatic conduit to the explosion; and a high-frequency component caused by a combination of ballistic impact, fountain collapse and pyroclastic flows. The onset of the high-frequency component coincided with the first impacts of the collapsing fountain and ballistic blocks with the ground as seen on video footage. This typically occurred 10-20s after the explosion onset and approximately marked the appearance of the first pyroclastic density currents, although the first visible appearance of the flows could be delayed relative to seismically inferred onset by a veil of fallout from an overhanging column. The timing of collapsing fountain and generation of pyroclastic density currents of Sim5 is in better agreement than Siml with seismic data and video observations.

Pyroclastic currents and their deposits The real explosions produced approximately radial pyroclastic surges that travelled at 30-60 m s - 1 on the upper flank, as well as high-concentration pumice-and-ash flows that travelled more slowly (c. 1 0 m s - 1 ) mainly within valleys for c. 3 minutes, reaching as far as 6km from the vent (Druitt et al, 2002). The pyroclastic currents produced by Siml, Sim2 and Sim3 (runouts c. 1800m peak frontal velocities c. 40-60 m s - 1 , average frontal velocities c. 30ms - 1 on steep flanks and c, 1 0 m s - l on lower slopes) reasonably approximated the observed surge runouts and velocities. Sim5

generated pyroclastic currents that moved a similar distance, at frontal velocities as much as 4 0 m s - 1 . Sim4 produced very thin currents at unrealistically low velocities. The observed pumiceand-ash flows on Montserrat had runout distances significantly extended by localized deep channels beyond approximately 500m from the vent (Fig. 1), and these phenomena could not be well simulated by our axisymmetric model. The temperatures of pumice-and-ash flow deposits, measured less than one day after deposition, indicate emplacement temperatures between 180 and 220 C (Druitt et al. 2002; Cole et al. 2002). Siml and Sim3 produced deposit (i.e. stagnated flow) peak temperatures of 100 to 135eC, less than the field values. Sim2 and Sim5, both associated with a less energetic plume and less air mixing, generated temperatures of 210 and 290 C, respectively. Sim4 produced temperatures far exceeding the observed range (430 C), reflecting lack of air mixing and rapid sedimentation of coarse particles.

DRE volume ejected and depth of evacuation All simulations, except Sim5. ejected slightly more DRE volume than the average of the observed explosions (3.0 x 105 m 3 ). but of the same order of magnitude, with values falling between 1.1 and 4.6 x 105 m3 (0.29-1.2 x 109 kg). The simulated explosions ejected 85-93% of the DRE available, VDREA. Siml reproduced the mass partioning of the real events most closely with 66% of VDREE entering the pyroclastic current (compared to the 63% observed). The average depth of conduit evacuation (for an assumed conduit diameter of 30 m) for the observed explosions, assuming an average of 50% vesiculation, was roughly 850m. Siml and Sim4 produced values closest to the real explosions, but Sim2 and Sim3 also produced results in reasonable agreement. For all main simulations, the depth at which the fragmentation wave stopped was probably affected by the solid boundary placed at Es = 0.70. Overpressure in the conduit for all simulations was constant with depth and therefore did not control the depth of evacuation.

Summary of comparisons To summarize, although Sim5 reproduced the ascent rate of the central plume most closely (especially that of 7 August), it failed to reproduce the overhanging plume shape of many observed plumes, and the associated curtain of fallout through which the pyroclastic currents pierced. Siml and Sim3 produced these features, but not the ascent rate pattern of the plume, with the field-based data mostly intermediate between Siml and Sim5 (Figs 15 and 16). Similarly, Sim5 produced an unstable fountain 450m high, in contrast to c. 1000m or more for Siml. Sim2 and Sim3. whereas field values range from about 400 to 750m above the vent. The corresponding Siml, Sim2 and Sim3 fountains collapsed between 25 and 30s into the simulations, producing pyroclastic currents at roughly 40s in contrast to their generation at 15-20s in Sim5 and also the observed explosions. The difference between Sim5 and the other three simulations is explained by Sim5's decreased conduit energy due to leakage of volatiles, which produced lower fountains and earlier collapse. The pyroclastic current runouts and speeds were reasonably matched by all simulations apart from Sim4, which was unrealistic in this and other respects. In comparison to deposit temperatures, Siml and Sim3 produced peak temperatures that were too low, significantly less than the field values and indicative of excessive air entrainment. Sim2 and Sim5, both associated with a less energetic plume and less air mixing, generated temperatures of 210 and 290 C, respectively, nearer field values. On balance it would seem that the field results, as exemplified by observations on 6 and 7 August, are best described as transitional between Siml and Sim5 (Figs 15 and 16). Siml, in which volatiles were transported from the magma reservoir to the upper conduit without loss, seems to be too energetic, although many qualitative aspects of the real events were captured by this simulation (Fig. 17). Likewise, Sim5 exhibited better scale effects and timing, but may

MODELLING DYNAMICS OF VULCANIAN EXPLOSIONS

Fig. 17. (a) Vulcanian explosion on 6 August 1997, about 91 s after onset, (b) Simulated explosion (Siml). Distribution of total paniculate volumetric fraction in the plume at 116s after onset. Contours as in Fig. 7a. (c) Juxtaposition of the eastern half of the 6 August explosion, and the simulation in (b). Note the similarities of pyroclastic current shape and runout distance, and the turbulent buoyant ash clouds generated above the current being drawn into the rising central plume. The simulation also reasonably approximates the shape of the rising central plume, and illustrates internal gradients of particle concentration (highest concentration shown by red).

343

344

A. B. CLARKE ET AL.

have been insufficiently energetic, perhaps because the volatile loss assumed in this case was excessive. To test this hypothesis, we have performed another simulation in which the leaking' conditions of Sim5 were used, but with volatile content somewhat increased. The results show that a volatile increase of <20% over Sim5 can create an overhanging plume with collapse height <700m consistent with the observations at Montserrat. The timing of this test simulation was intermediate between Siml and Sim5. In general these results validate the model procedure used to produce the unsteady dynamics of a Vulcanian explosion. They suggest strongly that volatile reduction in the conduit was a significant factor in influencing the dynamic behaviour of the Vulcanian explosions on Montserrat.

Shock waves Condensation of atmospheric moisture, indicative of shock waves moving away from the vent ahead of the erupting mixture, was visible on some videos of explosions. Vulcanian explosions are analogous to a shock tube (Woods 1995; Turcotte et al. 1990) in which high-pressure gas in a tube (the conduit) is separated from lower pressure gas (the atmosphere) by a diaphragm (the caprock) (Fig. 5). The instantaneous rupture of the diaphragm results in a series of compression waves rapidly coalescing into a normal shock wave that travels at supersonic speed into the region of lower pressure. At the same time, an expansion wave (fragmentation wave) propagates into the high-pressure region at the local speed of sound. The conduit pressure is P and the atmospheric pressure is P.dtm. The ratio of these pressures, P/Patm, is related to the initial Mach number of the shock wave, Ms, the ratio of specific heats for each of the two gases, 7atm (air) and 7m (magma mixture), and the acoustic velocities in each of the two gases, aatm and am, according to the following equation (Ishihara 1985):

(10)

This equation was solved for a range of Mach numbers for a highpressure region of mixture and a low-pressure region of atmospheric air (aatm = 340 m s-1, 7atm = 1 .4; am = 880 m s-1, 7m = 1.05) (Ishihara 1985; Dobran et al. 1993). The overpressure estimated above for these explosions, about l0MPa, suggests that shock waves propagated initially from the vent at Mach numbers of about 1.6, which is roughly 1400 m s-1. This velocity decays approximately exponentially with distance from the vent (Courant & Friedrichs 1948). The strength of such shocks, that is, the pressure ratio across the shocks, P2/Patm (see Fig. 5), is given by the following equation (Saad 1985): (11)

Assuming a conduit overpressure of l0MPa, shocks generated at Soufriere Hills Volcano should have been of strength, P2/Patm, roughly 2.8 of the eruptive mixture. The simulations show large atmospheric pressure gradients above the vent during the first seconds after explosion initiation. Further work is planned to simulate this phenomenon numerically.

Conclusions This paper constitutes a first attempt to use field data to estimate vertically varying pre-explosion conduit conditions and to combine these conditions with a pyroclastic dispersal model to reproduce the qualitative and quantitative behaviour of Vulcanian explosions. In general our results validate the model procedure used to replicate the unsteady dynamics of such explosions, and several simulations

qualitatively predicted much of this behaviour, including fountain collapse height and timing, initial plume ascent rate, plume ascent rate variation with time, total DRE volume of solids ejected, depth of conduit evacuation, generation of ash plumes associated with pyroclastic currents, and pyroclastic current velocities, temperatures and runout. Our simulations did not duplicate the multiple-jet character of many real events, because inhomogeneities involving conduit magma properties were not considered, and, because an axisymmetric topographical grid was used, replication of distal pumice-and-ash flow runout in localized channels was not attempted. Nevertheless, the fundamental features of the transitional explosive regime of simultaneous generation of an ascending buoyant plume, and emplacement of pyroclastic density currents, were captured with fidelity. Based on our conduit model and the multiphase axisymmetric simulations discussed in this paper, we can make several general conclusions. (1)

Vent parameters, such as mass flux, vertical mixture velocity and pressure, are characteristically highly transient in this type of short-duration explosion. Therefore, models employing the assumption of steady vent conditions cannot be reliably applied to this type of eruption. (2) Both a 30% decrease in assumed initial conduit overpressure and a 12% increase in meltwater content had significant effects on simulation results. Relative to the reference simulation, these changes influenced explosion duration, the total DRE volume of solids ejected, fountain collapse height, and the buoyancy of the central plume and development of ash thermals above pyroclastic currents. At higher water content the column is characterized by more complex and energetic dynamics. Decreasing pressure moved the simulation toward the collapsing regime. In all three particle size simulations, segregation between coarse and fine particles occurred during the plume collapse as well as during emplacement of the pyroclastic currents. (3) Removal of the fines fraction in Sim4 had the most dramatic effect of all variations tested. It decreased total DRE volume of solids ejected, significantly reduced pyroclastic current runout, increased the buoyant rise of the central plume, and virtually eliminated the development of ash cloud thermals from the pyroclastic currents. This simulation provides a clear example of the unrealistic results that can be obtained by ignoring the fine component of the eruptive mixture. (4) The field results, as exemplified by observations on 6 and 7 August, are best described as transitional between our Siml and Sim5. Siml, whose initial conditions best approximated the observationally constrained input data, appeared to be too energetic, although it captured many qualitative aspects of the real events. In this case, volatiles were assumed to be transported from the magma reservoir to the upper conduit without loss. In contrast. Sim5 exhibited better scale effects and timing, but may have been insufficiently energetic to produce some qualitative characteristics of the explosions such as the overhang, perhaps because the volatile leakage assumed in this case was excessive. A preliminary experiment suggests that less extreme leakage, leaving <20% additional volatiles, provides appropriate plume dynamics. The results strongly suggest that volatile reduction in the conduit was an important factor in influencing the dynamic behaviour of the Vulcanian explosions on Montserrat. We thank our many colleagues at Montserrat Volcano Observatory (MVO) for their vital assistance, with special acknowledgement to C. Bonadonna, C. Harford, R. Herd, R. Luckett and R. E. A. Robertson, our colleagues during the August explosions. R. Herd provided still photographs used in Figure 14. D. Lea and D. Williams provided video footage. J. McMahon provided key helicopter support beyond the requirements of duty. Support for monitoring operations was provided by the Department for International Development (UK), the British Geological Survey (BGS). the Seismic Research Unit of the University of the West Indies, and the US Geological Survey (USGS). A.C. and B.V. acknowledge support from the US National

MODELLING DYNAMICS OF VULCANIAN EXPLOSIONS

345

Science Foundation Grants EAR 96-14622, 28413 and 00-73761, and The Pennsylvania State University. A.N. and G.M. were assisted by the Istituto CNR-CNUCE, Pisa, and the Istituto Nazionale di Geofisica e Vulcanologia. A.N. acknowledges support from Gruppo Nazionale per la Vulcanologia INGV, Italy, project no. 2000-2/9. B.V. and T.D. were Senior Scientists at Montserrat in 1997 with BGS affiliation. BV was affiliated also with USGS Volcano Hazards Program. T. Koyaguchi, L. Wilson, A. Woods and P. Kokelaar reviewed and improved the manuscript.

where h is enthalpy, T is temperature, K is thermal conductivity, assumed to represent an effective value when the flow is turbulent, and Qk are the heat transfer coefficients between the gas and the /rth particle phase. Heat exchange among solid phases, viscous dissipation effects, as well as radiation are neglected given that they are small in comparison with convection, conduction, and gas-particle heat exchange (Valentine & Wohletz 1989).

Appendix: Multiphase pyroclastic dispersal model

Constitutive equations

In this appendix the transport equations solved by the transient multiphase flow model are briefly reported. A detailed description of them can be found in Neri (1998) and Neri et al (200\b).

The gas phase is assumed to follow the ideal gas law: Pg

_ Pg

~ R7i

where R is the gas constant of the water vapour, and the solid densities are:

Conservation equations

Pk = Ps

Conservation of mass. Gas phase:

The temperature of each phase is computed from its enthalpy as: hk

Solid phases:

where e is the volumetric fraction of a particular phase, p is the density, v is the velocity vector, / is time, subscript g refers to gas and subscript k refers to a solid phase.

where Cpg is the specific heat of the gas based on empirical correlations as a function of temperature (Reid et 0/.1986; Kadoya et al. 1986) and Cpk is the specific heat of particles and is assumed constant. The gas phase stress tensor is modelled by a subgrid-scale model using an effective viscosity, nge, which accounts for the turbulent dissipation at scales smaller than the computational grid size (Deardorff 1971). The gas viscous stress tensor is therefore:

Conservation of momentum. Gas phase: where I is the identity matrix and the effective gas viscosity nge is: /V = P>g + /% = V>g + C2sArAzpg(2tr(fg

• fg)$

where fig is the gas molecular viscosity, \igt is the effective gas turbulent viscosity, C$ is the Smagorinsky constant assumed equal to 0.1, Ar and Az are radial and vertical sizes of the computational cell, and the deformation tensor fg is:

Solid phases:

d

-V- VT*

The solid stress tensor for the kth particle is expressed in terms of an empirical solid viscosity rv^ (due to collisions among solid particles) and a coulombic repulsive component rc&, expressed as:

T* = TV,* - ra-I where the solid viscous stress tensor TV,,A is: where Pg is gas pressure, T is the viscous stress tensor, g is gravitational acceleration, Dgk is the gas-solids drag coefficient, and DkJ is the drag coefficient among particulate phases. Conservation of energy. d

TV,* =

* + (Vv*)T) - I(V • v*)I]

and the coulombic component gradient is:

Gas phase: where G(%) is a solid elastic modulus able to account for repulsive forces when low porosity values are reached in the mixture (Gidaspow 1994). Based on Miller & Gidaspow (1992) and Gidaspow (1994), particle viscosities were taken in the range:

= (0.5

Solid phases:

d

(Pas)

where larger values apply for coarser particles. It should also be noted that no influence has been considered between the gas and particle turbulence descriptions. Semi-empirical relationships for drag between the gas and solid phases are used. In the dilute regime, the drag coefficient is determined by a correction for the case of a single sphere. In the high concentration regime, the Ergun (1952) equation is used.

346

A. B. CLARKE ET AL, where the Nusselt expression depends on the particle concentration and Reynolds and Prandtl numbers as:

Nu k

=

where Reynolds and Prandtl numbers are: and

References where the Reynolds number is:

where CD is the particle drag coefficient, dk is the particle diameter, and pg is the gas density. Particle-particle drag has been observed to be important in laboratory experiments where particles of different sizes flow at different velocities (Soo 1967; Arastoopour et al. 1982; Aldis & Gidaspow 1989). The relationship adopted in the model is from Syamlal (1985) and it is an extension of the correlation of Nakamura & Capes (1976) for dilute mixtures to dense regimes.

where e is the restitution coefficient for a particle-particle collision (assumed equal to 1), a is an empirical coefficient which accounts for non-head-on collisions (here also assumed to be 1), the subscripts j and k represent parameters for particles of different sizes, and:

where the maximum solids volume fraction of a random closely packed mixture, EKJ, can be calculated as a function of the mixture composition and particle diameters as:

where k is the particle volume fraction of the kth phase at the maximum packing and:

Heat transfer among solid and gas phases is modelled by the generalization of a semi-empirical correlation accounting for different flow density and regimes and valid up to a Reynolds number of 105 (Gunn 1978). The heat transfer coefficient Qk between gas and the kth solid phase is related to the Nusselt number Nuk by:

ACHENBACH. E. 1972. Experiments on the flow past spheres at very high Reynolds numbers. Journal of Fluid Mechanics. 54. 565-575. ALOIS. D. F. & GIDASPOW, D. 1989. Combustion of a polydispersed solid using a particle population balance. Powder Technology. 57, 281-294. ALIDIBIROV, M. A. 1994. A model for viscous magma fragmentation during volcanic blasts. Bulletin of Volcanology. 56 459-465. ALIDIBIROV, M. & DINGWELL. D. 1996. High temperature fragmentation of magma by rapid decompression. Nature. 380. 146-149. AMSDEN, A. A. & HARLOW. F. H. 1974. KACH1NA: An eulerian computer program for multifield fluid flows. Los Alamos National Laboratory. Report LA-5680. ANILKLMAR, A. V., SPARKS. R. S. J. & STuRTEVANT. B. 1993. Geological implications and applications of high-velocity two-phase flow experiments. Journal of Volcanology and Geothermal Research. 56. 145-160. ARASTOOPOUR, H., LIN. S. C. & WEIL. S. 1982. Analysis of vertical pneumatic conveying of solids using multiphase flow models. American Institute of Chemical Engineers Journal. 28. 467-473. BARCLAY. J., CARROLL. M. R.. RUTHERFORD. M. J., MURPHY. M. D.. DEVINE, J. D., GARDNER. J. & SPARKS. R. S. J. 1998. Experimental phase equilibria constraints on pre-eruptive storage conditions of the Soufriere Hills magma. Geophysical Research Letters. 25. 3437-3440. BURSIK. M. I. & WOODS, A. W. 1996. The dynamics and thermodynamics of large ash flows. Bulletin of Volcanology. 58, 175-193. CAREY, S. N. & SPARKS, R. S. J. 1986. Quantitative models of the fallout and dispersal of tephra from volcanic eruption columns. Bulletin of Volcanology, 48. 109-125. COLE, P. D., CALDER, E.. SPARKS. R. S. J. ETAL. 2002. Deposits from domecollapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21. 231-262. COURANT. R. & FRIEDRICHS. K. O. 1948. Supersonic Flow and Shock Waves. Interscience, New York. DEARDORFF, J. W. 1971. On the magnitude of the subgrid scale eddy coefficient. Journal of Computational Physics. 7. 120-133. DENLINGER, R. P. & HOBLITT, R. P. 1999. Cyclic eruptive behavior of silicic volcanoes. Geology, 27(5). 459-462. DEVINE, J. D., RUTHERFORD. M. J. & BARCLAY. J. 1998. Changing magma conditions and ascent rates during the Soufriere Hills eruption on Montserrat. GSA Today, 8. 2-7. DOBRAN, F., NERI, A. & MACEDOMO, G. 1993. Numerical simulation of collapsing volcanic columns. Journal of Geophysical Research. 98. 4231-4259. DOYLE, F. J., COOK. W. H.. ELLER. R. C. ET AL. 1966. Analytical photogrammetry. In: THOMPSON. M. M.. ELLER. R. C.. RADLINSKI. W. A. & SPEERT, J. L. (eds) Manual of Photogrammetry. American Society of Photogrammetry, Falls Church. DRUITT. T. H. & MVO Staff 1998. The explosive eruptions of August 1997. Montserrat Volcano Observatory. Special Report 04. DRUITT, T., YOUNG, S., BAPTIE. B. ET AL. 2002. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano. Montserrat. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London. Memoirs. 21, 281-306. ERGUN, S. 1952. Fluid flow through packed columns. Chemical Engineering Processes, 48, 89. FAGENTS, S. A. & WILSON, L. 1993. Explosive volcanic eruptions-VII. The ranges of pyroclasts ejected in transient volcanic explosions. Geophysical Journal International, 113. 359-370.

MODELLING DYNAMICS OF VULCANIAN FORMENTI, Y. & DRUITT, T. H. 2000. Characterisation of Vulcanian explosions of Soufriere Hills Volcano, Montserrat, by video analysis. Abstracts and Addressses of the I A VCEl General Assembly 2000. Bali, Indonesia, 232. FUDALI, R. F. & MELSON, W. G. 1972. Ejecta velocities, magma chamber pressure and kinetic energy associated with the 1968 eruption of Arenal volcano. Bulletin of Volcanology, 35, 383-401. GARDNER, J. E., HILTON, M. & CARROLL, M. R. 1999. Experimental constraints on degassing of magma: isothermal bubble growth during continuous decompression from high pressure. Earth and Planetary Science Letters, 168, 201-218. GIDASPOW, D. 1994. Multiphase Flow and Fluidization: Continuum and Kinetic Theory Descriptions. Academic Press, New York. GUNN, D. J. 1978. Transfer of heat or mass to particles in fixed and fluidized beds. International Journal of Heat and Mass Transfer, 21, 467-476. HARLOW, F. H. & AMSDEN, A. A. 1971. A numerical fluid dynamics calculation method for all flow speeds. Journal of Computational Physics, 8, 197. HARLOW, F. H. & AMSDEN, A. A. 1975. Numerical calculation of multiphase fluid flow. Journal of Computational Physics, 17, 19-52. HOBLITT, R. 1986. Observations of the Eruptions of July 22 and August 7 1980, at Mount St. Helens, Washington. US Geological Survey, Professional Papers, 1335. ISHIHARA, K. 1985. Dynamical analysis of volcanic explosion. Journal of Geodynamics, 3, 327-349. JAUPART, C. & ALLEGRE, C. J. 1991. Gas content, eruption rate and instabilities of eruption regime in silicic volcanoes. Earth and Planetary Science Letters, 102, 413-429. KADOYA, K., MATSUNAGA, N. & NAGASHIMA, A. 1986. Viscosity and thermal conductivity of dry air in the gaseous phase. Journal of Physical and Chemical Reference Data, 14, 947-970. KIEFFER, S. W. 1981. Fluid dynamics of the May 18 blast at Mount St. Helens. In: LIPMAN, P. W. & MULLINEAUX, D. R. (eds) The 1980 Eruptions of Mount St Helens, Washington. US Geological Survey, Professional Papers, 1250, 379-400. MASSOL, H. & JAUPART, D. 1999. The generation of gas overpressure in volcanic eruptions. Earth and Planetary Science Letters, 166, 57-70. MELNIK, O. E. 1999. Fragmenting magma. Nature, 397, 394-395. MELNIK, O. & SPARKS, R. S. J. 1999. Nonlinear dynamics of lava dome extrusion. Nature, 402, 37-41. MELNIK, O. & SPARKS, R. S. J. 2002a. Dynamics of magma ascent and lava extrusion at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 153-171. MELNIK, O. & SPARKS, R. S. J. 2002b. Modelling of conduit flow dynamics during explosive activity at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 307-317. MILLER, A. & GIDASPOW, D. 1992. Dense, vertical gas-solid flow in a pipe. American Institute of Chemical Engineers Journal, 38, 1801-1815. MORTON, B. R., TAYLOR, G. I., & TURNER, J. S. 1956. Turbulent gravitational convection from maintained and instantaneous sources. Philosophical Transactions of the Royal Society of London: Series A, 234, 1-23. MURPHY, M. D., SPARKS, R. S. J., BARCLAY, J. ET AL. 1998. The role of magma mixing in triggering the current eruption at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3433-3436. MURPHY, M. D., SPARKS, R. S. J., BARCLAY, J., CARROLL, M. R. & BREWER, T. S. 2000. Remobilization of andesite magma by intrusion of mafic magma at the Soufriere Hills Volcano, Montserrat, West Indies. Journal of Petrology, 41, 21-42. NAIRN, I. A. & SELF, S. 1978. Explosive eruptions and pyroclastic avalanches from Ngauruhoe in February 1975. Journal of Volcanology and Geothermal Research, 3, 39-60. NAKAMURA, K. & CAPES, C. E. 1976. Vertical pneumatic conveying of binary particle mixtures. In: KEAIRNS, D. C. & BERGOUGNOU, M. A. (eds) Fluidization Technology. Hemisphere, Washington, DC, 159-184. NERI, A. 1998. Multiphase flow modelling and simulation of explosive volcanic eruptions. PhD Thesis, Department of Chemical and Environmental Engineering, Illinois Institute of Technology, Chicago. NERI, A. & GIDASPOW, D. 2000. Riser hydrodynamics: simulation using kinetic theory. American Institute of Chemical Engineers Journal, 46, 52-67.

EXPLOSIONS

347

NERI, A. & MACEDONIO, G. 1996a. Physical modeling of collapsing volcanic columns and pyroclastic flows. In: SCARPA, R. & TILLING, R. I. (eds) Monitoring and Mitigation of Volcano Hazards. Springer-Verlag, New York. NERI, A. & MACEDONIO, G. \996b. Numerical simulation of collapsing volcanic columns with particles of two sizes. Journal of Geophysical Research, 101, 8153-8174. NERI, A., PAPALE, P. & MACEDONIO, G. 1998. The role of magma composition and water content in explosive eruptions: 2. Pyroclastic dispersion dynamics. Journal of Volcanology and Geothermal Research, 87, 95-115. NERI, A., DI MURO, A. & Rosi, M. 2001a. Mass partition during collapsing and transitional columns by using numerical simulations. VSG Report No. 2001-1, Volcano Simulation Group, Instituto Nazionale di Geofisica e Vulcanologia. NERI, A., MACEDONIO, G., GIDASPOW, D. & ESPOSTI ONGARO, T. 2001b. Multiparticle simulation of collapsing volcanic columns and pyroclastic flows. VSG Report No. 2001-2, Volcano Simulation Group, Instituto Nazionale di Geofisica e Vulcanologia. PAPALE, P. 1999. Strain-induced magma fragmentation in explosive eruptions. Nature, 397, 425-428. REID, R. C., PRAUSNITZ, J. M. & POLLING, B. E. 1986. The Properties of Gases and Liquids, McGraw-Hill, New York. ROBERTSON, R., COLE, P., SPARKS, R. S. J. ET AL. 1998. The explosive eruption of Soufriere Hills Volcano, Montserrat, West Indies, 17 September 1996. Geophysical Research Letters, 25, 3421-3424. ROSENFIELD, G. H. 1959. The problem of exterior orientation in photogrammetry. Photogrammetric Engineering, 25, 536. SAAD, M. A. 1985. Compressible Fluid Flow. Prentice-Hall, Engelwood Cliffs. SCHMID, H. H. 1953. An analytical treatment of the orientation of a photogrammetric camera. Aberdeen MD, Ballistic Research Laboratories, Report No. 880. sELF, S., WILSON, L. & NAIRN, I. A. 1979. Vulcanian eruption mechanisms. Nature, 277, 440-443. SELF, S., KIENLE, J. & HUOT, J.-P. 1980. Ukinrek maars, Alaska, II. Deposits and formation of the 1977 craters. Journal of Volcanology and Geothermal Research, 1, 39-65. Soo, S. L. 1967. Fluid Dynamics of Multiphase Systems. Blaisdell-Ginn, Waltham. SPARKS, R. S. J. 1978. The dynamics of bubble formation and growth in magmas: a review and analysis. Journal of Volcanology and Geothermal Research, 3, 1-37. SPARKS, R. S. J. 1986. The dimensions and dynamics of volcanic eruption columns. Bulletin of Volcanology, 48, 3-15. SPARKS, R. S. J. 1997. Causes and consequences of pressurisation in lava dome eruptions. Earth and Planetary Science Letters, 150, 177-189. SPARKS, R. S. J. & WILSON, L. 1976. A model for the formation of ignimbrite by gravitational column collapse. Journal of the Geological Society of London, 132, 441-451. SPARKS, R. S. J. & WILSON, L. 1982. Explosive volcanic eruptions - V. Observations of plume dynamics during the 1979 Soufriere eruptions, St. Vincent. Geophysical Journal of the Royal Astronomical Society, 69, 551-570. SPARKS, R. S. J., WILSON, L. & HULME, G. 1978. Theoretical modeling of the generation, movement, and emplacement of pyroclastic flows by column collapse. Journal of Geophysical Research, 83, 1727-1739. SPARKS, R. S. J., BURSIK, M. L, CAREY, S. N., GILBERT, J. S., GLAZE, L. S., SIGURDSSON, H. & WOODS, A. 1997. Volcanic Plumes. Wiley, New York. SPARKS, R. S. J., YOUNG, S. R., BARCLAY, J. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills Volcano, Montserrat, West Indies: November 1995 to July 1996. Geophysical Research Letters, 25, 3421-3424. SYAMLAL, M. 1985. Multiphase hydrodynamics of gas-solids flow. PhD Thesis. Department of Chemical and Environmental Engineering, Illinois Institute of Technology, Chicago. THOMAS, N., JAUPART, C. & VERGNIOLLE, S. 1994. On the vesicularity of pumice. Journal of Geophysical Research, 99, 15633-15644. TURCOTTE, D. L., OCKENDON, H., OCKENDON, J. R. & COWLEY, S. J. 1990.

A mathematical model of Vulcanian eruptions. Geophysical Journal International, 103, 211-217. TURNER, J. S. 1979. Buoyancy Effects in Fluids. Cambridge University Press, Cambridge. VALENTINE, G. A. & WOLHLETZ, K. H. 1989. Numerical models of Plinian eruption columns and pyroclastic flows. Journal of Geophysical Research, 94, 1867-1887.

348

A. B. CLARKE ET AL.

VALENTINE, G. A., WOHLETZ, K. H. & KIEFFER, S. W. 1992. Effects of topography on facies and compositional zonation in caldera-related ignimbrites. GSA Bulletin, 104, 154-165. VOIGHT, B., HOBLITT, R. P., CLARKE, A. B., LOCKHART, A. B., MlLLER,

A. D., LYNCH, L. & MCMAHON, J. 1998. Remarkable cyclic ground deformation monitored in real time on Montserrat and its use in eruption forecasting. Geophysical Research Letters, 25, 3405-3408. VOIGHT, B., SPARKS, R. S. J., MILLER, A. D. ET AL. 1999. Magma flow instability and cyclic activity at Soufriere Hills Volcano, Montserrat. Science, 283, 1138-1142. WAITT, R. B., MASTIN, L. G. & MILLER, T. P. 1995. Ballistic showers during crater peak eruptions of Mount Spurr Volcano, summer 1992. In: KEITH, T. E. C. (ed.) The 1992 Eruptions of Crater Peak Vent, Mount Spurr Volcano, Alaska. US Geological Survey, Bulletin 2139. WILSON, L. 1972. Explosive volcanic eruptions - II. The atmospheric trajectories of pyroclasts. Geophysical Journal of the Royal Astronomical Society, 30, 381-392.

WILSON, L., SPARKS. R. S. J.. HUANG, T. C. & WATKINS. N. D. 1978. The control of volcanic eruption column heights by eruption energetics and dynamics. Journal of Geophysical Research. 83. 1829-1836. WILSON, L., SPARKS, R. S. J. & WALKER. G. P. L. 1980. Explosive volcanic eruptions - IV. The control of magma properties and conduit geometry on eruption column behaviour. Geophysical Journal of the Royal Astronomical Society. 63. 117-148. WOHLETZ, K. H.. MCGETCHIN, T. R.. SANDFORD, M. T. & JONES. E. M. 1984. Hydrodynamic aspects of caldera-forming eruptions: numerical models. Journal of Geophysical Research. 89. 8269-8285. WOODS, A. W. 1988. The fluid dynamics and thermodynamics of eruption columns. Bulletin of Volcanology. 50. 169-193. WOODS, A. W. 1995. A model of Vulcanian explosions. Nuclear Engineering and Design. 155. 345-357. WYLIE. J. J.. VOIGHT, B. & WHITEHEAD, J. A. 1999. Instability of magma flow from volatile dependent viscosity. Science. 285. 1883-1885.

Hazard implications of small-scale edifice instability and sector collapse: a case history from Soufriere Hills Volcano, Montserrat S. R. YOUNG 1 , B. VOIGHT2, J. BARCLAY3, R. A. HERD4, J.-C. KOMOROWSKI5, A. D. MILLER 6 , R. S. J. SPARKS7 & R. C. STEWART8 1 Montserrat Volcano Observatory, Mongo Hill, Montserrat, West Indies 2 Department of Geosciences, Penn State University, University Park PA 16802, USA 3School of Environmental Sciences, University of East Anglia, Norwich NR4 7TJ, UK 4 British Geological Survey, Keyworth, Nottingham NG12 5GG, UK 5 OVS-IPGP, Le Houlement 97113, Guadeloupe, West Indies 6 Geowalks, 23 Summerfield Place, Edinburgh EH6 8AZ, UK 7 Earth Sciences Department, University of Bristol, Queen's Road, Bristol BS8 1RJ, UK 8 Preparatory Commission for the CTBTO, PO Box 1250, A-1400 Wien, Austria

Abstract: During the 1995 to 1998 phase of dome growth at Soufriere Hills Volcano on Montserrat, we documented instability of the steep southern rim of English's Crater, known as Galway's Wall. The horseshoe-shaped English's Crater provided good evidence for previous sector collapses, and assessments undertaken in late 1996 anticipated the possibility of a partial sector collapse and a SW-directed explosion, hazards previously unrecognized on Montserrat. A change from predominantly endogenous to exogenous growth of the lava dome at the end of 1996 eased the stress on the southern sector. However, rapid dome growth in November and December 1997 led to severe reloading and eventual sector failure at the base of the buried Galway's Wall and in the adjacent hot-spring area. This failure resulted in the debris avalanche and lateral blast of 26 December 1997. Similar sector collapses at a number of small volcanoes in the Caribbean, as well as worldwide, are evidence that edifice instability develops commonly in dome-forming eruptions. The hazards from a sector collapse and a consequent lateral blast are extreme, and monitoring operations and disaster planning at such volcanoes should focus on these, as well as on the more common hazards of conventional pyroclastic flows associated with dome growth.

The eruption of Soufriere Hills Volcano on Montserrat began with phreatic activity in July to November 1995, followed by 29 months of essentially continuous dome growth (Young et al. 1998). Eruption of highly viscous, crystal-rich andesite began slowly at first, but increased rapidly in mid-1996, leading to a series of dome collapses and eventually the explosive activity of September 1996 (Sparks et al 1998). Thereafter, a period of endogenous, and then exogenous, growth was followed by increases in mass eruption rate, leading to large dome collapses on 25 June, 3 August and 21 September 1997. The latter two collapse episodes were followed by cyclic Vulcanian explosions (Druitt et al. 2002). Rapid dome growth resumed in late October 1997 and culminated in the events of 26 December (Voight et al. 2002; Sparks et al. 2002). Dome growth resumed immediately after these events and continued at a high rate until early March 1998, when it abruptly stopped. Lava extrusion did not occur for 20 months, although mild Vulcanian explosions, gravitational degradation of the dome and lahar activity continued (Norton et al. 2002). Dome growth restarted in November 1999. Geological investigations of the volcano (Roobol & Smith 1998) suggest that its c. 170 ka history (Harford et al. 2002) comprised dome-forming eruptions with generation of pyroclastic flows from dome collapse and Vulcanian to sub-Plinian explosive activity, but no evidence for more violent explosive activity. The morphology of the summit area (English's Crater) and Tar River valley to the east (Figs 1 and 2) suggests a previous sector collapse. This collapse must have occurred prior to the Castle Peak dome eruption (c. AD 1620; Young et al. 1996; S. R. Young, unpublished data), but after the last major dome forming episode (i.e. before c. 16.5 ka; Wadge & Isaacs 1988), as the scar was partially filled by the Castle Peak dome but cut older domes. Roobol & Smith (1998) suggest that a pyroclastic flow deposit with a radiocarbon age of c. 4 ka may be associated with the sector collapse. Massive, fragmental deposits of hydrothermally altered material near the top of the pyroclastic sequence in cliffs at the mouth of Tar River valley are interpreted as remnants of the debris avalanche deposit from this sector collapse. Detailed analysis of recently collected offshore data (Deplus et al. 1999) may reveal further details of previous debris avalanche events. Initial results indicate the presence of at least three debris avalanche deposits

offshore of Montserrat, one off the east coast near the mouth of Tar River valley (Harford 2000). The 1995 to present volcanic activity has been monitored closely, with substantial geological and geophysical data sets generated. Once dome growth began, hazard assessment concentrated on pyroclastic flow and tephra fall impacts on the island. However, growing evidence was obtained during late 1996 for stressing of the steep southwestern edifice wall (Galway's Wall, Fig. 2) by intrusion of magma into or beneath the dome during a period when there was very slow dome extrusion. This led to a reappraisal of hazards to take into account the possibility for catastrophic edifice failure, including volcanic debris avalanche, tsunami and lateral blast phenomena. Easing of stress on the wall due to a renewal of exogenous growth, and changes of the locus of dome growth away from the southern sector, temporarily reduced the likelihood of catastrophic collapse. A further switch in dome growth in November 1997 led to significant loading of dome and talus near the Galway's Soufriere area and again increased the likelihood of sector collapse. This collapse duly occurred on 26 December 1997 producing a debris avalanche with c. 4 km runout, and a high-energy pyroclastic density current (Sparks et al. 2002). The impact area covered 10 km 2 , and two villages on the southwestern coast of Montserrat were completely destroyed. An upsurge of interest in volcano collapse structures, especially prompted by the 18 May 1980 eruption of Mount St Helens (Lipman & Mullineaux 1981), has led to the recognition of many debris avalanche scars and deposits at various volcano types (see Siebert (1984, 1996) and McGuire (1996) for reviews). Large-scale flank instability is well documented at shield volcanoes, e.g. Hawaii (Denlinger & Okubo 1995) and at strato-volcanoes of various sizes, e.g. Mount St Augustine (Beget & Kienle 1992), Colima (Stoopes & Sheridan 1992) and Socompa (Wadge et al. 1995). Earthquakeinduced catastrophic failure of the flank of a composite dome (Unzen-dake) in Japan in 1792 led to formation of a tsunami and the death of 14500 people (Siebert et al. 1987). Sector collapse scars have also been investigated at a number of Caribbean volcanoes, e.g. Qualibou St Lucia (Roobol et al. 1983; Mattioli et al. 1995), Dominica and St Vincent (Roobol et al. 1983),

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 349-361. 0435-4052/02/$15 © The Geological Society of London 2002.

349

S. R. YOUNG ET AL.

350

62°10'W

N

St. Peter's

16°45'N -

Long Ground (MBLG)

English's Crater Chances Peak EDM target

Collapse Scar (1997) White River valley

la Soufriere Guadeloupe (Boudon et al 1984, 1987) and Mont Pelee. Martinique (Vincent & Bourdier 1989). The abundant evidence for collapse features suggests that small, composite dome volcanoes in the region are prone to flank instability (Boudon et al. 1999). Debris avalanche and associated hazards are reviewed by Siebert (1992, 1996) and Siebert et al. (1987). The mechanisms and mechanics of collapse are reviewed by Voight & Elsworth (1997). We present here information relating directly to the identification and monitoring of sector instability at Soufriere Hills Volcano in late 1996, eventually culminating in the catastrophic collapse of the southern sector on 26 December 1997 (Sparks el al 2002). This collapse led to the generation of a debris avalanche and was followed by a violent pyroclastic density current and a small tsunami. In view of the devastation produced by these events, we stress the need for careful consideration of sector collapse hazards for all dome-forming volcanoes.

Galway's Soufriere

Fig. 1. Map of Montserrat with locations mentioned in the text. Grey shading shows area of Figure 2. Solid black horseshoe is the English's Crater collapse scar: dashed black horseshoe is the December 1997 collapse scar. Contour heights are in feet (1 foot = 0.3048m).

Evidence for the nature of the volcanic edifice and Galway's Wall The pre-eruption morphology of the sector collapse scar at Soufriere Hills Volcano, named English's Crater (Fig. 2). was such that there were a number of old lava domes acting as buttresses to the "walls' that lay between them. Whilst the domes mainly comprise massive andesite, the walls comprise more fragmental material, mainly cemented pyroclastic flow and talus-apron deposits with tephra layers. Dense tropical vegetation obscured outcrops prior to the onset of the eruption, but the stratigraphy of the edifice was observable from early in 1996. Phreatic eruptions and acid gases stripped vegetation, revealing surface morphologies, and during December 1995, a rock avalanche occurred from the inside wall of the scar, exposing its fragmental and altered nature.

SECTOR COLLAPSE HAZARD

351

English's Crater \collapse scar

Gages Wall 1 Oct 96 lobe y 95-96 growth)

Fig. 2. Map of the summit area of Soufriere Hills Volcano (English's Crater and Chances Peak), as of late December 1996, with features mentioned in the text. A, site of the high-gain tiltmeter near the summit of Chances Peak; B, site of the extensometer and low-gain tiltmeter. Cracks I and II are marked; the Galway's Mountain crack lies a few tens of metres off the map to the SE. Map coordinates on the Montserrat grid are shown

During early 1996, field investigations and laboratory testing (Voight 1996), as well as computer modelling (Wadge et al 1996), concentrated on assessing two portions of the wall: a narrow sector that acted as a partial barrier to dome collapses into the capital, Plymouth (Gages Wall), and a wider and thinner sector to the south known as Galway's Wall (Fig. 2). Galway's Wall consisted of a narrow ridge about 400m in length, which connected the domes of Chances Peak and Galway's Mountain. The wall was precipitous on both sides, with a nearvertical inner face to English's Crater and an outer, southward face with an angle of 50°. The narrow ridge broadened into shoulders abutting the two old domes at either end (Fig. 3a). Galway's Soufriere, an active area of hot springs c. 500 m south of the base of Galway's Wall in the upper part of the White River valley, was particularly important to assessments of the stability of the wall. This extensive (0.1 km2) area of strongly hydrothermally altered rocks and hot springs had been unstable throughout the earliest stages of the eruption. The road leading to the soufriere began to show signs of slippage in late 1995, and sampling of the hot spring itself was made difficult at the same time by active landslides within the soufriere area. Movements of a metre or more occurred on slope-parallel slip planes crossing the road by late 1996 (Fig. 3b). Deformation at Galway's Wall Through the first six months of dome growth, activity was concentrated adjacent to Castle Peak dome (Fig. 2), with lava slowly

me filing with main axis

Galway's Wall filling English's Crater and spilling outward to the east. Galway's Wall began to be encroached upon by the growing dome in June 1996, when the focus of growth switched toward the south. Talus piled up against the inner side of Galway's Wall only to a height of a few metres before activity switched again to the east and NE in July 1996. No effects of lateral pressure of the lava dome on the wall were observed at this time. After the dome collapse and explosive activity on 17 September 1996 (Robertson et al. 1998), dome growth recommenced on 1 October and rapidly filled the new scar (Fig. 2). A strong volcanotectonic earthquake swarm beneath the volcano occurred on 1 November. This swarm initiated an intense, eight-day hybrid earthquake swarm, which, in turn, marked the start of a six-week period of hybrid earthquake swarm seismicity (Miller et al. 1998). The rate of dome growth diminished slowly thereafter until exogenous growth ceased on 26 November. Helicopter- and groundbased dome surveys undertaken in early December, coincident with an intense period of hybrid earthquake activity, showed endogenous swelling of an elongate area (c. 800 x 300 m) in the southern sector of the dome that had largely been emplaced in June 1996. This swelling caused vertical height changes of up to 100m adjacent to Galway's Wall (Fig. 2) and c. 2.5 x 106 m3 of lava was intruded at this time (Young et al. 1997). Swelling ceased on or before 13 December, approximately synchronous with the end of the hybrid earthquake swarms and with a switch to exogenous dome growth in the northern and eastern sectors. Deformation of Galway's Wall was first manifest when rock avalanches of superficial soil and weathered rock from the south

Fig. 3. (a) Galway's Wall, viewed from the south in late November 1996. Chances Peak is to the left (with communications mast) and Galway's Mountain to the right. Debris from rock avalanches can be seen at the base of the wall and several cracks are also visible (labelled), (b) Faulting across the access road to Galway's Soufriere, December 1996. (c) Large cold-rock avalanche from the face of Galway's Wall on 26 November 1996. The scar area is indicated near the top of the wall, and dust is seen where the first blocks have reached the base of the wall. This landslide was triggered by a large hybrid earthquake, felt by the photographer, 1 km to the south, (d) Crack II on the southern buttress of Chances Peak, just after its first sighting in early December 1996. A parallel crack, Crack I, was sighted a few days earlier. The width of the foreground of the photo is about 3 m. (e) Slip plane exposed at the back of Galway's Soufriere in December 1996 (indicated by arrows). Note also surficial landsliding around the soufriere area.

S. R. YOUNG ET AL.

354

face of the wall occurred in late October 1996. The installation of a broadband seismic network enabled refined monitoring of these avalanches through analysis of the signals they generated at nearby stations (see below). Degradation of the wall ranged from short periods (minutes) of near-continuous avalanching of friable material in gullies or from cracks that opened up on the wall, to discrete slab avalanches involving tens to 100000m 3 of rock (Fig. 3c). A talus apron soon grew at the base of the wall and one large rock avalanche developed a runout of over a kilometre. Avalanches could be correlated visually with individual hybrid earthquakes. The larger earthquakes could be felt, and their seismic signals heard over a scanner, at an observation post 1200m from the dome summit. The effects of the shaking could be seen over the whole wall; several avalanches might occur simultaneously, whilst dust was thrown up from numerous small cascades of rock. In addition to face-parallel and radial cracks in Galway's Wall itself, two large cracks (cracks I and II) were noted on the southeastern buttress of Chances Peak from late November (Figs 2 and 3d). These cracks extended across the buttress and one could also be seen on the inside wall of English's Crater with a vertical exposure of c. 50m. The cracks were oriented NE to ENE, approximately radial to the lava dome, and also parallel to a pre-existing structural grain (S. R. Young and others, unpublished data). Deformation at Galway's Soufriere in late 1996 was characterized by the development of a NW-SE oriented slide scarp along the northern part of the soufriere, along which its central portion dropped 1.5 to 2m (Fig. 3e). The nature of this slip plane appeared to be consistent with shallow slope instability rather than deepseated structure, although close inspection of this feature was not possible. Other manifestations of slope instability included an increased rate of landsliding within the soufriere and movement of cracks across the access road (see above).

Monitoring the instability Monitoring of the deformation of Galway's Wall was undertaken by the Montserrat Volcano Observatory (MVO) using various methods. These included seismic, deformation and visual monitoring. Safety considerations after October 1996 ruled out close visual observations and inspection, and placing of monitoring equipment on the wall itself.

Seismic monitoring Deformation at Galway's Wall began soon after the installation of a digital, mainly broadband, seismic network. The optimized network configuration and large bandwidth significantly improved discrimination of event types. Data from the older short-period network were also utilized, especially real-time seismic amplitude measurement (RSAM; Endo & Murray 1991). Event type counts (Fig. 4) were routinely utilized in appraisal of the status of the volcano. A novel method was developed to discriminate hot rockfalls from the lava dome and cold rock avalanches from the outside face of Galway's Wall (see below). Four main event types were discriminated, namely hybrid, volcanotectonic, long-period and rockfall events (Miller et al. 1998). Hybrid earthquakes dominated the activity during late 1996 (White el al 1998). A number of workers (e.g. Chouet 1996; Sparks 1997) have proposed that such hybrid events represent pressurization within the dome and upper conduit, and swarms of hybrid earthquakes were often clearly associated with cyclic dome pressurization and collapse (e.g. Voight el al. 1998, 1999). Hybrid earthquakes during late 1996 occurred in very well defined swarms (Fig. 5), although no deformation monitoring to record pressurization was in place until the end of this swarm activity. An increase in maximum amplitude for hybrid events (averaged over a 2-minute moving window) was noted through November and December 1996 (Fig. 5). Wave-form analysis of hybrid swarms and locations of the larger earthquakes in each swarm suggested a consistent source area at a shallow depth (1-4 km) beneath the dome. The wave-forms for many events were near-clones of one another, with several 'standard forms' being replicated. This suggested that the process was non-destructive and that nearly identical source processes were being repeated at several specific source sites. Thus maximum earthquake amplitude is a proxy for magnitude. It is estimated that the largest hybrid earthquakes had magnitudes of 3.5. Volcanotectonic earthquakes dominated the early, phreatic phase of the eruption (Aspinall et al. 1998) but were infrequent thereafter. A short, intense burst of volcanotectonic earthquakes at depths of c. 2-4 km beneath the dome occurred early in November 1996 within a hybrid swarm and prompted a change in style from small, closely spaced hybrid swarms to longer, more widely spaced ones. Apart from this 4-hour episode, volcanotectonic activity in late 1996 was diffuse. Long-period earthquakes were rare during

Galway's Wall avalanches

Long Period earthquakes

l-Nov-96 10-Nov-96

20-Nov-96

30-Nov-96

10-Dec-96

20-Dec-96

30-Dec-96

Fig. 4. Counts of earthquakes, dome rockfalls and Galway's Wall avalanches for Soufriere Hills Volcano during late 1996 and early 1997. The different earthquake types are described in the text. Shaded regions highlight correlations and anti-correlations.

SECTOR COLLAPSE HAZARD

Fig. 5. Temporal changes in amplitude of hybrid earthquakes located beneath the growing lava dome, plotted on a log scale as a proxy for magnitude (assuming fixed location; see text for details). Note the steady increase in the maximum magnitude of events and the bimodality of magnitudes as shown by the grey regions.

26-Oct-96

5-Nov-96

late 1996. These earthquakes were often associated with shallow degassing within the dome (Miller et al. 1998). Rockfall or rock avalanche activity is the last of the main group of seismic signals differentiated. Most rockfalls were caused by gravitational instability of the lava dome. Due to the rapid attenuation of these surface waves, the signal amplitude at different stations varied significantly as a function of the location of rockfall activity. Thus, during some periods of dome growth, the focus of activity could be pinpointed by the seismic network. Avalanches of cold rock from the outside face of Galway's Wall generated similar signals to dome rockfalls, but their occurrence could routinely be detected utilizing signal amplitude ratios at different stations. Visual confirmation of this method enabled real-time estimates of the rate and relative magnitude of rock avalanche activity on Galway's Wall throughout this period. However, it proved impossible to make the detailed field measurements needed to estimate rock avalanche volumes, so this precluded developing an empirical relationship between signal amplitude and duration and rock avalanche volume. The relative numbers of Galway's Wall rock avalanches versus dome rockfalls have been recalculated for the period from late October to late December 1996 (Fig. 4) using systematic maximumamplitude comparisons at two seismic stations: Long Ground (MBLG) and Galway's Estate (MBGE) broadband stations (Fig. 1). Threshold criteria for differentiation were set on the basis of observed Galway's Wall rock avalanche and dome rockfall events. Dome rockfalls produced a signal much larger in amplitude at Long Ground than at Galway's Estate, whilst Galway's rock avalanches produced larger signals at the Galway's Estate station. This method produced more consistent results than another, operator-influenced method used at the time (Young et al. 1997), although there are no significant discrepancies between inferences from the real-time and recalculated data sets. In the November to mid-December period, a strong anticorrelation was observed between hybrid earthquake swarms and rockfalls from the lava dome, whilst a correlation developed between hybrid earthquake swarms and cold rock avalanches from Galway's Wall (Fig. 4). Visual observations suggested that the cold rock avalanches were clearly the result of physical shaking by hybrid earthquakes (Fig. 3c). In May 1997, hybrid events produced surface accelerations as much as 0.2g on Chances Peak (B. Voight, unpublished data), and these events may not have been as strong as those in December 1996. Strong shaking also produced occasional rockfalls on the lava dome, but apparently Galway's Wall was more sensitive to these earthquakes, perhaps because of efficiency of wave transmission or wave-focusing effects.

355

15-Nov-96

25-Nov-96

5-Dec-96

15-Dec-96

Date

Deformation monitoring Local deformation around the wall of English's Crater was monitored by electronic tiltmeters and extensometer, by manual tape measurements of crack displacement and by some microgravity monitoring. Electronic distance meter (EDM) measurements were also attempted at this time to targets on Galway's Wall and Chances Peak, but were largely unsuccessful due to equipment and access limitations and poor atmospheric conditions. Two tiltmeter stations were installed on Chances Peak to monitor ground tilt within a few hundred metres of the growing dome and Galway's Wall. A high-gain Geomechanics tiltmeter with realtime telemetric links was installed close to the summit of Chances Peak in early December (Fig. 2, location A). In late December, a low-gain tiltmeter was installed together with an extensometer (Fig. 2, location B). The extensometer measured extension across Crack II, the westernmost of the two cracks that developed on the shoulder of Chances Peak (Figs 2 and 3d). The high-gain tiltmeter recorded ground tilt from 9 December 1996 until the end of February 1997 (Voight et al. 1998). Although instrumental noise is considerable, a period of strongly cyclic tilt can be identified between 13 and 19 December 1996, coincident with a period of vigorous exogenous dome growth (Fig. 2, '13 December dome'). The radial axis of the tiltmeter best illustrates the inflation and deflation of Chances Peak in response to pressurization during this period. Figure 6 shows the relationship between these tilt cycles and RSAM measurements. RSAM peaks in this case correspond to vigorous rockfall activity that occurred at or just before the maximum tilting; during this period, hybrid earthquakes were suppressed. The cycles are approximately 24 hours long, somewhat longer than those typically observed in 1997 (Voight et al. 1999). Manual crack measurements were undertaken, whenever conditions permitted, from 4 December 1996 to June 1997. Typically the measurements involved stretching a tape between nails driven into several trees on both sides of the crack. Crack I the more easterly crack (Fig. 2), became too dangerous to measure in early February 1997 and the area avalanched later that month. Crack II was measured manually and/or by electronic extensometer between early December 1996 and late June 1997. Figure 7 shows the displacement data for both cracks. A maximum of 10 days of crack opening preceded our first measurements, as clear observations of the area around 24 November revealed no evidence for surface cracks at that time. The rate of movement must have been relatively high for this initial 10-day period, given the degree of opening of the cracks already evident on 4 December. Rapid opening continued up to

356

S. R. YOUNG ET AL.

Fig. 6. Tiltmeter and RSAM records for the mid-December period, showing cycles of dome inflation (tilt peaks) and deflation (troughs) coinciding with RSAM peaks and troughs. Tiltmeter also shows overall inflationary trend. The cyclicity breaks down after the dome collapse explosive activity of 19 December. Times are local (GMT minus 4 hours). RSAM is the average of RSAM counts from seismic stations MGHT and MSPT. Tiltmeter is at Chances Peak (A on Fig. 2): x-axis (shown) is oriented radially to the dome.

the reinitiation of exogenous dome growth on 13 December. The opening rates of the cracks then decreased before accelerating again after mid-January 1997. The cracks opened at a rapid rate between mid-January and early March, and then remained more or less stable to June 1997, apart from limited widening in early May. A new crack was noted to the east of Galway's Wall on Galway's Mountain (Fig. 2) in early March 1997; measurements across this crack showed little consistent movement until readings ceased in June. The appearance of this crack coincided with vigorous dome growth in the southern sector, which occurred in March and April 1997 (Watts et al 2002). EDM measurements were attempted both to a previously fixed target on the southern flank of Chances Peak and to temporary targets on the northern buttress of Galway's Wall. Only one measurement to the temporary target was possible before it was lost due to avalanching. Measurements to the Chances Peak target from Galway's Estate (1.6km to the SW, MBGE on Fig. 1) were sparse due to poor field conditions and limitations of the instrument to shoot longer lines. However, a line-length shortening of 26 mm was recorded on this line between mid-October and early December 1996 (a rate of c, 215 mm a - 1 ) . The shortening rate reduced to

c. 40 mm a-1 from mid-December, similar to that recorded during early 1996. Visual monitoring Visual monitoring from a helicopter and from observation posts on the ground provided day-to-day qualitative information concerning the state of the lava dome and the upper edifice, and also enabled confirmation of the meaning of different types of seismic signals and the correlation of seismicity with rock avalanches. Qualitative assessment of pore fluids was made by visual observation of water seeps from Galway's Wall (which were rare and spatially scattered) and of water output at Galway's Soufriere (which remained stable and low), and proved useful in the absence of quantitative information. The unsteady build-up to sector collapse Seismic, deformation and visual observation data suggest that the deformation of Galway's Wall substantially decreased after the

Fig. 7. Crack extension data for the period from 1 December 1996 to 29 June 1997. Manual measurements of opening for Cracks I (grey squares) and II (black diamonds) are shown as individual points. Note the order of magnitude scale change between opening on these cracks. The continuous record for the electronic extensometer across Crack II is also shown, with a break during a period of telemetry malfunction.

SECTOR COLLAPSE HAZARD

onset of rapid, cyclic dome extrusion first observed on 13 December 1996. Dome rockfall activity increased over the next six days (Fig. 4) until the afternoon of 19 December, when the dome collapsed and formed a series of pyroclastic flows, which included pumice lapilli, in the Tar River valley. There were also some rock avalanches from Galway's Wall between 18 and 23 December (Fig. 4). Dome growth continued in the eastern sector at a high rate, accompanied by voluminous dome collapses and generation of pyroclastic flows towards the NE between 9 and 22 January (Young et al. 1998). Although substantial rock avalanche activity did not recur from Galway's Wall prior to its overtopping in March 1997, crack data (Fig. 7) suggest that stressing of the wall restarted in late January 1997 and continued until mid-March, when dome growth switched back to the SW sector. The late January to mid-March 1997 period was characterized by a lower dome growth rate (Sparks et al. 1998), some visual evidence for endogenous swelling, and intense hybrid earthquake activity (Miller et al. 1998). The dome talus reached the height of the top of the degraded Galway's Wall in February 1997. Dome rockfalls began to spill over Galway's Wall, eroding the top of the wall and cutting a deep notch in it (see Sparks et al. 2002, fig. 2a). The Galway's Soufriere area was progressively buried by block-and-ash flow deposits associated with these rockfalls and larger dome collapses at the end of March and in early April 1997 (Young et al. 1998). Then, beginning on 17 May, dome growth switched for a prolonged period away from the southern sector towards the NE. Dome growth in the southern sector resumed in early November 1997, with significant collapses generating block-andash flows down the White River valley on 4 and 6 November. Thereafter, new lava was extruded over the talus that buried Galway's Wall (see Voight et al. 2002, fig. 6). Dome growth at this time was vigorous (>5m 3 s - 1 ), although it slowed during the later part of December (Sparks et al. 1998). The dome reached its maximum elevation (c. 1030m above sea level) for any time during this first phase of the eruption (before November 1999), and loading of the dome and talus in the southern sector led to a condition that was increasingly unstable. Seismic activity was generally low, although large hybrid earthquakes occurred in late October and at the beginning of November. A gradual build-up in hybrid earthquake energy began on 22 December, and developed into a hybrid swarm on 24 December. The swarm included periods when individual events merged into continuous tremor. With increasing edifice loading and increasing seismicity, the potential for sector collapse was enhanced to levels similar to those late in 1996, although the geometry of the dome and edifice complex was now substantially different. Sector collapse took place at 07:01 GMT (03:01 local time) on 26 December (Sparks et al. 2002). A volcanic debris avalanche was generated by a landslip that involved the edifice between Galway's Soufriere and parts of Galway's Wall, and included substantial amounts of dome talus (Voight et al. 2002). The sector collapse facilitated rapid depressurization of the lava dome and explosive initiation of a high-energy pyroclastic density current (Woods et al. 2002; Ritchie et al. 2002). The entire southwestern sector of the volcano was devastated, involving an area of 10 km2 (Sparks et al. 2002).

Hazard management and anticipation of critical events Detailed hazard maps and hazard management on Montserrat concentrated until September 1996 on pyroclastic flow and ashfall hazards. The explosive activity of 17 September 1996 led to a reappraisal of volcanic hazards to include vertical explosive eruption phenomena. However, it was not until the onset of deformation of the southern edifice wall that assessment of sector collapse and associated lateral explosion hazards was made. The likelihood of sector collapse was informally assessed by MVO staff in late November 1996 as representing approximately a 30% probability for collapse within a three-month period. A revised risk map was

357

issued to the public on Montserrat at that time (Kokelaar 2002), with the zone of highest risk being around the southwestern flanks of the volcano and in the White River valley. A small but resolute population had remained in the village of St Patrick's (Fig. 1) until late November, but, with the signs of instability visible, they were persuaded to evacuate. A formal assessment of the probability of sector collapse using techniques described by Aspinall & Cooke (1998) was made at MVO on 21 December 1996. The likelihood of a sector collapse and subsequent lateral blast in the Galway's sector was then assessed at 15% in three months, suggesting a reduced risk in comparison to the previous assessment. An identical result was found in a subsequent formal elicitation on 7 March 1997. Identification of the potential hazard prompted efforts to upgrade the monitoring systems specifically to provide warning of an imminent sector collapse. Five real-time indicators were watched closely for signs of rapid increase: RSAM, hybrid earthquake magnitude, frequency of occurrence of Galway's Wall rock avalanches, crack extension and inflationary ground tilt. Earthquake types and locations were also closely monitored in real time. The upgraded systems proved very useful, but were recognized as not necessarily definitive. The problems of instrument site accessibility, the risk to staff involved in monitoring, and the limitations of available instrumentation, were severe and provided practical limits to what could be accomplished. A major managerial problem remained in that confident short-term (hours to days) warning for a sector collapse could not be guaranteed. Limit-equilibrium models were conducted in December 1996 for the purpose of investigating the stability of the Galway's Wall and Chances Peak area, and understanding the nature and significance of the rock avalanches that were occurring at that time (B. Voight, unpublished data). The analyses suggested that the shallow-slab failures of the wall that generated rock avalanches were primarily due to the dynamic loading imposed by the shallow strong hybrid earthquakes, rather than due to any fundamental instability of the wall itself. Loading from the static weight of dome rock that then existed behind the wall was viewed as subcritical, although a possible future influence of more complicated pressure patterns due to endogenous behaviour or changes in edifice geometry could not be precluded. The results were communicated to MVO by telefax on 14 December. Finite-element modelling of the Galway's Wall area was undertaken in late 1996 at the request of MVO (Wadge et al. 1998) in order to obtain some insight into possible failure mechanisms for that sector. Several models were tested, including a pressurized conduit model and one involving direct loading by the lava dome. All models suggested that the lower part of the wall and its base around Galway's Soufriere were the most likely failure points, should failure occur as a result of conduit pressure or dome loading. The models were necessarily idealized and homogeneous materials were assumed. The initial delay in identifying the hazard, the many months required to construct and run the models, and the difficulties in communication of the findings, all conspired to decrease the value of the modelling to real-time hazard assessment. Secondary hazards were also considered. A preliminary assessment of the tsunami threat from a sector collapse was made at MVO, using approximate methods developed in the literature. An improved assessment was later made with the assistance of a team of French scientists. The French island of Guadeloupe was most at threat from any tsunami generation, and ties between scientific staff on Montserrat and Guadeloupe were strengthened at this time. Numerical modelling of potential tsunami generation was rapidly undertaken using estimates of possible debris avalanche volumes, runout distances and velocities (Heinrich et al. 1998). These models inferred potentially hazardous wave heights along parts of the coastline of Montserrat, should a collapse occur, and also suggested only very small wave heights in Guadeloupe. Other Caribbean islands were not expected to be affected, although they were alerted to the potential at an early stage through the regional disaster agency (CDERA). The models proved useful and small (1.5 to 2 m) waves did come ashore along the central west coast of Montserrat following the 26 December sector collapse (Calder et al.

358

S. R. YOUNG ET AL.

1998; Sparks el al. 2002). It is interesting to note, however, that the waves were not generated by the debris avalanche, as had been modelled, but by the main pyroclastic density current. Monitoring and visual observations during December 1996 had not revealed signs of rapid deterioration of the situation that might conceivably have led to a short-term forecast of sector collapse. Nevertheless, a partly explosive dome collapse did occur on 19 December after cessation of the main phase of deformation associated with the Galway's Wall and Chances Peak area. Established procedures developed by MVO, based on conventional eruption indicators, led to a red-alert advisory from MVO during this dome collapse (Aspinall et al. 2002). It is impossible to know how close Galway's Wall was to failure during this period. With the burial of Galway's Wall after March 1997 and the burial and devastation caused by the explosions of August to October 1997, the opportunities for effective monitoring were much diminished. The visual indications of deformation were effectively lost, along with the loss of proximal monitoring equipment and telemetry; sites available for effective new monitoring stations did not exist. Consequently, no specific deformation precursors to the sector collapse of 26 December 1997 could be identified. The only relevant precursory information came from the seismic network, with the build-up in hybrid earthquake activity over the 36 hours prior to the collapse event. This build-up was noted by MVO, although the nature of the culmination of the seismicity could not be known. Two other factors influenced MVO thinking at the time. The first was an awareness of long-term cyclicity (six to seven weeks) for

periods of heightened activity, evident since May-June 1997 (Voight et al. 1999). Indeed, the occurrence of a series of explosive events after mid-September 1997 had been anticipated a month earlier. The last 'peak' of intensive activity had been in early November, so that there was some expectation by MVO of heightened activity during the last week of the year. The second was the relationship between hybrid earthquake swarms and dome collapses (Miller et al. 1998; Voight et al. 1998). Often a period of increasingly intense hybrid swarm activity preceded a dome collapse and, at the time, no major event had occurred during the first such swarm. From these factors, MVO considered that another dome collapse and runout of pyroclastic flows down the southwestern flanks of the volcano were the most likely culmination of these precursors, probably occurring after a few days of distinctive hybrid earthquake swarms of growing intensity. In order to analyse retrospectively the precursor seismicity, an analysis of the RSAM record from the single-component Windy Hill seismic station (MWHZ. Fig. 1) was undertaken using the materials failure forecast method (FFM; Voight & Cornelius 1991). We used the graphical method of Cornelius & Voight (1995) at three different time points during the generally accelerating build-up of hybrid earthquake activity prior to 07:01 GMT (03:01 local time) on 26 December. The 10-minute RSAM values from the Windy Hill station were averaged to hourly data and the inverse then plotted against time (Fig. 8). We have chosen three points at which the RSAM record shows a distinctive trend that might have been interpreted as build-up towards a climactic event.

to

Fig. 8. Application of the materials failure forecast method for the period prior to the sector collapse of 26 December 1997. Each plot shows the inverse RSAM record for MWHZ. All times are GMT. (a) Record as at 19:00 on 24 December 97 with graphical solution of FFM from the previous 8 hours of data, (b) Record as at 17:00 on 25 December 97 with graphical solution of FFM from the previous 12 hours of data, (c) Record as at 00:00 on 26 December 97 with graphical solution of FFM from the previous 5 hours of data, (d) Record as at 00:00 on 26 December 97 with graphical solution of FFM from the previous 19 hours of data. The collapse of Galway's Wall actually occurred at 07:01 GMT (03:01 LT) on 26 December.

SECTOR COLLAPSE HAZARD

The first analysis is at 19:00 (all times GMT) on 24 December (Fig. 8a), when RSAM had been increasing steadily for 8 hours. Taking the least-squares fit of the previous nine hourly RSAM figures, a conservative failure criterion (infinite RSAM) is considered to be met at 07:30 on 25 December, about 23.5 hours prior to actual sector collapse. The second analysis point is at 17:00 on 25 December (Fig. 8b), when RSAM had been increasing sporadically for the previous 12 hours. Extrapolation of the least-squares fit of the previous 13 points predicts infinite RSAM at 12:30 on 26 December, about 5.5 hours after actual failure. The final analysis point is at 00:00 on 26 December, when RSAM had been generally increasing for 19 hours and rapidly increasing for the previous 5 hours. The least-squares fit for the rapid build-up (Fig. 8c) predicts failure at 00:30, whereas the same fit for the slower, 19 hour build-up (Fig. 8d) gives a prediction of failure within half an hour of the actual failure at 07:01 GMT. The FFM method as applied to volcanic edifice collapse (Voight 1988) is probably more straightforward when deformation data such as displacements (from global positioning system or EDM), crack monitoring, or tilt are available. These kinds of data were only available at Soufriere Hills for the late 1996 period, when no accelerating build-up towards a potential event occurred. The actual collapse occurred when only seismic data were available, and the analysis above demonstrates the difficulty in using FFM for predictions, based on a secondary, noisy monitoring parameter (in this case RSAM). However, judicious use of the methodology, with suitable appreciation of its frailties and limitations, can potentially assist in real-time, systematic assessment of complex data in certain situations.

Discussion Small-scale sector collapse with associated magma depressurization and lateral blast generation constitutes a significant potential hazard at most dome-forming volcanoes. Where the volcanic edifice is inherently weak, loading of crater walls or flank areas by dome growth can cause destabilization and the potential for a sector collapse that includes both cold and hot material. In the case of Soufriere Hills Volcano, stressing of a steep flank by a growing dome and loading of a hydrothermally altered area at the base of the flank led to a sector collapse. This collapse then prompted further catastrophic disintegration of much of the remaining dome. The hazard implications for sector collapses are severe. Emplacement of a devastating debris avalanche in proximal areas with probable distal lahars or tsunami waves can be followed immediately by generation of violent pyroclastic density currents. All these events can occur in sequence, with little or no direct warning. Detailed monitoring of the lava dome and edifice, emphasizing parameters bearing on stability, are required to provide any chance of qualitative or quantitative forecasting for such events. However, the identification of the hazard and the confirmation of symptoms of potential collapse should prompt evacuation of any population at risk as soon as possible.

Influences on hazard assessment Due to the paucity of historical information on sector collapse and lateral blast eruptions, appraisals of sector-collapse hazard for Montserrat were necessarily based primarily on the improved understanding gained from geological investigations during the eruption. Nevertheless, experience elsewhere was also considered. The 3100 year BP eruption of la Soufriere on Guadeloupe (Boudon et al. 1984, 1987) is a well documented example at a similar volcano, which was considered as a basis for hazard assessment. Likewise, the published (and personal) experience from Mount St Helens and Bezymianny provided analogues useful for perspective. The sector collapse and lateral blast of Mount St Helens was also

359

used to provide an example in public lectures on Montserrat, to inform the public on the concern for Galway's Wall in late 1996, and to generate public support for evacuation zoning. Appraisal of edifice stability by engineering criteria (Voight 1996; Wadge et al. 1996, 1998; B. Voight unpublished data) suggested that the southern sector was the most likely to fail, and it explained the causes of the rock avalanches of 1996. All of the theoretical analyses were limited by the modelling simplifications required, and by the uncertainties involving distribution of materials and their properties. Although the models were useful in providing perspectives on various mechanisms, they were not as reliable as the information obtained from systematic observations and monitoring. The hydrothermal alteration in the area of Galway's Soufriere was an important factor in hazard assessment and is likely to have been important to the final collapse geometry, and indeed to its occurrence. The role played by increased pore fluid pressure within the hydrothermal system (Lopez & Williams 1993; Day 1996) remains uncertain. However, the process of progressive sealing of the hydrothermal system (Boudon et al. 1998) is favourable for raising fluid pressure and thus weakening the area. Qualitative monitoring of pore fluids seeping from Galway's Wall was undertaken during late 1996, which assisted with hazard assessment at that time, but no quantitative methods were attempted. Nevertheless, the existence of saturation within parts of the edifice is considered to be significant as regards both mechanical stability (Wadge et al. 1998), and mobility of the debris avalanche following collapse (Voight et al 2002). Despite the small number of well documented small-scale sector collapses worldwide, estimates of the various collapse and hazard parameters made during hazard assessments on Montserrat in late 1996 proved reasonable in retrospect. Both the collapse volume and the extent of the devastated area on 26 December 1997 were similar to those forecast on hazard maps in December 1996. Tsunami generation by small-scale edifice collapse has been documented for only a few similar volcanoes, e.g. Mount St Augustine in Alaska (Beget & Kienle 1992; Waythomas & Waitt 1998) and Unzen-dake (Latter 1981; Siebert 1984), but modelling undertaken for Soufriere Hills demonstrated the potential threat. Again, the actual phenomenon occurred on a scale quite comparable to that forecast in hazard assessments.

Volcano monitoring applied to sector collapse Beginning in early 1996, the Galway's Wall area was recognized as susceptible to sector collapse should substantial dome growth continue. Preliminary stability assessments were made at this time, materials were sampled from the base of the wall (by S. R. Young and B. Voight), and strength tests were conducted (Voight 1996). By late 1996, the stability of the southern sector was clearly an issue, and we attempted a number of methods of monitoring deformation in this sector. The success of this venture cannot easily be assessed due to the significant changes in overall geometry and circumstances that occurred between this period and the time of eventual sector collapse at the end of 1997. We did, however, collect useful data by various methods. Ground deformation monitoring by tiltmeter and EDM, and crack measurements by both manual methods and extensometer, proved to be the most effective methods in providing quantitative data that could be indicative of an imminent failure. We also found that visual observations combined with innovative use of routine seismic data assisted in tracking landslide frequency from the unstable sector. An acoustic emission approach (e.g. Stateham & Merrill 1979) was not attempted, although its implementation in such a potentially hazardous situation would have been problematic. We believed that the operating seismic arrays provided adequate information of this general type. The eventual sector collapse was monitored most effectively by the seismic network, although the exact nature of the deformations and culminating volcanic activity could not be interpreted from the

360

S. R. YOUNG ET AL.

collected seismic data. Because the collapse and eruption occurred at night, no visual or photographic documentation was available (as at Mount St Helens) to link the seismic record to observed events. Finally, we found field assessment of the nature of the volcanic edifice and potential debris-avalanche path most useful in parameterizing the quantitative models of avalanche runout and tsunami formation. Detailed digital terrain models of base topography and changes in dome morphology assisted both in the estimation of potential collapse volume and, in the case of dome bulging, in the assessment of hazard potential.

Conclusions Sector collapse may represent the most severe volcanic hazard during many andesitic dome-forming eruptions. Debris avalanche, tsunami and either lateral or vertical volcanic explosions (or both) are likely, with potentially devastating consequences. The geological record of Caribbean volcanoes and the recent history of the Soufriere Hills eruption on Montserrat suggest that sector collapse is a common feature of dome-forming eruptions in the region and probably worldwide, not the rare event it was once considered. Identification of potentially weak sectors and investigation of their character should be a part of any hazard assessment, preferably prior to eruption onset. Monitoring of seismic and deformation signals related to sector instability was possible on Montserrat in late 1996, although definitive criteria for imminent sector collapse could not be ascertained. By the time that the sector collapse occurred a year later, changes in volcano morphology and danger levels for monitoring meant that only seismic signals were recorded. Education of the public and crisis management officials in the dangers of sector collapse proved particularly challenging. The lack of visible signs of a worsening situation gave no indication of the potential catastrophe to the casual onlooker. The inability of observatory staff to guarantee any warning added to the difficulties of management for sector collapse hazards. We are grateful to the many members of the MVO staff who assisted in collection and interpretation of data used in this work; we thank especially M. Davies for crack measurements, R. Luckett and L. Pollard for seismological analysis, D. Williams and L. Lynch for technical support, and B. Darroux, L. Luke, A. Grouchy and J. McMahon for logistical support. We also thank USGS-CVO for logistical, instrumental and cerebral support, especially R. Hoblitt and A. Lockhart. The French modelling group provided rapid assistance in hazard appraisal, and we thank J.-L. Cheminee (IPGP) and Y. Caristan (CEA) for support. Comments from our reviewers and editors greatly improved the manuscript. Successful hazard management was achieved with the support of crisis managers on Montserrat and thanks to the spirit of the Montserratian population. Funding for MVO was mainly provided by DFID. Important support for B.V. was provided by the US National Science Foundation. For relevant authors, published by permission of Director, BGS (NERC).

References ASPINALL, W. P. & COOKE, R. M. 1998. Expert judgement and the Montserrat Volcano eruption. In: MOSLEM, A. & BARI, R. A. (eds) Proceedings of the 4th International Conference on Probabilistic Safety Assessment and Management PSAM4, 13-18 September 1998, New York City, USA, 3, 2113-2118. ASPINALL, W. P., MILLER, A. D., LYNCH, L. L., LATCHMAN, J. L., STEWART, R. C., WHITE, R. A. & POWER, J. A. 1998. Soufriere Hills eruption, Montserrat 1995-1997: Volcanic earthquake locations and fault plane solutions. Geophysical Research Letters, 25, 3397-3400. ASPINALL, W. P., LOUGHLIN, S. C., MICHAEL, F. V. ET AL. 2002. The Montserrat Volcano Observatory: its evolution, organization, role and activities. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 71-91.

BEGET, J. E. & KIENLE, J. 1992. Cyclic formation of debris avalanches at Mount St Augustine volcano. Nature. 356. 701-704. BOUDON. G., SEMET, M. P. & VINCENT, P. M. 1984. Flank failure and directed blast at Soufriere. Guadeloupe. French West Indies. Geology, 12, 350-353. BOUDON, G.. SEMET, M. P. & VINCENT. P. M. 1987. Magma and hydrothermally driven sector collapses: the 3100 and 11 500 Y. B. P. eruptions of La Grande Decouverte (La Soufriere) Volcano. Guadeloupe. French West Indies. Journal of Volcanology and Geothermal Research, 33, 317-323. BOUDON, G., VILLEMANT, B., KOMOROWSKI. J.-C. ILDEFONSE. P. & SEMET, M. P. 1998. The hydrothermal system at Soufriere Hills Volcano, Montserrat (West Indies): Characterization and role in the on-going eruption. Geophysical Research Letters. 25. 3693-3696. BOUDON, G., KOMOROWSKI. J.-C.. SEMET, M.. LE FRIANT. A. & DEPLUS, C. 1999. Frequent volcanic flank-collapses in the Lesser Antilles Arc: Origin and hazards. EOS, Transactions of the American Geophysical Union, 80, F1142. CALDER, E. S., YOUNG. S. R.. SPARKS, R. S. J. ET AL. 1998. The Boxing Day Collapse, 26 December 1997. MVO Special Report. 6. CHOUET, B. A. 1996. Long-period volcano seismicity: its source and use in eruption forecasting. Nature, 380, 309-316. CORNELIUS, R. R. & VOIGHT, B. 1995. Graphical and PC-software analysis of volcano eruption precursors according to the Material Failure Forecast Method (FFM). Journal of Volcanology and Geothermal Research, 64, 295-320. DAY, S. J. 1996. Hydrothermal pore fluid pressure and the stability of porous, permeable volcanoes. In: MCGUiRE. W. J.. JONES, A. P. & NEUBERG. J. (eds) Volcano Instability on the Earth and Other Planets. Geological Society. London, Special Publications. 110, 77-93. DENLINGER, R. P. & OKUBO. P. 1995. Structure of the mobile south flank of Kilauea volcano, Hawaii. Journal of Geophysical Research, 100. 24499-24507. DEPLUS, C.. FRIANT, A. L., BOUDON, G. ET AL. 1999. Large-scale debris avalanche deposits on the western flank of the Lesser Antilles arc revealed by the Aguadomar Cruise. EOS, Transactions of the American Geophysical Union, 80, F1152. DRUITT, T. H.. YOUNG, S. R.. BAPTIE. B. ET AL. 2002. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano. Montserrat. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London. Memoirs. 21, 281-306. ENDO, E. T. & MURRAY, T. 1991. Real time seismic amplitude measurement (RSAM): a volcano monitoring and prediction tool. Bulletin of Volcanology, 53, 533-545. HARFORD, C. L. 2000. The Volcanic Evolution of Montserrat. PhD thesis. University of Bristol. HARFORD. C. L., PRINGLE, M. S., SPARKS. R. S. J. & YOUNG. S. R. 2002. The volcanic evolution of Montserrat using 40 Ar 39 Ar geochronology. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London, Memoirs. 21. 93-113. HEINRICH, P., MANGENEY, A., GUIBOURG, S., ROCHE, R., BOUDON, G. & CHEMINEE, J.-L. 1998. Simulation of water waves generated by a potential debris avalanche in Montserrat. Lesser Antilles. Geophysical Research Letters, 25, 3697-3700. KOKELAAR, B. P. 2002. Setting, chronology and consequences of the eruption of Soufriere Hills Volcano. Montserrat (1995-1999). In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21, 1-43. LATTER, J. H. 1981. Tsunamis of Volcanic Origin: Summary of Causes, with Particular Reference to Krakatoa. 1883. Bulletin of Volcanology. 44, 467-490. LIPMAN, P. W. & MULLINEAUX. D. R. 1981. The 1980 Eruptions of Mount St. Helens, Washington. United States Geological Survey. Professional Papers 1250. LOPEZ, D. L. & WILLIAMS, S. N. 1993. Catastrophic volcanic collapse: relation to hydrothermal processes. Science. 260, 1794-1796. McGuiRE, W. J. 1996. Volcano instability: a review of contemporary themes. In: MCGUIRE, W. J.. JONES, A. P. & NEUBERG, J. (eds) Volcano Instability on the Earth and Other Planets. Geological Society, London, Special Publications, 110. 1-23.

SECTOR COLLAPSE HAZARD MATTIOLI, G. S., JANSMA, P. E., JARAMILLO, L. & SMITH, A. L. 1995. Sector collapse in island arc volcanoes: A digital topographic and bathymetric investigation of the Qualibou depression, St Lucia, Lesser Antilles. Caribbean Journal of Science, 31, 163-173. MILLER, A. D., STEWART, R. C, WHITE, R. A. ET AL. 1998. Seismicity associated with dome growth and collapse at the Soufriere Hills volcano, Montserrat. Geophysical Research Letters, 25, 3401-3404. NORTON, G. E., WATTS, R. B., VOIGHT, B. ET AL. 2002. Pyroclastic flow and explosive activity at Soufriere Hills Volcano, Montserrat, during a period of virtually no magma extrusion (March 1998 to November 1999). In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 467-481. RITCHIE, L. J., COLE, P. D. & SPARKS, R. S. J. 2002. Sedimentology of deposits from the pyroclastic density current of 26 December 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 435-456. ROBERTSON, R., COLE, P., SPARKS, R. S. J. ET AL. 1998. The explosive eruption of Soufriere Hills Volcano, Montserrat, West Indies, September 17 1996. Geophysical Research Letters, 25, 3429-3432. ROOBOL, M. J. & SMITH, A. L. 1998. Pyroclastic stratigraphy of the Soufriere Hills volcano, Montserrat: Implications for the present eruption. Geophysical Research Letters, 25, 3393-3396, ROOBOL, M. J., WRIGHT, J. V. & SMITH, A. L. 1983. Calderas or gravityslide structures in the Lesser Antilles island arc. Journal of Volcanology and Geothermal Research, 19, 121-134. SIEBERT, L. 1984. Large volcanic debris avalanches: characteristics of source areas, deposits, and associated eruptions. Journal of Volcanology and Geothermal Research, 22, 163-197. SIEBERT, L. 1992. Threats from debris avalanches. Nature, 356, 658-659. SIEBERT, L. 1996. Hazards of large volcanic debris avalanches and associated eruptive phenomena. In: SCARPA, R. & TILLING, R. (eds) Monitoring and Mitigation of Volcanic Hazards. Springer, Berlin, 541-572. SIEBERT, L., GLICKEN, H. & UI, T. 1987. Volcanic hazards from Bezymianny- and Bandai-type eruptions. Bulletin of Volcanology, 49, 435-459. SPARKS, R. S. J. 1997. Causes and consequences of pressurisation in lava dome eruptions. Earth and Planetary Science Letters, 150, 177-189. SPARKS, R. S. J., YOUNG, S. R., BARCLAY, J. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills Volcano, Montserrat, West Indies: November 1995 to December 1997. Geophysical Research Letters, 25, 3421-3424. SPARKS, R. S. J., BARCLAY, J., CALDER, E. S. ET AL. 2002. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 409-434. STATEHAM, R. M. & MERRILL, R. H. 1979. Acoustic emission techniques applied to slope stability problems. In: VOIGHT, B. (ed.) Rockslides and Avalanches, 2: Engineering Sites. Elsevier Scientific, Amsterdam, 577-593. STOOPES, G. R. & SHERIDAN, M. F. 1992. Giant debris avalanches from the Colima Volcanic Complex, Mexico: implications for long runout landslides (>1OO km) and hazard assessment. Geology, 20, 299-302. VINCENT, P. M. & BOURDIER, J. L. 1989. The primitive volcano of Mt Pelee: its construction and partial destruction by flank collapse. Journal of Volcanology and Geothermal Research, 38, 1-16. VOIGHT, B. 1988. A method for prediction of volcanic eruptions. Nature, 332, 125-130.

361

VOIGHT, B. 1996. Strength of wall rock of English's Crater, Soufriere Hills Volcano, Montserrat. In: WADGE, G. (ed.) The Soufriere Hills Eruption, Montserrat (Abstracts). Geological Society of London, 27 Nov 1996,29. VOIGHT, B. & CORNELIUS, R. R. 1991. Prospects for eruption prediction in near real-time. Nature, 350, 695-698. VOIGHT, B. & ELSWORTH, D. 1997. Failure of volcano slopes. Geotechnique, 47, 1-31. VOIGHT, B., HOBLITT, R. P., CLARKE, A. B., LOCKHART, A. B., MILLER, A. D., LYNCH, L. L. & McMAHON, J. 1998. Remarkable cyclic ground deformation monitored in real time on Montserrat and its use in eruption forecasting. Geophysical Research Letters, 25, 3405-3408. VOIGHT, B., SPARKS, R. S. J., MILLER, A. D. ET AL. 1999. Magma flow instability and cyclic activity at Soufriere Hills volcano, Montserrat, British West Indies. Science, 283, 1138-1142. VOIGHT, B., KOMOROWSKI, J.-C., NORTON, G. E. ETAL. 2002. The 26 December (Boxing Day) 1997 sector collapse and debris avalanche at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 363-407. WADGE, G. & ISAACS, M. C. 1988. Mapping the volcanic hazards from Soufriere Hills Volcano, Montserrat, West Indies using an image processor. Journal of the Geological Society, London, 145, 541-551. WADGE, G., FRANCIS, P. W. & RAMIREZ, C. F. 1995. The Socompa collapse and avalanche event. Journal of Volcanology and Geothermal Research, 66, 309-336. WADGE, G., WILLIAMS, C. & VOIGHT, B. 1996. Stability of the Gages Wall MVO Open-file Report 96/11. WADGE, G., WOODS, A. W., JACKSON, P., BOWER, S., WILLIAMS, C. A. & HULSEMANN, F. 1998. A hazard evaluation system for Montserrat. In: IDNDR, U. N. C. C. (ed.) Forecasts and Warnings, UK National Coordination Committee IDNDR, Project 3. Thomas Telford, London, 32. WATTS, R. B., HERD, R. A., SPARKS, R. S. J. & YOUNG, S. R. 2002. Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 115-152. WAYTHOMAS, C. F. & WAITT, R. B. 1998. Preliminary Volcano-Hazard Assessment for Augustine Volcano, Alaska. United States Geological Survey, Open File Report 98/106. WHITE, R. A., MILLER, A. D., LYNCH, L. & POWER, J. 1998. Observations of hybrid seismic events at Soufriere Hills Volcano, Montserrat: July 1995 to September 1996. Geophysical Research Letters, 25, 3657-3660. WOODS, A. W., SPARKS, R. S. J., RITCHIE, L. J., BATEY, J., GLADSTONE, C., & BURSIK, M. 2002. The explosive decompression of a pressurized volcanic dome: the 26 December 1997 collapse and explosion of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 457-466. YOUNG, S. R., HOBLITT, R. P., SMITH, A. L., DEVINE, J. D., WADGE, G. & SHEPHERD, J, B. 1996. Dating of explosive volcanic eruptions associated with dome growth at the Soufriere Hills volcano, Montserrat, West Indies. In: AHMAD, R. (ed.) Proceedings of the Second Caribbean Conference on Natural Hazards and Hazard Management, October 1996, Kingston, Jamaica. MVO Open-file Report 96/22. YOUNG, S. R., BARCLAY, J., MILLER, A. D., SPARKS, R. S. J., STEWART, R. C., DAVIES, M. A. & MVO STAFF. 1997. Deformation of the Galway's Wall and Related Volcanic Activity, November 1996 to March 1997. MVO Special Report 2. YOUNG, S. R., SPARKS, R. S. J., ASPINALL, W. P., LYNCH, L. L., MILLER, A. D., ROBERTSON, R. E. A. & SHEPHERD, J. B. 1998. Overview of the eruption of the Soufriere Hills volcano, Montserrat, July 18 1995 to December 1997. Geophysical Research Letters, 25, 3389-3392

The 26 December (Boxing Day) 1997 sector collapse and debris avalanche at Soufriere Hills Volcano, Montserrat B. VOIGHT1, J-C. KOMOROWSKI2, G. E. NORTON3, A. B. BELOUSOV4, M. BELOUSOVA4, G. BOUDON5, P. W. FRANCIS6, W. FRANZ7, P. HEINRICH 8 , R. S. J. SPARKS9 & S. R. YOUNG10 1 Geosciences, Penn State University, University Park, PA 16802, USA (e-mail: [email protected]) 2 Observatoire Volcanologique de la Soufriere (IPGP), Le Houelmont, Gourbeyre 97113, Guadeloupe 3 British Geological Survey, Key worth, Nottingham, NG12 5GG, UK 4Institute of Volcanic Geology and Geochemistry, Petropavlovsk-Kamchatsky, 683006, Russia 5 Institut de Physique du Globe de Paris (IPGP), 4 Place Jussieu, B 89, 75252 Cedex 05 Paris, France 6 Department of Earth Sciences, Open University, Milton Keynes MK7 6AA, UK (deceased) 1 Gannett-Fleming Engineers, Harrisburg, PA 17110, USA ! Laboratoire de Detection et de Geophysique, Commisariat a I'Energie Atomique,Bp 12, 91680 Bruyeres-le-Chatel, France 9 Department of Earth Sciences, Bristol University, Bristol, BS8 1RJ, UK 10 Montserrat Volcano Observatory, Montserrat, West Indies

Abstract: The southern sector of Soufriere Hills Volcano failed on 26 December 1997 (Boxing Day), after a year of disturbance culminating in a devastating eruptive episode. Sector collapse produced a c. 50 x 106 m3 volcanic debris avalanche, and depressurized the interior of the lava dome, which exploded to generate a violent pyroclastic density current. The south-directed growth of a lava lobe and build-up of lava-block talus, since early November 1997, brought the hydrothermally weakened sector to a condition of marginal stability. Limit-equilibrium stability analyses and finite-difference stress-deformation analyses, constrained by geomechanical testing of edifice and debris avalanche materials, suggest that the sector collapse was triggered by a pulse of co-seismic exogenous lava shear-lobe emplacement. Slip-surface localization was influenced by strain-weakening. The source region fragmented into avalanche megablocks, and further disruption generated a chaotic avalanche mixture that included variably indurated and coloured hydrothermally altered material, and much talus. The avalanche consisted of several flow pulses that reflected complexities of source disruption and channel topography. In the proximal zone, within 1.5km from source, many megablocks preserve pre-collapse stratigraphy. At major bends the avalanche separated into channelled and overspill flows. In the distal region, >2.5km from source, stacked sets of the main lithologies occur with a hummocky surface and abrupt flowage snouts, beyond which sparse hummocks occur in a thinly spread deposit. Textures suggest emplacement by laminar mass transport of partly saturated debris riding on a frictionally sheared base. Three-dimensional numerical simulations of emplacement governed by a Coulomb-type (Pouliquen) basal friction law imply low values of friction (<15°), consistent with geotechnical test data and the localized presence of pore-water pressures. The best-fit model suggests an emplacement time <3 minutes and a typical maximum velocity of about 4 0 m s - 1 , which are consistent with field estimates.

Following over a year of disturbance, the southern flank of Soufriere Hills Volcano failed on 26 December (Boxing Day) 1997, generating the most devastating episode of the entire eruption (Sparks et al. 2002). The complex series of events resembles, on a smaller scale, the debris avalanche and directed blast that occurred at Mount St Helens in 1980 (Lipman & Mullineaux 1981). At Soufriere Hills, an andesitic lava dome had grown over the unstable, hydrothermally weakened southern sector of the edifice. When this sector collapsed on Boxing Day, the interior of the lava dome was exposed and depressurized, and it exploded to generate a powerful pyroclastic density current that ravaged the southwestern flank and entered the sea. This paper focuses on the mechanics of the sector collapse and on the emplacement dynamics, characteristics and properties of the deposit of the resulting debris avalanche (Fig. 1). It is based primarily on fieldwork conducted at various times in 1998 and 1999, with some access limitations due to safety considerations related to ongoing activity. Some sampling and analyses were conducted as early as 1996.

Pre-collapse conditions: geological setting and precursory activity The Boxing Day collapse was the culmination of a series of events that affected the geometry and conditions of the southern flank (Sparks et al. 2002; Young et al. 2002). Only the main points need In honour of our fallen comrade, Peter Francis.

be summarized here. The eruption of Soufriere Hills Volcano involved the growth of an andesitic lava dome within the confines of English's Crater (Fig. 2), a horseshoe-shaped depression produced by a prehistoric flank collapse (Wadge & Isaacs 1988; Boudon et al. 1996). This depression, about 1 km wide, with interior walls 100150 m high and open to the east, had been partly filled by a young (c. 350 BP) lava dome known as Castle Peak, and a semi-circular moat existed between this dome and the walls of English's Crater (Fig. 2). The new lava extrusion began in mid-November 1995, and through 1996 was confined to English's Crater, although some pyroclastic flows travelled eastwards to the sea (Young et al. 1998). The Boxing Day sector collapse involved the flank including Galway's Wall and Galway's Soufriere (Fig. 2). Galway's Wall marked the southern edge of English's Crater, and extended about 600m from Chances Peak to Galway's Mountain (Fig. 3). The outer, southward-facing side of Galway's Wall was precipitous and composed of pyroclastic breccias and tufts related to the formation of the Chances Peak and Galway's domes (Young et al. 2002). South of Galway's Wall was the Galway's Soufriere fumarolic field (Figs 4 and 5), an area of active, acid hydrothermal activity at an altitude of 400-500 m within the upper White River valley (Boudon et al. 1998). Galway's Wall was not much affected by events at the volcano until June 1996, when active dome growth shed talus into the moat between Castle Peak and Galway's Wall. After an eastward-directed dome collapse followed by an explosion on 17 September 1996 (Robertson et al 1998), renewed growth occurred between October and December 1996. Fractures on the western buttress of Galway's Wall and within the wall itself were observed in late November, and continued to open into December 1996 (Young et al. 2002). During

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 363-407. 0435-4052/02/$15 © The Geological Society of London 2002.

363

364

B. VOIGHT ET AL.

Fig. 1. View northwards from the mouth of the White River, showing Boxing Day debris avalanche and tephra deposits on the south flank of Soufriere Hills Volcano as seen (a) about four days after deposition (photo MVO), and (b) in summer 1998 (photo K. West). Fergus Mountain on middle right, towns of Morris' and St Patricks to left of White River channel. Much of the ground surface (a) was veneered by pyroclastic density current deposits and associated fallout deposits, but in (b) most of these deposits had been eroded, and the debris avalanche deposits were etched in greater detail. Boxing Day deposits had extended the fan built by pyroclastic flows associated with Vulcanian explosions in October 1997 and the dome collapses of early November 1997.

this period, strong hybrid swarm seismicity occurred and triggered rockfalls from jointed tuffs exposed on Galway's Wall (Fig. 3). These earthquake swarms were interpreted as reflecting shallow intrusion. Montserrat Volcano Observatory (MVO) scientists expressed concern to officials and the public about the possibilities of collapse of Galway's Wall and an ensuing lateral volcanic blast, and as a consequence the area at risk was evacuated (Young et al. 2002). At about the same time, tsunami risk evaluations were carried out, qualitatively by B.V. at Montserrat, and quantitatively for the coast of Montserrat and neighbouring northern Guadaloupe by French scientists in collaboration with the MVO (Heinrich

et al. 1998). Later, at the time of the catastrophic events on 26 December 1997, there was no loss of life because the earlier evacuation had been maintained despite some public pressure to end it. In late March and April 1997, dome growth again shifted to the south, lava overtopped the low point on Galway's Wall, and blockand-ash flows eroded parts of it (Sparks et al. 2002. fig. 2a). Rock in the vicinity of Galway's Soufriere was partially buried by talus and block-and-ash flow deposits (Figs 6 to 8). From May through to October 1997 dome activity shifted away from Galway's Wall, but renewed growth here in November generated two major partial dome collapses on the southern flank, in total involving about 8 x 10 6 m 3

THE BOXING DAY SECTOR COLLAPSE

365

Fig. 2. (a) Topographic map of southern Montserrat, showing place names and features described in text. Seismic stations at Windy Hill (MBWH) and St Patrick's (MSPT) are shown by dots. (b) Detail of English's Crater and upper flanks of the Soufriere Hills dome complex (after Sparks et al. 2002).

of material. Further dome growth in November and December 1997 was rapid, a lobe of lava was squeezed over remnants of Galway's Wall, and an apron of dome talus extended to bury terrain near Galway's Soufriere (Fig. 9). By 8 December 1997 the dome had reached a volume of 102 x 10 6 m 3 , and by 25 December this had increased to 113 x 106 m3 (Figs 6-11; Sparks et al 2002). The short-term build-up to the Boxing Day collapse was rapid (Sparks et al 2002). At 14:30 (local time, LT) on 24 December, a

distinct hybrid earthquake swarm began (see Miller et al (1998) for discussion of types of seismicity) and it increased until 20:00 on 25 December when it merged into continuous tremor. The earthquakes increased in amplitude, but were smaller than those recorded in early November. The amplitude of the tremor peaked at 23:00 on 25 December, then declined until midnight when individual events could again be detected. Tremor resumed at 01:30 on 26 December and built in amplitude to the onset of the collapse at 03:01.

Fig. 3. View west along the axis of Galway's Wall in early December 1996. Chances Peak forms west buttress, White River valley is to the left, and the growing lava dome is to the right. At this time, the jointed tuffs of Galway's Wall were collapsing in a series of earthquake-triggered slab rockfalls, and MVO scientists became concerned about the possibility of a larger wall collapse that might trigger a lateral dome explosion. By March the dome lavas began to overtop and erode the wall. Active dome growth in this sector recommenced in late 1997 and culminated in the 26 December sector collapse (photo B.V.).

366

B. VOIGHT ET AL.

Fig. 4. Galway's Soufriere. (a) View NW into the soufriere on 7 December 1996. The soufriere extends into the hill between the White River valley and Galway's Wall (photo MVO). Portions of this area moved in slump landslides in December 1996. Brown deposit in river valley is distal flow snout of fresh, earthquake-triggered rockfall from the Galway's Wall, (b) The soufriere, looking northward from valley bottom, 26 March 1996, showing varicoloured. weak, highly altered tuffs in the valley and adjacent hill, with the Galway's Wall in the background (photo B.V.).

Chronology of the collapse Seismicity provides the main constraint on the timing of events. The main collapse event has been divided into six pulses. Sparks et al. (2002) suggest that the start time of the main collapse was 03:01.0 LT and the finish of major dome disintegration was 03:16.2. The spectral frequency of seismicity during the collapse was similar to that for the hybrids and tremor prior to the collapse, with the dominant frequency below 2.8 Hz. The form of the collapse signal was similar to that produced by the 6 November 1997 dome collapse and associated block-and-ash flow activity, but the signal amplitude in December was about five times greater. GOES-8 satellite data give information on the plume associated with the explosion and pyroclastic density current, but no information on the collapse (Mayberry et al. 1998). Only in the morning after the event did it become clear that a substantial sector collapse had occurred, accompanied by major destruction of the lava dome. Post-collapse morphology and collapse volume estimates The collapse and explosion produced two major scars (Figs 9b and 12b). The lower scar, formed by the sector collapse, was about 400 m wide. Its northern boundary formed an arcuate step roughly l00m high, located approximately along the southern, steeply sloping edge of the remnant of Galway's Wall (Figs 9b and 13b).

The floor of the scar, which represented the shear-slip surface of the collapse, extended to the south about 400-500 m in the White River valley where it was buried by fragmental deposits. At the southern part of the scar, a patch of discoloured deposits with steaming fumaroles was interpreted as material that buried remnants of Galway's Soufriere (Fig. 13b). The upper scar was a spoon-shaped amphitheatre within the dome, as much as 400m wide and extending about 450m north of the step (Figs 9b and 13b). The upper scar is interpreted as originating by a combination of gravitational failure and spontaneous, explosive quarrying processes involving gas-pressurized, highly crystalline but partly molten dome rock. The southern part of the upper scar may have been produced by retrogressive collapse involving dome material and fragments from Galway's Wall. Sparks et al. (2002) estimate roughly the volume of material involved in the failure, from the dimensions of the scars and by comparison with the pre-collapse topography of the Galway's Soufriere area and the dome: r. 20-30 x 10 6 m 3 of hydrothermally altered rocks from the Galway's Soufriere area; c. 5 x 10 6 m 3 of Galway's wall; c. 25 x 10 6 m 3 of the lava dome that had grown between 6 November and 26 December; and c. 30 x 106 m3 of dome talus deposited beyond Galway's Wall, reaching toward the Galway's Soufriere. Thus, overall the failure involved c. 80-90 x 106 m3 of material, with the sector collapse alone, from these estimates, accounting for roughly 50 x 10 6 m 3 . The uncertainty of these estimates may exceed 15%. Independent estimates, based on valley-fill deposits, are discussed below.

THE BOXING DAY SECTOR COLLAPSE

367

Fig. 5. Geology of the hill between Galway's Soufriere and Galway's Wall, observed from uppermost White River valley, (a) East side of White River valley between the soufriere and Galway's Wall on 26 March 1996. Sulphur-encrusted, partly altered pyroclastic breccia of prehistoric block-and-ash flow deposits. S.R.Y. is scale (photo B.V.). (b) Similar location and date, at boundary with Galway's Wall (to left of notch). A few metres from Galway's Wall, the north-dipping, layered pyroclastic breccias and tuffs of hill were offset 1 m by normal fault that displaced the sequence to the south. We speculate on the possibility of this hill as a prehistoric slump block (photo B.V.).

Volcanic debris avalanche deposit General description The avalanche deposit consisted of unconsolidated, poorly sorted volcaniclastic debris that buried about 2.7 km2 of the White River

valley and valley margins (Figs 13 to 15). The deposit spanned about 4km in length (uppermost parts are buried by fresh, postBoxing Day dome talus) and was about 200 to 700m wide. In the lower part of the valley, it was typically about 400m wide, and about 25-70 m thick. Topographic surveys of the White River valley area were carried out within three weeks of the collapse (Figs 16 and 17), and compared with a survey carried out after the

Fig. 6. Upper White River valley, as viewed from south. (a) In December 1996. Galway's Wall extended from Galway's Mountain on the right, to Chances Peak, where the radio tower marked a tiltmeter site. An access road led from the south coast to Galway's Soufriere. shown by pale colours in the river valley. with the alteration affecting also the mound-like hill between the valley and Galway's Wall (photo K. West), (b) By late April 1997. the river valley and much of the soufriere were being buried by block-and-ash flow deposits produced by collapses from the active lava dome (photo K. West). (c) By 8 December 1997. extensive southerly growth of the lava dome had generated talus that almost completely buried the hill shown in (a) and (b) (photo MVO). Minor changes in talus volume occurred between this date and the sector collapse of 26 December 1997. The pale-coloured landscape reflects the dusting of ash from numerous rockfalls.

THE BOXING DAY SECTOR COLLAPSE

369

Fig. 7. Upper White River valley, as viewed from southwest, (a) By 1 April 1997, the valley and soufriere were being partly buried by rockfalls from Galway's Wall, and by block-and-ash flow deposits from lava dome collapses. Access road at left centre. Notched hill is above soufriere, between valley and Galway's Wall (photo MVO). (b) By 8 December 1997, most features within the valley shown in (a) had been buried by a talus cone, setting the stage for edifice collapse about two weeks later. The talus was being fed by southward-moving lava at the top of the cone (photo MVO).

4-6 November 1997 dome collapse. The volume of material in the (lower) White River valley and delta was determined as 46 x 106 m3 (Sparks et al 2002). Most of the valley fill was the debris avalanche deposit, although block-and-ash flow deposits derived from the dome disintegration, and some subsequent explosions, covered much of the avalanche deposit and infilled depressions on its top. The debris avalanche deposit volume was substantially greater than the volume of the remnant collapse scar, but the residual scar, indented into the valley floor, did not reflect the full thickness and volume of material that had been lost. Considering the volume estimates from both the

source area and the valley deposits, and allowing for uncertainty, we estimated the debris avalanche volume as about 40-50 x 106m3. The uncertainties associated with the volume estimates prevented the volumetric bulking (increase in pore space) associated with avalanche emplacement to be determined, although bulking may have been minor inasmuch as much of the source material had significant initial porosity. The debris avalanche deposit extended throughout the White River valley, from the region of its partly infilled source scar, to its southern distal margin on the coastal fan of pyroclastic deposits at the mouth of the valley (Figs 13 to 16). It reached to within about

370

B. VOIGHT ET AL.

Fig. 8. Upper White River valley, as viewed from east, (a) By 22 April 1997, the river channel and Galway's Soufriere was buried by, mainly, block-and-ash flows from dome collapse. To the right of valley is notched hill, and farther to the right is Galway's Wall. Chances Peak is in the background (photo K. West), (b) By 8 December 1997, a talus cone had buried the topographic features of the uppermost valley. The talus was fed by a lava lobe that moved southward over the eroded Galway's Wall, seen at top right (photo MVO).

20m from the sea (Komorowski et al, 1999). Near the coast it outcropped as a broad veneer with mounds 0.5-1 m high and patches of multicoloured hydrothermally altered clastic debris. At about 400-800 m northward from the sea, particularly near the eastern margin, several pronounced, but crudely developed, topographic steps were formed in the debris avalanche deposit (Figs 1 , 1 8 and 19). In plan view the steps were curved such that positions near the channel centre were displaced farther down-valley than positions near the margins. Each step represented the front of a flow lobe, and the succession of steps was formed by imbrication of successive tongues of avalanche debris. The effect of this process, which

must have been rapid (see below), was to produce a stair-step profile near the eastern margin, with each step about 20-25 m high, succeeded by a rough bench (3-10 m relief), then another step, and another, for a total relief of 60-70 m above the distal sheet-like portion of the deposit nearer the coast. The pattern in plan view suggested that avalanche flow in the eastern portion of the channel was impeded by a ridge of older rock protruding above the channel floor, which had not been completely buried by previous 1997 deposits. Flow in the other part of the channel was unobstructed, and there the avalanche reached a greater distance, nearly to the sea.

THE BOXING DAY SECTOR COLLAPSE

Fig. 9. Maps of Galway's Wall area (a) before and (b) after the 26 December collapse (cf. Fig. 12).

371

372

Fig. 9. (continued)

B. VOIGHT ET AL.

THE BOXING DAY SECTOR COLLAPSE

373

Fig. 10. Topographic map of the Soufriere Hills dome complex just before the 26 December collapse (after Sparks et al. 2002). Area inside the dashed line was subjected to frequent topographic changes associated with dome growth and destruction. Elevations in metres.

The White River valley narrows at Fergus Mountain, about 1.5km from the source and 3 km from the sea. There, the valley bends to the SE, and another, sharper bend to the south occurs above Morris' village, at about 2km from the sea (Figs 14 and 15). The avalanche spilled out over the west valley wall at the second bend, emplacing deposits that extended nearly to Morris' (Fig. 19). A similar but smaller deposit spilled over the east valley wall, depositing a perched levee on the sloping flank of Fergus Mountain. These avalanche overspill deposits were detached from, and were 50-80 m higher than, the surface of the main avalanche deposits within the valley. Most of the avalanche deposit was cool to slightly warm upon emplacement, according to the first inspections a few days after emplacement (Sparks et al. 2002, table 2). However, some localized hot areas and fumaroles within the avalanche deposit were still active more than two years after emplacement; some reflect the activity of fumarolic areas buried by the deposit, but others appear rootless and reflect hot material in the deposit.

Debris avalanche textural fades and map units The debris avalanche deposit comprised a chaotic mixture of hydrothermally altered material, variably weathered volcanic breccias and volcaniclastic deposits, and fresh andesite rubble. Each of these types of materials was characterized by one or several distinctive colours, and thus some individual hummocks were of a single colour

that indicated a homogeneous material, whereas at others, colourmottling, lenses or patches were typical and indicated juxtaposition of different lithologies. In general, the hummocks were composed of fragments of source material that had been transported in a relatively intact, although to a variable extent disturbed, fashion. In Glicken's (1986, 1991, 1998) terminology such deposits, in which a vestige of the original stratigraphy or structure can be preserved, are referred to as avalanche block fades. In this paper we use the slightly modified term avalanche megablock facies or, more simply, megablock fades, in order to avoid possible confusion with the pyroclastic block-and-ash flow deposits that also occur on Montserrat. We use the term matrix fades for that unconsolidated part of the avalanche deposit in which original constituents have been completely mingled, e.g. completely disaggregated source material and entrained loose debris, conceivably containing most types of rock from the source area (Glicken 1998; Ui & Glicken 1986; Crandell et al. 1984; Ui et al. 2000). Fragments in both of these end-member facies typically are angular, and form unstratified, poorly sorted breccias and gravels. In order to avoid confusion, in the discussion that follows we reserve the term facies to general textural descriptions of avalanche deposits; we use the term unit to refer to the avalanche deposits we have mapped. The Boxing Day debris avalanche deposit primarily showed typical avalanche megablock facies, with stratigraphically coherent domains affected by localized crushing (Figs 20 to 22). With increasing distance from source, the avalanche deposit displayed increasingly common development of shearing textures and limited mingling of lithologies on a small scale. Shearing textures were

374

B. VOIGHT ET AL.

Fig 11. Digital elevation blocks of Soufriere Hills Volcano, pre-26 December 1997. Montserrat grid numbers are shown for reference. Note complex topography of summit area, (a) View N065, tilt 25 degrees, (b) View N290, tilt 20 degrees.

prominent at the base of avalanche megablock domains but were rarely exposed. Also rarely, the deposit displayed matrix facies, with homolithologic blocks set in a chaotic, very poorly sorted matrix, with clasts ranging from sand size to single lava blocks of l-2m diameter. Generally, the clasts consisted of both hydrothermally altered material from the old edifice, and dome lava incorporated from dome talus or parts of the collapsed active dome. Sparse clasts represented material from the hydrothermal plumbing system, often containing numerous fractures filled with gypsum, silica or sulphides. The marked colour and textural variations of the debris avalanche deposit represented different source rocks and alteration zones. The main altered-rock types included (1) bleached white deposits with gypsum and amorphous silica, (2) orange-yellow deposits, (3) reddish-brown deposits, and (4) yellowish-green depos-

its with pyrite and relict hydrothermal pore fluids. In general, types (1) to (4) may have constituted a succession at the source reflecting increasing depth, from young to old, and/or distance from the locus of alteration. The white deposits evidently were originally near the ground surface at Galway's Soufriere (cf. Figs 21 and 4). They contain mud-cracked palaeosols, relict roots and caliche surfaces with embedded branches, bent metal wires and even portions of the steel cable that formerly ran from Galway's Soufriere to Chances Peak. We identified and have mapped three main lithofacies of the debris avalanche deposit (Fig. 19). These are, informally, the Fumarolic Unit (with two subunits). Halo Unit and Talus Unit. The deposit is spatially divided into White River-channel and overspill elements, with the channelled deposits described as distal or proximal, respectively below and above the Fergus Mountain narrows

THE BOXING DAY SECTOR COLLAPSE

375

Fig. 12. Cross-sections through Soufriere Hills Volcano, from north to south, (a) before and (b) after the 26 December 1997 collapse (cf. Fig. 9). The pre-emption conditions included English's Crater and Galway's Soufriere. Additional materials comprise lavas, and talus and rockfall debris. In (b), the collapse first involved talus and some dome lava (dotted line), and underlying soufriere material, followed by explosive collapse of oversteepened, decompressed fresh dome lava, which led to generation of an energetic pyroclastic current. The positions of the failure planes are schematic.

region (Figs 14 and 19). There are two overspill deposits: the western one near Morris', and the eastern one perched on the flank of Fergus Mountain. The Fumarolic Unit was subdivided into two parts. The Soufriere Subunit comprised superficial deposits with materials showing intense epithermal alteration, derived originally from the Galway's Soufriere area (Fig. 4). The Soufriere Subunit constituted the white, orange or yellow suites (1 to 3 as described above) within the avalanche deposit, in texturally diverse deposits that ranged from matrix-rich lithified breccias to clast-supported unconsolidated breccias. The colours mainly reflected the original mineral associations observed in the highly acid, oxidized superficial part of the hydrothermal system of Galway's Soufriere, but may have also reflected stratigraphy or burial depth. Primary magmatic minerals were replaced by silica polymorphs, titanium oxide and sulphates (such as natroalunite, natrojarosite, gypsum and anhydrite) (Boudon

et al 1998; Chiodini et al 1996). These are the typical alteration products of unsealed acid-sulphate systems. Kaolinite, smectite and mixed-layer clay assemblages formed locally, with less acidic solutions (Boudon et al. 1998). Large patches of yellowish-green material (type 4) were located near the Fergus Mountain narrows and more proximally in the upper White River valley (Fig. 20). These were distinctive enough to map and are referred to as Chartreuse Subunit (Fig. 19). This material was inferred to represent the deeper, less-oxidising parts of the hydrothermal system, where magmatic or other influences created reducing conditions. We inferred that the colouration was due to reduced iron and sulphides, but detailed mineralogical and chemical studies have not yet been completed. The fact that locally there are repeated slices of avalanche material with stacked coloured alteration types suggests that a layered hydrothermal system existed that was intersected by the

376

B. VOIGHT ET AL.

Fig. 13. Proximal part of White River valley, north of Fergus Mountain. (a) By 3 May 1997, the valley and soufriere were filled by block-and-ash flow deposits from lava dome collapses, and these deposits also partly blanketed the bedrock shelf west of the river channel, where, at the foot of the western escarpment, they reached the access road to Galway's Soufriere (photo MVO). (b) View on 30 December 1997, showing hummocky. Boxing Day debris avalanche deposits (photo MVO). Fergus Mountain is in foreground, with the entire landscape dusted with tephra from the Boxing Day pyroclastic density current and associated ash cloud. Avalanche megablocks protrude from valley deposits. Yellowish coloration in deposits at right centre reflects fumarolic encrustations in deposits that have buried the decapitated Galways Soufriere. Beyond this, the sector collapse scar rises at Galway's Wall, with the notch in the wall leading to the dome collapse scar (see Fig. 9b).

collapse slip surface. As a result, the avalanche flow deposit in general displayed a vertical alteration order, with Chartreuse Subunit overlain by Soufriere Subunit materials. This suggests that depth of origin was a locally dominant factor in alteration zoning, although the spatial distribution of alteration could also be influenced by conditions that varied laterally as well as vertically (Giggenbach 1992). An irregular lower-intensity alteration halo was known to surround the soufriere and its substructure for several hundred

metres, extending especially in the direction of English's Crater (Boudon et al. 1998). The term Halo Unit was used to describe the avalanche materials from this source, which were characterized by less alteration than in the Fumarolic Unit. The primary sequences comprising the Fumarolic and Halo units may have been originally similar, with the distinguishing characteristics simply reflecting the degree of alteration. In the avalanche deposit the Halo Unit materials were typically grey, tan or brown consolidated breccias and tuffs, similar to rocks that were exposed in Galway's Wall and in

THE BOXING DAY SECTOR COLLAPSE

377

Fig. 14. Medial part of White River valley, viewed north, (a) 22 April 1997, with valley partly filled by block-and-ash flow deposits and materials reworked by lahars. Fergus Mountain at right, Morris' area to left on west bank. Soufriere Hills Volcano in background (photo K. West). Considerable additional filling occurred later (see Fig. 18a). (b) Boxing Day debris avalanche and coarse pyroclastic density current deposits mainly followed the drainage of the White River, but parts of these flows ramped over the western valley wall, toward Morris'. June 1998 (photo J.C.K.). Two prominent flow fronts in the avalanche deposit suggest two pulses of overspilling debris. On the opposite bank, the avalanche left a levee-like deposit of blocky avalanche debris on the flank of Fergus Mountain. Vegetation in (a) was removed by the violent pyroclastic density current, the grey deposits of which blanket the landscape in (b).

the hill of layered tuff between this and Galway's Soufriere (Fig. 5). These sites were located between Chances Peak and Galway's Mountain (Fig. 2), which constituted two of the five hornblendehypersthene dome complexes that formed the superstructure of Soufriere Hills Volcano (Rea 1974; Roobol& Smith 1998). The sequences in the vicinity of Galway's Wall derived from both complexes and mainly comprised block-and-ash flow deposits (MacGregor 1952; Rea 1974; Roobol & Smith 1998). The deposits in the hill between Galway's Wall and Galways Soufriere dipped northward (Fig. 5). We speculate that the hill might represent a slump block from a prehistoric sector collapse. The interpretation is supported by the

sense of slip on faults, the broad, amphitheatre-like valley morphology, and the recognition of probable old debris avalanche deposits in the White River valley and offshore from its mouth. If so, its slip surface was probably reactivated in the 1997 sector collapse. The Talus Unit comprised fresh, grey to reddish oxidized subangular breccia, representing mainly the talus cone of the 1997 dome lava that comprised part of the collapsed sector (Fig. 20d). Because the talus was mostly unconsolidated and of non-cohesive particles, its clast assemblages could not have remained intact as coherent megablocks. Neither, in general, were they thoroughly mingled with other material by the avalanche emplacement processes. Thus,

378

B. VOIGHT ET AL.

Fig. 15. Distal part of White River valley. (a) Mouth of White River at Old Fort Point on 31 March 1997. Morris' on left, Fergus Mountain and South Soufriere Hills on right, Galway's Wall in distance (photo MVO). (b) Same area a few days after 26 December 1997, with river valley containing deposits from the debris avalanche and coarse facies of the pyroclastic density current (photo MVO). Overspill deposits were emplaced above Morris' in foreground. Vegetation and buildings were largely destroyed by the violent pyroclastic density current, with its monochromatic grey tephra covering the landscape.

these deposits were neither avalanche megablock facies nor matrix facies in the original senses of these terms, but represented a new avalanche textural facies type. We designate it generically as noncohesive facies, defined as part of a debris avalanche deposit comprising originally non-cohesive source material. We speculate that rapid avalanche motion may have been complex, inasmuch as cohesive altered materials probably had deformation and transport properties quite different from cohesionless debris. Some intact dome lavas were originally interleaved with upper parts of the talus and therefore were included in the avalanche deposit, but they would be difficult to distinguish from talus. Most

of the fresh dome lavas from the northern scar were presumed to have become involved in the pyroclastic density current on Boxing Day (Sparks et al. 2002; Ritchie el al. 2002). However, discrimination of the coarse valley facies of the pyroclastic density current deposit from the Talus Unit of the avalanche deposit was sometimes difficult, because the source for both deposit types was practically identical. Furthermore, unconsolidated loose talus in the avalanche deposit was reworked and entrained in the energetic pyroclastic density current, and this reshaped the surface morphology of mounds of talus over which it passed. Mapping of generally undifferentiated blocky lava deposits is shown on Figure 19.

THE BOXING DAY SECTOR COLLAPSE

Fig. 16. Topography of the White River immediately after the 26 December 1997 collapse, based on surveys on 4 and 17 January 1998 (data from R. Herd).

379

Distance along section (m) Fig. 17. Transverse cross-sections across the White River valley show the original topography (dashed lines), the infill of material from all earlier eruptions, including the 4-6 November 1997 pyroclastic flow deposits (lower solid line), and the infill of deposits from the 26 December 1997 collapse and explosion (upper solid line).

Fig. 18. Comparative panoramic views of distal White River valley, (a) View in early December 1997, three weeks before the collapse. Fergus Mountain flank to right, villages of Morris' and St Patricks to left. Channelled block-and-ash flow deposits are derived from dome collapses on 4 and 6 November 1997. Note island surrounded by pyroclastic deposits at lower centre, representing a bedrock ridge from the original landscape (cf. Fig. I5a). As discussed below, this ridge influenced avalanche deposition on Boxing Day (photo B.V.). (b) Same area on 2 June 1998, showing debris avalanche deposit extending over fan nearly to the sea (photo J.C.K.). The current of avalanche debris in the central channel had continued in motion, while the eastern part came to rest on the ridge within the channel shown in (a). As this main current thinned, larger blocks became grounded, and a sheet-like deposit was formed at the avalanche terminus, with sparse, small hummocks. (c). Structure in distal lobate, hummocky avalanche deposit, June 1998 (photo J.C.K.). Eastern part formed a series of stacked lobes where the flow grounded against the ridge within the channel in (a). Coarse pyroclastic density current flow lines diverge about avalanche deposit. Overspill deposits at Morris' at upper left.

380

B. VOIGHT ET AL.

Fig. 18. (continued)

The volumetric proportions of the various units could not be determined from the non-dissected surface exposures alone, and the distribution of units at depth in the deposit is not yet known.

Surface relief, deposit morphology and composition, and structures The surface relief of the debris avalanche deposit, together with its internal geological fabric, provided information about the processes of avalanche formation, emplacement and deceleration. We distinguished three classes of primary surface features, which resulted from: (1) the incomplete disintegration of the failed part of the edifice; (2) avalanche transportation; and (3) avalanche deceleration and deposition. Hummocks are the characteristic morphologic feature of volcanic debris avalanche deposits (Voight et al. 1981; Siebert 1984, 1996; Glicken 1986, 1998; Ui & Glicken 1986; Ui et al 2000). At Soufriere Hills, hummocks over 100m wide protruded 0.5-30m above the mean elevation of the debris surface, and were commonly bounded by slumps or faults (Figs 13 and 20). Closed depressions as much as 50 m across and 20 m deep were also common features, although most such areas tended to be filled by younger deposits. Details for most hummocks in proximal areas were unrecorded, because safety considerations precluded close inspection. These constraints remained in effect throughout the year 2000. Many hummocks were buried or reshaped by later events. Occasional pyroclastic flow activity continued in the valley into 1999, and numerous severe storms occurred, so that intermittent severe erosion and redeposition have strongly modified the exposures and surface

morphology of the avalanche deposit. The lack of detailed new topographic maps over the debris avalanche deposit has impeded semi-quantitative morphological analyses of the numbers, shape orientation and size of hummocks. The largest hummocks occurred in the upper White River valley (<1.5km from source, north of the Fergus Mountain narrows), where they protruded above smooth-surfaced blocky pyroclastic flow deposits (Figs 13 and 20). Qualitatively, the scale of hummocky relief decreased down-valley, and individual hummocks in the distal area were more numerous, smaller, and less sharp-edged than those in the proximal region, with surface relief about 2-1 0m (Fig. 21). In comparison, the steep frontal snouts of distal flow lobes were typically about 10-20m high (Figs 1,18 and 2la). The change in surface morphology reflected the increasing damage sustained by the avalanche debris in moving down-slope and in squeezing through the Fergus Mountain narrows. Both distance of transport and time elapsed in transport may have been factors contributing to increased disruption. In the thinner, laterally spread, most-distal portion of the avalanche deposit that extended to within 20 m of the coast, numerous, increasingly separated hummocks, 2-5 m high, occurred, with each dominated by one or more of the main coloured lithologies. Some large proximal hummocks (overall >100m wide, c. 50m high) appear to be overlapping, rotated slide blocks containing relatively little-disturbed pre-collapse stratigraphy (Fig. 20a). These were similar to the toreva blocks of Reiche (1937), reported also from other volcanic debris avalanche sites (Wadge et al. 1995; Francis & Wells 1987; Belousov et al 1999), and may have reflected retrogressive failure of large fragments of the edifice, including Galway's Wall. The spatial position of these hummocks, and the

THE BOXING DAY SECTOR COLLAPSE

381

Fig. 19. Map of debris avalanche deposit showing map units and surface features.

fact that fumaroles on some were still active, suggested that they represented the least disrupted, but still moved, portions of the former Galway's Soufriere and surrounding alteration halo. Some fumaroles in avalanche debris over the site of the old soufriere may have masked hydrothermal systems that were still active, but in other cases, south of the former position of the soufriere, the fumaroles appeared to be rootless. At one location (location 1-3, Fig. 23; near Fig. 20a), lava talus overlying a rootless fumarole set in a megablock of Chartreuse Subunit material had been cemented by fumarolic sulphate precipitates. Other hummocks were bounded by normal faults and appeared to represent the horsts of simple extensional systems (Fig. 20c),

reflecting longitudinal stretching (hummock long-axes perpendicular to flow), or transverse stretching (hummock long-axes parallel to flow) (Voight et al. 1981). Structures such as imbricated thrusts or strike-slip faults formed in response to longitudinal compressional stresses. These occurred at channel constrictions (Fig. 20b). overspill ramps (Fig. 22b), in frontal ridges of avalanche debris near Morris' (Fig. 22b, d), and in imbricated flow snouts in distal channel deposits (Figs 18 and 24d). Field observations suggest that the avalanche thickened locally during flow by the overlapping of successive flow lobes. At the northern part of the narrows, Talus Unit materials occurred in thin lenses between other avalanche units, with the material overlying

382

B. VOIGHT ET AL.

Fig. 20. Debris avalanche morphology and structure in proximal and medial regions, (a) View towards east at overlapping tilted avalanche megablocks that preserve original stratigraphy, with yellow-green Chartreuse Subunit overlain by reddish brown Soufriere Subunit in June 1998 (photo J.C.K.). Block on right capped by breccia of fresh lava blocks. Active fumaroles within some blocks suggest that these represent moved parts of the Galway's Soufriere fumarolic field. (b) Stacked sequence near Fergus Mountain narrows in June 1998 (photo J.C.K.). Oxidized Soufriere Subunit pyroclastic breccia (I) overlain by a lens of lava block deposits (LBD), in turn overlapped by Chartreuse Subunit breccia (II). (c) Avalanche megablock facies at foot of Fergus Mountain overspill, June 1998 (photo J.C.K.). Soufriere Subunit materials. Horizontally bedded rock of Fergus Mountain flank exposed on right. (d) Detail at sample location 1-4,5 (see Fig. 23) of contact between pale megablock facies breccia and adjacent subrounded blocky deposits representing probable Talus unit, in March 1999 (photo A.B.B.).

the talus displaying prominent shear deformation textures in its basal part (Fig. 20b). The observations suggest that decelerating flow due to boundary resistance in the constricted frontal flow lobe led to its being overridden by faster moving material. Other complications in the collapse process, perhaps aided by the constriction at the Fergus Mountain narrows, induced at least three main flow pulses. These mainly followed the drainage of the White River, but parts of the flow ramped over the western wall of

the White River valley toward Morris', rather than manoeuvring around the sharp second bend below Fergus Mountain (Figs 19 and 22). Two prominent flow fronts occur in these overspill deposits, which suggests that two pulses of debris partly spilled over the valley walls (Figs 14b and 22d). Conceivably, part of the first flow pulse may have impacted against the west wall and partly infilled against it (Fig. 22b), thus assisting subsequent flow pulses to surmount the wall. The first overspilling flow reached several

THE BOXING DAY SECTOR COLLAPSE

383

Fig. 20. (continued)

hundred metres into the Morris' area, forming a prominent front, and the second reached a shorter distance, because it was less energetic or because the bank height was effectively raised c. 20 m by the first deposit. Each overspill deposit was composed of the typical vertical sequence of altered types (basal yellow-green Chartreuse Subunit, successively overlain by Soufriere Subunit types (3) reddish brown, (2) orange yellow, and (1) upper white (Fig. 22d)). The deposition here was interpreted as representing enmasse freezing of avalanche megablock facies materials. Also, at this valley bend, energetic block-and-ash flows from the dome explosion overtopped the White River valley walls, with the deposits extending well beyond the limits of the avalanche deposit (Figs 14b and 22). Some scattered large, pinkish, dense and fresh lava clasts on the avalanche deposit here reflect lag deposition from these energetic flows.

Directly across the valley from the Morris' overspill, the avalanche scoured the wall of the valley and left a levee-like deposit of blocky avalanche debris on the flank of Fergus Mountain (Fig. 19). The 10-50m thick levee-like deposit comprised mainly megablock facies and represented a single emplacement sequence with the typical green basal Chartreuse Subunit, upper Soufriere Subunit involving reddish-brown and orange-yellow types, and Talus Unit (Figs 20c and 22c). Isolated deformed blocks of the basal green and grey Chartreuse Subunit were scattered over the upper deposit surface, probably reflecting the dynamic impact of the avalanche against the valley walls. In the distal region, from below the second bend adjacent to Fergus Mountain to about 500m from the sea, at least three stacked sets of the main coloured lithologic types occur with preserved older stratigraphy, a hummocky topography (typically 3-10m relief).

384

B. VOIGHT ET AL.

Fig. 21. Debris avalanche morphology and structure in distal region, (a) Stair-stepped sets of the main coloured unit/subunit lithologies, shown by roman numerals, with frontal ridges 15-20m high. March 1999. Surface is hummocky with 3-10m relief (photo J.C.K.). (b) Distal hummock field in avalanche deposit in March 1999, showing multicoloured lithologies (cf. Fig. 4). Relief 2-5 m (photo J.C.K.). (c) Soufriere Subunit hummock on 8 March 1999, buried by flat-surfaced, coarse Boxing Day pyroclastic density current deposits containing reworked avalanche clasts. Relief 3-4m. Top surface is covered by fine pyroclastic density current deposits. Near sample location 8-1 (Fig. 23) (photo A.B.B.). (d) Multicoloured hummocky terrain displaying megablock facies (Fumarolic Unit), partly buried by flat-surfaced, olive-grey, pyroclastic density current deposit sequence (foreground). Post-depositional erosion channel on right. 8 March 1999 (photo B.V.).

and abrupt frontal snouts c. 20-40 m in height. Stoppage of the first prominent lobe, near the east channel margin, apparently was caused by the flow running up against a bedrock ridge exposed within the channel (Fig. 18; see General description section). This was followed soon after by a pile-up of successive flow fronts against the first. Each lobe contained the same stack of Chartreuse and Soufriere Subunits, with green-orange-white deposit types (Fig. 2la). The disposition of deposits suggested that the main current of avalanche debris in the central channel continued in motion while

the eastern lobe fronts came to rest (Figs 1 and 18). As this main current thinned and velocity waned, the larger blocks tended to become grounded, and the resulting large-block-depleted residual flow spread out to produce a broad, sheet-like deposit at the avalanche terminus, with sparse hummocks as much as a few metres high. In the distal region, parts of the avalanche deposit displayed numerous structures and diverse textures resulting from compressional, extensional and shear deformation of the avalanche as it moved and deposited (Figs 21 and 24). Megablock facies materials

THE BOXING DAY SECTOR COLLAPSE

385

Fig. 21. (continued)

were compressed in areas where avalanche material flowed around, or rode up against, previously deposited hummocks. In a few such places, compression led to thrust-sheet stacking of displaced stratigraphic units. Shear features include deformation of weak megablock facies domains into elongated slivers that surround more coherent material (Fig. 24c), alignment of such slivers, and stretched blocks of megablock facies carried within avalanche matrix facies. In the basal parts of the steep flow fronts at the distal eastern margin, and also at the head of the Fergus Mountain narrows, elongated, flattened megablock facies slivers suggest that the lithological units were squeezed and sheared under the moving overlying avalanche pile (Fig. 24d). Such shear textures reflected localized velocity gradients

in the flow, and were consistent with frictional energy dissipation mainly at and near the avalanche boundary surfaces. Textures produced by extensional deformation within the avalanche included the formation of fractures with jigsaw-fit (Glicken 1986, 1998; Ui 1983; Ui & Glicken 1986). Such cracks became more abundant, and with larger gaps, with increased distance from the avalanche source. The avalanche thus became increasingly fragmented, by greater separations of jigsaw-fit fractures, to produce a population of small clasts that themselves could fragment, by jigsaw-cracking, into smaller clasts. This process has been shown elsewhere to occur on all scales (Komorowski et al. 1991). Finally, clastic dykes several metres across, of dome-lavaderived debris that intruded megablock facies material, occurred

386

B. VOIGHT ET AL.

Fig. 22. Overspill areas, medial region. (a) Sweeping view to south of White River valley from above upper bend at Fergus Mountain. June 1998. Avalanche ramped over west bank, toward Morris ', and formed levee-like deposit over inside bend of east bank. Hummocky distal deposit at upper left shows frontal avalanche lobes. Valley centre filled by coarse pyroclastic density current deposits, and post-Boxing Day block-and-ash flow and lahar deposits. Pyroclastic density current deposits include arcuate deposit inside bend at lower left, and table-shaped overspill pyroclastic deposit in left-centre (photo J.C.K.). (b) Overspill areas viewed toward west in June 1998, showing complex pile-up of avalanche debris against and over the west channel bank (photo J.C.K.). Avalanche deposits on Fergus Mountain flank at lower right (for detailed view, see Fig. 20c). (c) Detail at east edge of Fergus Mountain overspill, January 1998 (photo B.V.). View to west. Sheared Soufriere Subunit pyroclastic sequence on left, talus unit on right. Near sample location 15-2 (Fig. 23). (d) View of Morris' overbank deposit, 8 July 1998, showing prominent hummocky flow front 20-40 m high (photo MVO). Ground surface in front of avalanche was scoured by the subsequent pyroclastic density current (grey deposits) and was veneered by a tar-like coating. A second flow front in the avalanche deposit is in sunlight a few hundred metres behind first flow front.

locally throughout the avalanche deposit. In the proximal region, dykes of dome-lava clasts intruded from below into megablocks of Charteuse Subunit materials (Fig. 24b). The lava clasts may have been derived from November 1997 block-and-ash flow deposits that capped the valley-fill sequence before Boxing Day. In the overspill location near Morris', rapid flow of the avalanche over the high

valley walls probably induced marked local dilatation and fracture of cohesive megablock facies, promoting the infilling. Evidence at some locations, such as near Figure 20d. revealed that some cracks were filled from above with Talus Unit debris. Such dykes could have resulted from avalanche flow over the relatively rough topography of the valley floor, causing cracks to open in stretched

THE BOXING DAY SECTOR COLLAPSE

387

Fig. 22. (continued)

avalanche megablock facies debris. The cracks were then filled by cohesionless lava particles, caving downwards under the influence of gravity and facilitated by jostling. Mutual relationships between pyroclastic current and debris avalanche deposit The sector collapse was immediately followed by a high-energy pyroclastic density current (Sparks et al 2002). Several depositional facies were associated with this pyroclastic current (Ritchie et al. 2002), and the general area of the White River valley and adjacent areas near St Patrick's and on the flank of Fergus Mountain were dominated by erosional features related to its passage (Figs 14 and

15). The debris avalanche deposit surface was scoured and striated. The debris avalanche was not generally hot. Its debris lay directly on top of the old vegetation and soil sequence, and buried leaves were unburnt. However, because it was scoured by the hot turbulent pyroclastic density current, wood that protruded from the avalanche deposit was charred. A black tar veneer several millimetres thick covered much of the landscape not buried by the avalanche deposit in the vicinity of the White River, and resulted from the nearly instantaneous distillation of the vegetation and organic soil by the hot current (Figs 14 and 22d). A second, highly energetic pulse of the pyroclastic current then passed abrasively over the tar and shaved-off the trunks of trees (Sparks et al. 2002). The tar was absent beneath the debris avalanche deposit where it was excavated in a number of distal locations near Morris'

Fig. 23. Sample locations of debris avalanche deposits and associated pyroclastic deposits. Photograph views denoted by arrows and figure number. Base map from Figure 19.

THE BOXING DAY SECTOR COLLAPSE

389

Fig. 24. Deformation structures in avalanche deposit (photos J.C.K.). (a) Dome lava clasts intruded as clastic dykes (arrows) into massive Soufriere Subunit materials in the Morris' overspill deposit, upper lobe, in June 1998. (b) Clastic dykes of dome lava clasts intruded into Chartreuse Subunit materials, probable toreva block, east side of proximal region of avalanche deposit in June 1998. Lava clasts were probably derived from November 1997 block-and-ash flow deposits forming the valley surface material before the sector collapse. (c) Shearing of soft avalanche blocks into elongate slivers that enclose more rigid blocks. Distal region in June 1998. (d) Flattened, sheared block texture in avalanche deposit in distal region in March 1999. Shearing prominent near basal part of 20m flow front. Chartreuse Subunit overlain in sharp contact by Soufriere Subunit. with dome lava Talus Unit on top.

(Fig. 22d), which showed that the debris avalanche had deposited here before the hot density current reached the lower slopes. The evidence for relative timing of these events is not definitive in the main channel, south of the Morris' overspill. In some other cases, pyroclastic currents overtook the debris avalanche while the latter was still in motion, e.g. Mount St Helens (Voight et al. 1981; Glicken 1986, 1998) and the Taapaca avalanche, Chile (J. Clavero & R. S. J. Sparks, unpublished data). Several post-avalanche-emplacement pyroclastic flow deposits occurred. As well as those associated with the Boxing Day collapse,

deposits of later pyroclastic flows and lahars have infilled some areas, burying debris avalanche and pyroclastic deposits in the White River valley (Fig. 22a, b). A fines-poor, coarse (clasts to 3 m) unconsolidated pyroclastic breccia was deposited from the Boxing Day density current as sheets and patches overlying the avalanche deposit. These were lithologically similar to dome lava talus that comprised part of the avalanche deposit. The two deposit types, in principle, might be distinguished on the basis of texture (generally less matrix in talus), energy of emplacement (highest for the pyroclastic density current) or temperature (cooler in talus). In places

390

B. VOIGHT ET AL.

Fig. 24. (continued)

the deposits could be traced into unambiguous pyroclastic current deposits away from the avalanche deposits. In some locations the blocky flows rode up or beyond the walls of the White River valley, resulting in coarse breccia deposits plastered against valley walls or on overbank benches, with a striking streamlined morphology in profile and plan (Figs 14 and 22a; cf. Ritchie et al 2002). Within the hummocky area of the distal avalanche deposit, massive pyroclastic deposits with a repetitive two-layer stratigraphy occurred. These draped over eroded hummocks and were ponded between them (Fig. 21c, d). A wavy-bedded and finely laminated, moderately sorted, fine-grained facies formed the upper layer, overlying a coarser, fines-depleted layer. These two layers were most prominent on top of hummocks and adjacent ridges, although either could occur separately and both had been eroded by subsequent events. The inter-hummock ponded deposits showed

extreme lateral textural and depositional variations, and lacked well developed laminae unless there was >5m distance between hummocks. In some places there were two coarse, poorly sorted massive layers, with a greenish lower layer, commonly inversely graded, rich in charcoal and containing yellow and orange hydrothermal chunks torn from the debris avalanche deposit, overlain by another massive layer rich in dome lava clasts with a pinkish, finer sandy matrix, and with prominent degassing pipes. Locally, a well sorted compact pinkish sandy deposit was found between two sets of the massive, poorly sorted layers containing clasts from the avalanche. On ridges only affected by the pyroclastic density current, and in some areas on the outer edge of the avalanche deposit, this well sorted deposit overlay a surface of striated tar. Calder et al. (1998) interpreted this well sorted deposit as the result of a surge-derived pyroclastic flow formed by

THE BOXING DAY SECTOR COLLAPSE

391

avalanche I (avalanche megablock facies) was characterized by HjL = 0.17, and the full Mount St Helens avalanche had H/L = 0.09 (Sousa & Voight 1995). H/L ratios have been observed to be volume-dependent (Fig. 25), with the observed HIL ratios ranging from 0.09 to 0.18 (average 0.13) for Quaternary volcanic debris avalanches between 0.1 and 1 km3 in volume, and from 0.04 to 0.13 (average 0.09) for avalanches > l k m 3 (Voight et al. 1985; Siebert 1996). Thus, from this simple measure, the Boxing Day debris avalanche had low mobility compared to those reported for large volcanic debris avalanches elsewhere (Fig. 25). This result is consistent with the relatively small volume of this deposit in comparison to, say, the Mount St Helens debris avalanche, and the tortuosity of the channel. Numerical modelling of the emplacement dynamics is considered below. Physical properties

Avalanche deposit materials Fig. 25. Ratio of vertical fall height (H) to horizontal runout distance (L) versus volume for volcanic (triangles) and non-volcanic (crosses) debris avalanches, with regression lines after Voight et al. (1985). Boxing Day avalanche (SHV) compared with Mount St Helens (MSH) avalanche I (block facies, open diamond), and entire avalanche (black diamond), based on Glicken (1986, 1998).

sedimentation from the low-concentration part of the pyroclastic density current (cf. Druitt et al. 2002). Deposition on sloping hummock flanks resulted in flows draining onto the ponded facies, thus creating complex spatial and temporal relationships although all events occurred within a brief timeframe. The ponded pyroclastic deposits still had a temperature of 57°C at about 40 cm depth in March 1999.

Dynamical constraints on avalanche emplacement A possible indicator on debris avalanche dynamics was the emplacement of deposits on the sides of the White River valley near the Fergus Mountain narrows. Simplistically, from conversion of kinetic energy to potential energy and neglecting friction, the deposit at Morris', and elevations 200-220 m and 60m above the surface of the channelled avalanche deposit, suggests a minimum flow velocity of 3 5 m s - 1 . The overspill may have been partly influenced by the rise of the peak avalanche flow wave through the constriction at the Fergus Mountain narrows. If so, the minimum velocity inferred by calculation is to some extent reduced. A reliable but limited constraint on avalanche dynamics was offered by the destruction of the MSPT seismic station at St Patrick's by the pyroclastic density current (Fig. 2a), at 03:03.3 LT (Sparks et al. 2002). The initial collapse began at 03:01.0. Since the overspilled part of the avalanche had stopped when overrun by the hot density current near Morris' (Figs 14 and 22d), the avalanche runout of c. 2.6km (measured from the toe of the slide scar, to Morris') occurred within 138s indicating a minimum average velocity of 1 9 m s - 1 . If the full channelled avalanche runout of c. 3.7km also occurred within this period, a possibility not excluded by the evidence, then the minimum average velocity was 2 7 m s - 1 . The mobility of volcanic debris avalanches can be empirically characterized by the ratio of maximum vertical fall height (H) to overall travel distance (L), in this case as measured from the rear scarp of the original slide mass to the snout of the deposit (Ui 1983; Voight et al. 1983, 1985; Siebert 1996). Using simplified models, this ratio can be cautiously interpreted in terms of the apparent coefficient of friction of the avalanche (Pariseau & Voight 1979). The highest part of the Boxing Day debris avalanche fell from a maximum height of about 1000m (near the summit of the dome), and its travel distance L as defined above was about 4500m suggesting H/L = 0.22. By comparison, at Mount St Helens the debris

Granulometry. The avalanche deposit was extensive and the material heterogeneous, varying from clay particles to megablocks >100 m wide. Representative sampling was impossible for much of this material, but we attempted to characterize it at least in broad terms. Because the avalanche deposit contained boulder-sized clasts unsampled for sieve analyses, the actual sorting of the deposits is commonly poorer, and the median diameters larger, than indicated by our data. Most samples represented avalanche megablock facies or non-cohesive facies materials, in large part derived from source clastic rock of the failed edifice, or unconsolidated talus deposits. The grain-size characteristics observed were largely influenced by the source materials and the textural modifications that occurred as a result of sector collapse and debris avalanche motion. Grain-size distributions were determined on samples (several kilograms) from diverse localities on the avalanche deposit (Fig. 23). Most samples were subjected to standard dry-sieve analyses, with sieves from 5 to -6 phi. When appropriate, optical scanning of aqueous dispersions (the 'microtrac' method) was used to ascertain grain sizes of the fine fraction, to 9 phi. Most samples were gravelly sand or sandy gravel, with up to 15% silt and clay grade particles (Fig. 26). Typical values were about 6-15wt% for combined clay (< 0.004 mm) and silt, 30-53 wt% sand, and 33-62 wt% for the >2mm fraction of pebbles, cobbles and organic debris.

Fig. 26. Percentage of gravel (>2mm), sand (2-0.063 mm), and silt + clay (<0.063mm) in samples of the debris avalanche deposit.

392

B. VOIGHT ET AL.

Fig. 27. Typical grain-size histograms of debris avalanche. Sample locations shown in Figure 23.

Histograms of grain-size distributions of the deposits had several distinctive characteristics (Fig. 27). In some cases the coarsest fractions (<—6 phi) were absent (as reflected by truncated histograms at coarse modes), but field observations indicated that in most cases this was the result of incomplete sampling owing to the limited sample size. The histograms were mostly polymodal. Modes around 0 to 1 phi and less than -3 phi were relatively common, and minima around -1 to -3 phi occurred in many samples. A frequency minimum between -1 and -2 phi occurred in samples of megablock facies material at Mount St Helens, Augustine and Shiveluch volcanoes (Glicken 1998; Siebert el al. 1995; Belousov et al. 1999), but such a tendency was undeveloped or only weakly developed in our samples. Such diffuse polymodal grain-size distributions constituted poor sorting and a wide range of median diameters. Median grain size typically varied from 0.6 to 2.9 phi, and exceptionally to -4.8 phi (sample 308-ld). Sorting coefficients (Inman 1952) ranged from 1.7 to 4.5. On a plot of median diameter versus sorting (Fig. 28), most samples fell into the coarse-grained part of the axial zone of the pyroclastic flow field of Walker (1971), suggesting these debris avalanche deposit samples had textures similar to pyroclastic flow deposits. Probably this correlation reflects the fact that much of the avalanche source material was derived from ancient pyroclastic flow deposits. Towards both the largest and the smallest median diameters, some improvement in sorting was observed. Within the White River valley, the sorting coefficients and median diameters maintained roughly similar values from the source to the distal end (Fig. 29). The biggest differences occurred as a result of source-rock lithological variation, with the most extreme values in fines-depleted coarse materials at one end of the spectrum, and clay layers or lenses on the other end, representing local clay-rich alteration products in Soufriere Subunit material. These variations were present at distances >3km from the crater, reflecting mainly the wider range of materials sampled and transported to the distal region.

Geotechnical tests. For some samples, a range of geotechnical tests was carried out. The most detailed tests were conducted on samples at location 15-2, representing overspill deposition on the west flank of Fergus Mountain (Fig. 23). The outcrop displayed Fumarolic Unit materials with sheared stratification. Sample 2A, taken as typical of the southern part of the outcrop, was mottled grey, brown and yellow clayey sand with gravel, Unified Soil Classification System (USCS) group SM, combined silt and clay content c. 40wt% (Fig. 30). (For a thorough account of these standard engineering descriptions and classifications of soils, see Bell 1993, Ch. 5.) Sample 2B was an overlying dark brown-grey silty sand (with gravel), USCS classification group SM, silt and clay

Fig. 28. Sorting and median diameter (Inman coefficients) for the debris avalanche deposits. Cross-hatched field shows pyroclastic flow deposits after Walker (1971).

THE BOXING DAY SECTOR COLLAPSE

393

Fig. 29. Granulometric characteristics of block facies of debris avalanche deposit versus distance from the volcano. (a) Median diameter. (b) Sorting. (c) Percentage gravel. (d) Percentage sand. (e) Percentage silt + clay. Granulometric definitions are given in caption of Figure 26.

content c. 31wt%. Sample 2C, representing a continuous layer in the outcrop, was an intact block of moist grey-brown plastic clay, USCS group CH, silt and clay content c. 96wt%. This clay block was cut out from the outcrop, wrapped in wet cloth, and transported to the laboratory in a sealed hard-plastic container. Sample 2A was a shovel sample, and laboratory shear strength tests were conducted on representative split-screened portions of it using procedures described in the Appendix. The specimen was consolidated and sheared, and then repeatedly sheared to provide a measure of displacement-weakening. Curves of shear stress against cumulative horizontal shear displacement showed significant reduction in strength (Fig. 31). Shear stress versus normal stress plots indicated linear Coulomb effective-stress strength relations over the normal stress range examined (Fig. 32), with peak strength indicated by: cohesion = 57.5 kPa, peak friction angle = 25 . Resi-

dual strength (large post-peak displacement) parameters were: cohesion = 33 kPa, residual friction angle = 17'. The term effective stress refers to the standard practice in geotechnical engineering of considering the differences between the total applied stresses and the pore fluid pressures (Lambe & Whitman 1969). Similar shear tests were carried out on clay-rich sample 15-2C. which showed contractive behaviour and profound reduction in strength (Fig. 31). Effective-stress strength relations were linear over the normal stress range examined (Fig. 32). with peak strength indicated by: cohesion = 77.8 kPa, peak friction angle = 13.5 . Residual strength parameters were: cohesion = 57.0 kPa. residual friction angle = 3.0C. These preliminary laboratory results indicate, to a reasonable approximation, the minimum frictional strength of Fumarolic Unit avalanche debris. The addition of coarse particles could increase

394

B. VOIGHT ET AL.

Fig. 30. Size-distribution curves for edifice (1996) and avalanche debris (1998) samples subjected to geotechnical testing.

the strength above these matrix values, particularly for volumetric proportions >25 % (Medley 1997), although the reported values may be reasonable estimates of the local strength at avalanche or megablock boundaries. The values may also provide lower-bound strength estimates for full-scale slip-surface failure of Fumarolic Unit fragmental deposits during the Boxing Day sector collapse.

Edifice materials Block samples of relatively weak intact tuff had been collected by B.V. and S.R.Y. at the base of Galway's Wall in March 1996, in anticipation of future instability (Voight 1996). Towards the end of 1996, the sampling site (near Fig. 5b) was buried by rockfall debris and dome talus (Fig. 6). Additional samples of weathered tuff were taken in March 1996 from the Gages Wall area of English's Crater (see Fig. 2), and samples of fresh (1995 lava) and altered (prehistoric lava) dome material were also acquired. In order to provide constraints for stability assessments, the block samples were subjected to laboratory triaxial rock-mechanics testing to provide the shear strength of intact material as a function of confining pressure, using procedures discussed in the Appendix. Additional testing was carried out to obtain intact tensile strengths and specimen dry bulk density. Note that strengths reported below are for intact materials, and values appropriate for the in situ jointed rock-mass would generally be less (Voight 2000). In general, the experimental Mohr-circle data indicated curved strength envelopes as a function of confining pressure. For Galway's Wall tuff, with dry bulk density of 1980 kg m-3 and tensile strength of 1.21 MPa, results are shown in Figure 33. The data gave good agreement with the Hoek & Brown (1980) failure criterion that predicts a parabolic envelope for brittle rocks subjected to compressive stress conditions, and approximates classic Griffiths theory in the region of tensile effective normal stresses.

Fig. 31. Shear stress versus displacement for reversal direct-shear tests on Soufriere facies debris avalanche samples. Note strength loss with displacement. (a) Remoulded sample 15-2A, clayey sand with gravel, normal stress 0.86 MPa. (b) Block sample 15-2C, highly plastic inorganic clay, normal stress 1.72 MPa.

Considering the full Galway's Wall dataset, the Coulomb parameters yielded a linear fit with cohesion = 2.19 MPa and friction angle = 33.0°, although this set of parameters underestimated strength at intermediate confining pressures, and overestimated strength in the tensile region (see Voight 2000). A piece-wise linear fit can provide reasonable Coulomb parameter approximations for different ranges of normal stress. Thus, for confining pressures >2.07MPa, the parameters cohesion = 3.65 MPa and friction angle = 31° provide an adequate fit. The weathered Gages Wall tuff was less dense and weaker, with dry bulk density of 1840 kg m-3 and tensile strength of 0.41 MPa. Coulomb parameters for moderate compressive confining pressures were, approximately, cohesion = 1.7 MPa, friction angle = 32°, although, as noted, the envelope was non-linear (Fig. 33). Corresponding data for fresh dome lava collected in the southwest moat inside English's Crater, and mildly altered lava from prehistoric dome-derived breccia exposed in Galway's Wall, were obtained. Approximate Coulomb friction angles are 55° and 44°, respectively. In addition, disaggregated materials were collected from weathered tuffs presumed to be representative of the outer-slope base of Galway's Wall, and also from weathered portions of Gages Wall (Fig. 1). Both materials were classified as grey-brown silty sand with gravel, USCS group SM (Fig. 30). Laboratory shear strength tests were conducted, using methods summarized in the Appendix. At all normal stresses, higher resistance was obtained on second-cycle shearing, probably reflecting enhanced interlocking and dilatation between sliding or rotating angular grains, whereas further displacements caused some reduction in strength (Fig. 34). Linear Coulomb

THE BOXING DAY SECTOR COLLAPSE

395

Fig. 32. Shear strength versus effective normal stress for Soufriere facies debris avalanche materials subjected to direct-shear tests, (a) Remoulded sample 15-2A, clayey sand with gravel. (b) Block sample 15-2C, highly plastic inorganic clay. Open symbols show peak shear strength; filled symbols show residual shear strength.

Fig. 33. Shear stress versus effective normal stress for intact samples of edifice materials subjected to triaxial and indirect-tension tests. Data plotted as Mohr circles, with curved envelope according to Hoek-Brown criterion, (a) Gages Wall tuff, altered. (b) Galway's Wall tuff.

effective-stress strength relations applied approximately over the normal-stress range examined (Fig. 35). For both sites, a single nominal set of parameters seemed appropriate, namely zero cohesion, peak friction angle = 35°, residual friction angle = 31°. These parameters may be appropriate for disaggregated material, or blocks bounded by uncemented joints or joint gouge within the tested range of normal stresses.

pated dome growth. A trial slip surface was postulated. and the shearing resistance required to equilibrate the mass of material on this surface (and other applied forces) was calculated by statics. The calculated resistance was compared with available shear strength to yield a factor of safety F, defined as the quotient of shear strength of the material and the shear stress required for static equilibrium. The value F= 1 represents a condition of incipient failure, and F > 1 represents stability. The procedure was repeated for other postulated slip surfaces, and the lowest F was found by iteration (Duncan 1996; Voight 2000). Laboratory tests on block samples of indurated tuffs and dome rock, and samples of disaggregated materials, as described above, were used along with literature data to constrain the stability analyses (Voight 1996): scaled reductions were used for rock-mass properties (Hoek 1983; Voight 2000). Seismic loading was simulated simply by a coefficient that represented the potentially destabilizing earthquake force due to horizontal acceleration of the material (Kramer 1996). The results of these stability analyses suggested that the outer part of Galway's Wall was marginally stable with respect to static loading for the dome geometry existing at that time. Stability decreased when pseudo-static earthquake loading was added, and for horizontal accelerations on the order of O.lg. shear failure on the upper south face of Galway's Wall was predicted by the analyses. The failures predicted were shallow, not deep-seated. Successive slope failures triggered by successive earthquake shocks were predicted to reduce wall thickness by exterior slabbing, rather than by massive deep-seated failure. The results did not preclude the latter possibility, given the uncertainties involving the forces caused by a shallow intrusion (Young et al. 2002). potential

Stability assessments and failure mechanisms

Studies made before the sector collapse Growth of the andesitic lava dome at Soufriere Hills caused decreasing structural stability of the southern sector of the volcano. The possibility of a future instability problem in this sector was recognized in early 1996, and clear warning signs of growing instability were recognized in November-December 1996 (Young el al 2002). Catastrophic sector collapse occurred a year later on 26 December 1997, during a period of enhanced seismicity and after two months of southward expansion of the lava dome and talus apron over the hydrothermally weakened area. Preliminary quantitative stability assessments, using two-dimensional limiting-equilibrium analyses, were made during hazards evaluations at Soufriere Hills in March 1996 and December 1996 (B. Voight, unpublished data). In such analyses, a representative topographic profile and cross-section was drawn, to represent existing conditions and also some future conditions related to antici-

396

B. VOIGHT ET AL.

Fig. 34. Shear stress versus displacement for reversal direct-shear tests on disaggregated edifice tuffs. Note small strength loss with displacement, (a) Sample 1: Gages Wall tuff, altered, (b) Sample 2: Galway's Wall tuff.

Fig. 35. Shear strength versus effective normal stress for edifice materials subjected to direct-shear tests, (a) Sample 1: Gages Wall tuff, altered, (b) Sample 2: Galway's Wall tuff. Open symbols show peak shear strength; filled symbols show residual shear strength.

through-going structural discontinuities within the wall, and assumed material properties. Indeed some radial cracks were observed and monitored (Young et al. 2002). However, it was recognized that deterioration of the wall by slabbing was most likely, and this could cause reduction of the wall height and consequently lead to overtopping by the growing dome, with rapid erosion of the friable, jointed wall rock. The deterioration of the wall in general followed the lines suggested by the analyses. Wall failures involved release of shallow slabs of jointed rock, triggered by the larger felt shocks in repetitive seismic swarms. Then, after several shifts in the locus of dome growth had reduced pressure against the south sector, lava overtopped the low point on Galway's Wall in March-April 1997, block-and-ash flows severely eroded parts of Galway's Wall, and Galway's Soufriere was partially buried by talus. In addition to these analyses, axisymmetric finite element modelling was conducted to explore the stress changes associated with dome growth and conduit pressurization, for the geometry existing in late 1996 (Wadge et al. 1998). The analysis assumed homogeneous elastic media, and was somewhat limited inasmuch as pore pressures, seismic loading, material zonation, and plastic deformation were not treated. Relative stability was interpreted from the ratios of elastic-media shear stresses to an assumed Coulomb relation.

outlined above. There are various ways in which these procedures can be manipulated, and the analyst needs to consider which methods are most accurate, and which of the accurate methods can be applied most easily (Duncan 1992, 1996; Fredlund 1984). A good comparative summary of the various methods is given by Bromhead (1986). The modified Bishop method was used in this work; the method satisfies moment and vertical force equilibria, and gives values of factor of safety F that fall within the range of equally correct solutions as determined by so-called exact methods (Duncan 1996; Bromhead 1986; Lambe & Whitman 1969). The method assumes a slip surface shaped as a circular arc, with material above this surface subdivided in a series of vertical slices for computation of body forces. As indicated previously, results for a specified slip

Limiting-equilibrium analyses made in 1998 Limiting-equilibrium stability analyses were carried out in January 1998, soon after the sector collapse, using the general procedures

Table 1. Assumed material properties for limiting equilibrium analyses Unit weight (kNm- 3 )

Cohesion (kPa)

Friction angle (deg)

Model A 1 2 3 4

20 20 20 20

1 10 1 5

39 42 39 45

Model B 1 2 3 4

20 20 20 20

1 10 1 5

37 42 38 45

Material

THE BOXING DAY SECTOR COLLAPSE

Fig. 36. North-south cross-section for Boxing Day collapse, showing materials and slip plane boundaries assumed in limit-equilibrium stability analyses. Material properties are given in Table 1. The water table, indicated by the dotted line at 'w', joins the ground surface at the toe of the slope.

surface are summarized as a factor of safety, with the value F= 1 implying incipient failure. In general, the F for three-dimensional analysis is slightly greater than the F for two-dimensional analysis (Duncan 1992). Table 1 lists material properties assumed, specific unit weights, and strength properties given by cohesions and effective-stress friction angles, varied selectively according to material type. The distributions of materials, and piezometric surface assumed, are shown in Fig. 36. A centre of rotation and radius was specified to represent the actual Boxing Day failure surface (assuming it to be circular), as inferred from construction of cross-sections (Fig. 12). The slip surface was assumed to be constrained by the south edge of the partly eroded Galway's Wall (Figs 36 and 9b). Results of analyses for two similar sets of model properties (Table 1) are discussed here. The parameters represented bulk, aver-

Fig. 37. Seismic data for 24-26 December 1997, for MBWH (Windy Hill, vertical component) seismic station (Fig. 2). (a) Number of earthquakes per hour, (b) Amplitudes of hybrid earthquakes as a measure of relative energy in units of m s - 1 . (c) Real-time seismic amplitude measurements (RSAM), in arbitrary units.

397

age properties assumed for these zones. However, we recognized that each zone was in fact heterogeneous and that local properties within a zone might differ substantially from the averages assumed. For model A the calculated safety factor F for the Boxing Day surface was 1.22 under static loading, and 1.01 for applied seismic loading with pseudo-static seismic acceleration coefficient of 0.1. Similarly, slightly weaker model B suggested F= 1.15 under static loading, and F=0.96 for a seismic coefficient of 0.1. These results could be interpreted to imply marginal stability for the static conditions considered. Small perturbations of applied loads, whatever the cause, could then bring the slope to a condition of failure. Such perturbations may be considered as the trigger mechanism, but the overall causes of failure are more complex and indeed include all the various factors that brought the slope to its condition of marginal static stability (Voight & Elsworth 1997, table 1). Prominent among these are the major topographic changes reflecting dome growth, wall erosion and talus deposition that occurred since 1996. as well as the history of hydrothermal alteration and possible episodes of prior slip. Potential triggers included the direct effects of seismic shaking, pore-fluid pressure changes induced by seismic shaking or other mechanisms, strain-weakening by deep-seated creep, and loading by lava extrusion. A combination of these cannot be excluded. For example, increase of pore-fluid pressure in Galway's Soufriere materials (Fumarolic Unit source materials) to an artesian condition could have resulted in instability. This magnitude of pore pressure would require an average piezometric surface above the original ground, but such a circumstance cannot be discounted given the local presence in or under the soufriere of weak lowpermeability clay layers, pre-existing shallow slip surfaces associated with landslides active in 1996 and earlier times, and the rapid loading of this area by dome-block talus in the weeks preceding the collapse. A low F (near unity) could also have promoted localized creep of weak materials, and the deformation tests previously described indicated that some Fumarolic Unit materials were indeed characterized by relatively low frictional strength and profound strain-weakening. Thus, the static value of F may have gradually decreased in the weeks preceding the failure. However, in addition, the coincidence of the Boxing Day sector collapse with a brief period of enhanced seismicity, from 14:30 on

398

B. VOIGHT ET AL.

Fig. 38. Numerical model of north-south cross-section for Boxing Day collapse, seismic loading, (a) Material zones: 1, dome talus and rockfall debris (purple); 2, fresh lava (yellow); 3, altered soufriere materials (brown); 4, old edifice materials (tan). White arrow marks the water table. Coulomb parameters assumed in zone 1: c' = lOkPa, = 40°; zone 3: c' = 50kPa, = 40°; other zones strong. Seismic loading coefficient, 0.15. Movement indicated by velocity vectors, arrows (max 0.037 m s . ) . (b) As (a), with deformation accentuated from strain weakening. Zone 3: c' = 20kPa, - 20°.

24 December onwards, was probably not accidental (Fig. 37). To judge from previous periods of enhanced hybrid swarm activity, the seismicity was probably accompanied by an enhanced pulse of effusion of gas-charged lava (Voight et al 1999). The hybrid swarm began with events every 20 minutes or so, and increased in intensity until the late evening on 25 December, after which the signal was effectively a tremor (Fig. 37; Sparks et al 2002). The individual hybrid events increased in amplitude as the swarm progressed, and, although even the largest of these events was smaller in magnitude than similar events recorded in November 1997, the sector geometry and its static loading was now at a more critical level. The amplitude of tremor peaked around 23:00 on 25 December, declined briefly, and then continued with ascending amplitude (roughly doubling the previous maxima) to the onset of the sector collapse at 03:01 (Fig. 37). The near-coincidence of distinct hybrid seismicity with the sector collapse was suggestive of a possible causative relation, and

supported the inclusion in the modelling of transitory earthquake loading by equivalent static forces. Strong-motion data on volcanoes are rare, although Voight et al (1983) reported information from Mount St Helens, and such data were acquired at Montserrat in May-June 1997, when ground accelerations measured on Chances Peak from shallow hybrid earthquakes were as much as O . l g (B. Voight, unpublished data). The accelerations associated with earthquakes in December 1997 may not have been this severe, but we nevertheless have explored the use of pseudo-static seismic coefficients of about 0.1 for analyses of the 26 December 1997 sector collapse.

Numerical deformation models The modelling work of 1996-1998, considered above, was of limited applicability in that stress-strain behaviour, development

THE BOXING DAY SECTOR COLLAPSE

399

Fig. 39. Numerical model of lava shearlobe emplacement at Soufriere Hills. (a) Lava in the upper part of the slope (yellow) is pushed southwards (to left) at a constant rate, and creates a velocity field (vector arrows, max 0 . 6 5 m s - 1 ) . (b) The velocity field creates a shear zone that extends to the toe of the slope and includes altered soufriere material. The deformed grid illustrates the displacements: the shear strain-rate field is shaded. Grid distortion scale is 190x. Zone at toe of slope is soufriere material that is susceptible to strain-weakening, with effects shown in Figure 40.

of plastic zones and strain-weakening, and dynamic loading, among other aspects, could not be adequately represented by the methods employed. It is preferable that the mechanism of failure should be elucidated by modelled strain localizations, rather than pre-specified as in limiting-equilibrium analyses, or vaguely interpreted from elastic models. The previous results did not preclude seismic shaking as a trigger mechanism, but the methods did not facilitate consideration of some other potential causes, such as enhanced conduit pressure, strain-weakening, and the invigorated emplacement of a fresh shear-lobe of lava. In order to illuminate these issues, we evaluated a series of two-dimensional explicit plain-strain finite difference models of the Soufriere Hills slope, using the Fast Lagrangian Analysis of Continua procedure (Cundall & Board 1988; Coetzee et al. 1998; Detournay & Hart 1999). The slope was subdivided into four zones, representing (1) fresh dome talus, (2) fresh lava, (3) altered soufriere material, (4) old

edifice, as shown in Figure 38, with the actual material parameter values different in various models. In general the bulk modulus was taken as 20GPa and shear modulus was l0GPa. The models could undergo plastic deformation when Coulomb yield limits were reached, which are different in the various zones. In some cases pseudo-static loads were used to simulate earthquake loading. Figure 38a shows assumed material zonation and displacement velocity vectors for the slope subjected to seismic loading, with Figure 38b showing enhanced deformation after strain-weakening. As the velocity and strain fields appeared to be more or less consistent with the deduced position of the actual failure surface at Montserrat (Fig. 12), the results did not preclude the hypothesis of seismic shaking alone as a trigger of the Boxing Day sector collapse. The lava shear-lobe emplacement hypothesis was treated next, whereby lava in the upper part of the slope was displaced southwards at a steady rate, creating a shear zone that extended to the toe

400

B. VOIGHT ET AL.

Fig. 40. (a) As Figure 39, with material in outer slope strain-weakened to: c' = 20 kPa, = 30° in zone 1 (talus and rockfall debris); and c' = 20 kPa, — 20° in zone 3 (soufriere material). Lava-lobe emplacement still continues as in Figure 39 (note grid distortion at upper right), but gravity-driven movements on outer slope dominate the grid distortion, shear strain-rate and velocity vector field. Grid distortion scale 24x, max. vector 5.0ms - 1 . (b) Gravity collapse of the outer part of the slope results in an oversteepened and decompressed face of fresh lava, which begins to collapse (grid distortion scale l . l x ) . Generation of the decompression-induced explosions is beyond the scope of the model.

of the slope and included altered soufriere material (see velocity field, Fig. 39a, and shear zone, Fig. 39b). In this model, shear-zone materials in zones 1 and 3 were assumed to strain-weaken, so that spontaneous, localized gravitational collapse might then occur in the outer slope (Fig. 40a). The deformation rates due to gravitational loading exceeded those associated with the southerly displacement due to continuing lava shear-lobe emplacement (Fig. 39b), thus promoting the gravity collapse. This, in turn, resulted in an oversteepened and decompressed face of fresh volatile-rich dome lava, which began to collapse, in part explosively, and to generate a violent pyroclastic density current (Fig. 40b). The details of the explosive initiation are beyond the scope of this type of model, but are explored by Fink & Kieffer (1993), Alidibirov & Dingwell (1996), Voight & Elsworth (2000) and Woods et al (2002). In general, these models suggest that increased loading and strain-weakening

deformation induced by lava lobe emplacement are viable trigger mechanisms for the Boxing Day sector collapse; some influence of superposed seismic loading is not excluded. Edifice stresses resulting from conduit pressure could also be modelled, although these stresses dissipate radially and a twodimensional model only provides an upper bound to the stresses. There were no field observations that suggested that propagation of brittle shear fractures occurred in a fashion that would connect an outer-flank collapse with stressed rock surrounding the pressurized conduit. Overall the analyses seemed to favour (1) collapse triggered by a pulse of rapid lava shear-lobe emplacement, with subsequent gravity-driven slip-surface localization influenced by enhanced loading, creep and strain-weakening, and/or (2) seismic shaking. We favoured a combination of the two mechanisms, inasmuch as lobe emplacement was accompanied by seismicity which may have

THE BOXING DAY SECTOR COLLAPSE

changed the distribution of stresses, and the seismicity itself was not particularly strong in comparison with previous seismic episodes and thus may not have been the exclusive trigger. Numerical modelling of dynamics of debris avalanche emplacement Assumptions

where 1 , 2 and D are characteristics of the material that in principle could be measured from deposits. Extrapolating Pouliquen's results to real avalanches, very approximately D = 7d. where d is the 'effective mean diameter' of flowing material. The equation provides a friction angle ranging between two values 8\ and 82 (61 < 62), depending on the instantaneous values of the velocity and thickness of the flow. The higher the velocity, the higher is the friction. Numerical emplacement models

The dynamics of the debris avalanche motion have been evaluated by numerical modelling (Heinrich et al. 2001). In these studies: (1) the avalanche was idealized as the flow of a homogeneous incompressible continuum due to its observed, macroscopic fluidlike behaviour; (2) bed erosion was neglected; (3) mass and momentum conservation equations were depth-averaged over the flow thickness, since the characteristic length of the avalanche was much larger than the thickness; (4) energy dissipation within the flow was neglected and the slope-parallel velocity was assumed approximately constant over the thickness (Savage & Hutter 1989); and (5) longitudinal gradients of the deviatoric stress were neglected throughout the flow. The governing equations used a slope-parallel and slope-normal co-ordinate system (Heinrich et al. 2001). Basal friction was modelled by Coulomb-type friction laws, where 6 is the friction angle between the rough bed and the mass. The friction angle could be assumed to be constant, independent of the shear rate (Savage & Hutter 1989; Naaim et al. 1997) or defined as a function of both the velocity (u) and the height of the flow (/?) (Pouliquen 1999). In the case of a constant friction angle, a constant ratio of the shear stress to the normal stress at the base was assumed, similar to a friction law for a rigid block on an inclined plane. Pouliquen (1999) argued that the constant friction assumption failed for granular flows over rough bedrock, and proposed an empirical friction coefficient = tan as a function of the Froude number (u/-\/(gh)) and the thickness h of the granular layer: (u,h ) = tan

401

+ (tan

- tan ) exp

The numerical model was based on a shock-capturing method, similar to those used to simulate shock waves in compressible flows, which appeared to be stable and accurate in other applications (Mangeney et al. 2000; Heinrich et al. 2001; Assier et al. 2000). The method is a one-dimensional Lagrangian approach, based on a highorder Godunov-type scheme as discussed by Heinrich et al. (in press). The landslide is initialized in the simulation by a parabolic shaped solid with volume of 50 x 10 6 m 3 , released from rest. A first series of numerical simulations was performed using the simple Coulomb friction law. for different values of the basal friction angle. The best agreement was obtained for 13 < < 15 (Heinrich et al. 2001). However, with this model a steep flow snout was absent, contrary to the field observations, and the model overspill areas located west of the White River valley did not correspond well to the field observations. In a second series of simulations, using Pouliquen's friction law. the observed phenomena were approximately reproduced for around 10 , 62 around 20 and assumed mean diameters of the order of 1 m (Fig. 41). By trial and error, the friction angles 8\ and 62 were chosen respectively at 11 and 25 . with d, the theoretical mean particle diameter, ranging from 1.35m to 2.6m. Model velocities were comparable with the field estimates as discussed above. For effective mean diameters d < 2.2m the avalanche overtopped the west bank of the White River valley and flowed for an excessive distance (Fig. 41c). For d > 2.2 m the deposit area was somewhat smaller than observed. The calculated runout distance was partly dependent on the mean diameter, with a difference of 300m obtained between solutions for d= 1.35m and 2.6m.

Fig. 41. (a, b) Flow evolution for Pouliquen's friction law with = 11 and 62 = 25 , and a mean particle diameter d=2.2m. (c) Flow position at / = 200 s for the same angles and and d = 1.35m. (d) Flow position at t = 200 s for the same rheological parameters (a) and (b) but using a new topography with a partial filling of the White River valley.

402

B. VOIGHT ET AL.

Fig. 42. Deposit thicknesses (m) calculated by Pouliquen's friction law with = 11° and 62 = 25° and a particle diameter d— 2.2m.

A good general agreement was obtained with deposit thicknesses (Fig. 42). West of the Fergus Mountain bend, the maximum model deposit thickness was about 10m, fairly close to that observed. In the valley, deposit thicknesses ranged from 7 to 60m for distances of 300 to 800 m from the sea, with a more pronounced frontal snout (c. 20m) than for models calculated by simple Coulomb friction. The simulated avalanche stopped 200m from the shoreline. A final simulation was carried out to estimate the influence of topographic changes in the valley. Starting with pre-eruption topography, the valley was then arbitrarily filled to 0.75 depth before releasing the avalanche volume. Numerical results are presented in Figure 41d for Pouliquen's law with friction angles between 11° and 25° and D = 2.2m. The results at t = 200s showed that overspill flow at the Fergus Mountain bend was accentuated, and resulting overspill deposit areas were very large. The volume passing the bend in the valley was then smaller, and so smaller thicknesses were obtained in the lower part of the White River valley. As a result of thickness reduction, the angle 82 was activated, which finally led to shorter runout. Thus, in this case, partly filling up the valley and smoothing topography did not lead to greater downstream velocities and runout, and smaller deposit thicknesses were obtained at the lower end of the valley. However, this counterintuitive result cannot be generalized, as less extreme filling might have reduced localized roughness of the channel floor, and yet not have facilitated an excessive overspill; in this case, intuition would have correctly suggested an increased runout.

Discussion of model results The best agreement was obtained for a model using Pouliquen's friction law, with an assumed effective mean diameter of 2.2m and

friction angles varying from 11° to 25°. This best-fit model-based particle diameter was greater than the median size of small-volume samples, and was perhaps also larger than field values even if much larger sample sizes were considered, although we must also recognize that fragmentation occurred during the flowage process (cf. Ui et al. 1986). In part this result may reflect the very small range of sizes considered in the original Pouliquen analysis, and also the need to extrapolate the size calibration procedure to very poorly sorted materials. Comparisons between flows calculated by Coulomb and Pouliquen's friction laws have shown the importance of the dependence of the friction angle on the Froude number and the flow thickness. The Pouliquen model gave results in better agreement with thicknesses observed in the field than those calculated by a simple Coulomb law. This result suggests a shear-rate dependence in the mechanical behaviour of debris avalanches, or at least the action of a second parameter. Irrespective of the friction law used, the empirical value of the apparent friction angle required to reproduce the significant mobility of the Boxing Day avalanche in Montserrat was low ( < 15°), in general agreement with the widely observed but poorly understood apparent excess mobility of large debris avalanches (Voight et al. 1985). At Montserrat, as at Mount St Helens (Voight et al. 1983), pore-water pressure in disaggregated, fumarolically altered debris probably partly accounted for enhanced mobility. For water-saturated materials, the apparent friction coefficient is a function of the actual friction coefficient and the pore-fluid pressure (Voight 1978, pp. 154-155; Sassa 1988; Voight & Elsworth 1997, pp. 11-14). Geotechnical testing of avalanche debris at Montserrat suggested effective-stress friction angles for sandy textured debris of 25-35° (peak) and 17-25° (residual), with minor cohesive strength. The apparent friction values for these materials in a flowing avalanche associated with high pore-water pressures would be much reduced, and more consistent with modelled values. In addition, local seams of clay-rich materials have measured low frictional strength (3-14°). Such materials are only found locally and probably do not account for bulk avalanche mobility, although it may be noted that failures in nature preferentially select the weakest materials. The ratio of fall height to runout distance (H/L) for the avalanche was about 0.22 (Fig. 25). This ratio can be crudely interpreted in terms of the apparent friction coefficient of the avalanche (Pariseau & Voight 1979), and thus the apparent friction angle is around 12-13°. This result suggests that much of the initial potential energy was consumed by basal friction (Hutter 1996), and that local topography variations did not too strongly influence runout. These results contrast with those of Dade & Huppert (1998), who suggest from scaling relationships that a constant-stress resistance law applies to long-runout mass movements. Our experimental and modelling results suggest that this resistance is not constant, but instead is related by friction coefficient to the varying thickness of the avalanche during emplacement. The Pouliquen relation suggests further that resistance may also be related to velocity. The Dade & Huppert result possibly can be rationalized by the view that an average resisting stress can be defined, and is approximately proportional to the average thickness of an avalanche. However, in detail, this thickness will vary considerably from point to point and from time to time during the emplacement process, and the resisting stresses will also vary in a corresponding way. Comparisons and contrasts Comparisons of the Boxing Day sector collapse and dome explosion with the well documented May 1980 edifice collapse and cryptodome explosion of Mount St Helens, USA, and the March 1956 collapse and explosion of Bezymianny in Russian Kamchatka, seem particularly relevant. All three complex events, edifice failures and explosions, are of Bezymianny-type (Gorshkov 1962; Siebert et al 1987; Voight & Elsworth 1997), meaning that an injection of magma had preceded and probably had prompted the sector collapse. A similar event at c. 3000 BP has been postulated for Soufriere Volcano in neighbouring Guadaloupe (Boudon et al. 1984).

THE BOXING DAY SECTOR COLLAPSE

In all these events, a sector collapse preceded and triggered the generation of energetic pyroclastic currents. At Mount St Helens, dacite magma was intruded asymmetrically early in the eruption, as a cryptodome into the interior of the stratovolcano (Lipman & Mullineaux 1981). The high-viscosity intrusion caused marked deformation of the north slope at rates of c. 2m day - 1 (Lipman et al 1981), which caused much concern for the USGS observatory team. In an interpretation one month in advance of the collapse, prompted by qualitative observations of the deformation, the possibility of a massive sector collapse and an accompanying pyroclastic explosion was recognized (Voight 1980, 2000; Decker 1981). On 18 May 1980, after several months of significant deformation and seismicity - but without recognized short-term precursors such as accelerating deformation or seismicity - sector collapse occurred retrogressively to generate a complex debris avalanche. This in turn facilitated a devastating laterally directed explosion that evolved contemporaneously with the later movement stages of the avalanche (Voight et al. 1981, 1983; Glicken 1986, 1998; Fisher et al. 1987; Sousa & Voight 1995; Hoblitt 2000). A moderately large (M = 5) earthquake occurred at the time of slope collapse and was generally interpreted to have triggered it. In the case of Bezymianny in 1956, andesitic magma was intruded into the edifice as much as six months before slope collapse, and frequent minor magmatic explosions had occurred. Massive deformation of the volcano slope by intruding magma preceded sector collapse of the edifice (Gorshkov 1959). Failure there too was associated with an earthquake, and collapse probably occurred retrogressively (Voight & Elsworth 1997) to generate a massive volcanic debris avalanche, and a violent laterally directed explosion (Belousov & Bogoyavlenskaya 1988; Belousov 1996). No detailed monitoring had been attempted, but large earthquakes were detected by the regional seismic network. In contrast, at Soufriere Hills Volcano a dome of pressurized andesitic lava built up over the south flank, and this, with talus shed by the dome, loaded a weakened (and eroded) area of altered rock that included active hydrothermal springs. Seismic precursors preceded the failure in the short term. The sector collapse undermined the pressurized dome, and thus the debris avalanche was succeeded by a destructive, laterally directed pyroclastic current. The collapse occurred at a mature stage in the eruption, which started about 2.5 years earlier, but signs of potential structural instability had developed over a year before and recognition of these signs had led to effective risk mitigation. The common factors in these three events are a relatively weak, partly altered edifice further destabilized by emplacement of silicic magma containing pressurized volatiles, and subjected to dynamic seismic loading. However, the differences are also substantial. It appears that the Soufriere Hills example falls into a separate class involving exogenous shear loading by extruding lava above a hydrothermally weakened flank, in contrast to endogenous stressing of the edifice by an intruding cryptodome (Donnadieu & Merle 1998; Voight 2000). Further, although large earthquake shocks occurred at Mount St Helens and Bezymianny, no such large individual triggering shocks were detected at Soufriere Hills; instead the sector collapse was preceded by a seismic swarm and tremor of many hours' duration. The Boxing Day events illustrate the potential for highly dangerous activity in all dome eruptions where hydrothermal alteration related to earlier episodes of volcanic activity has created a weakened edifice, and where high gas pressures in the interior of a dome lava and/or conduit can contribute to destabilization (Voight & Elsworth 2000), and can also generate violent pyroclastic currents if the edifice fails. The Socompa debris avalanche, northern Chile, as studied by our deceased co-author and friend Peter Francis and his associates (Francis et al. 1985; Francis & Self 1987; van Wyk de Vries & Francis 1996), offers other significant comparisons. The Socompa sector collapse (c. 7000 BP) reflected loading of weakened base material by the growing volcano (van Wyk de Vries et al. 1999), which caused outward spreading in the substrate prior to collapse. A somewhat similar process may have occurred at the base of the

403

rapidly growing talus cone at Montserrat. It is conceivable that loading under undrained conditions there (in saturated strata of low permeability) promoted creep movement and strain-weakening that influenced the flank failure. Some volcanoes, such as Mount St Augustine (Alaska), Shiveluch (Kamchatka, Russia) and Egmont (New Zealand) have collapsed repeatedly (Beget & Kienle 1992; Belousov et al. 1999; Ui et al. 1986; Palmer et al. 1991). Work by Boudon et al. (1999) and Deplus et al. (1999) has shown that volcanoes in the Carribbean arc also have been prone to sector collapse in the past. At least eight collapses have been recognized at Soufriere Volcano, Guadeloupe, where continuous vigorous hydrothermal activity, over thousands of years, has considerably modified the structure and mechanical properties of an edifice periodically affected by phreatic eruptions and less frequent magmatic dome eruptions. Indeed, the Boxing Day sector collapse at Soufriere Hills Volcano repeats, in general terms, a phenomenon that occurred there several times before, with English's Crater representing the scar of a major prehistoric debris avalanche, and with extensive hummocky avalanche deposits offshore (Boudon 2001). Discussion: evolution of the sector collapse Monitoring and mitigation The 26 December sector collapse at Soufriere Hills Volcano illustrated that a major edifice failure and directed explosion could occur, despite relatively intensive monitoring, without precursors sufficiently diagnostic to enable reliable short-term forecasting. Nevertheless, the potential for a collapse and explosion of this general kind at Montserrat was recognized over a year before its occurrence, which led to a precautionary evacuation. Consequently there were no casualties, despite the unmistakable violence of the events (Sparks et al. 2002). Further, the recognition of a pattern of enhanced volcanic activity every six to seven weeks from May 1997 onwards (Voight et al. 1998, 1999) enabled anticipation by the Montserrat Volcano Observatory team that a significant event or series of events might take place towards the end of December 1997. Further discussion of the hazards assessments and their implications are reported by Young et al. (2002). Causes of the collapse The 26 December sector collapse was caused by some combination of long-term alteration-weakening by hydrothermal activity, subsequent dome lava and talus loading of the (partly) water-saturated materials, and shaking associated with volcanic seismicity. Static conduit-magma pressurization seems unlikely as a major direct influence on failure, but might have been indirectly related, for instance as a factor in the production of hybrid seismicity (Voight et al. 1999). The rapid south-directed exogeneous growth of a shear-lobe of dome lava, and shedding of lava-block talus over the Galway's Wall and Galway's Soufriere area since early November 1997 provided quasi-static loading and perhaps also pore-fluid pressure enhancement that strongly contributed to failure. These loads may have been applied in undrained fashion (in the geotechnical sense; see Lambe & Whitman 1969) to pockets of watersaturated, weak, low-permeability, fines-rich alteration materials. Seismic loading may have augmented the destabilization forces in the 24 hours (especially the last few hours) prior to the sector collapse, in concert with the load changes caused by rapid effusion and endogenous dome lava emplacement. Thus, although the sequence of night-time failure events was not observed and is imprecisely known, an episode of rapid effusion accompanied by the shallow seismicity of 24-26 December probably provided the trigger for localized base shear-failure of the hydrothermally weakened hill of tuff between Galway's Soufriere and the deteriorated Galway's Wall. Possibly, this hill may have comprised a toreva block from a prehistoric sector collapse in the

404

B. VOIGHT ET AL.

White River valley, such that a partial slip surface already existed and merely required reactivation. Failure was probably progressive. With strain-weakening causing the shear resistance in the altered material to approach residual strengths, the slip surface propagated northwards through talus to the top of the lava dome, and broke out southwards through a toe at Galway's Soufriere while becoming detached on lateral margins. Strain-weakening behaviour is indicated by the laboratory tests, with the residual friction angle results varying widely, from 3° in clay to 31° in tuff. A typical friction angle for the avalanche debris may be between these extremes (e.g. sample 2A), but it should be recognized also that failure processes are selective, and thus the weakest units could strongly influence the failure, even if they did not represent a large volumetric proportion of the materials that failed.

is a function of both the actual friction coefficient and the pore-fluid pressure (Voight 1978; Sassa 1988; Voight & Elsworth 1997). Further, as discussed above, avalanche transport dynamics are strongly influenced by weak lithologies and weak zones. The best-fit dynamic model suggested an avalanche emplacement time of less than three minutes and typical maximum velocity of about 40 m s - 1 . These data are consistent with estimates based on field constraints, which suggest a probable emplacement time of 138s or more, dependent on location, and an average speed in excess of 19-35 m s - 1 . The numerical results illustrate the potential of such models to evaluate the capability of debris avalanches to surmount relief and to affect areas beyond the obvious drainage channels. Such modelling could be useful in anticipating potential effects of avalanches in future volcanic crises.

Flow behaviour

Conclusions

Most of the detached mass accelerated under the impetus of gravity, and moved rapidly as a disintegrating, shearing avalanche down the White River valley. In these movements the source region tended to fragment along weak layers and boundaries, forming avalanche megablocks, which subdivided further at weak boundaries during transport. The strength of the avalanche was effectively governed by the weakest lithologies and weak zones. Much of the megablock-facies material was more or less cohesive, with the overlying talus non-cohesive. This difference probably resulted in non-uniform avalanche flow behaviour, although the possibility is enigmatic to assess because of difficulty in identifying some talus materials from blocky pyroclastic flow deposits. The debris avalanche flow was unsteady, with several flow pulses implied by channel overtopping relations and by lobate forms on the deposit surface. Movement of following material was arrested as, at several locations, debris slowed in front, and transverse steps were formed similar to those described for volcanoes Socompa (Wadge et al 1995), Avachinsky (Castellana et al. 1995) and Shiveluch (Belousov et al. 1999). In turn, the sector collapse resulted in an oversteepened front of the fresh lava dome, which caused it to disintegrate, partly by gravity but also by release of external loads on the gas-pressurized lava, and to form a pyroclastic current that ravaged the south flank. The shear tended to be focused towards the base of the avalanche, accounting for the basal shear fabric, and for the transport of little-deformed megablock facies. Nevertheless, both compressional and dilational strains affected the avalanche debris at different places and times, and with increasing distance there was enhanced clast shattering, clast-size reduction, and mingling of coloured units. The textures mainly suggest emplacement by laminar flow, with interaction between megablock domains dominated by near-neighbour compression and shearing.

(1)

Emplacement dynamics Insights into the dynamics of avalanche emplacement were provided by three-dimensional numerical simulations, with the avalanche idealized as a flow of a homogeneous continuum governed by a basal friction law. The dominance of basal friction is fully justified by the fabric and structural observations discussed above. Numerical results showed that the observed distribution of debris, including overspill deposition, is well reproduced for an assumed Coulombtype (Pouliquen) friction law with a friction coefficient dependent upon the thickness and the velocity of the flowing mass. The low values of friction implied by numerical modelling are consistent with the fact that large avalanches travel farther than expected from overly simple models of slope failure. The mobility of this avalanche may have been enhanced by the presence of water-saturated, highly altered material from the Galway's fumarolic system, which constituted a large portion of the avalanche. For such water-saturated materials, the apparent friction coefficient

(2)

(3) (4)

(5)

(6)

Failure of the southern sector of Soufriere Hills Volcano on 26 December 1997 culminated in a devastating eruptive episode. Sector collapse produced a volcanic debris avalanche, and exposure of the depressurized face of the lava dome then resulted in generation of a powerful pyroclastic density current. A precautionary evacuation resulted in avoidance of casualties. The south-directed growth of a lava lobe and shedding of lavablock talus over the Galway's Wall and Galway's Soufriere area, particularly since early November 1997, provided static loading on hydrothermally weakened materials that brought the flank to a condition of marginal stability. Collapse was triggered by a pulse of co-seismic, exogenous, lava shear-lobe emplacement, with subsequent gravity-driven slip-surface localization influenced by strain-weakening. The source region tended to fragment along weak layers and boundaries to form avalanche megablocks, which were then transported and further disrupted by shear along the megablock boundaries and weak internal zones. Thus the strength of the avalanche was effectively governed by the locally weakest lithologies and weak zones. Shear was focused towards the base of the avalanche, accounting for the basal shear fabric, and for the transport of less-deformed megablock facies within the body of the flow. The resulting debris avalanche was sustained and consisted of several flow pulses that reflected complexities of the source disruption, varying flow properties of the older rocks and dome talus, and channel topography. At major bends, the avalanche separated into channelled and overspill flows, and in the distal region, stacked sets of the main lithologies occur with a hummocky surface and with abrupt flow-unit snouts. Numerical simulations support field observations that the avalanche dynamics were governed by basal friction, and suggest that this friction was dependent upon the materials, and also the thickness and the velocity of the flowing mass. The low values of friction implied by the simulations are consistent with geotechnical test data and the inferred localized presence of pore-water pressures.

We are primarily grateful to our colleagues of Montserrat Volcano Observatory (MVO) for considerable, continued intellectual and fieldwork support. We thank the UK Department for International Development for generous financial support of the volcano monitoring on Montserrat. Likewise the assistance of British Geological Survey (BGS) is acknowledged and appreciated. B.V., R.S.J.S. and S.R.Y. acknowledge support from BGS as Senior Scientists affiliated with MVO, and B.V. notes in addition, important assistance from several grants from the US National Science Foundation. Work by A.B.B. and M.B. was partly supported by grant RG1-172 of the US Civilian Research and Development Foundation awarded to A.B.B. and B.V., and also by the Alexander von Humboldt Foundation. Fieldwork by J.-C.K. was undertaken during time spent as a Senior Scientist with MVO. Supporting funds for J.-C.K. and G.B. were also obtained from the French PNRN (INSU-CNRS) research programmes, and the Observatoires Volcanologiques of the Institut de Physique du Globe de Paris (IPGP). Ph.H. was supported by the Commissariat a PEnergie Atomique, France.

THE BOXING DAY SECTOR COLLAPSE We recall with sadness and appreciation our colleague Peter Francis, a charter member with H. Glicken of Friends of Volcanic Debris Avalanches, with both being contributors of seminal works. We thank the helicopter pilots, particularly J. McMahon, A. Grouchy, and pilots from Bajan Helicopters (Montserrat Air Support Unit), for their skill and help, often beyond the call of duty. K. West and M. Feuillard offered photographs for our use, and many other photographs used were provided by staff of MVO. B.V. acknowledges helpful interchanges with colleagues D. Elsworth, H. R. Hardy, E. Kimball. H. W. Shen, and E. Oh. D. Hidayat, R. Herd, F. Donnadieu, and M. Volero helped with drafting. Critical reviews by R. E. A. Robertson and S. Self, and the exceptional editorship of P. Kokelaar, led to significant improvement of the paper. The assistance of J. K. McClintock in many matters is much appreciated. Published by permission of the Director, BGS (NERC).

Appendix Geotechnical tests For most direct shear tests on Sample 15-2A, well-mixed moist samples were pushed through a 2 mm sieve and moulded into the shear frame, which had a diameter of 6cm. A normal stress of 0.43 MPa was applied, the specimens were consolidated in a watersaturated state, and slow shearing accomplished at rates of, typically, 1.3x 10 - 4 mms - 1 . The specimens were then reverse-sheared through a number of cycles to, usually, cumulative displacements of 90-40 mm or more, to provide a measure of displacementweakening. These procedures were repeated for a normal pressure range of 0.86-1.72 MPa. Similar tests were carried out on clayrich block sample 15-2C, but in this case the block was cored with a thin-walled tube, and core specimens were extruded, trimmed, and inserted into the direct shear frame. Similar tests were also conducted on disaggregated, weathered tuffs from the Galway's Wall and Gages Wall areas of English's Crater (Fig. 2). For the block samples of tuff collected at Galway's Wall and Gages Wall, the blocks were cored (2.5 cm diameter) and duplicate right-cylinder specimens were subjected to laboratory triaxial testing to determine the shear strength of intact material as a function of confining pressure. Tests were conducted dry at room temperature, using loading rates of 138kPas - 1 , with confining pressures varied from 0.62-9.93 MPa. Additional testing was carried out to obtain intact tensile strengths by the method of cylinder edge-loading.

References ALIDIBIROV, M. & DINGWELL, D. B. 1996. Magma fragmentation by rapid decompression. Nature, 380, 146-148. ASSIER, S., HEINRICH, P.. SABATIER, P. C, SAVOYE, B. & BOURILLET, J. F. 2000. Numerical modeling of a landslide-generated tsunami: the 1979 Nice event. Pure and Applied Geophysics, 157. 1707-1727. BEGET, J. & KIENLE, J. 1992. Cyclic formation of debris avalanches at Mount St. Augustine volcano. Nature. 356, 701-704. BELL, F. G. 1993. Engineering Geology. Blackwell, London. BELOUSOV, A. B. 1996. Pyroclastic deposits of March 30 1956 directed blast at Bezymianny volcano. Bulletin of Volcanology, 57. 649-662. BELOUSOV, A. B. & BOGOYAVLENSKAYA, G. 1988. Debris avalanche of the 1956 Bezymianny eruption. Proceedings Kagoshima International Conference on Volcanoes, 460-462. BELOUSOV, A. B.. BELOUSOVA. M. & VOIGHT, B. 1999. Multiple edifice failures, debris avalanches, and associated eruptions in the Holocene history of Shiveluch volcano, Kamchatka, Russia. Bulletin of Volcanology, 61, 324-342 BOUDON, G. 2001. Instabilites gravitationnelles de l'are volcanique des petites Antilles. Atelier sur les Aleas Volcaniques les volcans antillais - des processus aux signaux. Paris, 18-19 Janvier. Institut National des Sciences de l'Univers, 4-5. BOUDON, G.. SEMET. M. P. & VINCENT. P. M. 1984. Flank failure-directed blast eruption at la Soufriere de Guadaloupe, French West Indies: a 3.000 year old Mount St Helens? Geology. 12. 350-353. BOUDON, G., VILLEMANT. B., KOMOROWSKI. J-C, ILDEFONSE. PH., HAMMOUYA. G. & SEMET. M. P. 1996. The hydrothermal system of Soufriere

405

Hills volcano. Montserrat (West Indies): Chemical, mineralogical. and microtextural signatures in fluids, altered rocks, and 1995-96 tephra. ///: WADGE. G. (ed.) The Soufriere Hills Eruption, Montserrat. Discussion Meeting of the Volcanic Studies Group of the Geological Society, 27 November 1996. Montserrat Volcano Observatory and Geological Society of London. 31-34. BOUDON. G., VILLEMANT. B.. KOMOROWSKI. J.-C.. ILDEEONSE. P. & SEMET. M. P. 1998. The hydrothermal system at Soufriere Hills volcano. Montserrat (West Indies): Characterization and role in the ongoing eruption. Geophysical Research Letters, 25. 3693-3696. BOUDON. G., KOMOROWSKI. J-C.. SEMET. M.. LE FRIANT. A. & DEPLUS. C. 1999. Frequent volcanic flank-collapses in the Lesser Antilles arc: origin and hazards. Eos, Transactions, American Geophysical Union. 80(46). F1142. BROMHEAD. E. N. 1986. The Stability of Slopes. Surrey University. London. CALDER. E. S.. YOUNG. S. R.. SPARKS. R. S. J. ET AL. 1998. The Boxing Day collapse, 26 December 1997. Montserrat Volcano Observatory Special Report 6. CASTELLANA. B.. DAVIDSON. J. P.. BELOUSOV. A. B. & BELOUSOVA. M. 1995. Milestones of geology of the Avachinskiy volcano. Kamchatka. Russia. Eos, Transactions, American Geophysical Union. 76. 537. CHIODJNI, G.. CIONI. R.. FRULLANI. A.. GUIDI. M.. MARINI. L.. PRATI. F. & RACO. B. 1996. Fluid geochemistry of Montserrat Island. West Indies. Bulletin of Volcanology. 58. 380-392. COETZEE. M. J.. HART. R. D . VARONA. P. D. & C U N D A L L . P. A. 1998. FLAC

Basics. Itasca. CRANDELL. D. R.. MILLER. C. D.. GLICKEN. H. X.. CHRISTIANSEN. R. L. & NEWHALL. C. G. 1984. Catastrophic debris avalanche from ancestral Mount Shasta volcano. California. Geology. 12. 143-146. CUNDALL. P. & BOARD. M. 1988. A microcomputer program for modeling large-strain plasticity problems. In: Numerical Methods in Geomechanics (Proceedings 6th International Conference. Innsbruck. Austria. April 1988). Balkema. Rotterdam. 2101-2108. DADE, W. B. & HUPPERT. H. E. 1998. Long-runout rockfalls. Geology. 26. 803-806. DECKER. R. W. 1981. The 1980 activity: a case study in volcanic eruption forecasting. ///: LIPMAN. P. W. & M U L L I N E A U X . D. R. (eds) The 1980 Eruptions of Mount St Helens. Washington. US Geological Survey. Professional Paper. 1250. 815-820. DEPLUS. C., BOUDON. G.. CHEMINEE. J-L. ET AL. 1999. Large-scale debris avalanche deposits on the western flank of the Lesser Antilles arc revealed by the Aguadomar Cruise. Eos. Transactions, American Geophysical Union. 80(46). 1152. DETOURNAY. C. & HART. R. (eds) 1999. FLAC and Numerical Modeling in Geomechanics. Balkema. Rotterdam. DONNADIEU. F. & MERLE. O. 1998. Experiments on the indentation process during cryptodome intrusions: new insights into Mount St Helens deformation. Geology. 26. 79-82. DRUITT. T. H. 1998. Pyroclastic density currents. ///: GILBERT. J. S. & SPARKS. R. S. J. (eds) The Physics of Explosive Volcanic Eruptions. Geological Society. London. Special Publications. 145. 145-182. DRUITT. T. H.. CALDER. E. S.. COLE. P. D. ET AL. 2002. Small-volume, highly mobile pyroclastic flows formed by rapid sedimentation from pyroclastic surges at Soufriere Hills Volcano. Montserrat: an important volcanic hazard. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano. Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21. 263-279. DUNCAN, J. M. 1992. State-of-the art: static stability and deformation analysis. American Society of Civil Engineers. Geotechnical Engineering Division Special Publication. 31. 222-265. DUNCAN. J. M. 1996. State-of-the-art: limit equilibrium and finite-element analysis of slopes. American Society of Civil Engineers. Journal of Geotechnical Engineering. 122. 577-596. FINK. J. H. & KIEFFER. S. W. 1993. Estimates of pyroclastic flow velocities resulting from explosive disintegration of lava domes,Nature. 363. 612-614. FISHER. R. V., GLICKEN. H. X & HOBLITT. R. P. 1987. May 18 1980. Mount St Helens deposits in South Coldwater Creek. Washington. Journal of Geophysical Research. 92(B10). 10267-10283. FRANCIS, P. W. & SELF. S. 1987. Collapsing volcanoes. Scientific American. 287. 90-99. FRANCIS, P. W. & WELLS. G. L. 1987. Landsat thematic mapper observations of debris avalanche deposits in the Central Andes. Bulletin of Volcanology, 50. 258-278.

406

B. VOIGHT ET AL.

FRANCIS, P. W., GARDEWEG, M., RAMIREZ, C. F. & ROTHERY, D. A. 1985. Catastrophic debris avalanche deposit of Socompa volcano, northern Chile. Geology, 13, 600-603. FREDLUND, D. G. 1984. Analytical methods for slope stability analysis. Proceedings 4th International Symposium on Landslides, Toronto, Vol. 1, 229-250. GIGGENBACH, W. F. 1992. Magma degassing and mineral deposition in hydrothermal systems along convergent plate boundaries. Economic Geology, 87,1927-1944. GLICKEN, H. 1986. Rockslide-debris avalanche of May 18 1980, Mount St Helens Volcano, Washington. PhD dissertation, University of California at Santa Barbara. GLICKEN, H. 1991. Sedimentary architecture of large-volume debris avalanches. In: FISHER, R.V. & SMITH, G. A. (eds) Sedimentation in Volcanic Settings. Society of Sedimentary Geologists (SEPM), Special Publications, 45, 99-106. GLICKEN, H. 1998. Rockslide-debris avalanche of May 18 1980, Mount St. Helens volcano, Washington. Bulletin Geological Society of Japan, 49 (2/3), 55-106. GORSHKOV, G. S. 1959. Gigantic eruption of the volcano Bezymianny. Bulletin of Volcanology, 21, 77-109. GORSHKOV, G. S. 1962. On the classification and terminology of Pelee and Katmai type eruptions. Bulletin of Volcanology, 24, 155-165. HEINRICH, P., MANGENEY, A., GUIBOURG, S., ROCHE, R., BOUDON, G. & CHEMINEE, J-L. 1998. Simulation of water waves generated by a potential debris avalanche in Montserrat, Lesser Antilles. Geophysical Research Letters, 25(19), 3697-3700. HEINRICH, P., PIATANESI, A., OKAL, E. & HEBERT, H. 2000. Near-field modelling of the July 17 1998 tsunami in Papua New Guinea. Geophysical Research Letters, 27(19), 3037-3040. HEINRICH, P., PIATANESI, A. & HEBERT, H. 2001. Numerical modelling of tsunami generation and propagation from submarine slumps: the 1998 Papua New Guinea event. Geophysics Journal International, 145(1), 97-111. HEINRICH, P., BOUDON, G., KOMOROWSKI, J-C., SPARKS, R. S. J., HERD, R. & VOIGHT, B. 2001. Numerical simulation of the December 1997 debris avalanche in Montserrat, Lesser Antilles. Geophysical Research Letters, 28, 2529-2532. HOBLITT, R. P. 2000. Was the 18 May 1980 lateral blast at Mt St Helens the product of two explosions? Philosophical Transactions of the Royal Society, London, A358, 1639-1661. HOEK, E. 1983. Strength of jointed rock masses. Geotechnique, 33, 187-223. HOEK, E. & BROWN, E. T. 1980. Empirical strength criterion for rock masses. American Society of Civil Engineers, Journal of Geotechnical Engineering, 106, 1013-1035. HOEK, E., KAISER, P. K. & BAWDEN, W. F. 1994. Support of Underground Excavations in Hard Rock. Balkema, Rotterdam. HUTTER, K. 1996. Avalanche dynamics, a review. In: SINGH, V. P. (ed.) Hydrology of Disasters. Kluwer, Dordrecht, 317-394. INMAN, D. L. 1952. Measures for describing the size distribution of sediments. Journal of Sedimentary Petrology, 22,125-145. KOMOROWSKI, J-C., GLICKEN, H. X & SHERIDAN, M. 1991. Secondary electron imagery of microcracks and hackly fracture surfaces in sandsize clasts from the 1980 Mount St Helens debris-avalanche deposit: implications for particle-particle interactions. Geology, 19, 261-264. KOMOROWSKI, J-C., NORTON, G., BOUDON, G. ET AL. 1999. The 1997 flankcollapse of Soufriere Hills volcano, Montserrat: Constraints on emplacement mechanisms from debris avalanche morphological and textural data. Eos, Transactions, American Geophysical Union, 80(46), 1152. KRAMER, S. L. 1996. Geotechnical Earthquake Engineering. Prentice Hall, Upper Saddle River, New Jersey. LACROIX, A. 1904. La Montagne Pelee et ses eruption. Masson et Cie, Paris. LAMBE, T. W. & WHITMAN, R. V. 1969. Soil Mechanics. John Wiley, New York. LIPMAN,P.W.&MuLLiNEAUX,D. R.(eds) 1981. The 1980 Eruptions of Mount St Helens, Washington. US Geological Survey, Professional Paper, 1250. LIPMAN, P. W., MOORE, J. G. & SWANSON, D. A. 1981. Bulging of the north flank before the May 18 eruption: geodetic data. In: LIPMAN, P. W. & MULLINEAUX, D. R. (eds) The 1980 Eruptions of Mount St Helens, Washington, US Geological Survey, Professional Papers, 1250, 143-156. MACGREGOR, A. G. 1952. Eruptive mechanisms: Mt. Pelee, the Soufriere, St. Vincent, and the Valley of Ten Thousand Smokes. Bulletin Volcanologique, 12, 49-74.

MANGANEY, A., HEINRICH, PH. & ROCHE, R. 2000. Analytical and numerical solution of dam-break problem for application to water floods, debris and snow avalanches. Pure and Applied Geophysics, 157(6-8), 1081-1096. MAYBERRY, G. C., ROSE, W. I. & BLUTH, G. 1998. Analysis of the ash cloud from the December 26 1997 (Boxing Day) dome collapse of Soufriere Hills Volcano, Montserrat using GOES-8 Imagery. Eos, Transactions, American Geophysical Union, 79, 1014. MEDLEY, E. 1997. Geological engineering of bimrocks (abstract). Association of Engineering Geologists, Annual Meeting. MILLER, A. D., STEWART, R. C., WHITE, R. A. ET AL. 1998. Seismicity associated with dome growth and collapse at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3401-3404. NAAIM, M., VIAL, S. & COUTURE, R. 1997. Saint-Venant approach for rock avalanche modeling. Saint-Venant Symposium, August 28-29, Paris. PALMER, B. A., ALLOWAY, B. V. & NEALL, V. E. 1991. Volcanic-debrisavalanche deposits in New Zealand - lithofacies organization in unconfined, wet-avalanche flows. In: FISHER, R. V. & SMITH, G. A. (eds) Sedimentation in Volcanic Settings. Society of Sedimentary Geologists (SEPM), Special Publications, 45, 89-98. PARISEAU, W. G. & VOIGHT, B. 1979. Rockslides and avalanches: basic principles and perspectives in the realm of civil and mining operations. In: VOIGHT, B. (ed.) Rockslides and Avalanches, 2: Engineering Sites. Elsevier, Amsterdam, 1-92. POULIQUEN, O. 1999. Scaling laws in granular flows down rough inclined planes. Physics of Fluids, 11(3), 542-548. REA, W. J. 1974. The volcanic geology and petrology of Montserrat, West Indies. Journal of the Geological Society, London, 130, 341-366. REICHE, P. 1937. The toreva-block - a distinctive landslide type. Journal of Geology, 45, 538-548. RITCHIE, L. J., COLE, P. D. & SPARKS, R. S. J. 2002. Sedimentology of deposits from the pyroclastic density current of 26 December 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 435-456. ROBERTSON, R. E. A., COLE, P., SPARKS, R. S. J. ET AL. 1998. The explosive eruption of Soufriere Hills Volcano, Montserrat, West Indies, 17 September, 1996. Geophysical Research Letters, 25, 3429-3432. ROOBOL, M. J. & SMITH, A. L. 1998. Pyroclastic stratigraphy of the Soufriere Hills volcano, Montserrat - Implications for the present eruption. Geophysical Research Letters, 25, 3393-3396. SASSA, K. 1988. Geotechnical model for the motion of landslides. Proceedings 5th International Symposium on Landslides, Lausanne, Vol. 1, 37-55. SAVAGE, S. B. & HUTTER, K. 1989. The motion of a finite mass of granular material down a rough incline. Journal of Fluid Mechanics, 199, 177-215. SIEBERT, L. 1984. Large volcanic debris avalanches: characteristics of source areas, deposits, and associated eruptions. Journal of Volcanology and Geothermal Research, 22, 163-197. SIEBERT, L. 1996. Hazards of large volcanic debris avalanches and associated eruptive phenomena. In: SCARPA, R. & TILLING, R. I. (eds) Monitoring and mitigation of volcano hazards. Springer-Verlag, Berlin, 541-572. SIEBERT, L., GLICKEN, H. & UI, T. 1987. Volcanic hazards from Bezymiannyand Bandai-type eruptions. Bulletin of Volcanology, 49, 435-459. SIEBERT, L., BEGET, J. E. & GLICKEN, H. 1995. The 1883 and late-prehistoric eruptions of Augustine volcano, Alaska. Journal of Volcanology and Geothermal Research, 66, 367-395. SLINGERLAND, R. L. & VOIGHT, B. 1979. Occurrences, properties, and predictive models of landslide-generated water waves. In: VOIGHT, B. (ed.) Rockslides and Avalanches, 2: Engineering Sites. Elsevier, Amsterdam, 317-397. SOUSA, J. & VOIGHT, B. 1995. Multiple-pulsed debris avalanche emplacement at Mount St Helens in 1980: evidence from numerical continuum flow simulation. Journal of Volcanology and Geothermal Research, 66, 227-250. SPARKS, R. S. J., BARCLAY, J., CALDER, E. S. ET AL. 2002. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 409-434. UI, T. 1983. Volcanic dry avalanche deposits identification and comparison with non-volcanic debris stream deposits. Journal of Volcanology and Geothermal Research, 18, 135-150.

THE BOXING DAY SECTOR COLLAPSE UI, T. & GLICKEN, H. 1986. Internal structural variations in a debrisavalanche deposit from ancestral Mount Shasta, California, USA. Bulletin of Volcanology, 48, 189-194. UI, T., KAWACHI, S. & NEALL, V. E. 1986. Fragmentation of debris avalanches during flowage. Evidence from the Pungarehu Formation, Mount Egmont, New Zealand. Journal of Volcanology and Geothermal Research, 27, 255-264. UI, T., TAKARADA, S. & YOSHIMOTO, M. 2000. Debris avalanches. In: SIGURDSSON, H. (ed.) Enclyclopedia of Volcanoes. Academic Press, San Diego, 617-626. VAN WYK DE VRIES, B. & FRANCIS, P. W. 1996. Catastrophic collapse at stratovolcanoes induced by gradual volcano spreading. Nature, 387, 387-390. VAN WYK DE VRIES, B., SELF, S., FRANCIS. P. & KESZTHELYI, L. 1999. Development of the Socompa debris avalanche (N Chile) from a spreading volcanic edifice. Eos, Transactions, American Geophysical Union, 80(46), 1142. VOIGHT, B. 1978. Lower Gros Ventre slide, Wyoming, USA. In: VOIGHT, B. (ed.) Rockslides and Avalanches, 1: Natural Phenomena. Elsevier, Amsterdam, 113-166. VOIGHT, B. 1980. Slope stability hazards, Mount St Helens volcano, Washington. File report, US Geological Survey Cascades Volcano Observatory, Washington (reproduced as appendix in: VOIGHT, B. 2000. Stability of andesite volcanoes and lava domes. Philosophical Transactions of the Royal Society, London, A 358, 1694-1703). VOIGHT, B. 1996. Strength of wall rock of Englishes Crater, Soufriere Hills volcano, Montserrat. In: WADGE, G. (ed.) The Soufriere Hills Eruption, Montserrat. Discussion Meeting of the Volcanic Studies Group of the Geological Society, 27 November 1996. Montserrat Volcano Observatory and Geological Society, London, 29-30. VOIGHT, B. 2000. Stability of andesite volcanoes and lava domes. Philosophical Transactions of the Royal Society, London, A 358, 1663-1703. VOIGHT, B. & ELSWORTH. D. 1997. Failure of volcano slopes. Geotechnique, 47, 1-31. VOIGHT, B. & ELSWORTH, D. 2000. Stability and collapse of hazardous gaspressurized lava domes. Geophysical Research Letters, 48, 1-4. VOIGHT, B., GLICKEN, H., JANDA, R. J. & DOUGLASS, P. M. 1981. Catastrophic rockslide avalanche of May 18. In: LIPMAN, P. W. & MULLINEAUX, D. R. (eds) The 1980 Eruptions of Mount St Helens, Washington, US Geological Survey, Professional Papers. 1250, 347-378.

407

VOIGHT. B., JANDA, R. J.. GLICKEN. H. & DOUGLASS. P. M. 1983. Nature and mechanics of the Mount St Helens rockslide-avalanche of 18 May 1980. Geotechnique, 33. 243-273. VOIGHT. B., JANDA, R. J., GLICKEN. H. & DOUGLASS. P. M. 1985. Nature and mechanics of the Mount St Helens rockslide-avalanche of 18 May 1980: Reply to discussion. Geotechnique, 35, 357-368. VOIGHT, B., HOBLITT. R. P., CLARKE, A. B.. LOCKHART, A. B.. MILLER, A. D.. LYNCH, L. & McMAHON, J. 1998. Remarkable cyclic ground deformation monitored in real time on Montserrat and its use in eruption forecasting. Geophysical Research Letters, 25. 3405- 3408. VOIGHT, B., SPARKS, R. S. J., MILLER, A. D. ET AL. 1999. Magma flow instability and cyclic activity at Soufriere Hills Volcano. Montserrat, B.W.I. Science, 283, 1138-1142. WADGE, G. & ISAACS, M. C. 1988. Mapping the volcanic hazards from Soufriere Hills volcano, Montserrat. West Indies using an image processor. Journal of the Geological Society, London. 145, 541-551. WADGE, G., FRANCIS. P. W. & RAMIREZ. C. F. 1995. The Socompa collapse and avalanche event. Journal of Volcanology and Geothermal Research. 66,309-336. WADGE, G., WOODS, A., JACKSON, P., BOWER. S.. WILLIAMS, C. & HULSEMANN, F. 1998. A hazard evaluation system for Montserrat. In: Forecasts and Warnings. UK Coordination Committee for IDNDR. Thomas Telford, 3.1-3.32. WALKER, G. P. L. 1971. Grain-size characteristics of pyroclastic deposits. Journal of Geology, 79, 696-714. WOODS. A. W., SPARKS, R. S. J., RITCHIE, L. J., BATEY, J., GLADSTONE, C. & BURSIK. M. I. 2002. The explosive decompression of a pressurized volcanic dome: the 26 December 1997 collapse and explosion of Soufriere Hills Volcano, Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21. 457-466. YOUNG, S. R., SPARKS, R. S. J.. ASPINALL, W. P.. LYNCH. L. L.. MILLER. A. D., ROBERTSON, R. E. A. & SHEPHERD, J. B. 1998. Overview of the eruption of Soufriere Hills Volcano, Montserrat, July 18 1995. to December 1997. Geophysical Research Letters, 25. 3389-3393. YOUNG, S. R., VOIGHT. B., BARCLAY, J. ET AL. 2002. Hazard implications of small-scale edifice instability and sector collapse: a case history from Soufriere Hills Volcano, Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London. Memoirs. 21. 349-361.

Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat R. S. J. SPARKS1, J. BARCLAY 2 , E. S. CALDER 1 , R. A. HERD 3 , J.-C. KOMOROWSKI 4 . R. LUCKETT 5 , G. E. NORTON 3 , L. J. RITCHIE 6 , B. VOIGHT 7 & A. W. WOODS8 1 Department of Earth Sciences, University of Bristol, Bristol, BS8 1RJ, UK 2 Department of Environmental Sciences, University of East Anglia, Norwich, NR4 7JT, UK 3 British Geological Survey, Keyworth, Nottingham, NG12 5GG, UK 4 Observatoire de Guadeloupe, Institut de Physique du Globe, Guadeloupe, French Antilles 5 British Geological Survey, Murchison House, West Mains Road, Edinburgh, EH9 3LA, UK 6 Centre for Volcanic Studies, University of Luton, Park Square, Luton, LU1 3JU, UK 7 Department of Geosciences, Penn State University, University Park, PA 16802, USA 8 BP Institute, Madingley Rise, Cambridge University, Cambridge CB3 OEZ, UK

Abstract: Growth of an andesitic lava dome at Soufriere Hills Volcano, Montserrat, beginning in November 1995. caused instability of a hydrothermally altered flank of the volcano. Catastrophic failure occurred on 26 December 1997. 14 months after the instability was first recognized. Two months before failure a dome lobe had extruded over the unstable area and by 25 December 1997 this had a volume of 113 x 10 6 m 3 . At 03:01 (local time) the flank rocks and some dome talus failed and generated a debris avalanche (volume 46 x 106 m 3 ). Between 35 and 45 x 106 m3 of the dome then collapsed, generating a violent pyroclastic density current that devastated 10km 2 of southern Montserrat. The failure of the flank and dome formed two adjacent bowl-shaped collapse depressions. The most intense activity lasted about 11.6 minutes. The hummocky debris avalanche deposit is composed of a mixture of domains of heterolithic breccia. The pyroclastic density current had an estimated peak velocity of 80-90 m s - 1 , and minimum flux of 10 8 k g s - 1 . The current was largely erosional on land with most deposition out at sea. Destructive effects included removal of houses, trees and large vehicles, and formation of a scoured surface blackened by a thin (3-4 mm) layer of tar. Two discrete depositional units formed from the pyroclastic density current, each with a lower coarse-grained layer and an upper fine-grained stratified layer. These deposits are overlain by an ashfall layer related to buoyant lofting of the current. Flank failure is attributed to loading of hydrothermally weakened rocks by the dome. The generation of the pyroclastic density current is attributed to failure and explosive disintegration of the dome, involving release and violent expansion of gases initially at high pore pressures.

In October 1996 the upper southern flank of Soufriere Hills Volcano, Montserrat, started to show signs of instability, by development of open fractures and rock avalanches associated with intense earthquake swarms. This instability raised concerns about the possibility of a sector collapse and an associated violent lateral blast. The anticipated phenomena took place 14 months later at 03:01 local time (LT) on 26 December 1997. This date is a holiday in the UK known as Boxing Day. Failure of the southern flanks of the volcano was followed by collapse of about 50% of the andesitic lava dome. As a consequence a debris avalanche formed and 10 km2 of southern Montserrat was devastated by a highly energetic pyroclastic density current (PDC). This paper documents these events, their effects and their products. There was no loss of life, as the area had been evacuated when the instability was first recognized in October 1996. The flank failure of 26 December 1997 contrasts with the larger flank failure of Mount St Helens in 1980, where sliding of the northern flanks of the volcano unroofed a pressurized cryptodome that immediately exploded and generated a laterally directed high-energy PDC or volcanic blast (Christiansen & Peterson 1981). In the case of Soufriere Hills Volcano, a large pressurized andesitic lava dome had built above a hydrothermally altered and unstable flank of the volcano. Failure of the flank rocks undermined the dome, which partially collapsed, disintegrated and generated a high-energy PDC. This paper complements other contributions describing the instability of the southern walls of the volcano and the debris avalanche (Voight et al. 2002), the sedimentology of the PDC deposits (Ritchie et al. 2002), modelling of the blast dynamics (Woods et al. 2002), remote sensing data (Mayberry et al. 2002) and hazards aspects (Young et al. 2002). Precursory activity Long-term evolution The events on 26 December 1996 were the culmination of long-term destabilization of the upper flank of the volcano and dome growth,

which can be traced back to October 1996. This section places the eventual failure of the southern flank in the context of the overall evolution of the eruption. The eruption of Soufriere Hills Volcano involved the growth of an andesitic lava dome within English's Crater (Fig. 1: Robertson et al. 2000). English's Crater was partly infilled by a young lava dome known as Castle Peak, which was extruded about 350 years BP (Young et al. 1998). A semi-circular depression or moat existed between the Castle Peak dome and the walls of English's Crater (Fig. la). Lava extrusion began in mid-November 1995 and throughout 1996 lava dome growth was confined to within English's Crater, with pyroclastic flows generated by dome collapse being discharged to the east down the Tar River valley. The area of eventual failure occurred between Galway's Wall and Galway's Soufriere. Galway's Wall constitutes the southern margin of English's Crater and extends about 600m from Chances Peak to Galway's Mountain (Fig. 1b). The outer, southward-facing side of Galway's Wall was precipitous and composed of pyroclastic breccias and tuffs related to the formation of the Chances Peak and Galway's domes (Voight et al. 2002). Galway's Soufriere was an area of active fumaroles beneath Galway's Wall, at an altitude of 400-500m above the White River valley (Fig. 1b). Faulting and fracturing of the Galway's Soufriere area had occurred prior to and in the early stages of the eruption. The exact date of the onset of this deformation is not known, but substantial changes in the area were first noted in October 1995. One prominent fault, trending NE. cut and down-faulted the road to Galway's Soufriere by about a metre to the SE. The inner side of Galway's Wall was not affected by the dome until June 1996 when active dome growth was focused in the SW side of English's Crater. Talus rapidly infilled the moat between Castle Peak and Galway's Wall. Activity from July to mid-September 1996 focused in the northern and eastern areas of English's Crater. A new period of dome growth, which began on 1 October 1996 following the explosive activity of 17-18 September 1996. had almost refilled the crater by the end of October. In late October 1996, small

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 409-434. 0435-4052/02/S15 r The Geological Society of London 2002.

409

410

R. S. J. SPARKS ET AL.

Fig. 1. (a) Topographic map of southern Montserrat showing main locations and features described in the text. The location of the seismic stations at Windy Hill (MBWH) and St Patrick's (MSPT) are also shown. Contours in hundreds of metres. (b) Detailed topography of English's Crater and upper flanks of Soufriere Hills Volcano with names of geographic features mentioned in the text.

avalanches on Galway's Wall first indicated instability. Fractures opened on Galway's Wall in late November. Dome growth slowed markedly in November 1996 (Sparks et al. 1998) to 0.5-1 m3 s - 1 . A remarkable pattern of alternating hybrid earthquake swarms and sluggish aseismic dome growth characterized activity, culminating in an eight-day intense hybrid swarm at the beginning of December 1996. During this period, there was pronounced instability of Galway's Wall. Fractures opened on the Chances Peak end of the wall and in the south-facing outer wall. Periods of avalanching occurred frequently across the entire wall, mostly concurrent with swarms of hybrid earthquakes. Direct observations showed that avalanches occurred simultaneously in several places during an intense hybrid earthquake. Avalanches from Galway's Wall were uncommon during aseismic periods, but increased rockfalls from the dome indicated that extrusion had accelerated then. Surveys in early December 1996 showed that part of the dome erupted in June 1996 had risen by at least 30m along an east-west zone adjacent to Galway's Wall. These observations suggested either a period of endogenous dome growth or shallow intrusion adjacent to Galway's Wall. These periods of deformation of Galway's Wall alternated with aseismic periods and dome

extrusion (Voight et al 1999). There was considerable concern about the possibility of wall failure, sector collapse and a lateral volcanic blast (Young et al. 2002). During the period December 1996 to late March 1997 the area of active dome growth moved to the north and east. Avalanche activity at Galway's Wall simultaneously diminished. Fractures developed on the Galway's Mountain end of Galway's Wall. In late March 1997, active growth shifted back to the south and a new lobe of lava started to extrude towards Galway's Wall. At this stage, the lava overlapped the lowest point on Galway's Wall and domecollapse pyroclastic flows spilled over the wall into the White River valley. Numerous pyroclastic flows eroded a gully a few tens of metres deep into Galway's Wall (Fig. 2a) and began burying Galway's Soufriere. From May 1997 to the end of October 1997 dome activity shifted to the north and east, away from Galway's Wall. Two series of repetitive Vulcanian explosions occurred in the August to October period (Druitt et al. 2002b). There were minor changes in the Galway's area, with occasional landslips. At the beginning of November 1997 dome growth shifted to the south. Two major dome collapses occurred on 4 and 6 November 1997. The associated pyroclastic flows

GENERATION OF DEBRIS AVALANCHE AND PYROCLASTIC DENSITY C U R R E N T

further infilled the White River valley, and formed a delta extending 400 m from the former coastline. The collapses in early November 1997 together produced about 8 x 10 6 m 3 of pyroclastic deposits.

411

This minor change cannot be linked to any inflation or deflation. A vigorous gas plume was observed above the dome on clear days in the weeks preceding the eruption, but no correlation spectroscopy measurements were being made in this period.

Short-term precursors Dome growth during November and December 1997 was rapid, at 7-8 m3 s-1 (Sparks et al. 1998), and was focused in the southern area adjacent to Galway's Wall. A lobe extruded laterally over Galway's Wall (Fig. 2b). There were numerous rockfalls, which formed a large apron of dome talus between the base of the lobe and Galway's Soufriere area. By 21 December 1997 (Fig. 2c) the dome summit had reached 1030m altitude and Galway's Wall and Galway's Soufriere were completely buried. The massive and blocky upper parts of the dome extended over and slightly beyond the position of the crest of Galway's Wall (Fig. 2c). Surveys (Fig. 3) indicated that 32 x 106 m3 of dome material extended beyond Galway's Wall and that the dome had reached a volume of 110 x 10 6 m 3 by 21 December 1997. The dome volume is estimated to have been 113 x 10 6 m 3 by 25 December. Volumes in this paper are given as bulk volumes uncorrected to dense rock values. During late November and December 1997 seismicity was low: rockfalls averaged about 70 per day, a few hybrid earthquakes occurred per day, and there were occasional long-period earthquakes (Fig. 4). Long-period earthquakes were typically short bursts of harmonic tremor. These lasted up to a few tens of seconds and were nearly monochromatic at approximately 2 Hz. Cyclicity in the seismicity, which had been pronounced earlier in the eruption (Voight et al. 1999), was barely discernible. There was, however, an expectation at the Montserrat Volcano Observatory (MVO) that activity would become more vigorous towards the end of December. A pattern of enhanced activity every six to seven weeks had been recognized since May 1997, from abrupt changes in tilt patterns, clustering of swarms of hybrid earthquakes, and occurrence of major collapses (Voight et al. 1998, 1999). The last such episode of enhanced activity had occurred on 4-6 November, so a major collapse towards the end of December was anticipated. The build-up to the flank failure and dome collapse of 26 December 1997 was rapid. The real-time seismic amplitude measurement (RSAM) chart and earthquake count data (Fig. 5) show seismic activity only marginally above background with a slight increase that can be traced back to 22 December. At 14:30 LT on 24 December a hybrid swarm began with events approximately every 20 minutes. The frequency of events slowly increased until approximately 20:00 LT on 25 December. The individual events also generally increased in amplitude as the swarm progressed (Fig. 5), but even the largest earthquakes were relatively small in amplitude, an order of magnitude smaller than those recorded in early November 1997. At 20:00 LT hybrid earthquakes were occurring too frequently to trigger the network and from this time onwards the signal was effectively continuous tremor. The amplitude of this tremor peaked at 23:00 LT on 25 December and declined until 00:00 LT when individual events could again be discerned. These individual events merged back into tremor at 01:30 LT on 26 December, which then built in amplitude up to the onset of the signal corresponding to the main event at 03:01 LT. Other monitoring data, gathered in the weeks between the end of explosive activity on 21 October and 26 December, do not reveal any marked changes. Deformation detected by electronic distance measurement and global positioning system did not show any deviation from established trends in this period, although the sampling frequency was insufficient to pick up short-term trends (days). About 40 mm of northward movement of a station on the western flanks of the volcano was recorded between mid-November and mid-January, but it is not known how this movement was partitioned between the periods before and after the 26 December activity. An electronic tiltmeter at the village of Long Ground showed irregular fluctuations during the precursor hybrid swarm on 25 December. There was a permanent positive offset on the x-axis of c. 3 rad, but no offset on the y-axis after 26 December.

Chronology Seismicity Since the flank failure of 26 December 1997 occurred during darkness, seismicity provides the main constraint on the timing of the activity. Seismic data reported here are taken mainly from the Windy Hill (MBWH) 1 Hz single-component station, which is part of the digitally telemetered "broadband' network (Fig. la). The seismic signal from the failure and ensuing events is shown in Fig. 6. Additional information is taken from the 'short-period', analogue

Fig. 2. (a) The lava dome in April 1997 viewed from the SW towards Galway's Wall. The new lobe can be seen in the centre of the photograph (L) and there is a deep erosional gully (G) incised into Galway's Wall by pyroclastic flows related to dome collapse in early April, (b) The 4 November lobe of the dome actively growing over Galway's Wall on 8 November 1997. with the shoulder of Galway's Mountain to the far right. The new spine-like lobe is growing in the collapse scar excavated in the dome during the major collapses of 4 and 6 November 1997. (c) The dome on 21 December 1997 viewed from the SW. The summit of the dome is at an altitude of about 1030m with the top of Chances Peak at 914m on the left. partly obscured by condensed steam drifting from the dome. Galway's Wall has been covered by the new dome and its talus (compare with (a)), and Galway's Soufriere area, in the foreground, has been buried by dome talus.

412

R. S. J. SPARKS ET AL.

Fig. 2. (continued).

network. The St Patrick's (MSPT) station (Fig. la) on this network stopped transmitting at 03:03.3 LT when destroyed by the pyroclastic density current. The main seismic signal was emergent and pulsatory and can be divided into six pulses (Fig. 6 and Table 1). Onset of the signal was gradual so that a starting time cannot be defined precisely. We have taken the start time of the initial collapse as 03:01.0 LT and the finish time as 03:16.2 LT. This main period was followed by less intense high-amplitude signals until the seismicity dropped to background levels at 03:32.1 LT. During this latter period, intervals of monochromatic tremor were recorded at 1.9 Hz, including three higher amplitude pulses. This gives a total duration of the main activity of 15.2 minutes, although the period of strongly pulsed high amplitude seismicity is only 11.6 minutes. The whole seismic anomaly lasted 31.1 minutes. The frequency of the main seismic signal was similar to that of the hybrids and tremor prior to the collapse, with dominant frequency below 2.8 Hz. The dominance of lower frequency energy was associated with the final two pulses of the main activity prior to a period of near-monochromatic tremor at 1.9 Hz. Monochromatic

tremor, best developed between 03:21 LT and 03:25 LT, was punctuated by three high amplitude signals of the same dominant frequency (Fig. 6). Each successive signal, occurring at 03:18.3 LT, 03:23.8 LT and 03:26.7 LT, was smaller than the previous one. The last high amplitude signal was followed by a rockfall signal. The tremor signal and the high amplitude peaks were similar to the signals generated by ash-venting and Vulcanian explosions respectively in the August to October 1997 period. They are therefore interpreted as post-collapse ash-venting and Vulcanian explosions.

Witness information People in several places on Montserrat heard roaring. The timing of these noises is consistent with the roaring being associated with ashventing, as interpreted from the seismic signals. Police at the village of Salem reported hearing two 'explosions' and then seeing flashes of light between 03:15 LT and 03:25 LT. The exact timing of these observations is unknown, and there were also separate reports of thunder and lightning. Darkness and low cloud cover limited direct

GENERATION OF DEBRIS AVALANCHE AND PYROCLASTIC DENSITY CURRENT

413

Fig. 3. Contour map of the dome (shaded area) on 17 December 1997 from a dome survey. The Montserrat grid is shown for reference and contours are in metres. The thick dashed line between Chances Peak and Galway's Mountain marks the position of Galway's Wall. observations. Buildings in the Trials area were burning at about 03:45 LT. At this time, an ash plume was seen from Garibaldi Hill, while a low-level ash plume drifted SW. A small amount of ash fell at Garibaldi Hill but none further north.

Satellite data The eruption plume was estimated to rise to 14.9km altitude, from GOES-8 satellite observations (Mayberry et al. 1998, 2002). The height estimates are constrained by comparing wind directions from Radiosonde data above Puerto Rico and Guadeloupe with

the plume dispersal direction, and by estimates of cloud-top temperature. Piarco FIC (Trinidad) reported ash at 12.2km at 03:55 LT and a British West Indies Airline pilot reported seeing ash at 11 km at 06:05 LT. The ash plume was rapidly dispersed to the SSE and SE at 20-28 km hr-1 (Mayberry et al. 1998, 2002), and passed over the central part of the eastern Caribbean and then out into the Atlantic. A lower level (<2.4km) plume moved slowly westwards. The ash plume was still visible over much of the region 24 hours after the event (Bonadonna et al. 2002). The distribution of ash on Montserrat (Ritchie et al. 2002) suggests that it originated from the plume derived from the PDC rather than from vertical explosions.

Effects of the collapse and pyroclastic current In the morning it was clear that a substantial failure of the flank between Galway's Wall and Galway's Soufriere had occurred, accompanied by a major collapse of the lava dome. The main features were: two collapse scars; a debris avalanche deposit infilling the White River valley and extending almost to the sea; a 70° sector of the volcano with an area of about 10km 2 devastated by a high energy PDC; and deposits formed by the PDC, extending to 600m beyond the original coastline over a 2km lateral distance (Fig. 7).

Collapse scars

Fig. 4. Seismic data for the period 25 October to 30 December 1997 for the Windy Hill seismic station (MBWH) showing the counts of number of events per day for different earthquake types.

Two major collapse scars formed (Figs 8 and 9). The lower scar was a 400m wide arcuate step with its backwall about 200m south of where the crest of Galway's Wall had been and sidewalls extending outwards about 400-500 m. Much of the floor of the lower scar was quite flat (Fig. l0a), although hummocky terrain formed on its southern open side. Fumaroles occurred on the floor of the scar. The headwall of the lower scar formed a 100m high step to the floor of the upper scar (Fig. lOa). The upper scar was a large scoop-shaped

414

R. S. J. SPARKS ET AL.

Fig. 5. Seismic data from 24 to 26 December 1997 at the Windy Hill seismic station (MBWH) showing: (a) rates of occurrence of hybrid earthquakes: (b) the relative amplitudes of the hybrid earthquakes as a measure of relative energy in units of m s: and (c) the real-time seismic amplitude measurement (RSAM) plot in arbitrary units of m s - 1 . The horizontal axis is marked in hours (24-hour clock Greenwich Mean Time) for each day. The pronounced spike on the RSAM is the collapse event starting at 03:01 LT.

amphitheatre within the dome, and extended about 300 m north of the original position of the crest of Galway's Wall. It cut the southern rim of the crater formed in the series of Vulcanian explosions from 22 September to 21 October 1997, and the southern margin of the dome lobe formed between 22 October and 1 November 1997.

The upper scar also cut into the eastern edge of Galway's Wall and the southern side of Chances Peak. A few days after 26 December 1997 the dome started to regrow in the upper scar and began to spill over the step feature. By the end of January the dome, growing at 9- 11 m 3 s - 1 . had partially

Fig. 6. The seismic signal of the activity of 26 December 1997 from the Windy Hill seismic station (MBWH) shown as ground velocity versus time. Six main pulses are identified (1-6). The start time is 02.55 LT on 26 December 1997. The amplitude scale of ground velocity is arbitrary.

GENERATION OF DEBRIS AVALANCHE AND PYROCLASTIC DENSITY CURRENT Table 1. Individual seismic pulse timings within main collapse signal (see Fig. 6)

415

Pulse no.

Start (LT)

End (LT)

1

03:01.8

03:02.3

30

2

03:02.8

03:04.6

108

Even build and decay around central peak, high amplitude at peak

with the pre-failure topography. The estimates were made utilizing simplified geometry and cross-sections. There is uncertainty, particularly in the estimate of the volume of the lower scar. The inventory includes: 20-30 x 106m3 of hydrothermally altered rocks from the Galway's Soufriere area; 5 x 10 6 m 3 of Galway's Wall; 25 x 106 m3 of the lava dome that had grown between 6 November and 26 December; and 30 x 106 m3 of dome talus that had extended beyond Galway's Wall over Galway's Soufriere area. The collapses involved a total volume of 80-90 x 10 6 m 3 , reflecting estimated uncertainties in these figures of about 15%. The volumes quoted here have not been converted to dense rock equivalent.

3

03:04.6

03:06.0

84

Similar to pulse 2, lower amplitude

The debris avalanche

4

03:06.3

03:10.2

234

Duration (s)

5

03:10.2

03:11.3

66

6

03:11.3

03:14.2

174

Comments

Relatively brief, low amplitude

Complex, moderate amplitude, no strong peak Short, moderate amplitude Consistent high amplitude for most of duration

infilled the upper scar, buried the step feature and had substantially filled the lower scar (Fig. 11). The dome ceased growth by 10 March 1998 with extrusion of a spine. Figure 11 shows the rim of the upper scar in March 1998, at which time the dome had largely filled the depression. We have estimated the bulk volumes of material involved in the collapses from the dimensions of the two scars and comparison

The debris avalanche deposit filled much of the White River valley (Figs 7 and l0b) and extended onto the coastal fan of pyroclastic deposits (Figs 7 and l0c), within 20m of the sea. The avalanche surmounted the northern side of the valley where it bends to the south at about 1.7km from the sea, and then emplaced deposits on the plains above the village of Morris' (Figs l0b and 12a). The avalanche also spilled out of the valley on the southern side, and emplaced a small deposit on the flank of Fergus Mountain (Fig. 7). Both of the overspill deposits are detached from the avalanche deposits in the valley, and occur 50-80 m above the upper surface of the valley-fill deposits. The debris avalanche deposits display a distinctive multicoloured heterogeneous patchwork of white, yellow, orange, red, green and grey (Fig. 12). Inspection of the deposits established that the variety of colours is a consequence of mixtures of several lithologies within the debris avalanche on scales of a metre to tens of metres. The debris avalanche deposit is a mixture of hydrothemally

Fig. 7. Map showing the distribution of the deposits of the 26 December 1997 activity in southern Montserrat.

416

R. S. J. SPARKS ET AL.

Fig. 9. Schematic NE-SW cross-section of the region affected by the collapse, with the profile of the dome prior to 26 December 1997. The pre-existing topography is shown by heavy lines. The inferred positions of the failure planes are shown by dashed curves and are schematic.

Fig. 8. Map of Galway's Wall area before (a) and after (b) the activity of 26 December 1997 based on information from video and photographs.

altered and variably weathered volcanic breccias and other volcaniclastic deposits, together with fresh andesitic breccias (Fig. 12b). The white, yellow and orange domains range from matrix-rich lithified breccias to clast-supported unconsolidated breccias, and they resemble the silicified and altered rocks originally exposed around Galway's Soufriere. Grey to orange and brown breccias are similar to the rocks that were exposed in Galway's Wall or in the talus of the new dome that had buried the area of failure. Upon first inspection a

few days after emplacement, the avalanche deposits were mostly cold. However, there were localized hot areas and fumaroles on the debris avalanche surface, which remained active for several months. The debris avalanche deposit displays characteristic hummocky topography with hummocks most prominent in the upper reaches of the White River valley (Fig.l2a). The heights of the hummocks decrease down the river valley. Observations in the area of Galways Soufriere made soon after the collapse found large hummocks up to 30m high surrounded by areas smoothed and filled by pyroclastic flow deposits (Fig. l0b). The proximal hummocks are interpreted as rotated blocks of undisrupted material, as are commonly found in other debris avalanche deposits (Ui 1983: Siebert 1984; Francis & Wells 1988). Hummocky topography becomes subdued down the valley, where the hummocks were smoothed and scoured. The general decrease in roughness of the avalanche surface is attributed to two processes: first, progressive disintegration and shearing of the rock masses reduced hummock size: and second, as detailed below, the surface was eroded by the following PDC. A comparison of the topography before and after the 26 December 1997 collapse indicates that the volume of material added to the White River valley was 46 x 10 6 m 3 (Fig. 13). Most of the valley-fill is the debris avalanche deposit, although PDC deposits locally cover it and infill the depressions between hummocks. The volume of the debris avalanche deposit is substantially greater than the volume of the lower collapse scar, which suggests that it contains a significant amount of dome talus. The debris avalanche deposit is 500m wide and 25-70 m thick in the lower reaches of the White River valley, and on the coastal fan the front of the debris avalanche deposit is 25 m high. More details on the characteristics of the debris avalanche and its deposit are given in Voight et al. (2002). The pyroclastic density current and its deposits The debris avalanche was immediately followed by a pulsatory high-energy PDC, which initiated by explosive disintegration of the

Fig. 10. Views from helicopter on 27 December 1997. (a) A view to the NE from above South Soufriere Hills looking into the lower collapse scar. Comparison with view shown in Fig. 2c shows a deep, flat-floored scar, with some steaming new fumaroles where the southern margins of the dome used to be. The newly exposed outcrops at A are part of Chances Peak. The notch at B is at about 650m altitude and the upper collapse scar within the dome is just visible behind the cloud. The backwall of the lower scar is the step discussed in the text. The newly exposed outcrops at C are the remnants of Galway's Wall below Galway's Mountain, (b) General scene of the White River valley viewed from off the west coast of Montserrat towards the east. The backwall of the lower collapse scar and Galway's Wall can be seen below A. The White River valley is partially filled with debris avalanche deposit (hummocky white-yellow-orange terrain), itself partly covered by grey blocky dome-collapse pyroclastic flow deposits. An area of debris avalanche deposit occurs just to the right of B on the north side of the valley. Fergus Mountain (around C) is stripped of vegetation, (c) View of the lower reaches of the White River valley and new coastal land (C) created by the dome collapses of 4 November and 26 December 1997. The pale-coloured hummocky topography of the debris avalanche filling the valley (D) almost reaches the sea. The grey area of new land with scattered large blocks beyond the old coastline comprises pyroclastic density current deposits. The position of the old coastline is marked by a dashed line.

GENERATION OF DEBRIS AVALANCHE AND PYROCLASTIC DENSITY CURRENT

417

Fig. 10. (continued)

Fig. 11. The dome in March 1998 viewed from the SW. The peak marked A is the rim of the upper collapse scar of 26 December 1997 at an altitude of 965 m. Growth of the dome between 28 December 1997 and 10 March 1998 had largely infilled this collapse scar, overgrown the remnants of Gahvay's Wall, and partly infilled the lower collapse scar (compare Fig. l0a). A late spine forms the summit of the dome at 1030m a.s.l.

Fig. 12. (a) A view from the location of Morris' village (Fig. 7) towards the dome showing the flow front of the debris avalanche deposit (left side of photograph) where it spilled out of the White River valley and deposited on the high ground above the village. Note the blackened ground surface in the foreground due to tar formed beneath the pyroclastic density current, and also the hummocky topography of the debris avalanche deposit partially filling the upper reaches of the White River valley, (b) Close-up of the debris avalanche deposit in the area above Morris' showing a common type of heterolithic matrixsupported breccia containing a variety of hydrothermally altered rocks and sparse fresh andesitic clasts. Hammer is 30cm long.

420

R. S. J. SPARKS ET AL.

Fig. 13. Topography of the White River valley immediately after the activity of 26 December 1997 from surveys on 4 and 17 January 1998. The stippled area marks the distribution of the debris avalanche deposit, valley-ponded PDC deposits and the new coastal fan (compare with Fig. 7). A series of cross-sections down the valley shows the original topography (dashed lines), the infill of material from all earlier flows up to and including the 4-6 November pyroclastic flow deposits (lower solid line) and the infill of deposits from the flank failure and dome collapse of 26 December 1997 (upper solid line).

pressurized lava dome as it collapsed. Here we use the general term pyroclastic density current following Druitt (1998). These currents are similar to the high-energy currents at Mont Pelee, Martinique in 1902, Bezymianny, Russia in 1956 and Mount St Helens, USA in 1980. Such currents have been described variously as lateral blasts, volcanic blasts and pyroclastic surges. Understanding of the processes in these energetic currents is still far from complete. We therefore avoid using the terms pyroclastic flow or pyroclastic surge, as these are commonly taken to imply high- and low-particle concentration conditions respectively. Energetic pyroclastic currents of this kind probably vary considerably in particle concentration, both in space and time, so that the more general term of pyroclastic density current avoids the problem of implying dominance of particular processes or flow properties in the descriptive nomenclature. There are considerable uncertainties in the volume of the PDC deposits, reflecting the fact that most of the material was deposited in the sea and also the uncertainties in how the dome talus was partitioned between the debris avalanche and the PDC. The volume of the debris avalanche (46 x 106 m3) is quite accurate, as it is based on surveys, but the proportion of dome talus incorporated into the debris avalanche is not well constrained. Comparing the debris avalanche volume with the estimated volume of the lower collapse scar indicates that 10-20 x 10 6 m 3 of dome talus was incorporated into the debris avalanche. Thus the PDC is estimated to have involved 35-45 x 10 6 m 3 , of which 25 x 106m3 was derived from the massive part of the dome. As outlined here and described in more detail by Ritchie et al. (2002), marked variations of deposit facies developed across the

area affected by the current. The facies terminology of Ritchie et al. (2002) is used. The main effect of the PDC was to wreak devastation over an area of about 10km 2 . Figure 7 shows the distribution of the area affected, the main zones and some key features. The devastated area involves a sector of 70° focused on the dome and extending from the village of Kinsale, in the north, to about 1 km south of the White River (Fig. 14). Part of the current swept around the northern face of the South Soufriere Hills and Fergus Mountain (Fig. 14a). It reached the summit of the South Soufriere Hills (Fig. 14a) and spilled into gulleys on the steep western face of the mountain. Some of the current passed over the saddle between Galway's Mountain and South Soufriere Hills and then transformed into a dense surgederived pyroclastic flow that travelled down Dry Ghaut (Druitt et al. 2002a) and almost reached the sea (Fig. 7). The PDC also swept over the summit of Chances Peak to the north and across the deep canyons of Germans and Gingoes Ghauts, which did not cause any significant current deflection. The whole area affected by the PDC shows substantial damage and erosional features. These effects are most severe in the area between St Patrick's and the White River valley (Fig. 15b) and decrease towards the southern and northern margins of the affected area. Here we describe the effects and deposits as they occur in traverses from the margins to the areas of greatest destruction. Only areas within 1 km of the coast have been visited on the ground, for reasons of safety. More detailed accounts are in complementary papers: the deposits (Ritchie et al. 2002). the surge-derived pyroclastic flow down Dry Ghaut (Druitt et al. 2002a). and the

Fig. 14. (a) South Soufriere Hills and White River valley area viewed from the west towards the volcano, one week after 26 December 1997. Damage caused by the PDC extends to the summit of South Soufriere Hills (S) while the entire north-facing side of Fergus Mountain (F) is severely damaged with all vegetation and much of the top-soil swept away. The debris avalanche deposits (D) partially fill the White River valley and the PDC deposits have extended the land to 600m over a 2km stretch of coastline. The dashed line marks the position of the former coastline adjacent to the sea-cliffs. Note that the coast had been extended directly beyond the mouth of the White River (area D) before 26 December 1997 by deposition from pyroclastic flows and lahars earlier in the eruption, (b) The area devastated by 26 December 1997 activity viewed from the SW. with the original location of St Patrick's village (P). which was swept into the sea by the PDC. The position of the former coastline is marked by a dashed line.

GENERATION OF DEBRIS AVALANCHE AND PYROCLASTIC DENSITY CURRENT

421

Fig. 15. Deposits of the 26 December 1997 activity, (a) PDC deposits at a locality in the village of Trials, showing the coarse-grained lower layer 1 and finer-grained upper layer 2 of unit I, with a sharp boundary between them. Length of spade is 40cm. (b) Breccia mound deposited by the PDC on the side of Fergus Mountain in an area dominated by erosion. The breccia mound comprises a streamlined hummock of layer 1 facies draped by a thin layer of fine-grained layer 2 (see also Ritchie et al. 2002).

GENERATION OF DEBRIS AVALANCHE AND PYROCLASTIC DENSITY CURRENT

mechanics of the explosive collapse of the lava dome (Woods et al 2002). In the northern margins around Kinsale, the deposit and damage effects come to an abrupt end at Aymer's Ghaut. Within a 100-200 m marginal zone the damage to most buildings is minor, confined to in-blown windows and displaced roof tiles. Wooden sheds were destroyed. The deposits consist of fine-grained ash and thickness varies from 0 to 10cm. In many places, two units can be recognized, each of which is interpreted as recording a discrete pulse of the current. Each unit comprises two layers (Fig. 15a). Layer 1 is relatively coarse-grained and fines-poor, and layer 2 is fine-grained and locally cross-stratified. A thin (3-4 cm) very finegrained ashfall deposit rich in accretionary lapilli caps the units. There is mostly a sharp grain-size break between the layers (Fig. 15a). From the slopes of Chances Peak only small trees (<10cm trunk diameter) were blown down. A similar marginal zone of small tree blow-down occurs on the west-facing slopes of Fergus Mountain and South Soufriere Hills. Along the coast through the village of Trials towards Gingoes Ghaut the levels of destruction increase rapidly. Roofs were partially or entirely removed (Fig. 16a), with evidence of them having been blown upwards. In concrete houses, with typical breeze-block walls 20cm thick, all windows and doors facing the current or at the sides of the house were blown inwards into the building interior, but windows and doors on the side facing away from the volcano were commonly blown outwards (Fig. 16b). Debris, including corrugated iron roofs and free-standing objects such as cars, was strewn across this area. Deposits remain fine-grained with the two units and bipartite internal layering of each unit mostly distinguishable. The deposits drape the landscape, with thickness typically in the range 10-20 cm, and they thicken further within houses. Some houses close to the northern side of Gingoes Ghaut have severe damage or collapse of walls facing the volcano and crumpled cars and other large objects were displaced by tens of metres. The destructive effects increase in intensity on the southern side of Gingoes Ghaut. This is best illustrated by a house and its environs located 200m from the coast and 100m to the south of Gingoes Ghaut. The property owner had a construction business and a yard containing trucks and a bulldozer adjacent to the house. The entire top floor of the house was removed and the debris strewn up to 50m downslope of the ruin (Fig. 16c). Steel reinforcement bars in the concrete walls were bent in the flow direction and windows and doors on the ground floor were blown inwards. Impact marks of large objects occur on the side of the house facing the volcano, in one case with a 1 m diameter rock embedded in the wall (Fig. 16c). Two trucks and a bulldozer were flung from the yard over the cliff onto the beach, and were severely deformed and damaged. Upslope of the house, trees (up to 50cm diameter) were broken at about 1 to 2m above the ground (Fig. 17a), but smaller trees (< 10cm trunk) and shrubs were merely bent over (Fig. 16b). Metal fence posts were bent away from the volcano (Fig. 16d). The PDC deposits in this area are typically 10-30 cm thick, reaching up to 40cm thick in the lee of convex breaks in slope and obstacles. The deposits are fine-grained with maximum juvenile clast sizes of 23 cm, contrasting dramatically with the evident ability of the current to pick up and displace large rocks and objects locally (Fig. 16c). At St Patrick's and Morris' destruction of buildings is complete. In general only the foundations of the houses remain, although a few basements survive where houses were built into a slope (Fig. 16e, f). Buildings swept away include St Patrick's Church, which was a stone-built structure, and the old sugar mill at Galway's Estate. Reinforcement bars and fence posts were blown down in radial directions from the volcano (Ritchie et al. 2002). There is

423

much less strewn debris in this area, and it is inferred that most loose debris, including vehicles and house fragments, were swept out to sea from this zone. Trees with trunk diameters up to 1m were removed entirely and only stumps (from 1 m high) or highly abraded roots remain (Fig. 17b,c). A storm several days after the devastation returned floating logs to the shoreline. The zone between St Patrick's and the shoulder of Fergus Mountain, across the White River valley, is dominated by erosional features (Fig. 14). The steep south-facing slopes of Fergus Mountain above the White River were stripped of vegetation and heavily scoured to a height of at least 300 m above the valley floor (Fig. 18a). The eroded surface displays pronounced furrows, with directions generally away from the volcano, but with local deviations indicative of some topographic influence. The debris avalanche deposit is likewise scoured and striated. Surfaces in the area between Gingoes Ghaut and the White River are not only scoured but also blackened. On close inspection the black colour was found to be a pervasive black tar about 1-4 mm thick (Fig. 18b). The tar does not occur beneath the debris avalanche deposit, making it clear that its origin relates to the PDC. This relationship also shows that the debris avalanche had stopped moving here before the hot density current reached the lower slopes. Preliminary data on the tar composition (I. D. Bull & R. P. Evershed, pers. comm.), in comparison to typical plant and soil material from southern Montserrat, indicate that the organic material comprises compounds consistent with a thermally induced breakdown of biopolymers and complex plant lipids. This supports the interpretation of its origin by rapid heating and destructive distillation of vegetation (grasses, small plants and shrubs). Field evidence indicates that the tar layer: (1) was formed over a surface that had been striated and grooved immediately before deposition; (2) is better preserved in the grooves; (3) flowed after condensation, including over steep walls; and (4) locally shows skid marks from small moving lava clasts. Deposition and local flow of tar were followed by a violent, short-lived phenomenon that cleanly removed all trees (up to several decimetres in diameter) in the area, leaving a clean shave at the boundary with the tar, and stripping the tar from all areas protruding above heights of about 1 cm or less. Locally blocks of older breccia-like units appeared set in a tar matrix, but on close inspection it was clear that the tar once covered all blocks that protruded above the deposit surface and that some efficient process then removed tar from protruding features, leaving it only in lower areas. These observations suggest that, before tar formation, the area was affected by turbulent erosive rapid flow that grooved the surface and burnt all vegetation, thus creating a combustion cloud charged with organic substances that soon (nearly synchronously) condensed to form a plastic tar layer. Then a second, highly energetic cloud passed, with basal clasts that formed skid marks before hardening of the tar and, more strikingly, shaved off all the trees. No significant lag deposit formed here. Deposits in the zone of greatest destruction locally form mounds, sheets and pockets (Fig. 15b). However, more extensive areas of deposition occur within the White River valley in the areas of relatively low slope and depressions on the irregular surface of the debris avalanche deposit. There are marked facies variations. The first-deposited, lowermost facies (layer 1) is a fines-poor breccia, which occurs as isolated mounds, sheets and pockets (Fig. 15b). Breccia deposits are strongly related to local topography and accumulations are common upstream and downstream of obstacles, such as mounds of the debris avalanche, large boulders and tree stumps (Ritchie et al. 2002). The second facies (layer 2) is finegrained, crudely stratified and is mostly made of poorly sorted ash with some small lapilli. These two facies are laterally equivalent to

Fig. 16. (overleaf) Structural damage due to the PDC of 26 December 1997. (a) A house in Trails village with its roof removed and windows and doors facing the current (from the right) blown inwards. (b) Downcurrent side of a house in Trials village with the window shutters blown outwards. (c) A house just to the north of Gingoes Ghaut about 200m from the coast with its top floor removed and impact structures where large blocks have hit the wall facing the volcano. One of the blocks remains embedded in the wall by Daniel Sparks. Note also the steel reinforcement bars bent in the flow direction, (d) A metal fence post (10cm in diameter) bent over in the downcurrent direction by the PDC at the site of St Patrick's village. Hammer 30cm long for scale, (e) Aerial view of St Patrick's village showing roads and foundations of houses with a blackened furrowed surface eroded by the PDC. (f) Ground floor foundations of a concrete villa in St Patrick's. Note the bending of the steel reinforcement bars in the flow direction, and a line of bent metal fence posts beyond the ruin.

424

R. S. J. SPARKS ET AL.

GENERATION OF DEBRIS AVALANCHE AND PYROCLASTIC DENSITY CURRENT

Fig. 16. (continued)

425

426

Fig. 16. (continued)

R. S. J. SPARKS ET AL.

Fig. 17. Effects on trees of the PDC of 26 December 1997. (a) Tree with basal trunk diameter of 30cm just south of Gingoes Ghaut (in the vicinity of the house shown in Fig. 16c) broken at about 1.5m above the ground. The branches and trunks of smaller trees and shrubs are bent but not broken, (b) Tree stump (40cm in diameter) at Reid's Hill Estate (Fig. 7) broken, abraded and burnt by the PDC. Flow direction from right to left. Hammer 30cm long, (c) Tree stump (45cm in diameter) in St Patrick's village broken and abraded flush with the ground by the PDC. The surface of the stump is polished and furrowed.

428

R. S. J. SPARKS ET AL.

Fig. 17. (continued)

the lower (layer 1) and upper (layer 2) deposits in the more marginal zones. The finer-grained fades always occurs above the coarsergrained breccia facies with a sharp grain-size break between the layers (see Gladstone & Sparks 2002). Breccia deposits of layer 1 are commonly not overlain by the fine-grained facies of layer 2 and the fine-grained facies of layer 2 occurs in places directly on the striated and eroded substrate. Both facies have been eroded, forming streamlined local accumulations where protected in the lee of obstacles. The third unit of ashfall deposit rich in accretionary lapilli in places drapes these deposits. More detailed descriptions of the main depositional features can be found in Ritchie et al. (2002). Poorly sorted, valley-confined, block-rich facies occur in the White River valley, with blocks up to 3m in diameter. These deposits infill low areas in the lower reaches of the valley, partially burying the debris avalanche deposit (Fig. l0b). They were visited only briefly and it is unclear whether these deposits represent a PDC facies or less energetic dome-collapse pyroclastic flows generated at a late stage in the 26 December 1997 activity. An important facies interpreted as deposited from surge-derived pyroclastic flows infills the bottom of all the valleys in the area of devastation and also in Dry Ghaut outside this area. This valleyponded facies is typically 1 to >3m thick. Surge-derived pyroclastic flows originate by rapid sedimentation from dilute surges on steep slopes and are therefore generated simultaneously with the deposits from the primary current. More details on the features and interpretation of these deposits can be found in Druitt el al. (2002a). The PDC extended the coastal fan along nearly 2 km of coast (Fig. l0c), out to a distance of 200m. The coastal fan deposits are fine-grained and massive, resembling the surge-derived pyroclastic flow deposits and the fine-grained upper surge layer. However, no sections were yet available for description and it is possible that coarser-grained deposits occur at depth. The volume of deposits of the PDC on land is estimated at only 1.8-3.2 x 10 6 m 3 . The estimated volume of the material that formed the PDC is much larger

(35-45 x 10 6 m 3 ) and it is inferred that most of this either entered the sea or was convected into the lofting plume of ash that was observed by satellites to reach 14.9km altitude (Mayberry et al. 2002). The submarine distribution of pyroclastic material off the coast indicates layers of new pyroclastic material locally up to 50m thick (K. Shufeldt pers. comm.). In the upper part of Galway's Estate there is a region of coarse well-sorted breccia that appears to rest on the PDC deposit. The layer can be traced into scattered angular glassy blocks on the density current deposits near the coast. This layer is interpreted as a Vulcanian fallout deposit related to the three explosions that followed the main dome collapse (Richie et al. 2002). Temperatures were measured in the PDC deposit in the days following their emplacement. Results indicate variable temperatures up to a maximum of 293 C (Table 2). Tsunami On the night of the collapse and PDC there were reports of a wave inundating the Old Road Bay area (Fig. la), located some considerable distance to the NW along the coast from where most of the PDC entered the sea. Investigations revealed that a wave had come inshore near the jetty. The wave was estimated to have been about 1 m higher than the road, which lay approximately 2 m above sea level, and to have moved up to 80m inland. Various objects, including a small wooden boat, a roof to a shelter and a stone table, were displaced several metres inland, and a large log was carried even further by the wave. There were also impact marks, up to a metre above ground level, on the sides of palm trees facing the sea. The grass was oriented in such a way as to indicate the retreat of the wave. There was no evidence for wave inundation higher than high-tide level anywhere else around the western coast of Montserrat.

Fig. 18. Erosional features caused by the PDC of 26 December 1997. (a) The shoulder of Fergus Mountain, south of the White River valley, has been stripped of dense forest and abraded and blackened, with prominent downslope striations marking the flow of the PDC across the mountain. The concentric bands represent the pyroclastic layering in the bedrock, (b) An area near Morris' village (Fig. 7) eroded with prominent furrows and covered by a thin black tar layer. The pale rock on the surface is about 10cm in diameter; a rucksack in the top left corner also gives scale.

R. S. J. SPARKS ET AL.

430

Table 2. Temperature measurements for deposits from the activity of 26 December 1997 Location Deposit type number and location

1 1 1 4 4 4 4 4 4 5 7 7 7

Depth of Time after Temperature measurement event (oC) (cm) (days)

Surge-derived PF, Dry 20 Ghaut at deposit front 25 35 PDC, WR delta area 30 60 30 30 50 30 Surge-derived PF in 50 German's Ghaut PDC over DAD in 20 White River valley 25 60

4

48

4 4 9 9 9 9 9 9 10

138 122 155 216 228 83 93 68 190

13

157

13 13

103 293

PDC, pyroclastic density current; PF, pyroclastic flow; WR, White River; DAD, debris avalanche deposit

The effects at Old Road Bay can be attributed to the peculiarities of wave behaviour along a coastline and to the abrupt change of coast direction at Old Road Bay. According to models (Heinrich et al. 1998) a mass flow with a volume of 25 x 106 m3 entering the sea at 50 m s-1 should produce a source-wave 5 m high. These conditions are close to those estimated for the mouth of the White River soon after 03:01 LT on 26 December. The coastal wave will decay in approximate accordance with the square root of the distance, and should be about 1m high at 10.7km distance at Old Road Bay, which is consistent with the observations. Two effects are thought to have contributed to the focusing of the wave here. First, the wave moving along the coast would have tended to refract towards the shore, especially where the near-shore shelves. Second, the wave met the shore head-on where the coast abruptly changed direction. Dynamical constraints In this section we use observations and models to estimate the energetics and dynamics of the events of 26 December 1997. The activity principally consisted of a debris avalanche that originated from failure of Galway's Wall and Galways Soufriere area, followed immediately by explosive collapse and disintegration of the dome, generating a high-energy pyroclastic density current. Debris avalanche The avalanche deposit was emplaced prior to the arrival of the PDC at approximately 03:03.3 LT, so a velocity of between 19 and 2 7 m s - 1 can be deduced (Voight et al. 2002). Another possible constraint on the debris avalanche is the emplacement of deposits on the sides of the White River valley above Morris' village, 60 m above the original valley floor. A velocity of 3 5 m s - 1 is deduced from conversion of kinetic energy to potential energy, neglecting friction and assuming that the debris avalanche during flow was not substantially expanded in comparison to the thickness of the final deposit in the valley. Heinrich et al. (1999, 2001) presented threedimensional numerical simulations of the debris avalanche emplacement using the pre-eruption topography. Their model treats the debris as obeying a Coulomb-type friction law. Their best-fit model suggests an emplacement time of less than 3 minutes and a peak velocity of 4 0 m s - 1 . The models also indicate that during flow the upper surface of the avalanche was well below the top of the walls of the White River valley.

Ash plume Data from the GOES-8 satellite constrain the ash plume to have reached approximately 14.9km a.s.l. A satellite image at 03:15 LT (at the end of the period of most intense activity) indicates a wide source for the plume, and it is inferred that it was a typical distributed plume formed by buoyant lofting from the PDC (Sparks et al. 1986; Carey et al. 1988). Satellite observations (Mayberry et al. 1998, 2002) show that the plume had two separate parts: an ash-rich higher part and an ash-poor lower part rich in condensed steam, which was displaced to the west as recorded by satellite images. The second, water-rich plume is inferred to be the consequence of the PDC entering the sea. Plume-rise models (Sparks et al. 1997; Calder et al. 1997) can be applied to infer a flux of approximately 6500 m3s-1 for a plume 14km high. If the flux was maintained for 11 minutes then the total volume of airborne ash would be 4.5 x 10 6 m 3 , or 15-20% of the presumed volume of dome material that collapsed to form the PDC. The pyroclastic density current There are several constraints on the dynamics and timing of the PDC. The upper parts of the current climbed almost to the summit of South Soufriere Hills from the saddle between Galway's Mountain and South Soufriere Hills (at which point the PDC transformed into the surge-derived pyroclastic flow in Dry Ghaut). The trimline, representing the upper surface of the current, descends in altitude along the side of Galway's Mountain and intersects the saddle at about 600m a.s.l. Here the current spilled over the saddle and down into Dry Ghaut. The current collided with the South Soufriere Hills and climbed about 150m to the summit at 750m. Using the simple conversion of kinetic to potential energy from the estimate of height climbed, neglecting friction, gives a velocity of 55 m s - 1 . This velocity estimate is for the higher parts of the current in a peripheral area. The St Patrick's seismic station (MSPT in Fig. la) was destroyed at 03:03.3 LT. Two arguments help constrain a minimum velocity from this observation. First, field observations show that the debris avalanche was emplaced before the PDC arrived. The best-fit models of Heinrich et al. (1999, 2001) indicate a period of 50 seconds for the avalanche to reach the bend in the White River valley 2 km from the inferred toe of the failed region. Thus the time of generation of the PDC could not have been much before 03:01.9 LT. Thus the maximum time for the current to move from the dome to the seismic station (a distance of 5km) is 85 seconds, yielding a minimum average velocity of slightly less than 6 0 m s - 1 . The onset of the first pulse in the seismic signal was between 03:01.8 LT and 03:02.3 LT. If it is assumed that the onset of the first pulse represents the initiation of the PDC. then minimum velocities of between 55 and 8 3 m s - 1 can be inferred. The head of a turbulent gravity current travels at about 70% of the speed of the following current (Simpson 1987), so these estimates suggest minimum internal speeds of 80-120ms - 1 . The first pulse was, however, smaller than the second pulse (Fig. 6), so that the peak current speed may have been attained at a later stage. This interpretation is consistent with the observation that PDC deposits in the zone of intense damage are eroded. Damage to different kinds of building structure was extensively studied by nuclear scientists and engineers in the 1950s and 1960s, and can be related to impact pressure. Valentine (1998) reviewed damage to buildings by nuclear explosions to provide comparative data on high-energy volcanic flows. He proposed that the impact pressure in a pyroclastic current can be compared with the pressures experienced by structures in a nuclear blast. He provided tables of damage to different kinds of structures at different impact pressures, which can be interpreted in terms of velocity, u if the flow density, p, is known (impact pressure = pu2) We assume here that the mass fraction of gas in the flow, n is in the range 0.1-0.2. An approximate expression for the density of the mixture is density = p/nR T, where P is the pressure, R = 265 J kg-1 K-1, and the temperature of the flow

GENERATION OF DEBRIS AVALANCHE AND PYROCLASTIC DENSITY CURRENT T = 500 K. Using these values, we calculate a density of about 6 k g m - 3 , which is about five times that of the ambient air. The solid volume fraction, x, in the flow is then given approximately by the expression: x = (l-

n)pjpnRT

where p is the density of the solid. For n = 0.1-0.2, the solid volume fraction therefore has a value of order 0.01. The values of parameters chosen to make this estimate are based on models of the current (Bursik & Woods 1996; Woods et al. 2002), as discussed further below. In the Kinsale area, the damage indicates velocities of

431

about 30-40 m s -1, the damage at Germans Ghaut suggests velocities exceeding 40m s - 1 , and the total destruction in the St Patrick's area suggests velocities well in excess of 60ms - 1 . These estimates are similar to the minimum velocities inferred from observational data above. Estimates of current velocity are now compared with the ash flow model of Bursik & Woods (1996) that considers the sustained flux of a dilute turbulent suspension from a point source. We have adapted that model to investigate a radially expanding flow over a sector of 70° down a cone with a 10° slope, which is the average slope of Soufriere Hills from the dome to the sea. To test the sensitivity of the results, we have examined variations in the initial speed,

Fig. 19. Calculations using the model of Bursik & Woods (1996) to estimate current velocity and the associated dynamic pressure over the first 4km from the source. In all calculations, the gas content is 0.0005 and the initial temperature 1000 K. Unless stated otherwise, the mass flux is 108 k g s - 1 , the initial velocity is l 0 0 m s - 1 , the initial radius is 500m, the slope angle is 10°, and the model does not allow particle fallout, (a) Comparison of predictions for velocity (solid curves) and dynamic pressure (dashed curves) in the case in which fallout is included with the reference model without fallout, (b) Comparison of model predictions for mass fluxes of 1, 2 and 4 x 108 k g s - 1 . (c) Comparison of model predictions for initial velocities of 75, 100 and 125ms - 1 . (d) Comparison of model predictions for initial radii of 250, 500 and 750m. (e) Comparison of model predictions for slopes of angle 5°, 10° and 15°.

432

R. S. J. SPARKS ET AL.

mass flux, radius and gas content of the current, as well as the slope of the terrain. For a conservative volume for the solids in the current of 30 x 10 6 m 3 sustained over no more than 12 minutes, a flux of order 108 k g s - 1 is estimated, but, given the pulsations in the seismic signal, considerably higher peak fluxes are probable, and we present calculations for fluxes of 108, 2 x 108 and 4 x 108 k g s - 1 . Given the uncertainties in the data, this range should cover the actual current conditions. The models assume an initial temperature of 1000K, initial velocities in the range 75-125ms - 1 , initial radii in the range 250-750 m and slopes in the range 5-15°. The initial mass fraction of water in the flow is assumed to be 0.0005, based on the blast model for a dome with 10% porosity at a pressure of 5 MPa (Woods el al. 2002). However, the calculations with a gas content of 0.001 show little difference. The reference calculation for comparison has source velocity of 100ms - 1 , source radius of 500m and gas content of 0.0005. The model assumes a suspension with a particle settling velocity equivalent to particles with a mean size of 300 m and density of 2500 kg m - 3 . The grain size chosen for the model is consistent with the average grain size of the deposit (Ritchie et al. 2002). With a flux of 108 k g s - 1 the reference model current reaches the coast at about 4km distance with a speed of 120ms - 1 , and has a dynamic pressure of order 4 x 104Pa (Fig. 19). However, the model is not fully consistent with the field data in that the 26 December 1997 PDC was largely erosional and there was relatively little sedimentation on land. We have therefore also run a calculation in which the particles do not sediment from the current, in order to determine the sensitivity of the predictions of current velocity on the sediment load. The results (Fig. 19a) suggest that the velocity is a little larger without sedimentation, but that the difference lies well within the range of uncertainty of the source conditions. The model predicts that the current without sedimentation travels further from the source before buoyant lift-off at a distance well beyond the coast. The absence of sedimentation sustains the current's excess density for longer, and thus allows a longer runout for the same flux. In the other calculations (Fig. 19b-e) we assume that there is negligible sedimentation and examine the variation in model predictions with source mass flux (Fig. 19b), source velocity (Fig. 19c), source radius (Fig. 19d) and slope of the terrain (Fig. 19e). The range of source conditions, with associated observational uncertainties, leads to a range of plausible current velocities and hence dynamic pressures. However, in all calculations, over much of the travel distance, the model currents achieve a balance between buoyancy and inertial resistance, and so the velocities and pressures gradually converge with distance from the source. Indeed, the variations in model predictions are within about 25%. The models (Fig. 19a-e) predict velocities in the range 80120ms - 1 , comparable to those inferred from other observations. Using this well-mixed-dispersion model to calculate the velocity of a dilute current should give a good leading order estimate of current properties. However, models for the explosive expansion of the dome (Woods el al. 2002) predict that the resulting initial suspension (the PDC) would have been vertically stratified in terms of particle size and mixture density. The full effects of such stratification have not yet been fully explored, but, as described with some preliminary modelling and experiments in Woods el al. (2002), it appears that stratification probably causes some differences relative to models that assume a homogeneous (well mixed) suspension.

Discussion The activity on 26 December 1997 was by far the most dramatic and intense episode of the eruption. It took place 2.5 years after the eruption began and 14 months after the first signs of instability were detected in both the SW face of Galway's Wall of English's Crater and in the upper flank of the volcano below the wall. The growth of the lava dome ceased 2.5 months afterwards in March 1998. The flank failure on 26 December 1997 thus illustrates that: (1) highly hazardous activity can occur at a late stage in domebuilding eruptions; (2) precursory instability can be manifested over

a long period of time; (3) a major flank failure can occur with little warning; (4) even immediate precursors may not be sufficiently diagnostic to allow accurate forecasting. However, the possibility of a major flank failure was anticipated more than a year before, which led to precautionary evacuation. As a consequence there were no casualties. The recognition of a pattern of enhanced volcanic activity every six to seven weeks from May 1997 onwards also allowed the MVO to anticipate that a significant collapse might take place towards the end of December 1997. While the hybrid earthquake swarm that preceded the flank failure and dome collapse of 26 December 1997 was more energetic than seismicity in the preceding few weeks, it was similar in character to many others that occurred earlier in the eruption (Miller el al. 1998). For example, the seismic swarms associated with the collapse in early November 1997 had individual earthquakes an order of magnitude more energetic than the largest earthquakes in the 25 December swarm. The build-up to the 25 June 1997 collapse, on the northern side of the dome, involved four successive hybrid swarms, starting on 21 June 1997. The major collapse occurred during the fourth swarm. Thus it could not be certain that such a hybrid swarm would inevitably end in a major collapse. The flank failure on 26 December 1997 can be attributed to some combination of dome loading and conduit pressurization, which affected a sector of the volcano already weakened by hydrothermal activity. Wadge et al. (1998) presented calculations on the stress distribution in a section through the dome, across Galway's Wall and through the Galway's Soufriere area. The models predicted that the rock mass in the Galways Soufriere area was the most susceptible to failure, rather than Galway's Wall itself. The rapid growth of the dome over Galway's Wall and Galway's Soufriere from the beginning of November 1997 loaded the area additionally. The relative roles of variations in rock strength, pre-existing structures, pressurization in the conduit, fluid pressure in the hydrothermal system and dome loading in creating conditions for failure require further investigation. The seismic data for the 24-hour period leading up to the collapse can be interpreted as the onset of fracturing leading to failure. The peak seismic energy occurred two hours before failure and the slight decline in seismic activity can be interpreted as registering irreversible weakening. The failure surfaces must have been at low angles to produce the bowl-shaped lower collapse depression (Fig. 9). We conclude that the flank failure undermined the dome, which led to its collapse and consequent generation of a high-energy PDC. The PDC was highly energetic, causing destruction over an area of 10km 2 . The velocity in some places reached at least 6 0 m s - 1 and theoretical models indicate peak velocities in the order of 100 m s-1. The current was pulsatory over a period of about 12 minutes. We interpret the pulses as the consequence of the unsteady retrogressive failure of the dome. A pulsating current is also indicated by the deposits and erosional features. In marginal zones two depositional units, each with a coarse-grained lower layer and fine-grained upper layer, can be recognized. In the central zone of greatest destruction the current was largely erosive, but locally there are remnants of PDC deposits, particularly where topography or obstacles provided some protection. These localized deposits show the two layers characteristic of the two depositional units elsewhere. The local deposits in the central zone are also strongly eroded. The highly energetic PDC requires a significant explosive component during the collapse and initial disintegration of the dome. Close to the source the current was sufficiently energetic and thick to reach the summit of South Soufriere Hills, to spill over the saddle between South Soufriere Hills and Galway's Mountain to generate a surge-derived pyroclastic flow in Dry Ghaut (Druitt et al. 2002a), and to reach the summit of Chances Peak. The current also spilled out of the Galway's Soufriere area over valley walls up to 100 m high to inundate the areas to the south of the White River. The deeply incised valleys of Gingoes and Germans Ghauts are up to 100m deep in their upper reaches below Chances Peak, but these valleys had no substantial influence in diverting the current, as deduced from tree blow-down directions (Ritchie et al. 2002). These observations imply that the current was several hundred metres thick

GENERATION OF DEBRIS AVALANCHE AND PYROCLASTIC DENSITY even close to the dome. Woods et al. (2002) develop a model for the generation of the PDC by collapse and disintegration of a dome with high internal pore pressures. The collapse of the dome on 26 December 1997 excavated deep into it, where pore pressures are estimated to have been high. Woods et al. (2002) and Melnik & Sparks (1999) calculate internal dome pressures of several megapascals. As the pressurized dome disintegrated the released gases expanded explosively to generate a thick density-stratified suspension, which then subsided rapidly under gravity to form the current. The activity on 26 December 1997 contrasts with the 1980 eruption of Mount St Helens, USA. In the case of Mount St Helens, a cryptodome was emplaced asymmetrically in the interior of the edifice within an active hydrothermal system at an early stage in the eruption. Failure of the edifice generated a debris avalanche followed immediately by a high-energy PDC. In the case of the Soufriere Hills a pressurized lava dome built up on top of the edifice above an area of active hydrothermal springs and altered rock. When the edifice failed, it undermined the dome, resulting in a similar sequence to Mount St Helens of a debris avalanche followed by a high-energy PDC. The common factors in the two eruptions are an edifice predisposed to failure, with further instability caused by hydrothermal activity and by emplacement of magma containing pressurized gases. Although the volume of the Soufriere Hills flank failure is only 5% that of Mount St Helens, the PDC was still very destructive. The flank failure and dome collapse of 26 December 1997 illustrated the potential of highly dangerous activity in all dome eruptions where hydrothermal activity related to repeated dome eruptions creates weakened unstable edifices and where high gas pressures in the interior of domes can generate devastating pyroclastic currents when they fail. The authors thank the Department for International Development, which has provided outstanding financial support for the monitoring work on Soufriere Hills Volcano, Montserrat. R.S.J.S. acknowledges the support of a NERC Professorship and NERC grant GR3/11683. We are grateful for the skill and help of the helicopter pilots, particularly J. McMahon and A. Grouchy. Reviews of the paper by P. Kokelaar, S. Self and J. Vallance are much appreciated. Published by permission of the Director, BGS (NERC).

References BONADONNA, C., MAYBERRY, G. C., CALDER, E. S. ET AL. 2002. Tephra fallout in the eruption of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, 1995 to 1999. Geological Society, London, Memoirs, 21, 483-516. BURSIK, M. I. & WOODS, A. W. 1996. The dynamics and thermodynamics of large ash flows. Bulletin of Volcanology, 58, 175-193. CALDER, E. S., SPARKS, R. S. J. & WOODS, A. W. 1997. Dynamics of coignimbrite plumes generated from pyroclastic flows of Mount St Helens, August 7th, 1980. Bulletin of Volcanology, 58, 432-440. CAREY, S. N., SIGURDSSON, H. & SPARKS, R. S. J. 1988. The fluid dynamic behaviour of particle-laden plumes. Journal of Geophysical Research, 93, 15314-15328. CHRISTIANSEN, R. L. & PETERSON, D. W. 1981. The 1980 eruption of Mount St. Helens: Chronology of the 1980 eruptive activity. In: LIPMAN, P. W. & MULLINEAUX, D. R. (eds) The 1980 Eruptions of Mount St Helens, Washington. US Geological Survey, Professional Papers, 1250, 17-30. DRUITT, T. H. 1998. Pyroclastic density currents. In: GILBERT, J. S. & SPARKS, R. S. J. (eds) The Physics of Explosive Volcanic Eruptions. Geological Society, London, Special Publications, 145, 145-182. DRUITT, T. H., CALDER, E. S., COLE, P. D. ET AL. 2002a. Small-volume, highly mobile pyroclastic flows formed by rapid sedimentation from pyroclastic surges at Soufriere Hills Volcano, Montserrat: an important volcanic hazard. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, 1995 to 1999. Geological Society, London, Memoirs, 21, 263-279. DRUITT, T. H., YOUNG, S. R., BAPTIE, B. ET AL. 2002b. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306.

CURRENT

433

FRANCIS, P. W. & WELLS, G. L. 1987. Landsat thematic mapper observations of debris avalanche deposits in the Central Andes. Bulletin of Volcanology, 50, 258-278. GLADSTONE, C. & SPARKS, R. S. J. 2002. The significance of grain size breaks in turbidites and pyroclastic density current deposits. Journal of Sedimentary Research, in press. HEINRICH, P., BOUDON, G., KOMOROWSKI, J.-C., SPARKS, R. S. J., HERD, R. A. & VOIGHT, B. 2001. Numerical simulation of the December 1997 debris avalanch in Montserrat, Lesser Antilles. Geophysical Research Letters, 28, 2529-2532. HEINRICH, P., MANGENEY, A., GUIBOURG, S., ROCHE, R., BOUDON, G. & CHEMINEE, J. L. 1998. Simulation of water waves generated by a potential debris avalanche in Montserrat, Lesser Antilles. Geophysical Research Letters, 25, 3697-3701. HEINRICH, P., MANGENEY, A., BOUDON, G. & ROCHE, R. 1999. Modeliser un raz de maree cree par un volcan. L'eruption a Montserrat, le 26 Decembre 1997. La Recherche, 318, 66-71. MAYBERRY, G. C., ROSE, W. I. & BLUTH, G. 1998. Analysis of the ash cloud from the December 26 1997 (Boxing Day) dome collapse of Soufriere Hills Volcano, Montserrat using GOES-8 Imagery (abstract). Transactions of the American Geophysical Union, 79, 1014. MAYBERRY, G. C., ROSE, W. I. & BLUTH, G. J. S. 2002. Dynamics of volcanic and meteorological clouds produced on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, 1995 to 1999. Geological Society, London, Memoirs, 21, 539-555. MELNIK, O. & SPARKS, R. S. J. 1999. Nonlinear dynamics of lava extrusion. Nature, 402, 37-41. MILLER, A. D., STEWART, R. C., WHITE, R. A. ET AL. 1998. Seismicity associated with dome growth and collapse at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3401-3404. RITCHIE, L. J., COLE, P. D. & SPARKS, R. S. J. 2002. Sedimentology of deposits from the pyroclastic density current of 26 December 1997 eruption at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, 1995 to 1999. Geological Society, London, Memoirs, 21, 435-456. ROBERTSON, R. E. A., ASPINALL, W. P., HERD, R. A., NORTON, G. E., SPARKS, R. S. J. & YOUNG, S. R. 2000. The 1995-98 eruption of the Soufriere Hills volcano, Montserrat. Philosophical Transactions of the Royal Society, 358, 1619-1637. SIEBERT, L. 1984. Large volcanic debris avalanches and associated eruptive phenomena. In: SCARPA, R. & TILLING, R. I. (eds) Monitoring and Mitigation of Volcano Hazard. Springer, Berlin, 541-572. SIMPSON, J. E. 1987. Gravity Currents. Ellis Horwood, Chichester. SPARKS, R. S. J., RICE, C. J. & MOORE, J. G. 1986. The giant eruption cloud of the May 18th 1980 explosive eruption of Mount St. Helens. Journal of Volcanology and Geothermal Research, 28, 257-274. SPARKS, R. S. J., BURSIK, M. L, CAREY, S. N., GILBERT, J. S., GLAZE, L., SIGURDSSON, H. & WOODS, A. W. 1997. Volcanic Plumes. John Wiley, Chichester. SPARKS, R. S. J., YOUNG, S. R., BARCLAY, J. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills volcano, Montserrat: November 1995 to December 1997. Geophysical Research Letters, 25, 3421-3424. UI, T. 1983. Volcanic dry avalanche deposits: identifications and comparison with nonvolcanic debris stream deposits. Journal of Volcanology and Geothermal Research, 18, 135-150. VALENTINE, G. A. 1998. Damage to structures by pyroclastic flows and surges, inferred from nuclear weapons effects. Journal of Volcanology and Geothermal Research, 87, 117-140. VOIGHT, B., HOBLITT, R. P., CLARKE, A. B., LOCKHART, A. B., MILLER, A. D., LYNCH, L. & McMAHON, J. 1998. Remarkable cyclic ground deformation monitored in real time on Montserrat and its use in eruption forecasting. Geophysical Research Letters, 25, 3405-3408. VOIGHT, B., SPARKS, R. S. J., MILLER, A. D. ET AL. 1999. Magma flow instability and cyclic activity at Soufriere Hills Volcano, Montserrat, B.W.I. Science, 283, 1138-1142. VOIGHT, B., KOMOROWSKI, J.-C., NORTON, G. E. ETAL. 2002. The 26 December (Boxing Day) 1997 sector collapse and debris avalanche at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, 1995 to 1999. Geological Society, London, Memoirs, 21, 363-407.

434

R. S. J. SPARKS ET AL.

WADGE, G.. WOODS, A. W., JACKSON, P.. BOWER. S.. WILLIAMS. C. A. & HULSEMANN, F. 1998. A hazard evaluation system for Montserrat. In: Forecasts and Warnings. UK National Coordination Committee IDNDR, Thomas Telford, Project 3, 3.1-3.32. WOODS, A. W., SPARKS, R. S. J., RITCHIE, L. J. ET AL. 2002. The explosive decompression of a pressurized volcanic dome: the 26 December 1997 collapse and explosion of Soufriere Hills Volcano. Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, 1995 to 1999. Geological Society. London. Memoirs, 21, 457-466.

YOUNG. S. R.. SPARKS. R. S. J., ROBERTSON. R.. LYNCH. L., MILLER. A. D.. SHEPHERD. J. & ASPINALL. W. A. 1998. Overview of the Soufriere Hills Volcano and the eruption. Geophysical Research Letters. 25. 3389-3392. YOUNG. S. R.. VOIGHT. B.. BARCLAY. J. ET AL. 2002. Hazard implications of small-scale edifice instability and sector collapse: a case history from Soufriere Hills Volcano. Montserrat. In: DRLTTT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, 1995 to 1999. Geological Society. London. Memoirs. 21. 349-361.

Sedimentology of deposits from the pyroclastic density current of 26 December 1997 at Soufriere Hills Volcano, Montserrat L. J. RITCHIE 1 , P. D. COLE1 & R. S. J. SPARKS 2 1

Centre for Volcanic Studies, University of Luton, Luton LU1 3JU, UK (e-mail: [email protected]) Department of Earth Sciences, University of Bristol, Bristol BS8 1RJ, UK

Abstract: At 03:01 (local time) on 26 December 1997, major sector collapse followed by collapse of the andesitic lava dome occurred at Soufriere Hills Volcano, Montserrat. The collapse of the dome involved explosive disintegration and formation of a highly energetic pyroclastic density current (PDC), which was dispersed principally to the SW and devastated an area of 10 km 2 . The deposits of the PDC are divisible into valley-confined and unconfined facies. The latter is characterized by two bipartite units (Units I and II), both of which are composed of a fines-poor layer (layer 1) typically overlain by a finer-grained, fines-rich layer (layer 2). The sequence is interpreted as recording strongly pulsatory (unsteady) flow and is capped by Unit III, an accretionary lapilli-rich fallout layer. There are pronounced variations of lithofacies, thickness, grain size and sedimentary structures related to local topography. The PDC was highly erosive: it sculpted isolated mounds of deposit and heavily scoured the pre-existing substrate. Lithofacies are granulometrically distinct, with median diameter (Md ) increasing as sorting coefficient ( ) decreases. Lithofacies characteristics depend strongly on azimuth over a 70° sector, with major lateral (crossflow) changes at similar radial distances from the dome. The deposits are similar to those produced in the blast eruptions of Mont Pelee in 1902 and Mount St Helens in 1980. We infer that particle size sorting occurred during explosive expansion of the collapsing lava dome, such that the resulting PDC was initially stratified in both grain size and density. The marked lateral and vertical variations in grain size of the deposits indicate efficient further development of density stratification and grain-size sorting during transport, due to air entrainment and sedimentation.

The collapse of the Soufriere Hills lava dome on 26 December 1997 resulted in the most voluminous and intense volcanic activity of the 1995-1999 phase of the eruption. A pulsatory high-energy pyroclastic density current (PDC) was generated by the collapse. The scale and nature of devastation indicate that the current was much more energetic and destructive than the dome-collapse and fountaincollapse pyroclastic flows that had occurred previously (see Cole et al. 2002; Druitt et al. 2002b). The deposits of the PDC show strong similarities to those generated in eruptions that have been identified as volcanic blasts, such as at Bezymianny, Russia in 1956 (Belousov 1996) and Mount St Helens, USA in 1980 (Hoblitt et al. 1981). Volcanic blasts are devastating phenomena, most dramatically demonstrated at Mont Pelee, Martinique by the destruction of St Pierre, on 8 May 1902 (Lacroix 1904). Deposits formed by these blasts provide insights into transport and depositional mechanisms; those on Mont Pelee have been documented by Fisher et al. (1980), Fisher & Heiken (1982), Boudon & Lajoie (1989) and Bourdier et al. (1989). The eruption of Mount St Helens on 18 May 1980 allowed further detailed examination of blast deposits (Hoblitt et al. 1981; Walker 1983; Fisher et al. 1987; Fisher 1990; Druitt 1992). The need to analyse the effects and understand the processes involved in the entrainment, transport and deposition of material within volcanic blasts is highlighted by the severe devastation caused, while the study of the deposits also allows refinement of the chronology of the eruption. In this paper we describe the deposits and landscape modification due to passage of the PDC, and use the deposit lithofacies to constrain its transport and depositional mechanisms. We first describe the stratigraphy of the deposits and characteristics of lithofacies preserved, with their lateral variations. Second we describe sedimentary structures, erosional features and directional data in conjunction with detailed grain-size characteristics of the deposits. Finally we present a model for the transport and depositional mechanisms, related to theoretical modelling of the lava-dome collapse (Woods et al. 2002).

26 December (Boxing Day) 1997 collapse The chronology and dynamics of the 26 December (Boxing Day) 1997 collapse are discussed in detail by Sparks et al. (2002) and are only briefly summarized here. The collapse followed rapid growth on the southern side of the dome in November and December 1997,

after the September-October 1997 period of explosive activity (Young et al. 1998; Sparks et al. 1998; Robertson et al. 2000). Sector collapse of the hydrothermally altered flank at 03:01 local time (LT) undermined the dome, which then failed. Explosive disintegration of the pressurized dome generated a highly energetic PDC that devastated about 10km 2 over a 70° sector of the southwestern flank of the volcano (Fig. 1). The size of the resulting scars indicated that the dome collapse was approximately four times larger than previous collapses and that the entire event involved a total volume of material of 80-90 x 10 6 m 3 (Sparks et al. 2002). The debris avalanche from the sector collapse travelled 4km down the White River valley (Fig. 1) and terminated on a preexisting fan of pyroclastic material at the coast. Its deposits (an estimated volume of 46 x 10 6 m 3 ) comprise hummocky lobes of hydrothermally altered material, and occur along the length of the White River valley (Voight et al. 2002). Two lobes of the debris avalanche deposit, one NW of the village of Morris' and another on the east side of the White River valley (Fig. 1), formed where the avalanche surmounted the valley sides. The main PDC was directed southwestwards from the dome along an axis 1.5km north of the White River valley, and a large proportion of it entered the sea (Sparks et al. 2002). On land the PDC was largely erosive and generally formed thin deposits, as documented in this paper. An ash plume that was derived from the PDC ascended to 14.9 km and was tracked by the National Oceanic and Atmospheric Administration (NOAA) as it was blown towards the SSE (Bonadonna et al. 2002). According to seismicity, the main explosive collapse was sustained for 15.2 minutes and was followed by three high-amplitude signals interpreted as registering Vulcanian explosions (Sparks et al. 2002). The whole event had a duration of 31.1 minutes. On the southern side of the White River valley the PDC ran up 150m elevation to the summit of South Soufriere Hills (Fig. 1), which implies a minimum velocity of 55 m s-1 (Sparks et al. 2002). Velocities estimated from the destruction of a seismic station at St Patrick's (Fig. 1), and damage to buildings (Sparks et al. 2002), are in the range of 50-70 m s - 1 . Applications of flow models (Bursik & Woods 1996; Woods et al. 2002) indicate that peak PDC velocities may have exceeded 9 0 m s - 1 . Description of lithofacies The area devastated by the collapse is characterized by deep valleys (ghauts) radiating from the southern and western flanks of

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 435-456. 0435-4052/02/$15 © The Geological Society of London 2002.

435

436

L. J. RITCHIE ET AL.

Fig. 1. Topographic map of southern Montserrat showing the area devastated by the 26 December 1997 pyroclastic density current (PDC) and the extent of the associated deposits.

Soufriere Hills Volcano. The largest drainage is the White River valley, which extends south from the dome (Fig. 1). Prior to the eruption, the upper slopes of the area were heavily vegetated and several villages were located near the coastline to the west and SW. Dome collapses and explosions earlier in 1997 generated pyroclastic flows and surges that travelled down the White River valley (Cole et al. 2002) and built a fan of pyroclastic deposits extending 1 km along and 400m seaward of the original coastline. The general distribution of the main lithofacies is shown in Figure 1, and the main sedimentological and granulometric characteristics of each of the lithofacies are detailed in Figure 2. To describe the deposits and effects of the PDC we have divided the devastated area into six regions (Fig. 3) on the following basis. First, the PDC went out to sea, so there is no information on the most distal areas. Second, lithofacies distributions are strongly controlled by topography, so there are significant differences between those deposits occurring in valleys and those on interfluves. Some regional boundaries are defined by marked changes in grain size across major valleys. Third, we observed that lithofacies, structural damage and erosion depend strongly on azimuth, such that across the 70 sector of the affected areas there are major lateral changes at similar radial distances from the dome. The strong azimuthal variations in lithofacies make development of a scheme of proximal to distal variations difficult. Hence we describe axial to peripheral areas across the 70 o sector, with the main axis of the PDC extending through the upper reaches of the White River valley and across the southern part of Region 3 (Fig. 3). We divide the deposits of the 26 December 1997 PDC primarily into those that are valley-confined and those that are unconfined.

Uncon fined lithofacies Deposits constituting unconfined lithofacies cover an area from the summit of the volcano to Aymer's Ghaut in the west and to the summit of South Soufriere Hills in the south (Fig. 1). Three stratigraphic units (I. II and I I I ) are defined (Fig. 2). The deposits are similar to those documented at Mount St Helens (Hoblitt et al. 1981) and our nomenclature follows previous schemes (Hoblitt et al. 1981; Druitt 1992. 1998).

Unit I, layer 1. Layer 1, at the base of the sequence, typically ranges from coarse breccia to coarse ash and has little fine ash. It is up to 3m thick and characteristically shows normal grading (Fig. 4a), although reverse to normal grading occurs locally (Fig. 2). Layer 1 is brown to grey and is predominantly composed of clasts of poorly vesiculated juvenile dome rock, which increase in abundance and maximum size (up to 50cm) from peripheral towards axial areas. Sparse accidental pumice clasts. thought to have been incorporated from underlying deposits of the Vulcanian explosions of August to October 1997 (Young et al. 1998: Druitt et al. 2002b). are also present. At localities downcurrent of the debris avalanche deposits (e.g. locality 70, Fig. 3). layer 1 contains up to 40% of accidentally incorporated hydrothermally altered clasts. Within 2-3 km of the dome, the layer is rich in accidental lithics and elongate fragments are imbricated with long axes oriented 180-200 . Poorly defined stratification occurs locally at the top of layer 1. The contact with the substrate is typically sharp and erosional (Fig. 4b.c).

SEDIMENTOLOGY OF DEPOSITS FROM THE PYROCLASTIC DENSITY CURRENT

437

Fig. 2. Composite section and summary of the sedimentological characteristics of the unconfined facies of the pyroclastic density current (PDC) deposits (N, normal grading; R, reverse grading; S, symmetrical grading; P, planar stratification; C, cross-stratification). Median diameter (Md0) and sorting coefficient ( ) are mesures as defined by Inman (1952).

Unit I, layer 2. Layer 2 is predominantly composed of ash wit subordinate lapilli, up to 5 cm in size, and is finer grained than layer 1. It typically ranges in thickness from 1 to 10cm, although locally it is up to 30 cm thick, and is grey and massive to normally graded with local reverse and reverse-to-normal symmetrical grading (Fig. 2). Stratification is generally better developed than in the underlying layer 1, with cross-bedding and dune-forms present (Fig. 4b, c). Contacts with the underlying layer 1 are typically sharp with a distinct decrease in grain size (Fig. 4a), but locally they are gradational.

Unit II, layer 1. This layer ranges from breccia to lapilli-bearing coarse ash. Its maximum thickness is 160cm in axial areas, but it thins rapidly to 2-7 cm in peripheral areas (Fig. 2). The layer is normally graded at distances of 3-6 km from the dome and reversely graded at more axial localities <3 km from the dome. Lithics up to 1 m in diameter are of both angular, glassy, poorly vesiculated dome rock and hydrothermally altered lithologies. Locally the layer shows marked thickening into topographic depressions (Fig. 5a, b).

dome rock up to 3 cm (Fig. 2). Where the layer is thinner than 5 cm it is typically stratified and has similar characteristics to layer 2 of Unit I.

Unit HI. Unit III is a normally graded layer, 4-6 cm thick, composed of fine ash with abundant lenticular accretionary lapilli typically 5 mm in diameter but up to 11 mm in diameter. Angular clasts of moderately vesicular juvenile dome rock, ranging in size from 1 to 5 cm, are typically scattered across the surface. Unit III is interpreted as the fallout layer from ash clouds generated by the PDC. The accretionary lapilli within this unit are taken as evidence of abundant moisture in the cloud and the constant thickness of the layer over topographic irregularities suggests a simple fallout mechanism of deposition. The scattered angular clasts are interpreted as registering the three Vulcanian explosions that followed the generation of the PDC.

Valley-confined Unit II, layer 2. Layer 2 is rich in fine ash. It is up to 7cm thick, normally graded and contains clasts of poorly vesicular, juvenile

lithofacies

The valley-confined deposits are subdivided into block-poor deposits, which were found in all major drainages and topographic

438

L. J. RITCHIE ET AL.

Fig. 3. Topographic map of the devastated area showing region boundaries based on deposit characteristics and damage effects. Numbered dots represent sample localities mentioned in the text and arrows illustrate the mean flow directional data collected from bent steel reinforcement bars on buildings, steel fence posts and. where no structures were present, tree blow-down, streamline features and striations.

depressions across the area, and block-rich deposits, which occur primarily in the White River valley. Block-poor deposits. A major block-poor deposit formed in Dry Ghaut (Fig. 1). The current that formed this drained east from above the White River valley and terminated 200m from Landing Bay on the east coast of the island. It moved outside the area inundated by the parent PDC, in a southeasterly direction governed by the valley. The deposits are fine-grained and massive with

maximum thicknesses of 1.5m. They show typical valley-filling form with flat upper surfaces. The block-poor deposits are attributed to high-concentration granular flows that were derived by sedimentation from pyroclastic surges. These surge-derived pyroclastic flows are described in detail by Druitt el al. (2002a). Similar block-poor valley-filling deposits occur in all the valleys (e.g. Gingoes and Germans Ghauts) in the area affected by the PDC, and are also interpreted as surge-derived pyroclastic flow deposits in that they derive from the lower concentration parts of the PDC.

Figure 4. (a) Unit I. layers 1 and 2 in Region 4 Upper Galway's Estate, locality 67 (Fig. 3). Note the pronounced grainsize break between the layers, (b) Unit I. layers 1 and 2 and Unit III in Fairfield. Region 1. locality 61 (Fig. 3). Note the cross-stratification in the hollow accompanied by a thickening of layers 1 and 2. (c) Unit I. layers 1 and 2 and Unit III. Stratification and infill in a topographic depression in Trials. Region 1. locality 32 (Fig. 3).

SEDIMENTOLOGY OF DEPOSITS FROM THE PYROCLASTIC DENSITY CURRENT

439

Fig. 4. (continued)

Block-rich deposits. These deposits resemble the block-and-ash flow deposits produced previously by dome collapse on Montserrat (Cole et al. 1998, 2002). They are up to several metres thick and partially infill topography on the hummocky surface of the debris avalanche deposit in the White River valley and locally above the village of Morris' (Fig. 1).

Regional and local lithofacies variations We now describe the lithofacies variations with respect to the defined regional areas (Fig. 3) and, within each region, to local topography. On a large scale, deposits drape the topography throughout the area (Fig. 6). However, on a local scale, marked lithofacies and thickness variations relate to small-scale topography. Some of the thickening in valleys such as Gingoes and Germans Ghauts is due to deposition from surge-derived pyroclastic flows (Druitt et al. 2002a). On a local scale (0.01-10 m), rapid changes in thickness (0.1 to 1 m) occurred in one of two different ways related to topography. First there was marked thickening in depressions and valleys irrespective of their

orientation, which we refer to as ponding (Fig. 7a). Second, thickening occurred in the lee of obstacles and convex breaks in slope (Fig. 7b). Thickness relationships from peripheral to axial regions are illustrated in Figure 8a and b. Both layers 1 and 2 tend to increase in thickness from peripheral to axial areas. Normal grading in the deposits is well developed, especially in Unit I, layer 1. Median diameters plotted against height above the local ground level show this trend (Fig. 9). The strength of grading apparently is unrelated to the thickness of the deposit or location (Fig. 9) Region 1 Region 1 is the most peripheral of the devastated area (Figs 1 and 3). Deposits are more continuous but thinner than in all other regions, with a mean total thickness of 10cm (Fig. 2). Within Unit I, layer 2 is typically finer grained and less well sorted than layer 1 (see Fig. 2). Layer 1 shows normal grading, and near the northerly limits of the deposit it becomes discontinuous and is confined to small (10cm wide and 2-5 cm deep) topographic depressions (Fig. 4c). The

440

L. J. RITCHIE ET AL.

Fig. 5. (a) Unit II, layer 1 in Region 4 Upper Galway's Estate, locality 68 (Fig. 3). The deposit is a topographically confined sheet and is absent on the interfluve to the right, which is covered by Unit I, layers 1 and 2 (person for scale). (b) View of the edge of the Unit II. layer 1 deposit. locality 68 (Fig. 3). Note the absence of impact craters on the surface of Unit II, layer 1 (person for scale middle right). Box shows the location of (a).

deposits are typically massive, but commonly become stratified adjacent to, or within, small-scale topographic irregularities (Fig. 7). In particular, stratification occurs downcurrent of small obstacles and at the base of small depressions 10-30 cm across (Fig. 10). Stratification typically consists of several alternating fines-rich (median diameter, Md0 = c. 2.5) and fines-poor (Md0 = c. 1.4) layers, each of which is up to 1 cm thick. Occurrence of Unit II, layer 1 is intermittent throughout this region. The layer generally has sharp contacts with underlying Unit I, layer 2 (Fig. lla). Unit II, layer 2 is thickest in the NE of the region within 2 km of the dome, and has a gradational contact with Unit II, layer 1 (locality 75, Fig. 3). A remarkable feature of Region 1 is that deposits of Units I and II are absent on surfaces at elevations >1.5m above the local ground level. Several houses with flat concrete roofs 1.5-2m above the local ground surface lack Unit I and II deposits. Unit III is ubiquitous and occurs up to 6cm thick (Fig. 2), with accretionary lapilli up to 11 mm in diameter.

Region 2 Region 2 is dominated by two interfluves. Reid's Hill and Spring Estate, and is bounded on either side by Germans and Gingoes Ghauts (Fig. 1). The deposits show little overall variation in thickness (mean 20cm) and in general drape the topography with a flat upper surface. In this region all layers from the three depositional units were present locally (Fig. 11b). Unit I, layer 1 is typically normally graded, and in a few places has local poorly defined internal stratification. Substantial local lateral variations in median diameter occur; for example in one locality the layer coarsens from 2.4o to 0.8o over a horizontal distance of 50cm. Unit I, layer 2 shows well developed stratification adjacent to small breaks in slope or depressions of only a few centimetres in depth (Fig. 4b,c). The layer is typically normally graded within these depressions. In pronounced depressions the layer is typically stratified, with alternating well sorted, fines-rich

Fig. 6. Schematic transect from NW to SE across the devastated area, illustrating the draping nature of the deposits, and measured sections with grain-size histograms. Locality numbers of sections are marked on Figure 3.

442

L. J. RITCHIE ET AL.

Fig. 7. Diagrams illustrating thickening characteristics in relation to local topography in the pyroclastic density current (PDC) deposits. Locality numbers correspond to Figure 3. (a) Ponding. (b) Thickening into convex breaks of slope.

(Md0 — c. 3.1) and fines-poor (Md0 — c. 1.5) layers up to 1 cm thick (Figs 7bii and 4b, c). Bordering deep valleys, one or two 4-5 cm thick fines-rich ash layers (Md0 = c. 2.7) typically occur at the top of layer 2 of Unit I. The lower of these layers is typically massive and the upper layer is stratified. A 1 cm layer of grey, fines-poor ash (Unit II, layer 1) intermittently overlies these layers, which are interpreted to be part of Unit I, layer 2. These layers, observed throughout Region 2. correlate with similar layers in the northwestern limits of Region 3.

Deposits of Unit II are thickest on Spring Estate in the NE of this region (Fig. 12). Layer 1 is normally graded. containing clasts up to 3 cm of moderately vesicular grey juvenile dome rock. It has a gradational contact with the overlying finer grained layer 2. Elsewhere in the region, layer 1 occurs as an intermittent deposit <1 cm thick (Fig. 11b). Layer 2 occurs only on Spring Estate (localities 72 and 75); it is normally graded, containing clasts up to 3cm in size, and has poorly developed stratification. Unit III is up to 4cm thick and is normally graded, with accretionary lapilli up to 5 mm in diameter. Region 3 This region extends from Germans Ghaut in the NW to the White River valley in the east and includes an overspill lobe of the debris avalanche deposit. This area is predominantly characterized by PDC deposits that are patchily developed. The deposits are typified by low-profile mounds and ridges metres to tens of metres long.

Fig. 8. (a) Thickness at all localities plotted against distance from west to east shows a rapid increase in axial regions. (b) Unit I. layers 1 and 2 have similar thicknesses in peripheral areas, but layer 1 tends to thicken more in axial areas relative to layer 2, which is thickest to the NW of the White River valley.

Fig. 9. Median diameter versus height above base of the deposit to illustrate strong normal grading. All samples are Unit I. layer 1 except those marked otherwise. Tie-lines join samples from the same locality.

SEDIMENTOLOGY OF DEPOSITS FROM THE PYROCLASTIC DENSITY CURRENT

443

Fig. 10. Well developed stratification developed downflow of slope break above a brick fragment (B) in Region 1, Fairfield, locality 55 (Fig. 3). Note the thickening of Unit I, layers 1 and 2 downflow of the break.

sheet-like deposits, and accumulations localized in hollows and around obstacles. These localized and often isolated depositional structures occur in areas dominated by non-deposition or erosion. Unit I, layer 1 deposits typically form lensoid mounds up to 40 m long and 3m high, as well as accumulations both upcurrent and downcurrent of obstacles such as building foundations, boulders and tree stumps (Figs 13, 14 and 15). More detailed descriptions of these features are given in the 'Structures and bedforms' section. Layer 1 is typically a coarse breccia largely composed of blocks, up to 1 m in diameter, and lapilli with little fine ash. The layer is massive, although normal grading is well developed throughout the region. Clasts commonly show imbrication with long axes oriented at approximately 220°. Elongate accumulations of layer 1 breccia are also orientated with long axes at about 220° (Fig. 15a). In nonsheltered areas Unit I, layer 1 is commonly absent and layer 2 rests directly upon an eroded surface (Fig. 14). Layer 2 is up to 30cm thick and shows poorly developed stratification. Contacts between layers 1 and 2 are mostly sharp (Fig. 14), although gradational contacts occur in sheltered locations. Bordering Germans Ghaut, one or two layers of finer grained ash (Md = 2.8) are present at the top of Unit I, layer 2 (Fig. 11c). The upper one is typically stratified but usually ungraded, whereas the lower layer is massive, locally with symmetrical or reverse grading. These layers correlate with deposits at the same stratigraphic level described previously in Region 2 (Fig. 11c). Unit II, layer 1 is present in the western parts of the region as an intermittent layer of fines-poor, coarse ash up to 2cm thick. However, in the hummocky terrain on top of the debris avalanche deposit, Unit II, layer 1 comprises a coarse breccia commonly concentrated around larger boulders and hummocks, with accumulations both upcurrent and downcurrent of such obstacles. Unit III is up to 4cm thick and confined to the northwestern parts of the region where there is thinning and fining of Unit I, layers 1 and 2 (Fig. l1c).

Region 4 Region 4 extends from 1 to 3 km from the dome (Fig. 3). The deposit is discontinuous with considerable thickness variations over lateral distances of as little as 2m. The deposits are more continuous in the northeastern parts of the region. Unit I, layer 1 is typically a normally graded, coarse breccia although local reverse-to-normal grading also occurs. Imbrication of elongate clasts with long axes

oriented 230° occurs locally. Contacts between layers are mostly sharp (Fig. 4a) although some are gradational. Unit I, layer 2 is fines-rich (Md — 2.7) and thickest in the lee of large boulders and pronounced convex breaks in slope (Fig. 7). Stratification, although poorly developed, occurs where the layer is thickest. At locality 68 (Fig. 3) Unit II, layer 1 rests directly upon an eroded surface and shows marked thickening into a depression tens of metres wide (Fig. 5). Large boulders up to 1m in diameter are present on the upper surface of the deposit and elongate clasts exhibit a preferred orientation of 228 C . Impact craters associated with the large blocks were not found. Region 5 Region 5 is east of the White River valley (Fig. 3) and is notable for the extensive erosion on the west-facing valley wall where almost all vegetation, including trees, has been removed. A valley-confined surge-derived pyroclastic flow was emplaced in Dry Ghaut to the east (Druitt et al. 2002a). Deposits are of limited extent due mainly to the predominance of steep slopes, although layers 1 and 2 of Unit I are both found at O'Garras' on the coast (Fig. 1). Both layers notably contain hydrothermally altered lithic fragments. Unit III is absent from Region 5. Region 6 Region 6 includes the White River valley, extending 4 km south to the coast (Fig. 3). Safety considerations prevented detailed examination of PDC deposits there, so the following account is preliminary. The valley is filled primarily by debris avalanche deposits (Sparks et al. 2002; Voight et al. 2002), which extend onto the pre-existing pyroclastic fan at the coast. The debris avalanche deposit is partially mantled by predominantly coarse-grained PDC deposits up to 10m thick. The PDC deposits contain moderately vesicular to nonvesicular blocks of andesitic dome rock, up to several metres in diameter, set in an ash matrix with a layer of fines-poor coarse ash at the base. Locally the PDC deposits contain abundant (>40%) hydrothermally altered clasts that are similar to those in the debris avalanche deposit (Voight et al. 2002). These coarse-grained and massive deposits could not be specifically correlated with the layered deposits in other regions and are mapped as a discrete lithofacies in Figure 1. They closely resemble dome-collapse pyroclastic flow

444

L. J. RITCHIE ET AL.

Fig. 11. Measured sections, with corresponding grain-size histograms, through the pyroclastic density current (PDC) deposit across the regions: (a) Region 1; (b) Region 2; (c) Region 3; (d) Region 4.

deposits (Cole el al. 2002). Prominent boulders and hummocks on the surface of the debris avalanche deposit provided loci for the deposition of PDC deposit both upcurrent and downcurrent. Smoothed and striated surfaces on the debris avalanche deposit indicate that the PDC caused considerable scouring. PDC deposits occur at the coast, with the pre-existing fan widening from 1 to 2km (Fig. 1). On the coastal edge of the fan, massive PDC deposits, up to 1m thick, show reverse grading with blocks up to 15cm in diameter. The massive deposits are divided into two or three subunits by fine-grained reversely graded layers (Fig. 16). The subunits have characteristics similar to layers 1 and 2 of the standard ignimbrite sequence of Sparks et al. (1973). They pass laterally into thinner, finer grained and normally graded layer 2 deposits, although the correlative unit, if any, is unknown. Gas segregation pipes were formed locally. Numerous surface craters, thought to be formed by degassing from pipes, occur up to a few metres in diameter and are encrusted with a sulphurous deposit.

Interpretation of depositional features Stratification Stratification occurs throughout, but is especially well developed in the base of topographic depressions. Stratification and development of cross-stratification are more prominent with distance from the volcano. Well developed cross-stratification is diagnostic of traction occurring during sedimentation. Where the currents crossed topographic irregularities, such as small depressions. flow separation is likely to have occurred. This flow separation will develop turbulent eddies that promote tractive working of the sedimented material. Druitt (1992) relates cross-stratification in the deposits at Mount St Helens to a dilute current, which allows tractive working of the sediment load. Arnott & Hand (1989) showed in their experiments that development of lamination is supressed and a massive deposit is formed when sedimentation rate is sufficiently

SEDIMENTOLOGY OF DEPOSITS FROM THE PYROCLASTIC DENSITY CURRENT

445

Fig. 11. (continued)

rapid. The increasing prominence of stratification in peripheral areas indicates that tractional sedimentation was more dominant. This in turn implies a decreasing rate of suspended-load fallout and decreasing particle concentration with distance. Grading Normal grading occurs throughout the deposits, unrelated to location or thickness, and is particularly well developed in Unit I,

layer 1. Normal grading is attributed to deposition from a turbulent particulate current (Middleton 1967) with sufficiently low concentration to allow segregation of coarser grains to the leading part and base of the current. Middleton (1966) noted that rapid deposition of sediment and grading within the deposits occur due to the decline of velocity of the body of the current with time at a fixed location. Reverse and symmetrical reverse-to-normal grading is observed in the deposits, particularly in Unit I, layer 2, and is interpreted as being indicative of high-concentration suspensions deposited under waxing and fluctuating flow conditions respectively (Middleton 1967).

446

L. J. RITCHIE ET AL.

Fig. 12. Complete stratigraphic section of the deposits at Spring Estate. Region 2 locality 72 (Fig. 3).

There typically is a grain-size break and a sharp contact between layers 1 and 2 of Units I and II, which we interpret as the boundary between deposits from the body of the current (layer 1) and the turbulent overlying wake (layer 2). The sharp contact is interpreted to represent an erosive phase between the deposition of layers 1 and 2.

Structures and bedforms Within Regions 6, 5, 3 and the lower reaches of Region 2, numerous localized accumulations of coarse layer 1 breccia are present. These accumulations typically form isolated mounds in areas dominated by erosion and/or non-deposition. In this section we describe some of the best examples. On a flat shoulder of Fergus Mountain, just east of the White River valley in Region 5 (locality 69, Fig. 3), a large breccia mound occurs and in section has the shape of an aerofoil (Fig. 15b). Its long

axis is oriented north-south (359 ). oblique to the inferred current direction. The structure is 70m long and varies in width from 25m at its northern end to 17 m in the south. The northeastern, upcurrent margin is 3 m high and almost vertical while the downcurrent end tapers to the west at an angle of 14 . The mound is composed mainly of coarse (Md > -5.0). moderately sorted ( =1.9) layer 1 breccia. It is up to 2.7m thick with a drape, up to 30cm thick, of layer 2, which thins to 10cm at the southern end of the structure. Clasts within layer 1 are up to 1 m in diameter, but are typically 60cm or less. They are of fresh, grey juvenile dome rock, together with abundant (>50% volume) hydrothermally altered rock. Imbrication of elongate large clasts and wood fragments is well developed within the mound, and long axes of imbricated clasts are oriented approximately 200 as arrow on Fig. 3). In Region 3, west of the debris avalanche overspill lobe at Morris' (locality 48, Fig. 3), is an isolated elongate mound 40m long, 7m wide and 3m high, with its long axis oriented at 222 (Fig. 15a). It thins at both ends (NE and SW). with the downcurrent end thinning as a tapered wedge. The sides slope at 30 to the SE

Fig. 13. (a) Deposit accumulated downcurrent of a boulder in Region 3 locality 46 (Fig. 3). Unit I, layer 1 rests upon pre-eruption ashfall deposits associated with previous block-and-ash flows that travelled down the White River valley in 1997. Note the scattering of Vulcanian fallout clasts on the surface of the deposit, (b) Accumulation of Unit I. layer 1 upcurrent of a boulder in Region 3 near locality 48 (Fig. 3). (c) Accumulation of Unit II. layer 1 up- and downcurrent of a boulder in Region 3 near locality 48 (Fig. 3). (d) Accumulation of Unit I, layer 2 in the lee of a demolished building, locality 47. Morris' (Fig. 3) (person for scale).

SEDIMENTOLOGY OF DEPOSITS FROM THE PYROCLASTIC DENSITY CURRENT

Fig. 13. (continued)

447

448

L. J. R I T C H I E ET AL.

Fig. 14. Relationships of Unit I, layers 1 and 2 and Unit III in the lee of boulders. Layer 1 is preserved directly behind the boulder or within a metre downcurrent. Sections perpendicular to the long axes of the features show that erosion may have occurred obliquely to the long axis of the structure.

and 18 to the NW. It is composed of coarse layer 1 breccia with subangular to subrounded clasts of grey, juvenile andesite and hydrothermally altered dome rock up to 60cm, but predominantly 10-30 cm. The whole mound is draped by up to 25cm of layer 2 deposit. The surrounding area is devoid of the coarse layer 1 material, and layer 2 occurs on the pre-eruption surface. Unit I, layer 1 breccia accumulations are typical in the lee of obstacles, such as severed tree trunks and building debris (Fig. 13d). The dimensions of the accumulation vary according to the size of the obstacle, but are generally 2-3 m long, 1-1.7m wide and 5070 cm high (Fig. 13). Layer 1 is typically 0-25 cm thick, composed of poorly sorted ( = 2.5), coarse ash (Md = 2.0). Layer 2 is 0-4 cm thick, well sorted ( = 1.7), fine ash (Mdo = 3.0). Two different occurrences in the stratigraphy of the deposits downcurrent of the boulder are identified. First, Unit I, layer 1 occurs downcurrent of

the boulder with Unit I. layer 2 directly above (Fig. 14a). Second. Unit I. layer 2 occurs directly downcurrent of the boulder and rests upon the pre-eruption substrate. An erosional contact with Unit I. layer 1 occurs about 30cm downcurrent of the boulder (Fig. 14b). The edge of the feature that is facing the volcano is typically truncated, exposing both layers 1 and 2. and Unit III is absent from this side of the mound. Clasts up to 11 cm in size protrude from the surface of the truncated edge.

Granulometry Grain-size distributions were determined for 270 samples collected from 76 localities (Fig. 3). Samples were dry sieved at 1 intervals

Fig. 15. (a) Symmetrically streamlined breccia of Unit I. layer 1 in Region 3. Morris' area, locality 48 (Fig. 3). Flow direction is from top right to bottom left. Note the absence of coarse blocks in the vicinity of the feature. Foreground consists of layer 2 deposits resting directly on eroded surface (person for scale), (b) Northern end of the large streamlined breccia mound in Region 5 on the south side of the White River valley, locality 69 (Fig. 3) (helicopter on far right for scale).

SEDIMENTOLOGY OF DEPOSITS FROM THE PYROCLASTIC DENSITY CURRENT

449

Fig. 15. (continued)

to 63 m (4 ). Those samples with a fine ash (<63 m) content >15% were further analysed with a laser particle sizer to 12 . The grain sizes of coarse breccias (coarser than —5 ) were determined from photographs. Grain-size statistical parameters of Inman (1952) are used. Median diameters (Md ) of deposits range from -5 to 5 . Md versus sorting coefficients ( ) show good distinction between layers 1 and 2 of Unit I (Fig. 17). At any one locality layer 1 is always coarser than layer 2, but samples of layer 2 in axial areas can be coarser than layer 1 samples from peripheral areas, resulting in the overlap in Figure 17a and b. Unit III is finer grained than the other layers. Grain-size data from the Mount St Helens blast deposit of 18 May 1980 were chosen for comparison with 26 December 1997 samples. They are plotted with Walker's (1983) fields for pyroclastic flows and surges in Figure 18. The 26 December 1997 samples are similar to the Mount St Helens Md and data, plotting within Walker's pyroclastic surge and flow fields.

Fig. 16. Strongly reverse-graded fine to coarse breccia deposited in Region 6 on the coastal fan edge.

Mean values of the grain-size parameters Md and were calculated for layers 1 and 2 for each region (Fig. 19a). Layers 1 and 2 become finer grained, and the contrast in Md between layers 1 and 2 decreases, from axial to peripheral regions. Tie-lines connecting layers 1 and 2 for each region (Fig. 19b) show that the difference in grain size (Md and ) between layers 1 and 2 decreases between Regions 4 (axial) and 1 (peripheral). A transect 5km from the dome, traversing (from axial to peripheral) across the White River valley to Aymer's Ghaut in a NW direction, shows the variation in Md of Unit I layers 1 and 2 (Fig. 20a). This graph illustrates that within each region Md of layer 1 remains similar, and that the most pronounced change occurs across Gingoes Ghaut. Generally the grain size of layer 1 increases markedly towards the White River valley, while layer 2 shows little systematic change (Fig. 20a). Transects from vent to coast, parallel to Aymer's and Gingoes Ghauts for Regions 2 and 3 respectively, show little change in median diameter (except Region 1, which becomes

450

L. J. RITCHIE ET AL.

Fig. 17. (a) Median diameter (Md ) versus sorting coefficients ( ) for all samples with fields for Unit I, layers 1 and 2 and Unit III. (b) Median diameter (Md ) versus wt% of fine ash (<63 m) for all samples showing fields for Unit I, layers 1 and 2 and Unit I I I .

Fig. 18. Median diameter (Md ) versus sorting coefficient ( ) for all samples (Unit I, layer 1 and 2 and Unit I I I ) compared to samples from similar stratigraphic units from the 18 May 1980 blast deposit at Mount St Helens (data from Hoblitt el al. 1981). The pyroclastic flow (solid lines) and surge fields (dashed lines) are from Walker (1983).

finer grained) with distance from the dome (Fig. 20b). Generally, radial transects show less variation in grain size than transverse transects from axial to peripheral areas. Contoured Md , mean and mode maps for both layers 1 and 2 of Unit I (Fig. 21) show a distinct lobe of coarser deposit extending SW from the upper reaches of Region 4 to Region 3. In general the deposit becomes finer grained northwestwards from this lobe, as

opposed to radially from the dome, reflecting the strong azimuthal variations. Unit I, layer 1 shows an island of finer grained deposit in the western part of Region 1, around Fairfield, which is not apparent in Unit I, layer 2 (Fig. 21). Contoured Md , thickness and sorting maps for Unit III (Fig. 22) show that this layer is thickest in peripheral Region 1, where accretionary lapilli reach their maximum size (Fig. 2).

SEDIMENTOLOGY OF DEPOSITS FROM THE PYROCLASTIC DENSITY CURRENT

451

Fig. 19. (a) Averages of median diameter (Md ) and wt% fine ash (<63 m) for all samples in Regions 1, 2, 3 and 4. Tie-lines join Unit I, layer 1 and corresponding layer 2 for regions, (b) Averages of median diameter (Md ) and sorting coefficient ( ) for all samples in Regions 1, 2, 3 and 4. Tie-lines join Unit I, layer 1 and corresponding layer 2.

Impact of the current (erosional features and flow direction indicators) Features indicative of erosion are widespread across the whole region, but they are most pronounced in axial areas. The walls of the White River valley are severely scoured along their entire length, with removal of all vegetation on either side of the valley, but particularly to the SE. Region 3, especially around the St Patrick's and Morris' area, is severely eroded, with scour marks and striations in the soil and on concrete roads. A 1-3 mm thick veneer of carbonized vegetation and baked soil covered much of this area (Sparks et al. 2002), and was scoured. The mean orientation of striations in axial areas was about 220 . Directional data were collected from water pipes (10cm diameter), bent steel reinforcement bars on buildings (1.5cm diameter), and hollow steel fence posts (6cm diameter) that had been pushed over by the current (Fig. 23a). Where these structures were absent, tree blow-down directions were recorded. In addition, projectiles, such as sheets of corrugated iron up to 3 m in length that were wrapped around debris and trees, were also used. The data (Fig. 3)

Fig. 20. (a) Median grain-size diameters for layer 1 and layer 2 of Unit I with distance along a NW to SE transect parallel to the coastline, 5 km from the dome (solid line in Fig. 3). The positions of Gingoes and Germans Ghauts, two major valleys cutting the area, are shown. Layer 1 increases in median diameter rapidly, whereas layer 2 shows no systematic variation, (b) Transects constructed in Regions 1, 2 and 3 parallel to Gingoes and Germans Ghauts (dashed lines in Fig. 3). The data show little variation in median grain size of layer 1 (Unit 1) within the regions, but distinct differences between the regions, which is attributed to the effects of valleys in channelling the coarse-grained lower parts of the current.

show a pattern emanating from the dome in a general SW direction, although on a local scale there is some variation in orientation. In particular, downed fence posts bordering valleys show variability in both the orientation and degree of flattening (Fig. 23a). A line of 13 flattened fence posts with their bases c. 2.5m apart, on the northern side of Gingoes Ghaut, shows a mean orientation of 215 C while the direct line to the dome is 245°. The posts vary in orientation from 193 to 250°, a total variability of 57° (Fig. 23b).

Fig. 21. Contoured median (Md ), mode and mean maps for Unit I, layers 1 and 2 show a radial lobe of coarse material fining either side as opposed to a concentric pattern of fining with increasing distance from the dome.

SEDIMENTOLOGY OF DEPOSITS FROM THE PYROCLASTIC DENSITY CURRENT

453

Fig. 22. Contoured isopach and median diameter (Md ) map for Unit III showing a thickening away from the dome, accompanied by a reduction of median diameter. Debris avalanche deposit depicted in grey.

The effects on trees varied across the area. In peripheral areas (Region 1) larger trees were only defoliated and some trees and shrubs bent over. In Region 2 large trees (>50cm trunk diameter) were typically broken at 1-2 m height, whereas small (typically <10cm trunks) were bent over without breaking. In the axial zone (Region 3) trees (diameters up to 80cm) were broken at their base and the remaining stump severely abraded (Sparks et al. 2002).

Dramatic variations in structural damage occur between the regions. In Region 1 buildings mostly remain standing, with damage resulting from impacted projectiles and secondary burning. Roofs were blown off and windows were blown out on the downcurrent side and blown inwards on the upcurrent side. In Region 2 structures above ground level were totally removed with only basezment levels remaining intact (Sparks el al. 2002). In Regions 3 and

Fig. 23. (a) Fence posts (1.3m length) bordering Gingoes Ghaut in Region 1, locality 59, view looking NE (Fig. 3). (b) Diagram illustrating variation c directional data in the same line of fence posts adjacent to Gingoes Ghaut.

454

L. J. RITCHIE ET AL.

4 buildings were totally removed down to their foundations and the debris presumably was deposited out at sea. Some debris from these buildings was strewn on the downcurrent side of the foundations. There is evidence for at least two erosive phases during passage of the PDC. In Region 1, Unit I rests upon pre-eruption ash. whereas in Regions 3 and 4 the deposit rests upon a striated surface. Therefore the pre-existing ash and ground surface were strongly scoured by the PDC in the axial regions prior to deposition. The presence of thick (3 m), streamlined isolated mounds of coarse Unit I deposits in areas that are otherwise devoid of breccia is evidence of non-deposition and or erosion after deposition of Unit I (e.g. Fig. 15a, b). In Regions 3 and 4 especially, the deposits constituting Unit I are confined to the lee of obstacles and convex breaks of slope. Furthermore, in some localities, particularly in Region 3, layer 2 deposits are found directly upon an eroded surface with no underlying layer 1 which is confined to the lee of obstacles. This suggests that erosion occurred locally between the deposition of Unit I, layer 1 and deposition of layer 2. It is not clear whether the layer 2 here is part of Unit I or Unit II.

Discussion The explosive disintegration of the dome on 26 December (Boxing Day) 1997 is interpreted as a volcanic blast and both the deposits and the destructive effects strongly resemble those from 'blast eruptions' such as the 8 May 1902 eruption of Mont Pelee, Martinique, the 30 March 1956 eruption of Bezymianny, Kamchatka, and the 18 May 1980 eruption of Mount St Helens, USA. Grain-size parameters such as median grain size and sorting coefficients of layers 1 and 2 are similar to corresponding layers 1 and 2 from Mount St Helens (Fig. 18). We now consider what inferences can be made about the nature and dynamics of the PDC of 26 December 1997. We draw together sedimentological observations of the deposits with other information (seismic data) on the eruption from Sparks et al. (2002). We also consider the interpretation in the context of the model of explosive disintegration of a collapsing dome with a pressurized interior (Woods et al. 2002). Two key features of the PDC deposits are the substantial lateral variations across the sector inundated by the current and the bipartite divisions of the deposit, as found in other blast deposits (Hoblitt el al. 1981; Druitt 1992, 1998).

Bipartite layering of deposits Two distinct depositional units have been identified in the deposits from the PDC (Units I and II). We consider these to be the products of two separate depositional events. Each unit shows a bipartite division with a lower, typically normally graded, coarsegrained, fines-poor and moderately sorted layer (layer 1) overlain by a typically massive to stratified finer grained layer (layer 2). The origin of bipartite layers (1 and 2) with such markedly different grain sizes and mostly with a sharp contact is attributed to the dynamics of gravity currents. Wilson & Walker (1982) attributed formation of a coarse-grained, fines-poor layer 1 (also termed a ground layer) to the entrainment of air into the head of a dense pyroclastic flow. The upward movement of air in the flow head, combined with its rapid expansion, leads to strong fluidization effects, with sedimentation of coarse and dense particles to form layer 1 and with elutriation of fine particles into the turbulent wake of the current. Wilson & Walker's model was developed for the Taupo Ignimbrite, which they interpreted as deposits from a concentrated flow with the body of the flow depositing layer 2 ignimbrite over the layer 1 (ground layer). However, layer 2 in volcanic blast deposits differs in that it is a landscape-draping deposit derived from a low-concentration turbulent suspension (Druitt 1992). The presence of normal grading, which is also diagnostic of deposition from a low-particle turbulent current, is

characteristic of Units I and II. The same bipartite system is also characteristic of pyroclastic surge deposits formed by detachment from dome-collapse block-and-ash flows (Cole et al. 2002), so the development of two layers does not necessarily require a highconcentration flow. The concept of flow-head entrainment of air nevertheless remains valid for a dilute PDC. Consideration of experimental and theoretical studies of entrainment and velocity structure of gravity currents (Kneller et al. 1999) provides further pertinent information on current dynamics. Laboratory studies show that mixing at the flow head creates a turbulent and dilute wake above the body of a gravity current (Hallworth et al. 1996). The main body is denser than the wake, due to admixing of the ambient fluid to form the wake. Thus gravity currents naturally divide into two components. Kneller el al. (1999) investigated the velocity structure of gravity currents and demonstrated that they have a velocity maximum and turbulence minimum in their interior. Thus the current is divided into regions: the underlying body and overriding wake. In a PDC this division should be accentuated both by heating of entrained air. producing a thicker, more expanded wake with segregation of fine particles into it. and by any initial stratification, as postulated by Woods et al. (2002). Thus layer 1 might be attributed to the deposit of the flow body and layer 2 to the deposit of the wake. Studies of sediment gravity currents have utilized box models in which it is assumed that there is uniform vertical mixing (e.g. Bonnecaze et al. 1993; Dade & Huppert 1996). These models have also been applied to pyroclastic density currents (Dade & Huppert 1996; Bursik & Woods 1996) and have had some success in describing general attributes of sedimentary gravity current deposits in the laboratory and in large-scale natural examples. The box models can explain normal grading in deposits, but cannot yet predict the formation of two discrete layers with a sharp grain-size break. We thus surmise that the development of a turbulent dilute and fine-grained wake is a critical process in the PDC.

Erosion Analysis of the effects of erosion and destruction, combined with grain-size variations, shows that the PDC was directed SW along an axis 1.5km to the north of the White River valley. The axial areas are characterized by the most intense erosional effects, but are also where the thickest deposits occur. On a local scale, it seems that topography plays a major role in the nature of sedimentation, influences small-scale thickness variations, and accounts for rapid facies variations. The apparent thickness variations may be the result of penecontemporaneous erosion causing removal of the surrounding deposit with preservation of the deposit in sheltered areas. Another interpretation is that topographic irregularities cause strong fluctuations of the local flow conditions, with the development of turbulent eddies. Thus conditions for deposition occur adjacent to regions of non-deposition or erosion on very local scales. Fluctuations in flow conditions such as flow separation may also account for the development of the stratification that typically accompanies the thickening. One possible explanation for the erosional and depositional relationships observed is that the current involved strong waxing and waning pulses as a consequence of unsteady retrogressive failure of the dome over a 10-15 minute period. Seismic data support six pulses, with the second pulse being the strongest (Sparks et al. 2002). The stratigraphy supports two depositional events, forming Unit I and Unit II, together with at least two erosional events. Each seismic pulse may relate to specific erosional and or depositional events. In the axial zone the PDC appears to have been initially erosional then depositional followed by further erosion, with remnants of the initial deposit preserved in topographically protected areas, such as depressions and in the lee of obstacles such as houses, mounds and blocks. Within any one short-lived intense pulse, local variations in flow conditions may have allowed simultaneous deposition of breccia sheets, mounds.

SEDIMENTOLOGY OF DEPOSITS FROM THE PYROCLASTIC DENSITY CURRENT

lee-side accumulations and regions of non-deposition or erosion. In this interpretation, layer 1 was deposited as localized accumulations and bedforms in a largely erosional environment, and then draped by layer 2 in the waning phase of the pulse. Features sculpted by highly erosive currents include striations and abrasion of objects. Erosional structures similar to those documented here have been described at Mount St Helens (Kieffer & Sturtevant 1988; Fisher 1990). Of particular interest are features that appear to have been sculpted by an erosive component travelling obliquely to the principal depositional body of the current (Fig. 14a and b). Streamlined breccia deposits have long axes parallel to the inferred direction of flow, but some accumulations up- and downcurrent of boulders are truncated on their southeastern side. A similar feature that may be the result of an erosive current travelling in a different direction to the depositing current is observed in Region 5 with the development of a coarse breccia mound (shown in Fig. 15b). This structure has a long axis oriented parallel to the White River valley, with a steep, nearly vertical side facing almost perpendicular to this. The surrounding area is heavily scoured, with remnant patches of layer 2 facies deposits devoid of coarse breccia. It is envisaged that the breccia mound was deposited by a current travelling down the White River valley, and that the steep side was sculpted either by helicoidal eddies oriented subparallel to the flow direction or by parts of the PDC that had surmounted South Soufriere Hills and then flowed towards the valley.

Stratification of the PDC Understanding of the sedimentation of the PDC deposits is complicated further by the notion that the current was strongly stratified as a direct consequence of the explosive expansion of the internally pressurized dome (Woods et al. 2002) and not solely as a result of flow-head mixing processes and simple particle settling. Axial deposits indicate occurrence of strong stratification of grain size and density in the current that is so close to the source that segregation in an established current to generate the stratification seems implausible. A number of lines of evidence for a stratified initial dispersion exist within the deposits. (1) There is well developed normal grading. (2) Very coarse layer 1 deposits are capped by significantly finer grained layer 2 deposits in axial areas (such as breccia mounds capped by layer 2). (3) There is significant fining of the deposits with distance from source together with the persistence of the bipartite layering. (4) The marked changes in the grain size of layer 1 formed as the PDC expanded across the major valleys are also consistent with a strongly stratified current. The decrease in median grain size of layer 1 from SE to NW across deep valleys is consistent with the coarser grained lower parts of the current being diverted into the deep valleys and the upper, finer grained parts being less affected by them (Fig. 20). (5) Dry Ghaut contains surge-derived pyroclastic flow deposits, which merge into PDC deposits at the head of the valley in the saddle between Galway's Mountain and South Soufriere Hills. The deposits are fine-grained with few clasts greater that 1 cm (Druitt et al 2002). It was the upper parts of the PDC that spilled over the saddle about 1.5km from the dome, and these observations indicate that this part of the current was both finegrained and dilute. Coarse-grained components of the PDC were focused down the White River valley and across the axial region of dispersal (Regions 3 and 4). The PDC was able to spill out of the Galway's Soufriere area both to the north around the shoulder of Chances Peak and to the SE across the saddle between South Soufriere Hills and Galway's Mountain (Fig. 1). As the current was directed principally to the SW it would have been difficult for it to climb these elevated areas. Thus the PDC must have been hundreds of metres thick within less than a kilometre of the dome. A collapsing mass of dome rock could not produce such a thick cloud without energetic explosive expansion of the disintegrating mass. Woods et al. (2002) have developed a model for the expansion of a pressurized dome and have established that a dome with 10% porosity and pressurized gases (to a few megapascals) can easily generate such an expanded

455

cloud on timescales of a few seconds. A key feature of the model by Woods et al. (2002) is the strong density and grain-size stratification that is generated by explosive expansion. The coarse particles lag behind the fines and so the expanded mixture becomes finer grained and more dilute radially upwards and outwards. The multiple layering of Unit I that occurs only in areas bordering the valleys is interpreted as having been developed locally where the deep valleys generated billowing clouds with lobes that overlapped and so caused unsteady deposition. A similar interpretation was made by Fisher (1990), who described multiple layering in the 18 May 1980 Mount St Helens deposits. Directional data along the margins of valleys have shown significant deviations in direction over as little as 2.5m, which is consistent with the formation of turbulent vortices as material diverted by the valleys billowed out of their confines. The model of an initially stratified current due to explosive expansion (Woods et al. 2002) can explain the strong azimuthal variations of thickness, grain size, lithofacies characteristics, erosion and destruction. Radial explosive expansion creates an initial dispersion that is radially stratified in grain size and concentration. The denser and coarser grained interior then moves under gravity down the axis with some influence of the White River valley in channelling this part of the current. The higher density of this part results in more powerful, higher velocity and more erosive flows in the axial region. The upper, finer grained and more dilute parts of the current more readily spread into the peripheral areas and are less influenced by topography. They are also less energetic and erosive and produce finer grained deposits. Azimuthal variations are accentuated by capture of coarse-grained basal parts of the stratified flow by the deep valleys.

Conclusions (1)

(2)

(3)

(4)

(5)

The PDC deposits record evidence for at least two depositional events (Units I and II). Each event produced two layers, the lowermost fines-poor and the upper fines-rich. A third unit (Unit III), which caps the sequence, is a fallout layer from the plume associated with the PDC. The deposits formed range in grain size from coarse breccias to fine ash layers. Deposition was inherently patchy, particularly in axial areas, but was more continuous and uniform towards the peripheral areas. Cross-stratification becomes more pronounced from axial to peripheral regions and is locally related to topographic irregularities, indicating that depositional conditions are sensitive to the underlying substrate. The marked lateral and vertical variations in grain size and lithofacies are attributed to the explosive expansion of the collapsing lava dome, which resulted in the PDC initiating with stratification in both grain size and density. There is evidence for at least two erosional events related to the pulsatory waxing and waning of the PDC during the collapse. Erosion occurred immediately prior to deposition and also between deposition of Unit I, layer 1 and Unit II, layer 2. It accentuated the patchy nature of the deposits by sculpting bedforms tens of metres in size. Directional indicators such as flattened fence posts, bent steel reinforcement bars and damaged trees show a radial pattern from the dome, but locally there are variations. These may be due to helicoidal vortices acting within the current or to the effects of topography such as valleys.

The authors would like to thank the personnel at Montserrat Volcano Observatory for field assistance, and the pilots of St Lucia Helicopters and Bajan Helicopters for skilful access to the field. We gratefully acknowledge the Department for International Development, which provided considerable financial support for the monitoring work on Soufriere Hills Volcano. R.S.J.S. acknowledges a NERC professorship and support of NERC grant GR3/11683. We also thank P. Kokelaar and R. Hiscott for their helpful and constructive reviews and comments.

456

L. J. RITCHIE ET AL.

References ARNOTT, R. W. C. & HAND, B. M. 1989. Bedforms, primary structures and grain fabric in the presence of suspended sediment rain. Journal of Sedimentary Petrology, 59, 1062-1069. BELOUSOV, A. 1996. Deposits of the 30 March 1956 directed blast at Bezymianny volcano, Kamchatka, Russia. Bulletin of Volcanology, 57, 649-662. BONADONNA, C., MAYBERRY, G. C., CALDER, E. S. ET AL. 2002. Tephra fallout in the eruption of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 483-516. BONNECAZE, R. T., HUPPERT, H. E. & LISTER, J. R. 1993. Particle driven gravity currents. Journal of Fluid Mechanics, 250, 339-369. BOUDON, G. & LAJOIE, J. 1989. The 1902 Peleean deposits in the Fort Cemetery of St Pierre, Martinique - a model for the accumulation of turbulent nuee-ardentes. Journal of Volcanology and Geothermal Research, 38, 113-129. BOURDIER, J. L., BOUDON, G. & GOURGAUD, A. 1989. Stratigraphy of the 1902 and 1929 nuee-ardente deposits, Mt Pelee, Martinique. Journal of Volcanology and Geothermal Research, 38, 77-96. BURSIK, M. I. & WOODS, A. W. 1996. The dynamics and thermodynamics of large ash flows. Bulletin of Volcanology, 58, 175-193. COLE, P. D., CALDER, E. S., DRUITT, T. H., HOBLITT, R. P., ROBERTSON, R. E. A., SPARKS, R. S. J. & YOUNG, S. R. 1998. Pyroclastic flows generated by gravitational instability of the 1996-97 lava dome of the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters. 25, 3425-3428. COLE, P. D., CALDER, E. S., SPARKS, R. S. J. ET AL. 2002. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 231-262. DADE, W. B. & HUPPERT, H. E. 1996. Emplacement of the Taupo Ignimbrite by a dilute turbulent flow. Nature, 381, 509-512. DRUITT, T. H. 1992. Emplacement of the 18 May 1980 lateral blast deposit ENE of Mount St. Helens, Washington. Bulletin of Volcanology. 54, 554-572. DRUITT, T. H. 1998. Pyroclastic density currents. In: GILBERT, J. S. & SPARKS, R. S. J. (eds) The Physics of Explosive Volcanic Eruptions. Geological Society, London, Special Publications, 145, 145-182. DRUITT, T. H., CALDER, E. S., COLE, P. D. ET AL. 2002a. Small-volume, highly mobile pyroclastic flows formed by rapid sedimentation from pyroclastic surges at Soufriere Hills Volcano, Montserrat: an important volcanic hazard. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 263-279. DRUITT, T. H., YOUNG, S. R., BAPTIE, B. ET AL. 2002b. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306. FISHER, R. V. 1990. Transport and deposition of a pyroclastic surge across an area of high relief: The 18 May 1980 eruption of Mount St. Helens, Washington. Geological Society of America Bulletin, 102, 1038-1054. FISHER, R. V. & HEIKEN, G. 1982. Mt Pelee, Martinique: May 8 and 20 1902, Pyroclastic flows and surges. Journal of Volcanology and Geothermal Research, 13, 339-371. FISHER, R. V., SMITH, A. L. & ROOBOL, J. M. 1980. Destruction of St Pierre, Martinique, by ash-cloud surges, May 8 and 20 1902. Geology, 8, 472-476.

FISHER, R. V., GLICKEN. H., & HOBLITT. R. P. 1987. May 18 1980, Mount St. Helens Deposits in South Coldwater Creek, Washington. Journal of Volcanologv and Geothermal Research, 92. 10267-10283. HALLWORTH, M. A.. HUPPERT.. H. E.. PHILLIPS. J. C. & SPARKS. R. S. J. 1996. Entrainment into 2-dimensional and axisymmetric turbulent gravity currents. Journal of Fluid Mechanics. 308, 289-311. HOBLITT, R. P., MILLER, C. D. & VALLANCE. J. W. 1981. Origin and stratigraphy of the deposit produced by the May 18 directed blast. In: LIPMAN. P. W. & MULLINEAUX. D. R. (eds) The 1980 Eruptions of Mt St Helens, Washington. US Geological Survey. Professional Papers, 1250. 401-419. INMAN, D. L. 1952. Measures for describing the size distribution of sediments. Journal of Sedimentary Petrology. 22. 125-145. KIEFFER, S. W. & STURTEVANT, B. 1988. Erosional furrows formed during the lateral blast at Mount St. Helens. May 18 1980. Journal of Geophysical Research. 93, 14793-14816. KNELLER. B. C.. BENNETT. S. I. & MCCAFFREY. W. D. 1999. Velocity structure, turbulence and fluid stresses in experimental gravity currents. Journal of Geophysical Research. 104. 5381-5391. LACROIX. A. 1904. La Montague Pelee et ses Eruptions. Masson et Cie. Paris. MIDDLETON. G. V. 1966. Small scale models of turbidity currents and the criterion for auto-suspension. Journal of Sedimentary Petrology. 36. 202-222. MIDDLETON. G. V. 1967. Experiments on density and turbidity currents. I I I . Deposition of sediment. Canadian Journal of Earth Sciences. 4. 475-505. ROBERTSON. R. E. A.. ASPINALL, W. P.. HERD, R. A.. NORTON. G. E.. SPARKS. R. S. J. & YOUNG. S. R. 2000. The 1995-1998 eruption of the Soufriere Hills Volcano. Montserrat. WI. Philosophical Transactions of the Royal Society of London. 358. 1619-1637. SPARKS. R. S. J.. SELF. S. & WALKER. G. P. L. 1973. The products of ignimbrite eruptions. Geology. 1. 115-118. SPARKS, R. S. J., YOUNG. S. R.. BARCLAY. J. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills Volcano, Montserrat: November 1995 to December 1997. Geophysical Research Letters. 25. 3421-3424. SPARKS. R. S. J.. BARCLAY. J.. CALDER. E. S. ET AL. 2002. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs, 21. 409-434. VOIGHT. B., KOMOROWSKI. J.-C. NORTON. G. E. ET AL. 2002. The 26 December (Boxing Day) 1997 sector collapse and debris avalanche at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21, 363-407. WALKER, G. P. L. 1983. Ignimbrite types and ignimbrite problems. Journal of Volcanology and Geothermal Research. 17. 65-88. WILSON. C. J. N. & WALKER. G. P. L. 1982. Ignimbrite depositional facies - The anatomy of a pyroclastic flow. Journal of the Geological Society. 139, 581-592. WOODS. A. W., SPARKS, R. S. J.. RITCHIE. L. J.. BATEY. J.. GLADSTONE. C. & BURSIK. M. 2002. The explosive decompression of a pressurized lava dome: the 26 December 1997 collapse and explosion of Soufriere Hills Volcano, Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London. Memoirs. 21. 457-466. YOUNG. S. R.. SPARKS, R. S. J.. ASPINALL. W. A.. LYNCH. L.. MILLER. A. D.. ROBERTSON. R. E. A. & SHEPHERD. J. 1998. Overview of the eruption of Soufriere Hills volcano. Montserrat. 18 July 1995 to December 1997. Geophysical Research Letters. 25. 3389-3392.

The explosive decompression of a pressurized volcanic dome: the 26 December 1997 collapse and explosion of Soufriere Hills Volcano, Montserrat A. W. WOODS1, R. S. J. SPARKS2, L. J. RITCHIE3, J. BATEY2, C. GLADSTONE1 & M. I. BURSIK 4 1

2

BP Institute, Madingley Rise, University of Cambridge, Cambridge CB3 OEZ, UK (e-mail: [email protected]) Centre for Environmental and Geophysical Flows, Department of Earth Sciences, University of Bristol, Bristol BS8 1RJ, UK 3 Centre for Volcanic Studies, Luton University, Luton LU1 3JU, UK 4 Geology Department, SUNY, New York 14260, USA

Abstract: We calculate the gas flux through a pressurized and permeable lava dome as magma slowly ascends to the surface and degasses. This provides an estimate of the mass of gas and the associated energy that may be released following the explosive collapse of such a dome. We estimate that during the 26 December 1997 dome failure of Soufriere Hills Volcano, up to 1016 J of energy was released. We then develop a model of the dome break-up driven by the pressure jump across a fragmentation wave. We predict that the fragmentation front advances into the dome at a speed of order l 0 m s - 1 , while the gas and solid fragments issue from the dome at speeds of order 50-60 m s - 1 . This fragmented material forms a dense fountain of gas and solid fragments. This may rise 100-200 m above the dome and then spread out to form a dense pyroclastic density current, with initial density of order a few hundred kilograms per cubic metre. We anticipate that such flows will become stratified in density and mean grain size. Some new experiments on stratified gravity currents illustrate that the flow propagation speed is similar to a well mixed current of the same total buoyancy, although the flow tends to become laterally stratified, with the dense lower region out-running the fluid higher in the current.

Amongst the most hazardous kinds of volcanic phenomena are the high energy pyroclastic density currents associated with large-scale collapse of andesitic and silicic lava domes. In the past literature, such currents have been variously called nuees ardentes, volcanic blasts and ash hurricanes. Examples include the 1902 nuees ardentes of Mont Pelee, Martinique, which destroyed St Pierre (Lacroix 1904), the ash hurricane of the 1952 eruption of Mount Lamington, Papua New Guinea (Taylor 1958), and the volcanic blast of Mount St Helens in 1980 (Hoblitt el al 1981). The latest example of the phenomenon occurred on 26 December 1997 at Soufriere Hills Volcano, Montserrat, when the andesitic lava dome was disrupted by failure of an upper flank of the volcano (Sparks et al. 2002). The initial conditions in these violent and dangerous volcanic events are not well established. In the case of Mount St Helens, failure of the northern flanks of the volcano involved a gas-pressurized cryptodome and hydrothermal system that clearly exploded simultaneously with collapse and disintegration of the rock mass. In the case of Soufriere Hills a high energy pyroclastic density current devastated 10km 2 of southern Montserrat (Sparks et al. 2002). Velocities in the flow are estimated to have been 80-100ms- 1 and the current was strongly erosive over large areas. The current was initiated by disruption of about 5 x 10 7 m 3 of the andesite dome. The juvenile components in the deposits consist of poorly vesicular dense clasts and the disruption did not uncover the conduit. Earlier in the eruption, two series of cyclic Vulcanian explosions, producing pumiceous ejecta, occurred when the conduit was uncovered by major dome collapses (Druitt et al. 2002). These observations indicate that explanation of the energetics of the current are to be found in the process of disruption of a large pressurized dome. The purpose of this work is to explore the dynamics of such an event. Our study is based on the hypothesis that the dome was internally pressurized, as has been previously postulated to explain various phenomena in dome-generated pyroclastic flows (Sato et al. 1992; Fink & Kieffer 1993; Alidibirov 1994). As a dome grows, it can develop internal gas pressures far in excess of atmospheric and well in excess of the static pressure associated with the dome weight (Sparks 1997; Melnik & Sparks 2002). As volatiles are exsolved from the ascending magma, and the resulting gas phase separates from the magma by rising through the magma, gas pressures in the dome can be very high. If the dome subsequently collapses and fragments, this gas will be released, together with any fine-grained material generated during the collapse process. Observations at Montserrat (Sparks et al. 2002) suggest that when sections of the

dome, of typical dimension 200m, fail, the resulting decompression of gas can raise paniculate material hundreds of metres above the dome, thus providing a source for subsequent pyroclastic density currents (Fig. 1). As in other deposits from high energy pyroclastic density currents (Hoblitt et al. 1981), the deposits of the 26 December 1997 current show marked lateral and vertical facies variations (Ritchie et al. 2002; Fig. 2). Particles are strongly sorted and segregated into a coarser-grained lower layer, a medium- to fine-grained upper layer and a very-fine-grained ash layer. Furthermore, there are marked lateral (flow-transverse) grain size variations across the region inundated by the current (Fig. 2). Although some of these sorting

Fig. 1. Schematic of the processes involved in explosive dome collapse and the subsequent formation of a gravity-driven pyroclastic density current.

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 457-465. 0435-4052/02/$15 C The Geological Society of London 2002.

457

458

A. W. WOODS ET AL.

Fig. 2. Map of the distribution of the pyroclastic density current deposits of 26 December 1997 on Montserrat. showing contours of median grain size and profiles of grain size in the coarse lower layer and the fine upper layer. The finer deposits are spread widely whereas coarse deposits are confined to the central axis of dispersal. Note the large jumps in grain size which mark changes in grain size in the lower coarse layer across major valleys: the fine upper layer is much less influenced by such topography.

features may result from transport processes in a well-mixed current (Dade & Huppert 1996; Bursik & Woods 1996; Druitt 1998), here we suggest that the flow may become strongly stratified in density and particle size as the flow develops from the fountain (Valentine 1987). First, we develop a model of the propagation of a fragmentation wave as it moves into a pressurized, disrupted lava dome. We start by developing a model of the pressure profile through a steadily degassing lava dome and we estimate the energy associated with the compressed gas, available to drive the expansion as the dome rock disintegrates following collapse. Using this picture of the precollapse dome, we then examine the propagation of a fragmentation wave into a collapsed sector of the dome. We calculate the speed of the wave and of the mixture as it rises from the dome. The decompressed gas and particulate material form a dense fountain, which collapses from above the dome to form a pyroclastic density current. We show that, owing to differential sedimentation speeds, this current is likely to become strongly stratified in density and grain size as it propagates laterally. We present some preliminary experimental data that illustrate some of the important dynamical features of such stratified gravity currents. Throughout the model development, we focus on the implications of the work for the 26 December 1997 dome collapse of Soufriere Hills Volcano, but in the concluding discussion we examine its application to other historical lateral blast eruptions, including the 18 May 1980 eruption of Mount St Helens, USA.

Model of dome pressurization Observations of numerous eruptions that form lava domes indicate that there are extended periods of effusive dome growth, with intermittent short-lived violent activity. During the effusive phase, the dome builds up with degassed magma, while the exsolved volatiles separate from the magma and travel through the permeable dome to the atmosphere (Eichelberger el al. 1986; Melnik & Sparks 1999). There is growing evidence both from models of dome growth and field observations that considerable overpressures develop in such lava domes (Jaupart 1998; Sato el al 1992; Fink & Kieffer 1993; Sparks 1997; Melnik & Sparks 1999). In the uppermost parts of volcanic conduits, gas exsolution increases viscosity by orders of magnitude and triggers crystallization, which can also increase pressure. Models of the coupling between gas escape and viscous conduit flow indicate that overpressures of several megapascals can develop at the conduit exit, where overpressure is defined as the difference between pressure and local lithostatic pressure (Melnik & Sparks 2002). Such overpressures at the base of an actively growing dome are expected to be much larger than the static pressure due to dome weight and this overpressure is dissipated by escape of gas through the porous structure of the dome. There is good evidence for high gas pressures within the dome of Soufriere Hills and for a link between the degassing process and the triggering of dome collapse (Voight el al. 1999). Long-period earthquakes were occasionally observed to be linked to escape of gas jets from the dome surface and

EXPLOSIVE DECOMPRESSION OF PRESSURIZED VOLCANIC DOME

459

sometimes preceded rockfalls. Luckett et al. (2002) have shown that most rockfalls have long-period components that support a causative involvement of degassing in triggering collapses. Many of the major dome collapses on Montserrat occurred at the end of a hybrid seismic swarm and at the peak of inflation recorded on tiltmeters (Voight et al. 1999). These relationships can be interpreted as caused by gas pressure build-up in the upper conduit and interior of the dome, leading to a pulse of growth and instability. If the dome is modelled as a sector of a sphere, of solid angle , then the cross-sectional area available for the flow at a radius r is A(r) = r2. For quasi-steady flow, the effusion rate Q is related to the gas flux, Qg, according to Qg = nQ where n is the exsolved gas mass fraction. If the dome has permeability K then the mass flux of gas through the dome is:

where p is the pressure, is the vapour viscosity, and p is the gas density. For typical pressures and temperatures of the dome, the gas density may be simplified to the approximate relation (Tait et al. 1989):

where R is the gas constant, which has value 462 Jkg - 1 K-1 for water vapour. Here T is the temperature, which we assume has a value of 850°C, although the calculations are relatively insensitive to this parameter given the other uncertainties in the modelling. If a quasi-steady flow regime becomes established, then the gas supplied with the continuing flow of magma will migrate through the dome and vent at the surface. For such a steady flow, the gas pressure on the outer surface of the dome will be equal to or larger than atmospheric. In order to find the minimum mass of gas stored in the dome in such a steady flow regime, we apply the boundary condition that at the outer surface of the dome, r = ra say, the pressure equals atmospheric, pa. It is this pressurized gas in the dome interior that will drive fragmentation and expansion of the dome material in association with collapse. For a steady flux of gas, Qg, Equation 1 then implies that: l/2

p(r)=p(ra)

+

(3)

where RT p(ra)2K ra

(4)

The magnitude of the parameter determines the gas distribution within the dome, and we now describe the various constraints on the value of (3. Although some of these parameters are difficult to determine from field evidence, measurements on samples of dome rock indicate that the dome material has a matrix with permeability in the range 10 -14 to 10 -12 m2 if pervasive fractures in the dome are also considered (Melnik & Sparks 2002). The bulk volatile content of the magma has been estimated to be of order 2 wt% (Barclay et al. 1998) and at the time of the dome failure on 26 December 1997, the dome extrusion rate had a value of about 8 m3 s-1 (Sparks et al. 1998). Together, this implies that there is an associated gas flux of about 400 kg s - 1 . An independent source of data on the volatile flux is available from correlation spectroscopy (COSPEC) data measured during the eruption. These data show fluxes of SO2 mostly in the range 5-15 kg s-1 (Young et al. 1998). If the S content of the magma is in the range 100-1000 ppm then the water flux will be 200-1000 times greater than SO2. Thus the associated flux of water vapour is estimated to be hundreds to thousands of kilograms per second, in accord with the above independent estimate. Using the above parameter values, we may determine , and then Equation 3 provides a relationship between the gas flux, Qg, and the pressure at the top of the conduit, P(ro), where r0 is the radius of the conduit and ra is the radius of the dome surface. Figure 3 illus-

Fig. 3. Pressure at the centre of a permeable dome as a function of the gas flux propagating through the dome. In this calculation, the dome is taken to be a sector of a sphere with solid angle 2 and curves are shown for dome permeability 10 - 1 2 , 10 - 1 3 and l0 -14 m 2 .

trates the relationship between the pressure at the centre of the dome, just above the conduit, P(r 0 ), and the total gas flux issuing from the dome, Qg, for the range of parameter values thought to be typical in the Montserrat dome during the latter part of 1997. In these calculations, we have assumed that the gas is predominantly water vapour, that the initial temperature is 850°C and that the solid angle associated with the dome sector is 2. The vapour viscosity is taken to be 2 x 10 -5 Pa s. By comparison with the above field estimates of the gas flux, we estimate that the pressure at the centre of the dome was of order 5-10MPa prior to the dome collapse. These overpressures are consistent with recent models of conduit flow, in which it has been shown that the effects of volatile exsolution and the associated microlite crystallization in the ascending magma lead to large magma viscosity and hence pressure in the upper part of the conduit (Sparks 1997; Melnik & Sparks 1999). As expected, as the overpressure increases, the model predicts an increase in flux, and for a given overpressure, the model predicts greater fluxes in domes of greater permeability. Figure 4 illustrates the variation of the pressure with radius in the dome for three representative values of the permeability, in the case of a 300m radius dome, with a dome extrusion rate of 8 m 3 s - 1 . It is seen that the pressure remains at very high values, > 106Pa

Fig. 4. Pressure as a function of radius in a permeable dome of radius 10 < r < 300 m. Curves are given for dome permeabilities of value K= 10 -12 ,10 -13 and 10 -14 m2.

A. W. WOODS ET AL.

460

throughout the dome, except in a narrow boundary layer, of width 5-10 m, near to the surface, where it decreases towards atmospheric pressure. As a result of this pressure distribution, if there is a collapse or landslide, regions of very high pressure just below the outer part of the dome may be exposed to much lower pressures. This may trigger a fragmentation wave as these large local pressure gradients can disrupt and fragment the dome rock. We will explore this process in more detail later in the paper. Note also that the pressure near the centre of the dome increases rapidly as the radius decreases; the precise value depends on the radius at the top of the conduit, but typically has a value 5-10 MPa. We now examine the gas mass and associated energy stored within the model dome in order to develop some insight into the scale of the phenomenon. Gas and energy available for dome fragmentation We now use the above model of the gas pressurization in the dome to determine its energy and gas content, and hence determine the potential for explosive fragmentation and disintegration following a trigger such as a landslide or collapse. The gas mass in the dome, G is given by the integral: G=

pr2

dr

(5)

Here, we take the porosity of the dome to be = 0.1, based on measurements presented in Robertson et al. (1998) and Melnik & Sparks (2002). The density p may be expressed in terms of the pressure, using the radial profiles calculated above (Fig. 4). For the three examples shown in Figure 4, the total gas content of the dome is estimated to be 1.7 x 107, 5.2 x 107 and 1.6 x 10 8 kg respectively, for the three different cases in which the permeability has values 10 - 1 2 , 10 - 1 3 and 10 - 1 4 m 2 . In each of these cases, the mass of solid magma in the dome is 1.3 x 10 11 kg, and so the net mass fraction of gas in the dome ranges from about 1.3 x 10 - 4 to 1.2 x 10 -3 . This is a key result in terms of the subsequent dynamics of the fragmented mixture, as we shall establish later in the paper. Such pressurized domes have one or two orders of magnitude less exsolved volatile per unit mass of magma than is typically present in a steady Plinian explosive eruption, and this can lead to a substantially greater the density of the pyroclastic gravity currents produced. The energy E that is available to fragment the dome materials and subsequently to disperse the rock fragments following disintegration of the dome, may be expressed in terms of the thermal energy of the gas together with that fraction of the solid dome material which remains in good thermal contact with the gas. For the fragmentation wave, the mass fraction of solid remaining in good thermal contact with the gas is small owing to the short time available for solid-gas heat transfer during the passage of the fragmentation wave. However, the flow that subsequently develops following fragmentation evolves over times of order tens of seconds. A much more substantial mass of fine-particulate material, corresponding to particles smaller than about l00 m, can remain in thermal equilibrium with the gas during the flow (Woods & Bursik 1991). Since this quantity is poorly constrained by field observations, we take it to be a parameter in the model. The internal energy of the gas stored in the dome may be expressed as

E=

pc Tr2 dr

(6)

where pc T is the energy per unit volume. In this calculation the mass of solid material that remains in thermal equilibrium with the gas during the explosion is included through the definition of the bulk specific heat (Wilson et al. 1978; Woods 1988): c =fcp + cvw

(7)

where cp denotes specific heat of the particulate, cvw is the specific heat at constant volume of the water vapour, and is the mass of

solid per unit mass of gas. Equation 6 may be rewritten in the simpler form: E=

dr

(8)

J

where 7, the ratio of specific heats at constant volume and constant pressure, depends on the mass fraction of the solid, , which remains in thermal equilibrium with unit mass of the gas: 7 = (fcp + cpw)(fcp + cvw)

(9)

Here cpw is the specific heat at constant pressure of the water vapour. For simplicity, in these calculations, we assume that 7 is constant, consistent with our approximation that the water vapour behaves as a perfect gas (Equation 2): this simplification is justified because any differences in the relevant properties of water vapour within the dome have only a small effect compared to the uncertainty in the value of / (Equation 9). For the evolving decompressed flow, after passage of the fragmentation wave, values of may lie in the range 104-102, depending on the number of fine fragments in the flow. However, later in the paper, we show that during the passage of the fragmentation wave, a much smaller mass of solid is in contact with the decompressing gas. We therefore examine values of in the range 1
For the case of a dome of permeability 1 0 - 1 2 m 2 , as shown in Figure 4, the energy available for the flow may therefore vary from about 2.5 x 1013 to 2.5 x 1016 J. depending on the grain size of the fragments and the timescale of the flow. The lower value corresponds to the energy primarily associated with the pressurized gas, with very little heat transfer from the hot solid. The larger value corresponds to thermal equilibrium between a very large mass of solid material and the gas. as might arise in the subsequent flow. In that case, the solid can buffer the temperature of the gas through heat transfer, leading to a much more energetic phenomenon. It follows from the above calculations of gas content in the dome that, with a larger permeability, the mass of stored gas is less and hence the energy available for fragmentation is less. The fragmentation wave We now examine the decompression and disintegration of the dome following a collapse that disrupts the surface of the dome, exposes regions of high pressure to the atmosphere (Fig. 4) and triggers the explosive decompression. A number of calculations describing certain features of explosive eruptions involving fragmentation have previously been presented in the literature. Kieffer (1981) described the lateral blast at Mount St Helens on 18 May 1980 by adapting a model for the steady-state expansion of a high-pressure jet issuing from a jet engine. Turcotte et al. (1990) and Woods (1995) developed one-dimensional shock-tube models of a transient explosion confined within a conduit. There are also a number of time-dependent numerical simulations of the initial stages of flow in an explosive eruption (Neri & Dobran 1994; Valentine & Wohletz 1989). However, such models do not capture the initial fragmentation process, and it is this phenomenon which we now examine. The fragmentation process follows the collapse of a sector of the dome and involves the break-up of solid material under a large local differential pressure gradient as the fragmentation wave propagates through the mixture (Bennet 1974: Alidibirov 1994; Fig. 5).

EXPLOSIVE DECOMPRESSION OF PRESSURIZED VOLCANIC DOME

461

that is able to stay in thermal equilibrium with the gas is of order dt/d ~ 10 - 3 to 10 - 2 . This may be comparable to or greater than the mass of gas (Equation 5), and results in an additional source of thermal energy being available to drive the fragmentation process. The assumption of dynamic equilibrium between the gas and the solid fragments introduced in the previous models is a considerable simplification of the process in that the smaller particles, with less inertia, are likely to be swept along by the expanding gas phase and pass between the larger fragments. This may lead to some vertical stratification of the ensuing flow, and we discuss this later in the paper. However, as a first model of the fragmentation, we assume that the gas and solid fragments accelerate together as a mixture across the fragmentation front, reaching speed u. We also assume that the fragmentation front advances into the collapsed sector of the dome with speed v. We denote the initial volume fraction occupied by the gas to be 0. We assume that after disintegration, the pressure of the disrupted material falls to atmospheric and we denote the volume fraction occupied by the gas by 1. Mass conservation across the fragmentation surface may then be written as two coupled equations. First, for the gas (Alidibirov 1994):

v)p1

v oA) = ] (u +

(11)

where denotes the initial density of the gas. Mass conservation for the solid fragments requires: v(l -

Fig. 5. Schematic of the fragmentation wave advancing into the dome as material decompresses and expands. The control volume illustrated is used in the model of Equations 11 to 14.

The critical differential pressure required to fragment the dome depends on the detailed structure and planes of weakness in the dome. Experimental studies (Alibidirov & Dingwell 1996) indicate that natural geological materials fragment at differential pressures of a few megapascals. It is likely that there will be a wide range of differential pressures that lead to failure of different parts of the dome. However, the maximum degree of fragmentation occurs when the pressure falls to atmospheric across the fragmentation wave. The heterogeneity of the dome may lead to dispersion of the fragmentation wave as the weaker parts of the collapsed dome fail first, when subjected to relatively small pressure differentials. Such complexity is beyond the scope of the present work and, in our model, we examine the dynamics of the end-member fragmentation wave, in which the pressure falls to atmospheric across the wave. This limiting case will be most representative for the relatively slow fragmentation process in which the decompressed material ahead of the fragmentation front adjusts to atmospheric pressure. We develop a simplified quantitative picture of the fragmentation wave as it advances into the collapsed part of the dome (Figs 1 and 5). We draw from earlier models of fragmentation waves (Alidibirov 1994), but include some important developments for the present application. In those models it was assumed that the gasparticle mixture remains in dynamic equilibrium, but that the gas and solids are thermodynamically isolated with the gas expanding adiabatically across the fragmentation front. It is this expansion and cooling of the gas that provides the energy for the fragmentation and initial impulsive acceleration of the solid material (Equations 8 and 10). In the dome-collapse events of interest herein, there is a considerable range of particle sizes, and some gas-solid heat transfer is expected. Indeed, for a fragment of diameter d, and a fragmentation wave of speed v, the time for the wave to pass the fragment is d/v, and in this time a region of thickness d1 = d2 / K on the outer surface of the solid fragments will cool. Although this may represent a very small fraction of the total solid mass, it may be comparable to the much smaller mass of gas. For example, for a fragmentation wave of speed 10 m s - 1 , a mean grain size of 10 -4 to 10 - 3 m, and thermal boundary layer thickness, dt ~ 10 -7 m, the fraction of the solid mass

) = (u + v)(1 -

(12)

)

The fragmentation process is driven by the expansion of the gas following failure of the solid matrix confining the gas. If a mass f of solid material is in local thermodynamic equilibrium with unit mass of gas, then the effective ratio of specific heats, 7, is given by Equation 9. In this case, the expanding gas satisifies the modified relation: )

P /P = ( /

(13)

for quasi-adiabatic flow, where 7 accounts for the heat transfer between the solid and gas. The overall energy balance across the front has the form: v

(cvv + c s )(To - T1) =

u + v(

+ (1 -

)u2/2

(14)

where the left-hand side denotes the heat released by the expanding gas, the first term on the right-hand side denotes the pressure work on displacing the air ahead of the fragmenting material, and the second term denotes the kinetic energy of the mixture of gas and solid fragments. Note that the first term on the right-hand side of Equation 14 accounts for the work done by both the gas and solid fragments. In Equation 14, the expressions Cvy and cs denote the specific heat of the vapour at constant volume, 1400 J k g - 1 K-1 and the specific heat of the solid fragments, l 0 0 0 J k g - 1 K - 1 . After some algebra, this model may be simplified to an expression for v2 and hence an expression for the speed of (i) the fragmentation wave as it moves into the collapsed dome material, and (ii) the ascending fountain of dense particle-laden gas. Figure 6 illustrates the variation of the speed of these two quantities as a function of the initial dome pressure, assuming that the mixture decompresses to atmospheric pressure. The dome is assumed to have an initial porosity of 0.1, and curves are shown for 7 = 1.33,1.16 and 1.033 corresponding to (a) negligible heat content from the solid, (b) a comparable amount of heat from the solid and the gas and (c) about ten times more heat from the solid than from the gas. Figure 6 illustrates that the speed of the fragmentation front decreases as the dome pressure increases, while the speed of the decompressed gas and particle mixture increases with dome pressure. For small pressure drops, the change in density of the mixture of gas and fine particles is small, and therefore the mixture speed is quite small. As a result, the fragmentation front can propagate rapidly into the dome without generating too much momentum in the decompressed material. For larger pressure drops, however, the mixture speed increases owing to the greater mass of gas. To limit the total momentum production, the rate of decompression of the dome thus decreases.

462

A. W. WOODS ET AL.

Fig. 6. Speed of the fragmentation wave that propagates into the dome (solid line), and speed of the expanding mixture of decompressed ash and solid particles produced from the dome (dashed line) as a function of the pressure at a radius of 10m from the centre of the dome. Curves are shown for values 7 = 1.033, 1.16 and 1.33, corresponding to a range of different solid masses in thermal equilibrium with the expanding gas.

model, we have adapted the model of Bursik & Woods (1996). as described in Sparks et al. (2002). Field evidence indicates that the flow was largely erosive, that they travelled beyond the coast, about 4km from the dome, and that there was relatively little net sediment deposition. The highly erosive nature of the current indicates high shear stress at the base associated with the high speed of the flow. We have therefore modified the model by assuming that particles remain suspended in the flow. This leads to predictions that are consistent with the current reaching the shoreline, about 4 km from the dome (Sparks et al. 2002). More detailed analyses of the deposits and the flow dynamics, however, lead to a somewhat more complex picture of the flow dynamics and particle sedimentation. Valentine (1987), following work of Middleton & Southard (1978), has shown that the ability of the current to maintain particles in turbulent suspension depends on the Rouse number, /?„, defined as the ratio of the particle fall speed and the frictional or shear velocity of the current. Rn = n'//3v*, where 3 = 0.4, v* is related to the mean current speed by the relation v* ~ 3vj\n(3QB!R) where B/R is the ratio of the boundary layer thickness of the flow and the surface roughness length, taken to have a value of about five (Valentine 1987). and v is the mean flow speed. Here, u* denotes the fall speed of the particles, given by the quadratic drag law: ,= (^Y :

W /

Except for small dome pressures, the velocity of the fragmentation wave varies relatively weakly with the initial dome pressure, in the range of about lO^Oms" 1 for the parameters used in Figure 6. As the initial dome pressure increases, the speed of the gas and particles does increase significantly owing to the decompression of a much greater mass of the gas phase. Figure 6 also illustrates how the speed of the gas-particle mixture increases and that of the fragmentation front decreases as more heat is supplied from the solid to the gas (compare 7 = 1.333 and 1.0333). Figure 6 suggests that the fragmentation wave would require about 20-30 s to advance into a collapsed sector of a dome of size 250-300 m, and that the gas and particles would emerge with speeds of order 50-60 ms" 1 . The time of rise of such a dense particle-laden jet is only of order 5-6 s before it reaches its maximum height, and then generates a laterally spreading flow. As a result, the fountaining process may become quasi-steady as the failed sector of the dome disintegrates.

(.5)

where a is the particle density, g the acceleration due to gravity, d the clast diameter, p the bulk density of the flow and c the drag coefficient. In using Equation 15, we have assumed that all the finer-grained particles that are suspended in the flow contribute to the drag experienced by the particles which are just suspended by the flow. This is valid if the majority of the particles are much smaller than the critical size for turbulent suspension, and provides an end-member model of the process. If, for a given particle size, Rn > 2.5, then such particles cannot be suspended in the flow. In contrast, for Rfl < 2.5, particles are just suspended in the flow, but tend to be concentrated at the base of the flow. Only much smaller particles, for which Rn < 1, become well-mixed throughout the flow. Figure 7 illustrates the variation of Rn with current speed in a current of density 100kgm~ 3 , for three different particle sizes. It is seen that for flows with speeds of order 60-lOOms" 1 (Sparks et al. 2002), particles of size 0.01-0.1 m are only just suspended in the flow. Such particles will therefore tend to become concentrated at the base of the flow. Only those particles that are much smaller

Particle transport and segregation The mixture of fine-grained ash and gas rises from the fragmenting dome as a negatively buoyant fountain, decelerating under gravity. The initial conditions are given by the fragmentation wave model. As seen above, the model predicts that for internal pressures at the base of the dome of several megapascals, the speed of the mixture is about 60ms" 1 while the void fraction of the flow is about 70-80%, and this implies an initial flow density of the order of 100-500 kgm~ 3 . The motion of this negatively buoyant fountain may be described in terms of the gas thrust region of eruption column models (Woods 1988; Sparks et al 1997). Since the mixture is much denser than the air, it rapidly decelerates to rest under gravity, attaining a maximum height of a few hundred metres. Material then spreads and sinks back to the ground to form a radially spreading gravity current. For a flow issuing from a region of radius 100-150m, with a flow speed of 60 ms" 1 , the volume flux will be about 2-4 x 10 6 m 3 s -1 . With a mixture density of a few hundred kilograms per cubic metre, the mass flux will be of order 2-6 x 108 kgs" 1 . This is consistent with the independent estimate of the mass flux derived by comparison of observations of the pyroclastic density current with the predictions of a well-mixed model of the flow (Sparks et al. 2002). In order to follow the motion of such a pyroclastic density current produced from the collapsing material, as a leading order

Fig. 7. Variation of the Rouse number. Rn. as a function of the flow speed and particle size, in a flow of bulk density lOOkgirr 3 , with clasts of density 2500 kg m~ 3 . Calculations are shown for particles of size 0.001. 0.01 and 0.1 m. Flow speeds for the 26 December event have been estimated to have values of the order 80-100ms~' from field observations and flow calculations (Sparks et al. 2002).

EXPLOSIVE DECOMPRESSION OF PRESSURIZED VOLCANIC DOME

463

Fig. 8. (a) Schematic of the laboratory apparatus used in the analogue laboratory experiments. (b) Photograph of a laboratory experiment in which a stratified gravity current propagates along the base of a flume. The current is produced from a two-layer source with the upper to lower layer buoyancy ratio being four. The current becomes stratified laterally as the lower layer runs ahead of the upper layer.

than 0.001 m will be well-mixed in the flow. Particles with size in excess of about 0.1 m cannot be kept in suspension and will either be transported in traction or be deposited. This prediction is consistent with the observation that the deposits in the regions affected by the main pyroclastic density current of 26 December 1997 are coarse breccia mounds and sheets (Ritchie et al. 2002). The vertical stratification of the current in both density and particle size, which results from the concentration of intermediate sized particles at the base of the flow, is likely to develop rapidly as the material in the collapsing fountain spreads laterally. This is because the fall speed of these intermediate sized particles through the flow is comparable to the flow speed (Fig. 7). The stratification may also be enhanced by any particle size sorting produced during the disruption of the dome by the fragmentation wave. In particular, as the mixture initially decompresses and expands, the smaller clasts may be carried upwards around the larger clasts, producing a more dilute and finer grained region at the top of the fountain. Although the well-mixed model may provide a reasonable leading order description of the flow, and, in particular, the predictions of flow speed and dynamic pressure are in broad accord with field observations (Sparks et al. 2002), the vertical stratification is likely to have a substantial impact on the flow structure and particle deposition patterns (Valentine 1987). In the next section we describe some preliminary experimental results of these effects.

water (Fig. 8a). The gravity current was produced from a finite source of fluid initially placed behind a lock gate. The source fluid consisted of two superposed layers of equal volume but different salinity and hence density. Figure 8b shows a photograph of the motion of the evolving current, wherein the lower layer salinity was four times that of the upper layer. The lower layer initially propagated much more rapidly, running ahead of the upper layer and thereby forming a laterally segregated flow. Indeed, the current appears to separate horizontally, with the denser fluid outrunning the less dense layer. The detailed structure of the flow depends on the buoyancy and volume contrast between the two layers. Indeed, experiments suggest that with a small buoyancy difference between the two layers, the two tend to mix as they propagate; however, for larger buoyancy contrasts, with buoyancy ratio in excess of a factor of about two to three, two laterally distinct currents tend to form, as shown in Figure 8. Analysis of the speed of these experimental currents, in comparison to the well-mixed currents, is of interest. Figure 9 illustrates

Experiments modelling a stratified density current As indicated above, we expect the flow to become vertically stratified, with a dense underflow overlain by a more dilute cloud of fine particles and gas. Such vertical stratification in particle size and density may impose some important effects on the propagation and sedimentation from the flow, with the denser lower part of the flow running ahead of the more dilute upper portion, and being much more constrained by topography. In order to demonstrate the potential effect of such stratification on the flow, and hence the nature of the deposits, we have conducted two series of preliminary analogue laboratory experiments, involving stratified saline and stratified particle-laden currents, and we report on these below. Saline currents In these experiments, we examined the motion of a stratified saline gravity current propagating along a 3 m flume filled with fresh

Fig. 9. Distance propagated by a stratified saline current consisting of two layers with 1% and 4% salt, and a control well-mixed current with 2.5% salt. The stratified flow propagates somewhat faster during the initial stages of motion, since the lower layer is relatively dense, but subsequently the well-mixed current advances more rapidly. In both cases, the background fluid is pure fresh water.

464

A. W. WOODS ET AL.

how the position of the nose of the current varies with time, for a two-layer source with a buoyancy ratio of 4:1, in comparison to the speed of an analogous well-mixed current, containing the same total mass of salt and hence total negative buoyancy. Initially, the stratified flow travels somewhat faster owing to the greater negative buoyancy of the lower layer. However, the lower layer rapidly thins and slows down since it is of smaller volume than the well-mixed current (Fig. 8b). Therefore, ultimately the analogous well-mixed current advances more rapidly. This experimental result is of considerable interest, since it suggests that although the flow morphology and deposition from a stratified gravity current may be distinct from a well-mixed current, to leading order, the overall speed of the flow is similar. These initial experiments confirm the phenomenological picture we have developed, that the stratification produced by the initial blast wave can lead to lateral segregation of the flow, and vertical stratification of the deposits. We are presently exploring the dynamics of these stratified laboratory currents in more detail, and will report on these elsewhere. Particle-laden currents In a second set of experiments, stratified particle-laden currents of fresh water were released into a flume filled with fresh water. The stratified current consisted of two particle sizes, such that the lower half of the current involved particles of both sizes, while the upper half of the current involved only particles of the smaller size. The stratified source condition was established by mixing particles of both sizes throughout the fluid behind the lock gate, and then allowing the particles to settle for that time necessary for the larger particles to sediment to a prescribed depth in the source fluid behind the lock gate. Owing to the difference in the fall speed of the two particle sizes, during the time required for the larger particles to sediment through the upper part of the source fluid, only a thin zone, less than 0.001 m deep, was depleted of the small particles at the top of the upper layer. The initial concentration of the larger particles was chosen so that the total particle load in each current at the start of the experiment was the same. The deeper the lower layer, the smaller the density contrast between the layers. The morphologies of the flows were quite similar to those of the stratified saline currents, with the dense lower layer initially running ahead of the overlying fluid. However, following the initial slump, the flow slows down more rapidly than the well-mixed current. From this we can infer that in the stratified flow, the coarse-grained deposit tends to form closer to the source compared to that produced by a comparable well-mixed current. This is reminiscent of the very well sorted deposits formed from the 26 December 1997 dome collapse at Soufriere Hills Volcano (Fig. 2; Ritchie et al. 2002). We are presently examining these effects in more detail. Discussion The idea that the pyroclastic density flow produced by the initial explosive expansion of pressurized gases develops strong density and grain size stratification provides some interpretations that are not easily explained by a homogeneous flow (cf. Valentine 1987). The grain size and lithofacies of the 26 December 1997 pyroclastic density current deposits show marked lateral variations within the affected areas. The current deposit is coarsest grained and shows strong erosion and destruction laterally between the peripheral areas (Ritchie et al. 2002). There are marked changes in the grain size of the lower layer 1 deposits across the deep valleys oblique to the flow, whereas layer 2 deposits are little affected by the valleys. The blowdown directions of trees and fence posts are not affected by deep valleys except right at the valley margin. The upper parts of the current also passed over a high col near the dome and only produced fine-grained deposits on the other side. All these observations can be explained by a strongly stratified current, with the dense coarsegrained parts being channelled by topography down the central zone

whereas the finer-grained upper parts were able to spread over a much wider area, with much less influence of topography. The calculations of fragmentation wave dynamics also have some interesting application to other blast-forming eruptions, and, in particular, the 18 May 1980 lateral blast of Mount St Helens, Washington, USA. The initial stages of that event involved a major landslide. This landslide disrupted the cap rock enclosing a cryptodome of volatile-rich magma which was originally stored just below the summit of the volcano (Kieffer 1981). As the landslide travelled down the flank of the volcano, and the magma decompressed, a massive lateral blast-flow developed. This travelled over 25km from the volcano before lifting off as a buoyant cloud (Kieffer 1981; Sparks et al. 1986). Estimates of the total energy release associated with the exploding cryptodome are of order 10 14 to 1015 J. Similar calculations to those presented above, but accounting for the much larger gas content of the magma in the cryptodome, suggest that the decompression would lead to velocities of the particle and gas mixture of order 100ms –1 (Fig. 5). which could therefore ascend several hundred metres before supplying a dense gravitydriven flow, in accord with recent interpretations of the flow deposits and observations of a ground-hugging flow (Druitt 1992. 1998: Bursik et al. 1998).

Conclusions (1)

Using a new model for the steady degassing of a permeable dome, we estimate that the pressure within the dome at Soufriere Hills Volcano, prior to the collapse on 26 December 1997. was of order 5-10 MPa. We also estimate that the energy associated with this gas moving through the degassing dome was as large as 1016 J and the mass of this gas was about 10 –3 to 10 –4 times the mass of the solid that was disrupted. (2) We predict that there was a relatively thin region near the dome surface across which the pressure fell rapidly towards atmospheric. A sudden disruption or collapse of this outer part of the dome may then expose the high-pressure material to the atmosphere, leading to formation of a fragmentation front across which the dome material disintegrates. (3) Using a new model to describe the dynamics of such a fragmentation wave, we predict that the decompressed gas and particles rise from the dome with speeds of order 60 m s – 1 . while the fragmentation wave moves into the dome at speeds of 10–20 m s – 1 . The erupting material may thus form a fountain a few hundred metres high before falling back to ground and generating a pyroclastic density current. (4) A well-mixed model of these pyroclastic density currents suggests that the flow is initially very dense owing to the relatively small gas content of the material that disrupts from the decompressed dome. However, calculations suggest that as the flow moves away from the source, the larger clasts will accumulate preferentially near the base of the current, forming a density-stratified flow. New analogue laboratory experiments indicate that such stratified currents can lead to very different flow structures, with the dense, relatively coarsegrained underflow running ahead of the more dilute overlying cloud and naturally leading to stratified deposits. Our work is supported by the NERC. R.S.J.S. acknowledges the support of a NERC Research Professorship. The paper was improved significantly by incorporation of comments from G. Valentine. L. Wilson, an anonymous referee and the tireless editor P. Kokelaar.

References ALIBIDIROV, M. A. 1994. A model for viscous magma fragmentation during volcanic blasts. Bulletin of Volcanology. 56. 459-465. ALIBIDIROV. M. A. & DINGWELL. D. B. 1996. Magma fragmentation by rapid decompression. Nature. 380. 146-148.

EXPLOSIVE DECOMPRESSION OF PRESSURIZED VOLCANIC DOME BARCLAY, J., RUTHERFORD, M. J., CARROLL, M. R., MURPHY, M. D., DEVINE, J. D., GARDNER, J. & SPARKS, R. S. J. 1998. Experimental phase equilibrium constraints on pre-emption storage conditions of the Soufriere Hills magma. Geophysical Research Letters, 25, 3437-3440. BENNET, F. D. 1974. On volcanic ash formation. American Journal of Science, 274, 648-661. BURSIK, M. I. & WOODS, A. W. 1996. The dynamics and thermodynamics of large ash flows. Bulletin of Volcanology, 58, 175-193. BURSIK, M. I., KURBATOV, A., SHERIDAN, M. & WOODS, A. W. 1998. Transport and deposition in the May 18 1980, Mount St Helens blast flow. Geology, 28, 155-158. DADE, B. & HUPPERT, H. E. 1996. Emplacement of the Taupo Ignimbrite by a turbulent flow. Nature, 381, 509-512. DRUITT, T. H. 1992. Emplacement of the 18 May 1980 lateral blast deposit east-northeast of Mount St Helens, Washington. Bulletin of Volcanology, 54, 554-572. DRUITT, T. H. 1998. Pyroclastic density currents. In: GILBERT, J. S. & SPARKS, R. S. J., (eds) The Physics of Explosive Volcanic Eruptions. Geological Society of London, Special Publications, 145, 145-182. DRUITT, T. H., YOUNG, S. R., BAPTIE, B. ET AL. 2002. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306. ElCHELBERGER, J. C, CARRIGAN, C. R., WESTRICH, H. R. & PRICE, R. H.

1986. Non-explosive silicic volcanism. Nature, 323, 598-602. FINK, J. H. & KIEFFER, S. W. 1993. Estimates of pyroclastic flow velocities from explosive decompression of lava domes. Nature, 363, 612-615. HOBLITT, R. P., MILLER, C. D. & VALLANCE, J. W. 1981. Origin and stratigraphy of the deposit produced by the May 18 directed blast. In: LIPMAN, P. W. & MULLINEAUX, D. R. (eds) The 1980 Eruptions of Mount St Helens, Washington. US Geological Survey, Professional Papers, 1250, 401-419. JAUPART, C. 1998. Gas loss from magmas through conduit walls during eruption. In: GILBERT, J. S. & SPARKS, R. S. J. (eds) The Physics of Explosive Volcanic Eruptions. Geological Society, London, Special Publications, 145, 73-90. KIEFFER, S. 1981. Fluid dynamics of the May 18 1980 blast at Mount St Helens. In: LIPMAN, P. W. & MULLINEAUX, D. R. (eds) The 1980 Eruptions of Mount St Helens, Washington. US Geological Survey, Professional Papers, 1250, 379-400. LACROIX, A. 1904. La Montagne Pelee and ses eruptions. Masson et cie, Paris. LUCKETT, R., BAPTIE, B. & NEUBERG, J. 2002. The relationship between degassing and rockfall signals at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 595-602. MELNIK, O. & SPARKS, R. S. J. 1999. Non-linear dynamics of lava-dome extrusion. Nature, 402, 37-41. MELNIK, O. & SPARKS, R. S. J. 2002. Dynamics of magma ascent and lava extrusion at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoir. MIDDLETON, G. V. & SOUTHARD, J. B. 1978. Mechanics of sediment movement. Short Course 3, Society of Economic Palaeontologists and Mineralogists, Eastern Section, 6.37-6.41. NERI, A. & DOBRAN, F. 1994. Influence of eruption parameters on the thermofluid dynamics of collapsing volcanic columns. Journal of Geophysical Research, 99, 11 833-11 857.

465

RITCHIE, L., COLE, P. & SPARKS, R. S. J. 2002. Sedimentology of deposits of the pyroclastic density current of 26 December 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 435-436. ROBERTSON, R., COLE, P. ET AL. 1998. The explosive eruption of Soufriere Hills volcano, Montserrat, West Indies, September 17 1996. Geophysical Research Letters, 25, 3429-3432. SATO, H., FUJII, T. & NAKADA, S. 1992. Crumbling dacite dome lava and generation of pyroclastic flows at Unzen Volcano. Nature, 360, 664–666. SIMPSON, J. 1997. Gravity Currents. Cambridge University Press, Cambridge. SPARKS, R. S. J. 1997. Causes and consequences of pressurisation in lava dome eruptions. Earth and Planetary Science Letters, 150, 177-189. SPARKS, R. S. J., MOORE, J. G. & RICE, C. J. 1986. The initial giant umbrella cloud of the May 18 1980 explosive eruption of Mount St Helens. Journal of Volcanology and Geothermal Research, 28, 257-274. SPARKS, R. S. J., BURSIK, M. I, CAREY, S. N., GILBERT, J., GLAZE, L., SIGURDSSON, H. & WOODS, A. W. 1997. Volcanic Plumes. Wiley, Chichester. SPARKS, R. S. J., YOUNG, S. R. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills volcano, Montserrat, West Indies, November 1995 to December 1997. Geophysical Research Letters, 25(18), 3421-3424. SPARKS, R. S. J., BARCLAY, J., CALDER, E. S. ET AL. 2002. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 409-434. TAIT, S., JAUPART, C. & VERGNIOLLE, S. 1989. Pressure, gas content and eruption periodicity of a shallow crystallising magma chamber. Earth Planetary Science Letters, 169, 301-310. TAYLOR, G. A. M. 1958. The 1951 eruption of Mt. Lamington, Papua. Bureau of Mineral Resources of Australia, Geology and Geophysical Bulletin, 38, 1-117. TURCOTTE, D., OCKENDEN, H., OCKENDEN, J. & COWLEY, S. 1990. A mathematical model of Vulcanian eruptions. Geophysical Journal International, 103, 211-217. VALENTINE, G. A. 1987. Stratified flow in pyroclastic surges. Bulletin of Volcanology, 49, 616-630. VALENTINE, G. & WOHLETZ, K. 1989. Numerical models of Plinian eruption columns and pyroclastic flows. Journal of Geophysical Research, 94, 1867-1887. VOIGHT, B., SPARKS, R. S. J., MILLER, A. D. ET AL. 1999. Magma flow instability and cyclic activity at Soufriere Hills Volcano, Montserrat, British West Indies. Science, 283, 1138-1142. WILSON, L., SPARKS, R. S. J., HUANG, T. C. & WATKINS, N. 1978. The control of volcanic eruption column heights by eruption energetics and dynamics. Journal of Geophysical Research, 83, 1829-1836. WOODS, A. W. 1988, The dynamics and thermodynamics of eruption columns. Bulletin of Volcanology, 51, 69-91. WOODS, A. W. 1995. A model of vulcanian explosive eruptions. Nuclear Engineering and Design, 155, 345-357. WOODS, A. W. & BURSIK, M. I. 1991. Particle fallout, thermal disequilibrium and volcanic plumes. Bulletin of Volcanology, 53, 559-570. YOUNG, S. R., ROBERTSON, R. ET AL. 1998. Monitoring SO2 emission at the Soufriere Hills volcano: Implications for changes in eruptive conditions. Geophysical Research Letters, 25, 3681-3684.

Pyroclastic flow and explosive activity at Soufriere Hills Volcano, Montserrat, during a period of virtually no magma extrusion (March 1998 to November 1999) G. E. NORTON1,2, R. B. WATTS3, B. VOIGHT 4 , G. S. MATTIOLI5, R. A. HERD 1,2 , S. R. YOUNG1, J. D. DEVINE6, W. P. ASPINALL 7 , C. BONADONNA3, B. J. BAPTIE8, M. EDMONDS9, C. L. HARFORD 3 , A. D. JOLLY10, S. C. LOUGHLIN 7 , R. LUCKETT 11 & R. S. J. SPARKS3 1 Montserrat Volcano Observatory, Mongo Hill, Montserrat, West Indies 2 British Geological Survey, Keyworth, Nottingham NG12 5GG, UK (e-mail: [email protected]) 3 Department of Earth Sciences, University of Bristol, Bristol BS8 IRJ, UK 4 Department of Geosciences, Pennsylvania State University, University Park, PA 16802, USA 5 Department of Geosciences, University of Arkansas, Fayetteville, Arkansas, USA 6 Department of Geological Sciences, Brown University, Providence RI 02912, USA 7 Aspinall & Associates, 5 Woodside Close, Beaconsfield, Bucks HP9 1JQ, UK 8 British Geological Survey, Edinburgh EH9 3LA, UK 9 Department of Earth Sciences, Cambridge University, Cambridge CB2 3EN, UK 10 Geophysical Institute, University of Alaska, Fairbanks, Alaska 99775, USA. 11 International Seismological Centre, Pipers Lane, Thatcham, Berkshire RG19 4NS, UK

Abstract: Dome growth at Soufriere Hills Volcano halted in early March 1998. After dome growth ceased, seismicity reduced significantly, but activity related to dome disintegration and degassing of magma at depth continued. A sustained episode of pyroclastic flows on 3 July 1998 marked the single largest collapse from March 1998 to November 1999. This led to a remarkable episode of dome collapses, low-energy explosions and ash-venting that resulted in the regular production of ash plumes, commonly reaching 1.5-6 km above sea level (a.s.l), but sometimes up to 11 km a.s.l., and the development of a small block-and-ash cone around the explosion crater. During the period of this residual activity, higher levels of activity occurred approximately every five to six weeks. This periodicity was similar to the cycles observed during active dome growth during 1995 to 1998, and probably had a similar cause. The relatively high level of observed activity caused continued concern regarding volcanic hazards and their potential to impact upon the resident population. Vigorous magma extrusion resumed in November 1999. The activity of the intervening period is attributed to the continued cooling and degassing of the dome, conduit and deep magma body, the impact of rising volcanic gases in the volcanic edifice, and limited magma flow in the conduit.

The eruption of Soufriere Hills Volcano on Montserrat began on 18 July 1995 with phreatic explosions from Castle Peak, a prehistoric (c. 350 years BP) andesitic dome sited within the horseshoeshaped English's Crater (Fig. 1). After an initial phreatic phase, most of the following 32-month period involved activity related to the growth and collapse of a viscous lava dome (Young et al. 1998a). On 17 September 1996 a magmatic explosive eruption followed a 9-hour period of dome collapse involving c. 9.5 x 106 m3 (dense rock equivalent; DRE) of the lava dome (Robertson et al. 1998; Calder et al. 2002). High magma production rates (7-10 m3 s – 1 ) from May to October 1997 led to major gravitational collapses (e.g. Loughlin et al. 2002) and two series of Vulcanian explosions (Druitt et al. 2002). On 26 December 1997, catastrophic failure of the southwestern wall of English's Crater and flank of the volcano resulted in removal of 55 x 106 m3 of the lava dome and talus (Calder et al. 2002; Sparks et al. 2002; Voight et al. 2002). This collapse, the largest of the eruption at the time of writing, was followed by rapid new growth of the dome. By early February 1998, the dome had restored the volume destroyed by the 26 December 1997 collapse and for the period 26 December 1997 to midFebruary 1998 was estimated at a rate of 6m3 s – 1 . Growth of the dome continued through the end of February 1998, but ceased on about 10 March 1998, at which time the total amount of magma erupted since 1995 was 300 x 106 m3 (DRE), and the volume of the dome at this time was 113 x 106m3 (Fig. 2a). During magma extrusion from November 1995 to March 1998, cyclicity in the volcanic activity was noted on several timescales and in several datasets (Voight et al. 1998, 1999; Denlinger & Hoblitt 1999). Major dome collapses or periods of explosivity were thought to occur approximately every six to ten weeks (Voight et al. 1999). On a shorter timescale, cycles of between 8 and 14 hours were often noted in seismicity, tilt data and explosions or increased pyroclastic flow activity (Voight et al. 1998). This latter pattern persisted to early March 1998, although by this time the peaks in activity were barely recognizable.

Major dome collapses during the period of magma extrusion were generally associated with intense precursory seismic activity. For example, the collapse on 26 December 1997 occurred after nearly 36 hours of a swarm of hybrid earthquakes (Sparks et al. 2002). The recognition of these short- and intermediate-scale patterns of activity enhanced the forecasting capability of the Montserrat Volcano Observatory (MVO). After the cessation of magma extrusion in March 1998, seismic activity decreased markedly, and the dome appeared to be stable. A sudden large dome collapse on 3 July 1998, without any recognized seismic or other precursory activity, restarted a series of hazardous dome collapses and explosions at a level thought to be unusual for the period following a major phase of dome-building extrusion. There was some evidence that magma may have been slowly ascending up the conduit, and surface deformation recorded by an array of six continuous global positioning system (CGPS) receivers showed ongoing inflation or uplift of the volcanic edifice during the period from February 1998 to November 1999 (Mattioli et al. 2000), suggesting that pressure in the magma chamber and conduit may have been increasing. This led to concerns about whether vigorous magma production at the surface would resume. In mid-November 1999, after 20 months of this residual activity, a second phase of lava extrusion commenced. This paper outlines the phenomena that were observed at Soufriere Hills Volcano during the intervening period of virtually no magma extrusion from March 1998 to November 1999. There are few accounts in the literature that chronicle pauses in magma production or the ends of such dome-building eruptions (e.g. Nakada et al. 1999; Mori et al. 1996), and thus few cases with which to compare the activity under discussion. In general, the scientific literature contains many descriptions of the onsets of eruptions and precursory activity (e.g. Christiansen & Peterson 1981; Chouet et al. 1994), but much less information on dormancy of lava domes or on criteria for recognizing the end of lava dome eruptions. The observations made during the period of virtually no

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 467-481. 0435-4052/02/$15 © The Geological Society of London 2002.

467

468

G. E. NORTON ET AL.

Fig. 1. Map of Montserrat, showing main towns and locations mentioned in the text.

ACTIVITY DURING A PERIOD OF VIRTUALLY NO MAGMA EXTRUSION

Fig. 2. Map showing the main features of the dome complex as it appeared on (a) 10 March 1998, (b) 13 July 1998 and (c) 26 January 1999.

469

470

G. E. NORTON ET AL.

magma extrusion at Soufriere Hills Volcano have improved our understanding of the evolution of andesitic dome-building eruptions, particularly for hazards assessment and risk management purposes. Of particular importance are the observations that major collapses of the lava dome and Vulcanian explosions with fountaincollapse pyroclastic flows were possible without extrusion of new magma, and with no discernible precursory seismic activity.

made during the period of dome growth from November 1995 to March 1998, and includes an overview of the data collected during the period of virtually no magma extrusion from March 1998 to November 1999. Discussion of these latter data is included in later more detailed sections.

Seismic monitoring Monitoring methods Several methods were used to monitor the activity of Soufriere Hills Volcano (Aspinall et al. 2002) during the period under discussion, including seismic, ground deformation, volcanological, and gasmonitoring techniques. This section summarizes the observations

The main monitoring method consisted of seismic data collection and analysis (Miller et al. 1998). Five main types of seismic signal were recognized at Soufriere Hills Volcano: volcanotectonic earthquakes, long-period earthquakes, hybrid earthquakes, rockfall or pyroclastic flow signals, and explosion signals. Elevated levels of volcanic tremor were also measured periodically at Soufriere

Fig. 3. Key monitoring data for 1 January 1998 to 30 November 1999. (a) The number of rockfalls and volcanotectonic earthquakes recorded each week by the MVO digital seismological network. (b) Changes in line length between a reflector site on the northern wall of English's Crater to an instrument site 3 km to the north of the volcano at Windy Hill (Fig. 1). Each point on the graph represents the average of ten direct measurements of slant distance done with a Leica TC1 100 Total Station (see Jackson et al. (1998) for details of the technique used at the MVO). The number of dome collapses, explosions or ash-venting episodes per week is also shown. (c) SO2 flux as measured by COSPEC ( x 1 0 6 k g d a y – 1 ) and SO 2 concentration from diffusion tubes (ppb).

ACTIVITY DURING A PERIOD OF VIRTUALLY NO MAGMA EXTRUSION

471

Hills Volcano. Volcanotectonic earthquakes had impulsive starts and then rapidly decreased in amplitude. They were predominantly high-frequency signals (>2 Hz). They were common at the start of the eruption and occurred sporadically during March 1998 to November 1999, often in swarms. Long-period earthquakes had a more emergent start and generally low frequency content (1-2 Hz), and were common during active dome growth. Hybrid earthquakes had impulsive starts but contained a significant amount of lowfrequency energy. These signals were often associated with periods of rapid dome growth, and hybrid swarms were recognized as precursors to major dome collapses during dome growth. Both long-period and hybrid earthquakes were rare during the period of virtually no magma extrusion. Rockfall or pyroclastic flow signals were the most common types of seismic signals. They had an emergent start and a gradual decrease in amplitude towards the end of the signal; they contained a mixture of frequencies. Pyroclastic flow signals were simply long-duration and high-amplitude rockfall signals. Explosions had quite characteristic signals. They often had a long-period component (1-2 Hz) at the start, followed by a highamplitude, mixed-frequency signal related to pyroclastic flows (Druitt et al. 2002). After the pyroclastic flow signal had died away, the long-period signal remained, but at a lower amplitude than the initial signal. This tremor was often related to ash-venting, and sometimes continued for many tens of minutes. The main types of seismicity recorded during the period of virtually no magma extrusion were volcanotectonic earthquakes, rockfalls, tremor and explosion signals (Fig. 3a, b).

Deformation monitoring Several different methods were used to measure the deformation of the flanks of the volcano, the main methods being global positioning system (GPS) geodesy, electronic distance measurement (EDM), and tilt measurements. During the period of virtually no magma extrusion, GPS data were downloaded daily from three Leica (MVO) and three Trimble (University of Puerto Rico) CGPS receivers stationed around the volcano. In addition, over 20 sites were occupied occasionally by temporary GPS surveys to provide a wider coverage of the deformation field of the volcano. EDM was used frequently in the early stages of the eruption to monitor the deformation of the flanks of the volcano, but three out of four networks were destroyed as a result of volcanic activity during 1997, and reinstallations of near-field reflectors were repeatedly destroyed. Electronic and dry-tilt measurements were taken periodically during the ongoing eruption. Electronic tilt was particularly successful during periods of rapid dome growth when cyclical activity with time periods of typically 8 to 14 hours was measured on the tiltmeters, showing repetitive inflation (with associated hybrid earthquake activity, and culminating in pyroclastic flows or explosions) and deflation (Voight et al. 1999; Denlinger & Hoblitt 1999). Tilt stations close to the summit of the volcano were destroyed in August 1997; stations at a greater distance away from the dome showed little or no variation. In general, large deformations of Soufriere Hills Volcano were confined to areas close to English's Crater, although measurable, volcanically induced deformation was recorded as far away as 8 km from the dome. In addition, a relatively large differential movement of + 10 cm was noted, increasing the distance between a site at Long Ground, 2 km to the ENE of the volcano, and a site at White's Yard, 3 km to the NE of the volcano (see Fig. 1), from June 1996 to December 1998. Since these sites are only 733m apart, this movement could not have been accommodated elastically. Ground inspection revealed a fracture that crossed the survey line, trending at 068°. There was insufficient evidence to determine whether this surface fracture was related to volcanic deformation or to localized mass movement disrupting the ground surface. Apart from minor, short-term variations, three certain trends were observed in the GPS data collected between March 1998 and November 1999: (i) eastward movement of stations to the east of

Fig. 4. Hermitage Estate CGPS coordinate changes, 1 January 1998 to 1 December 1999, showing (a) latitude, (b) longitude, and (c) vertical co-ordinate changes. All data are residuals plotted relative to a best-fit curve through the entire dataset since June 1996.

the volcano; (ii) higher rates of movement for stations within 2 km of the dome; and (iii) significant vertical displacement at all GPS sites, implying inflation or uplift of the edifice relative to the Earth's centre of mass (Fig. 4). The rate of movement of an EDM reflector on the northern wall of English's Crater (Fig. 1) fell dramatically in March 1998 when dome growth stopped (see Fig. 3b). The rate of shortening of the line from the reflector to the instrument site at Windy Hill to the north dropped by an order of magnitude, from a peak of about 3 cm per month between late January and early March 1998 to less than 1 cm per month from early March to late May 1998. From late

472

G. E. NORTON ET AL.

May 1998 to early September 1998 the line lengthened considerably, and then shortened again until the reflector site was destroyed in early October 1998.

Volcanological monitoring Observations of the volcanic activity constituted a large proportion of the monitoring effort. Many different techniques were used, but the main parameters that were measured included the measurement of dome and deposit volumes (and consequently rates of extrusion), the geology and petrology of the new deposits, and measurements of ash cloud heights. These observations will be described in more detail below.

Gas monitoring Two methods of monitoring gas emissions were used regularly throughout the eruption. The first method used a correlation spectrometer (COSPEC) to measure the daily output of SO2 from the volcano. SO2 fluxes during the period of dome growth fluctuated, but generally showed an increase in flux as the rate of magma extrusion increased (Young et al. 1998b). There was also evidence for shortterm fluctuations of SO2 flux in concert with short-term cycles in seismicity (Watson et al. 2000; Young et al. 2000). The maximum SO2 flux rate recorded during dome growth was 2.5 x 106 kg day – 1 , although the long-term average flux was c. 0.5 x 106 kg day – 1 (Young et al. 1998b). COSPEC data were not collected during periods of high magma extrusion rate from July 1997 to February 1998, or at various other times due to instrument malfunction or when precluded by inaccessibility due to volcanic activity. SO2 was also measured at ground level using diffusion tubes that were left at sites around the volcano for about two weeks (Norton 1997). Trends in these data closely tracked long-term averages of COSPEC data so that, when COSPEC measurements were not possible, diffusion tube results provided a useful proxy for SO2 output from the volcano. SO2 flux continued to vary during the period of virtually no magma extrusion (Fig. 3c), but generally remained comparable to gas fluxes recorded during the period of dome growth. This is somewhat surprising given the previously noted positive correlation between extrusion rates and gas fluxes, which would have suggested that gas fluxes in the period of virtually no magma extrusion should have been very low. The highest values of SO2 emission from March 1998 to November 1999 were usually associated with pyroclastic flow activity or periods of ash-venting (Fig. 3b). SO2 flux then tended to decrease after ash-venting episodes. One other method of measuring gas concentrations has been used occasionally on Montserrat: Fourier transform infrared spectroscopy (FTIR; Oppenheimer et al. 2002). Using this method the ratios of different gases were measured, such as the ratio of hydrogen chloride to sulphur dioxide. During the period of virtually no magma extrusion, FTIR was used during July through October 1998, and again in January 1999, and the results indicated that the ratio of sulphur to chlorine gases had increased by more than one order of magnitude in comparison with measurements first done in 1996 during magma extrusion (Oppenheimer et al. 2002).

Chronology from March 1998 to November 1999 March to June 1998: cessation of dome growth and slow degradation of the dome On 1 March 1998, the summit of the lava dome was measured at 1011 m above sea level (a.s.l.) and was blocky with a small number of short spines. These spines grew rapidly during the ensuing week, and, on 9 March 1998, the most prominent spine was measured at a

height of 1027 m a.s.l. Seismic activity, which had been low since early February 1998, declined further in March, and volcanotectonic earthquakes and rockfall signals became the most common types of seismicity recorded. Hybrid earthquakes had been the dominant type of earthquake during dome growth (Aspinall et al. 1998; Miller et al. 1998). Theodolite measurements during clear conditions on 5 April 1998 indicated that the highest point on the dome was the top of the large spine that had grown rapidly before 10 March 1998 in the SW sector of the dome. This spine had attained a final elevation of 1031 m a.s.l. (Fig. 2a). A survey of the dome complex conducted on 10 March 1998 indicated that the total dome volume was 113 x 106 m3 (DRE). Visual observations of the dome complex from 10 March 1998 to 3 July 1998 indicated no major changes in the morphology, implying that magma extrusion had ceased. Slow degradation of the upper parts of the dome by occasional rockfalls from early March to late June 1998 formed deep gullies on the eastern and southwestern flanks. The average number of rockfalls decreased from 390 per week in February 1998 to 44 per week during March 1998 (Fig. 3a). Some of the larger rockfalls had runouts of as much as 1 km, and reached the base of the talus slope. Some small pyroclastic flows occurred down the eastern flank, the largest of which travelled down a narrow ravine and reached a runout distance of 1.8 km. These collapses each lasted a few minutes, generated minor amounts of convecting ash, and left fine-grained, thin deposits (<5 m thick) in a well developed chute extending from the upper flanks. Active fumaroles developed in a V-shaped cleft on the eastern side of the dome and in a gully between the western edge of the 26 December 1997 scar (Fig. 2a) and the area of new growth within the scar. Earthquake activity from early April to late June 1998 consisted principally of small amounts of volcanotectonic seismicity located at depths of 2.5-3.5 km below the summit of the dome. Two exceptions to this were a small swarm of 12 high-amplitude hybrid earthquakes that occurred on 6 May 1998. and a volcanotectonic earthquake swarm on 4-5 June 1998. Ground deformation measurements (EDM) on the northern flank of the volcano indicated that the rate of deformation had slowed markedly between early March 1998 and late May 1998 (Fig. 3b).

Major dome collapse of 3 July 1998 A sudden increase in activity occurred on 3 July 1998, with a major dome collapse to the east. No precursory seismic activity was recognized, although the average number of rockfall signals had increased slightly from 0-4 per day through most of June 1998 to 8-13 per day three days prior to the collapse. The seismic signal caused by the collapse started with a very high-amplitude, short-lived onset at 03:02 local time (all times are local time = UTC - 4 hours); this was followed by approximately 2.5 hours of pyroclastic flow signals. Two high-amplitude pulses at 03:50 and 04:30 are evident in the seismic data, although neither was as large as the seismic signal at the initial failure. The main collapse period was followed by a period of heightened rockfall activity and an increase in the number of volcanotectonic earthquakes, which lasted until 14:07. The volcanotectonic earthquakes were probably caused by depressurization of the system during and after the collapse. The ash plume generated by the collapse rose to 9-14 km a.s.l. and the top of the cloud moved to the ENE. The lower-level plume moved to the NW. Most of the ash fell as accretionary lapilli (diameters as much as 2-4 mm in northwestern Montserrat) and the ash layer reached a thickness of 2 mm c. 6 km NW of the volcano. Pyroclastic flows were observed in the Tar River valley at 04:22 and had reached the sea 3km to the east by 04:50. The associated pyroclastic surge was well developed, scoured the south side of the Tar River valley and extended 700 m to the north of the valley (Fig. 5). A small explosion occurred later in the afternoon at 14:07.

ACTIVITY DURING A PERIOD OF VIRTUALLY NO MAGMA EXTRUSION

473

Fig. 5. Map of deposits from the 3 July 1998 dome collapse.

Ballistic blocks up to 1.5m diameter, observed in a field of impact craters 1 km SE of the dome summit, suggested that this (or some component of the main collapse) might have been a small Vulcanian explosion. The collapse is estimated to have removed 15-19 x l06 m3 (DRE) of the lava dome, leaving a canyon-like feature extending deep into the SE flank of the dome, with chutes leading down to the south edge of the Tar River valley (Fig. 2b). The pyroclastic fan at the mouth of the Tar River valley was extended 350 m to the north, and c. 6 x 106 m3 (DRE) of material was added. A substantial volume of material must have been deposited offshore of the Tar River valley. After the collapse, strong fumarolic activity was observed along a clearly defined NW-trending linear fracture, 50-100 m long, at the base of the new scar. This fracture was located in a general position that had been recognized since September 1997 as the site of vigorous fumarolic activity, and the rocks in this vicinity were probably hydrothermally altered. Significant pulse-like ash-venting continued for two to three weeks, and fumarolic activity was also evident on the southern and northern flanks of the dome. Watts et al. (2002) argue that the dome consists of a shell of loose blocky talus surrounding a main core of massive lava. The combination of hydrothermally altered rocks in the area of the 3 July 1998 collapse, and the presence of loose blocky talus, created a weakened area in the dome, which collapsed without precursory seismic activity. However, there was also intense rain prior to and during the collapse, and it is possible that rain infiltrated into the dome interior through the steep talus slope on the eastern flanks and contributed to the initiation of dome failure. This is not unprecedented, since rainfall-induced dome collapses have previously been recognized at Merapi Volcano, Indonesia (Ratdomopurbo & Poupinet 2000; Voight et al. 2000). Measurement of SO2 flux with a COSPEC in early July 1998 gave values of 1.3-3.0 x 10 6 kgday – 1 between 5 July and 11 July, with one occasion when it rose to >4 x 10 6 kgday – 1 on 13 July 1998. These levels represented the highest measured SO2 fluxes since the onset of the eruption. The flux rate declined more or less steadily throughout the rest of July to an average of 1 x 10 6 kgday -1 (Fig. 3c). Similarly, at a diffusion tube site 4.5km west of the volcano, SO2 from March to June 1998 was generally below 25 ppb (Fig. 3c); this rose suddenly to over 200 ppb at the same site averaged over the two-week period from 29 June 1998 to 13 July 1998. This high flux suggests that high gas pressure may have existed within the dome and upper conduit prior to the collapse, and this

gas pressure may have contributed to the failure. Steam generated by rainwater percolating into the dome interior may have further enhanced the gas pressures within the dome. Gas pressure acts at the boundaries of potential failure blocks within the dome, with an uplift force that reduces frictional resistance, and an outward push that increases destabilizing forces. The net effect of gas pressurization is thus to increase instability, leading to dome collapses, as discussed by Voight & Elsworth (2000). The northern EDM line from a reflector on the northern wall of English's Crater to Windy Hill lengthened by 8cm between May and August 1998, although the data are sparse between May and July 1998. This may indicate a decrease of loading stresses on the wall of English's Crater following the 3 July 1998 collapse (Fig. 3b). In the weeks following the collapse, no fresh magma apparently reached the surface, despite the wholesale depressurization of the vent area due to the loss of dome above it. August to November 1998: small dome collapses Earthquake and rockfall activity remained at an elevated level immediately after the 3 July 1998 collapse, but then gradually declined through July, with the exception of a volcanotectonic earthquake swarm on 25 July 1998. The early part of August was quiet with few rockfalls or earthquakes. Active fumaroles on the southern side of the dome and associated rockfalls gradually undermined the dome and led to two episodes of pyroclastic flow activity down the southwestern flanks of the volcano during the second week of August. The flows travelled 1.8km from the dome and each episode was followed by about an hour of continuous rockfall activity. A steep lava buttress overhanging the 3 July 1998 scar collapsed in mid-August and generated a series of pyroclastic flows, which reached the Tar River pyroclastic fan, 3 km east of the dome. An intense period of ash-venting, which lasted from 19 to 21 August, was correlated with nearly monochromatic volcanic tremor and high SO2 fluxes (>1 x 10 6 kgday –1 measured by COSPEC; concentration of >300ppb measured by diffusion tubes). By late August, the level of activity had again declined, with low seismicity and a general reduction in levels of SO2 flux to an average of about 0.5 x l 0 6 g d a y – 1 (Fig. 3c). Gradual degradation of the lava dome continued during September 1998, with three periods of pyroclastic flow generation caused by small-volume collapses to the east, west and north.

474

G. E. NORTON ET AL.

A swarm of volcanotectonic earthquakes on 26 September 1998 was followed by periods of ash-venting and associated volcanic tremor, which continued almost daily until the end of the month. SO2 fluxes increased towards the middle of September 1998 in tandem with observations of enhanced ash-venting from the dome fumaroles (Fig. 3c). Measurements on the EDM line from the northern wall of English's Crater to Windy Hill made in September 1998 indicated that the line had again begun to shorten (Fig. 3b). The reflector was destroyed by pyroclastic flows in early October 1998. The CGPS site at Hermitage Estate (Fig. 1), located approximately 500m from the dome, showed its most rapid horizontal acceleration to the NE from early September to the end of December 1998 (Fig. 4a, b). Small-volume dome collapses increased in frequency during October 1998, leading to further degradation of the dome. Five main collapses occurred on 13 October at 08:01 (to the NE and NW), 18 October at 09:16 (direction uncertain), 20 October at 22:41 (to the west), 26 October at 00:51 (to the east and south) and 31 October at 04:18 (to the south and west), each producing deposits up to 3 km from the dome. Ash plumes reached 2-8 km a.s.l. and deposited ash predominantly to the west and NW of the volcano, but light tephra fallout was also experienced in the north of the island. Volcanotectonic earthquake swarms located at 3-4 km depth were associated with the collapses on 13 and 18 October, but there was no anomalous volcanotectonic seismicity associated with the other three collapses. Volcanic tremor and associated ash-venting were recorded after each collapse and lasted for a few minutes to several hours. By late October 1998, collapse of dome lava down Gages valley had exposed the incandescent interior of the dome and excavated a deep gully, which extended to the southern side of the dome. A large fracture, oriented NE-SW, developed in the SW part of the dome. There was an apparent decrease in SO2 fluxes towards the end of October, and a single measurement on 2 November 1998 gave 0.7 x l 0 6 k g d a y – 1 (Fig. 3c). November 1998 began with a swarm of volcanotectonic earthquakes on 1 November, the largest of which were felt throughout the island. The frequency of small-volume dome collapses and the runout distances of the resulting pyroclastic flows increased in early November, reaching 3 km from the dome to the east on 3 November, and 5 km from the dome to the SW on 5 November. Volcanotectonic earthquakes again followed very rapidly from the collapse on 5 November. The largest collapse occurred at 06:07 on 12 November and probably involved a dome volume of 2.5 x 106 m3 (DRE). This resulted in pyroclastic flows that reached the sea to the east, west and south. This collapse may have had an explosive component, as suggested by simultaneous generation of pyroclastic flows in three different directions, possibly by fountain collapse. A convective ash plume rose to 8km a.s.l. The collapse was followed by vigorous ash-venting, which caused continued tephra fallout to the NW of the volcano. As a consequence of this collapse and the previous collapses in October and November, a deep gorge was cut into the dome running approximately ENE-WSW. The gorge was about 150m deep and 30m wide, and it split the dome into two peaks (Fig. 6). In the weeks following the large 12 November 1998 collapse, there were a few small pyroclastic flows and some periods of low-amplitude seismic tremor coupled with ash-venting.

December 1998 to May 1999: minor explosions with ash-venting Small-volume (<10 6 m 3 ) dome collapses continued into December 1998, but then the style of activity changed to unequivocally explosive behaviour. At the onset of a small collapse at 10:34 on 19 December, powerful black jets of ash and rock burst from the east side of the dome, preceding the generation of a pyroclastic surge to the east. This was the first direct evidence of explosive activity, although it was unclear whether it preceded or followed the dome collapse. The resulting deposit was small in volume and almost entirely confined to a pre-existing incised channel within the Tar

River valley. On 21 December, at the onset of a sudden, large seismic signal, dense black jets of ash and vigorously convecting ash clouds were observed coming from the main vent in the 3 July 1998 scar. Ballistic blocks were observed to heights of about 80 m above the vent. Very vigorous ash-venting continued for over 30 minutes after the initial explosion, but no pyroclastic flows were generated. No pumiceous fallout occurred from these explosions. The expulsion of ballistics suggests that new magma may have risen to shallow levels in the conduit, plugged it. and then been explosively expelled. Periods of volcanic tremor occurred almost daily during December and were observed to coincide with episodes of ashventing of varying vigour. It is interesting to note that this period of explosive activity followed immediately after the period of most rapid horizontal displacement of the northeastern sector of the edifice as recorded at the Hermitage CGPS site (Fig. 4a. b). A spectrum of activity was observed throughout this period, from brief episodes of weak steam-venting, to extremely vigorous, pulse-like ash-venting commonly following dome collapses, to small explosions generating plumes up to 6km a.s.l. After a volcanotectonic earthquake swarm on 4 to 5 December 1998, the number of volcanotectonic earthquakes declined towards the end of the month while the number of long-period signals increased. The hypocentres of the volcanotectonic earthquakes became more diffuse, and small clusters of earthquakes were located with epicentres 3 km to the NW. 1 km to the SE and 1 km to the NE of the vent, as well as directly below the volcano. A series of small explosions occurred during the first half of January 1999. These explosions often generated small-volume pyroclastic flows, possibly by fountain collapse, which reached up to 3 km to the east and south of the dome. The flow deposits were usually very thin (<1 m thick throughout their length), fine-grained (<1 mm diameter grain size) and closely followed topographic features such as pre-existing incised channels in older deposits. Ash clouds rose to 6km a.s.l.. and weak, pulse-like ash-venting continued for 15 to 30 minutes after each explosion. Each explosion caused a pressure wave detected by an infrasonic sensor 2 km to the east of the dome, produced an audible sound within 6 km of the volcano and generated a seismic signal with distinct long-period energy. Brief periods of ash-venting, lasting 20-30 minutes and correlating with seismic tremor, occurred throughout January. These became shorter and weaker towards the end of the month. During this period, the average SO 2 flux was elevated, at >10 6 kg d a y – 1 . The dome had a volume of 77 x 10 6 m 3 (DRE) at the end of January 1999. with its highest point (977 m a.s.l.) above the southern flank (Fig. 2c). The gorge from the 3 July 1998 and OctoberNovember 1998 collapses cut up to 100m deep into the pre-1995 floor of English's Crater and had removed a minimum of 5 x 106 m3 of pre-1995 rock from this area. The part of the dome to the north of the gorge comprised three main buttresses above the northwestern, northern and northeastern flanks (Fig. 7). and contained two-thirds of the total dome volume in January 1999. Episodes of ash-venting or small explosions from within the gorge, accompanied by small pyroclastic flows with runouts of <2km, occurred about once every 48 hours between 5 and 12 February. The explosion on 5 February was the largest in the February 1999 sequence, and the accompanying ash cloud rose rapidly to over 5km a.s.l. Pyroclastic flows from this event travelled to the east down the Tar River valley. Although the COSPEC results showed a generally decreasing SO 2 flux throughout most of January 1999. the levels remained elevated until February 1999 (Fig. 3c). There were increased levels of activity from 1 March to 31 May 1999. with 54 small explosions or ash-venting episodes during these three months. The largest explosions produced ash clouds up to approximately 9km a.s.l., with fountain-collapse pyroclastic flows and associated lightning and tephra fallout. These occurred at a rate of a little less than one per day in late March, with the number of explosions increasing in the first week in April 1999 (Fig. 3b). and then decreasing again until mid-May 1999. The first week in April was the most intense period of activity during the time of virtually no magma extrusion. Ash-venting episodes throughout this period were at times very intense. In particular, a fumarole away from the

ACTIVITY DURING A PERIOD OF VIRTUALLY NO MAGMA EXTRUSION

475

Fig. 6. Photographs of Soufriere Hills Volcano taken in late December 1998, showing the dome complex after the series of collapses culminating in the 12 November 1998 scar. (a) View from the west showing (1) Gages Mountain; (2) the dome; (3) the 12 November 1998 scar; (4) Chances Peak; and (5) the Gages pyroclastic fan. (b) View from the east showing (1) the 12 November 1998 and 3 July 1998 collapse scar; (2) an explosion crater with (3) a tuff ring being formed; (4) the northern part of the dome; (5) the southern part of the dome and (6) the southern wall of English's Crater (photographs by R. Robertson).

main dome complex on the southern wall of English's Crater was so vigorous that a small cone of tephra had built around the vent in just a few days. Gradual degradation of the dome was caused by the explosions, and pyroclastic flows reached to the Tar River pyroclastic fan or to less than 2km from the dome in other directions. Seismicity was dominated by small explosion signals and ashventing. Many of the explosion or ash-venting episodes had impulsive starts with a gradual decline in amplitude towards the end of the signal. The signals had a long-period component and each explosion generated a pressure wave. Volcanotectonic earthquakes (10-70 per week) and rockfall signals (30-160 per week) continued throughout this period, with no seismic build-up before the ash-venting episodes and explosions.

Fist-sized blocks were collected from fountain-collapse pyroclastic flows to the east of the volcano, from explosions on 1 March, 1, 8 and 14 April 1999; tephra fallout was also collected from these explosions. Some of this material was moderately vesiculated and petrological analysis indicated the presence of fresh green hornblende with no reaction rims, although the majority had been extensively oxidized. By comparison with magma erupted during active dome growth, this observation suggested that some of the magma may have ascended rapidly from the magma chamber (Devine et al. 1998a, b; Murphy et al. 1998), although it is possible that the samples with fresh hornblende may have been entrained into the pyroclastic flows or eruption columns from deposits resulting from the 1995-1998 activity. No new dome extrusion was observed in the

476

G. E. NORTON ET AL.

Fig. 7. Photograph of the northern flank of Soufriere Hills Volcano in January 1999 showing (1) Perches Mountain; (2) the three main buttresses of the dome that constitute the northern section of the dome complex: (3) Chances Peak; and (4) the pyroclastic flow fan (photograph by R. Robertson).

following weeks. The simplest explanation is that the explosions were evacuating the upper conduit and enabling it to be refilled from below, although the pressure driving the flow was not great enough to cause extrusion and dome building.

May to November 1999: earthquake swarms, continued dome collapses and small explosions An intense volcanotectonic earthquake swarm occurred on 22 May 1999, the first since 5 December 1998, and also the most intense for at least a year. There were 121 volcanotectonic earthquakes in the swarm, which lasted from 07:00 to 17:00, with peaks in activity between 07:00 and 08:00, and between 13:00 and 15:00. A small ash cloud to 2km a.s.l. was produced at 13:21 at the time of maximum intensity of earthquake activity. The volcanotectonic earthquakes were located consistently between 2.9 and 3.9km depth, either directly under the dome, or on a trend 1 km east-northeastwards away from the dome. Depths of previous swarms during the period of virtually no magma extrusion (mostly between October and December 1998) were also at about 3 km, and predominantly under the dome, although several swarms had epicentral clusters in other areas; e.g. 1 km to the WNW or 1 km to the ESE of the dome. Focal mechanisms for volcanotectonic earthquakes throughout late 1998, and from the swarm on 22 May 1999, were oblique in nature (thrust) rather than strike-slip, thus implying an increase in pressure at the depth of origin. Towards the end of the swarm on 22 May 1999, the earthquakes became more strike-slip oriented, possibly implying a change in stress later in the day. These earthquakes showed some similarities to the seismicity from 1992 to 1995, prior to the start of the eruption in July 1995 (J. Shepherd, pers. comm.). The earthquake swarm was followed within 12 hours by a dome collapse at 02:43 on 23 May 1999. The ash cloud from the pyroclastic flows reached to a height of about 6 km a.s.l. Tephra fallout was predominantly dispersed to the west of the volcano, with accretionary lapilli to 3mm diameter reported 4km from the vent. New pyroclastic flow deposits were seen to the east, with a substantial blocky flow deposit in the main valley as far as the sea, 3 km from the dome. A new pyroclastic surge deposit had also reached to 1.5km from the dome and had spread 500m north of the main block-and-ash deposit. The nose of the block-and-ash flow deposit on the delta was pinkish in colour, with a few large blocks greater

than 1 m diameter. A pyroclastic surge deposit was also observed on the southern flanks of the dome with a runout of less than 1 km. SO2 fluxes after the dome collapse remained low (0.1-0.6 x 106 kgday – 1 from 24 to 29 May 1999 compared to 0.1-0.5 x l06 kgday – 1 from 15 to 22 May 1999). A dome collapse occurred on 5 June 1999 at 17:45, with a plume up to 4 km a.s.l. Pyroclastic flows were seen entering the sea 3 km to the east, and flows were also seen travelling less than 2km to the NE of the dome. Tephra-fallout deposits 7km to the NW of the volcano reached 1 cm in thickness and included accretionary lapilli. Approximately 1.5 x 10 6 m 3 had collapsed from the northeastern part of the dome. A more substantial dome collapse with a volume of about 4.2 x 10 6 m 3 DRE occurred on 20 July 1999. This collapse generated pyroclastic flows that covered most of the pyroclastic fan at the end of the Tar River valley, as well as pyroclastic surges 1.5 km to the SE over a topographic barrier. This latter area had not been previously affected by pyroclastic surges during the period of active dome growth. The ash plume from the pyroclastic flows reached to 11 km a.s.l. Volcanic activity remained at an enhanced level for some time after 20 July 1999, including two small explosions and two volcanotectonic earthquake swarms before the end of July. Activity in August was generally at a lower level, with reduced gas emissions and occasional small dome collapses with associated pyroclastic flows. The flanks of the lava dome slowly became stabilized, as talus and loose blocks were shed from all areas. Volcanic activity levels increased again in early September 1999, with a substantial explosion on 3 September 1999 and subsequent dome collapses and enhanced rockfall activity in the following seven days. Volcanic activity reduced to a lower level by the beginning of October 1999. Minor ash-venting episodes and an increase in rockfall signals occurred towards the end of October 1999. At the end of October, a series of phreatic explosions was followed by an intense hybrid earthquake swarm on 3-8 November 1999. The hybrids were mostly located beneath the volcano at depths of 1.5-3 km below English's Crater. Magmatic explosions on 8 and 9 November 1999 produced steamy ash clouds that rose up to 6-8 km a.s.l. The tephra-fallout deposit contained small (<0.5cm) pumice lapilli. These explosions were followed by periods of low-frequency tremor, but there was no associated domecollapse activity. On 27 November 1999, a new dome was observed inside the gorge in the centre of the 1995-1998 dome, marking the start of the second phase of magma extrusion.

ACTIVITY DURING A PERIOD OF VIRTUALLY NO MAGMA EXTRUSION

Activity types The main types of volcanic activity that took place in the period March 1998 to November 1999 were pyroclastic flows, ash-venting episodes, and Vulcanian explosions. The following paragraphs summarize the essential features of these phenomena. It must be noted, however, that a continuum of types of activity occurred throughout this period, such that, for example, some dome-collapse pyroclastic flows may have had an explosive onset, and some ashventing episodes may have been vigorous enough to be termed explosions.

Pyroclastic flows Pyroclastic flows during the period under review were formed by both dome collapse and fountain collapse. Cole et al. (1998) and Calder et al. (1999, 2002) distinguished two types of dome-collapse pyroclastic flows at Soufriere Hills Volcano: (i) discrete collapses that involve the shedding of material as a single pulse; and (ii) sustained collapses that involve the incremental collapse of a significant portion of the dome. Dome-collapse pyroclastic flows during the period March 1998 to November 1999 were mainly single, discrete events. The flows often lasted only a few minutes and produced blocky (up to c. 5 m diameter boulders) deposits of fresh and hydrothermally altered dome material. Runout distances were generally <2km, although the larger flows extended as far as 5km from the dome (Fig. 8). The collapses were largely responsible for the excavation of deep scars and gullies in the dome. Vigorously convecting ash plumes, which rose to 1.5-14 km a.s.l., were generated by these collapses. The 3 July 1998 collapse was exceptional during this period as it lasted for over 2.5 hours and involved a series of successive collapses. The seismic record for this collapse shows a sudden highamplitude onset followed by a decrease in energy, and then at least two further high-amplitude pulses. This implies a sudden failure of an outer part of the dome, followed by successive failures of the interior.

Fig. 8. Map of the pyroclastic flow and surge deposits from Soufriere Hills Volcano from April 1996 to March 1998 and from March 1998 to November 1999.

477

With the onset of ash-venting and unequivocal low-magnitude explosions in December 1998, fountain-collapse pyroclastic flows started to occur. These flows had runouts of >2 km and generated relatively fine-grained deposits with very few blocks larger than 1 m diameter. Some blocks collected from the nose of the flows were semi-vesicular. The flows were not erosive and left only thin deposits along valley bottoms. Highly expanded pyroclastic surges were often associated with the flows, particularly close to the vent. Observations of the initiation of some of these flows indicated that they were generated by partial collapse of poorly developed eruption columns. In comparison with the Vulcanian explosions during the period of high extrusion rate in 1997 (Druitt et al. 2002), the explosions from December 1998 to October 1999 were relatively low-energy with fountain collapse probably occurring no more than 100m above the vent, although few direct observations were made.

Ash-venting Steam-venting continued throughout the period of virtually no magma extrusion. At times, vigorous emission of ash-laden steam occurred from various locations on the lava dome. This phenomenon is here termed ash-venting. Ash-venting episodes were discrete events that typically lasted about 30 minutes, but on rare occasions the episodes lasted several hours. The first unequivocal episode of ash-venting occurred on 19 August 1998, in the scar produced by the 3 July 1998 collapse. Jets of steam and ash were discharged by a particularly vigorous fumarole on the back wall of this scar. Ash-venting was intense for about 24 hours, but waned over the following days. Some of the fumaroles were temporarily buried by rockfalls within the scar, and venting declined to low levels by the end of August. Periods of vigorous ash-venting occurred again towards the end of September 1998, and were then correlated with low-amplitude seismic tremor and increased SO2 fluxes. The increased frequency of small-volume pyroclastic flows in October and November 1998 was accompanied by periods of vigorous ashventing, which often lasted for several hours.

478

G. E. NORTON ET AL.

Periods of volcanic tremor, which coincided with episodes of ash-venting, were recorded daily throughout December 1998. Spectrograms of the tremor associated with ash-venting were often sharply peaked at between 1 and 2 Hz. These peaks generally remained constant over the duration of the tremor period, but occasionally the frequency of the dominant spectral peak changed as the amplitude of the tremor was also observed to change. This strong correlation with active ash-venting led to the interpretation that the vent itself may have been acting as the source for the tremor, or that the venting may have been generating some resonance in the body of the lava dome, near the vent orifice.

Small-magnitude Vulcanian explosions Small-magnitude Vulcanian explosions, which generated buoyant plumes above the volcano, were unequivocally observed first in early December 1998, but the dome collapses in November 1998 may also have had explosive components. While these initially produced discrete blasts, which threw metre-sized ballistic blocks up to 80m above the vent, they were unable to generate substantial convecting eruption columns with most ash clouds reaching no more than 6km a.s.l. By January 1999, the explosions were generating fountain-collapse pyroclastic flows, which reached the sea in the Tar River valley, 3 km east of the dome. The plumes initially rose rapidly to between 2 and 3km a.s.l. and then convected slowly to between 3 and 6 km a.s.l. The explosions originated from a crater located in the remnants of the 3 July 1998 scar, and caused detectable pressure waves. A blocky tuff ring formed around the developing explosion crater and gradually increased in width and depth through 1999. A boom was heard at the initiation of some explosions, and spectacular thunder and lightning were associated with some eruption plumes. The explosions were often accompanied by periods of intense ash-venting which were sometimes audible as a jet engine-like noise. The most active period of explosions was between the start of March 1999 and the end of May 1999, with 54 explosions or periods of ash-venting between 1 March and 31 May 1999. Most of the explosions reached between 1.5 and 6km a.s.l., which is low relative to the Vulcanian explosions during the period of high extrusion rate in 1997 when heights of between 3 and 15km a.s.l. were measured (Druitt el al. 2002). The explosions during the period of virtually no magma extrusion produced no pumiceous fallout, and the blocks collected from fountain-collapse pyroclastic flows were only semi-vesiculated. The fountain-collapse flows were only rarely distributed radially in several directions around the vent, with most being directed to the east or west by the gully in the dome, implying that fountain collapse occurred at a low elevation above the vent. Thus the explosions during this period must have occurred at low magma production rates, and with only limited vesiculation of the magma within the upper levels of the conduit.

Discussion This paper documents a period of relative dormancy between two phases of magma extrusion. It is clear, however, that the volcano was not inactive. Indeed, apart from a four-month period from March to July 1998, the activity was never at a particularly low level: there were numerous, relatively small explosions, and pyroclastic flows impacted areas that had not been affected previously (Fig. 8). Throughout this period, critical issues were: (i) whether the activity that was being observed indicated continuing, but less energetic, ascent of magma; (ii) whether energetic magma extrusion at the surface would restart; and (iii) what precursors might be expected prior to the onset of any renewed dome growth. This type of residual volcanic activity between two phases of dome growth has not been commonly described in detail in the literature, and so it is important to make comparisons between the types of phenomena that were observed during this period of

dormancy and those that have been described at the ends of other dome-building eruptions. Episodes of growth at dome-building eruptions tend to end with declining growth rates (over weeks to months), formation of spines, and reduction in seismic activity to near-background levels. The 1951-1956 eruption of Mount Lamington ended with slow dome growth accompanied by slow extrusion of a series of spines (Taylor 1958). A sharp decrease in seismicity accompanied the cessation of dome growth at the end of the 1991-1992 eruption of Mount Pinatubo (Mori el al. 1996). Low levels of earthquakes, gas flux and ground deformation all marked the end of the 1990-1995 eruption of Mount Unzen (Nakada el al. 1999). Intense activity at the end of dome growth appears to be rare and, when it occurs, is most often in the form of minor phreatic explosions or dome collapses. For instance, small phreatic explosions occurred after cessation of dome growth at the 1976 eruption of St Augustine volcano (Swanson & Kienle 1988) and in 1992 at the end of the eruption of Pinatubo (Daag et al. 1996). Eruptions that ended with magmatic explosions are not common, but examples include Galeras (Stix el al. 1997), Lascar (Matthews el al. 1997) and Irazu (Krushensky & Escalante 1967). At Galeras, cessation of dome growth in 1991 was followed by 7.5 months of low seismicity and low gas flux before sudden Vulcanian explosions began on 16 July 1992 (Stix et al. 1997). However, no further magma extrusion had occurred at Galeras at the time of writing. At Soufriere Hills Volcano, in the three months before magma extrusion paused in March 1998, the rate of dome growth was still quite high, although growth rates appeared to have been gradually decreasing throughout February and early March 1998 (Watts et al. 2002). In March 1998. the number of seismic signals diminished to near-background levels, and the character of the seismicity changed: volcanotectonic earthquakes and rockfalls became dominant relative to the hybrid earthquakes that had been prominent throughout dome growth (Aspinall et al. 1998: Miller et al. 1998; Voight et al. 1999). Changes in dome morphology from March to July 1998 were minor and related only to minor disintegration of the dome. Hence it appeared that magma ascent had ceased. However, data from the CGPS sites, processed using absolute point positioning techniques and final orbit clock products (see Mattioli et al. (1998) for discussion), showed a signal of uplift or inflation relative to the Earth's centre of mass (Mattioli et al. 2000). A major dome collapse on 3 July 1998 unsealed the cooling dome and conduit, and may have been a result of pressurization due to cooling and degassing of the stagnant magma column, possibly triggered by rainwater infiltration into the steep talus slope. After this collapse, SO 2 flux was exceptionally high, with values up to >4.0 x 10. 6 kgday –1 , and typically between 0.5 and 1.5 x 10 6 kgday – 1 , indicating a sudden increase in gas output relative to the previous three months. These levels were even higher than those measured during the period of active dome growth. Exposure of the vent area, and local development of steep slopes within the collapse crater, led to further dome collapses and ash-venting. Eight months after dome growth had ceased, mild Vulcanian explosions restarted, possibly in November 1998 and unequivocally in December 1998. These explosions continued, along with episodes of vigorous ash-venting and occasional dome collapses, until the renewal of vigorous magma extrusion in November 1999. The high gas fluxes and high ratio of sulphur to chlorine throughout this period (Oppenheimer et al. 2002) pointed to continued, on-going degassing of the dome, conduit and deeper magma body. There was some evidence that SO2 flux often peaked after an explosion or ash-venting, and that the flux decayed gradually thereafter (Fig. 3c). This suggests that the vent and conduit were opened during the explosion or venting episode, and then gradually resealed with time, possibly by slow ascent of new magma in the conduit. The general progression of the activity from dome collapses through ash-venting to small explosions from July 1998 to November 1999, and the overall decrease in SO2 flux through this period, may also indicate a longer-term gradual cooling, degassing and partial resealing of the conduit.

ACTIVITY DURING A PERIOD OF VIRTUALLY NO MAGMA EXTRUSION There is some evidence for weak peaks in dome collapse or explosive activity occurring approximately every five to six weeks. This is most discernible in the frequency of explosions and dome collapses (Fig. 3b). This periodicity is similar to, although not as strongly defined as, the six to ten week cyclicity observed during active dome growth in 1995-1998 (Voight et al. 1999), and may have been driven by similar repeated pressurization processes in the dome, conduit and deeper magma body. The key difference between the periodicity seen during active magma extrusion and that during dormancy is the lack of precursory hybrid earthquake activity in the latter. On the contrary, volcanotectonic earthquake activity sometimes followed major collapses, indicating a response to depressurization of the conduit. This lack of precursory activity made forecasting the timing of individual collapses and explosions more difficult, although the recognition of the weak cyclicity gave the MVO some ability to forecast periods of likely enhanced activity. The lack of precursory hybrid seismicity may reflect a lesser degree of pressurization than that which occurred in more active periods. Previously, it had been recognized that the development of hybrid seismicity responded to a pressure threshold (Voight et al. 1999), which may not have been surpassed in March 1998 to October 1999. There were some indications that magma may have been ascending slowly at various times during the period of dormancy, particularly after 3 July 1998. Petrological evidence indicated that fresh material might have been erupted from the vent during the explosive activity in March and April 1999. It was not clear, however, that the samples collected from the distal ends of pyroclastic flow deposits necessarily reflected magma newly erupted from depth. An alternative explanation is that they were reamed from the conduit wall or dome during the explosions, or even incorporated into the pyroclastic flows from earlier (1995-1998) deposits during transport. Other evidence for slow ascent of magma includes the cumulative volume of material expelled in scores of explosive eruptions. For example, in the period of the most intense explosive activity from 1 March 1999 to 31 May 1999, given an average eruption column height of 3.5km a.s.l. for each of 54 explosions or periods of ash-venting, the mass erupted per explosion can be calculated using the method and estimated parameters given in Druitt et al. (2002, equation 1). This requires an average eruptive volume of about 2000 m3 (DRE) per explosion, thus giving an average magma production rate of 0.01 m3 s–1 in this three-month period. It is also possible that slow extrusion of magma in the vent area may have occurred during periods of poor or limited visibility, and that any dome material was expelled from the vent during subsequent explosions. These lines of evidence suggest that magma may have been rising in the conduit throughout parts of the period of virtually no magma extrusion, although with less energy than from November 1995 to March 1998, such that the magma emerged only rarely, if ever, from the vent. Similarly, the volcanotectonic earthquake swarm on 22 May 1999 was similar to the seismicity prior to the onset of the ongoing eruption in 1995, and this may also have been an indication of magma movement at depth. We suggest that movements immediately after 3 July 1998 were very sluggish, due to the high viscosity of largely degassed magma, and also to the narrowing of the conduit from crystallization at the wall. Over the following months, the mean conduit viscosity gradually reduced as newer magma entered the vent, and ascent became slightly easier. Thus, by late 1998, activity became increasingly explosive as conduit volatile content could not be readily dissipated. This trend continued in 1999 as a gradual prelude to the more vigorous dome building that commenced in November 1999. However, the upward trend was not continuous, but included some minor peaks and valleys, with a noticeable peak occurring in April 1999. Possibly these variations reflected periodic repressurization of the magma chamber related to the multiple-week cyclicity. The above hypothesis can be considered in the light of conduitflow physics, as expressed simply by the expression:

479

where Q is the flow rate, P is the overpressure (total pressure minus magmastatic pressure) driving the conduit flow, r is the conduit radius, is the bulk magma viscosity, and L is the length of the conduit. Over the period 26 December 1997 to mid-February 1998, the flow rate was approximately 6 m 3 s – 1 . The chamber overpressure at depth L = 5000m was of the order of 10-15 MPa during the period of vigorous flow (Voight et al. 1999), the outlet pressure was about 5 MPa at the base of the 250-m-high dome, and the net overpressure driving conduit flow was therefore about 5-10 MPa. By March 1998, flow had ceased, suggesting that chamber pressure had declined and was then counterbalanced by pressure due to the weight of the dome. The chamber overpressure was thus about 5 MPa, net overpressure was zero, and Q was zero. With the 3 July 1998 dome collapse, however, the dome counterweight was removed, such that the net overpressure driving conduit flow was again c. 5 MPa. In a viscous system, the applied stress would have required flow to restart. If we assume that the rate Q = 6 m3 s–1 in January 1998 was associated with an overpressure of 10 MPa, then the flow in July driven by P = 5MPa should have recommenced at the rate 3 m3 s – 1 , all other factors remaining equal. However, no such large flow rate was observed, and this is consistent with the idea that the magma viscosity may have increased substantially during the stagnant interlude. Increasing viscosity by a factor of 100 reduces Q to 0.03m 3 s – 1 . Likewise, if conduit radius decreased by only 10% as a result of congealing at the conduit wall, the flow rate would be decreased by a further 34% or Q = 0.02 m 3 s – l . These values concur with estimates of magma production rate during periods of increased explosive activity, for example, during March to May 1999. Finally, we note that various parameters such as overpressure and mean viscosity may have varied from time to time over the period of virtually no magma extrusion, so that some subperiods may have been more or less active than others. The volcanotectonic seismicity during the period of virtually no magma extrusion was most likely related to variations in the pressurization in the conduit at 3-4 km depth, coupled with localized structural adjustments to changes caused by the preceding magma movements and extrusion. CGPS data from early 1998 to late 1999, however, are best fitted by an inflating point-source at approximately 6 km depth, embedded in a simple elastic half-space (Mogi 1958). The derived depth is similar to that obtained by Mattioli et al. (1998) for GPS data from 1995 to mid-1997, although the dilatation is opposite in sign and about one-half the magnitude. A strong indication of vigorous renewed magma ascent was registered in early November 1999, when the first hybrid swarm since May 1998 occurred, to be followed by small Vulcanian explosions with pumiceous fallout. On 27 November 1999, a substantial new dome was observed to be forming over the vent reamed out by the numerous explosions since December 1998. The continuation of substantial activity at the surface, despite the apparent pause of magma extrusion, poses several problems for attempting to determine whether the eruption has ended, once a phase of magma extrusion ceases. The question is of vital importance to hazard management and government officials. If one defines an eruption purely in terms of the sustained arrival of magma at the Earth's surface, then the first modern eruption of Soufriere Hills Volcano can be said to have ended in early March 1998, and a new one started in November 1999. However, pyroclastic flows and surges with accompanying ash clouds continued to occur and large collapses were experienced in July and November 1998, and June and July 1999, and Vulcanian explosions with ash plumes up to 9 km a.s.l. occurred frequently from December 1998 to November 1999. Thus, if one defines an eruption in terms of observable phenomena, then the volcano continued in eruption through the period being reported here. In this case, the results of monitoring data collected were also equivocal since, for example, COSPEC measurements showed gas fluxes were higher than during periods of magma extrusion, and the eastern flanks of the volcano continued to show evidence of ground deformation. The experience of volcanic activity at Soufriere Hills Volcano demonstrates that hazardous activity can continue for extended periods even after magma has apparently ceased to erupt at the

480

G. E. NORTON ET AL.

surface. With hindsight, it might be argued that the high level of observable activity, continued deformation and high gas fluxes, from March 1998 to November 1999, were an indication that the volcano was only in temporary repose, and that further vigorous magma extrusion was quite possible. Indeed, this was always recognized as a possibility in the probabilistic risk assessments (always examined for the 'next' six-month period), although the 'consensus' probability of a restart declined as time went on. In general it was suspected that the observed activity was more likely to have been due to cooling and degassing of the dome, conduit and deeper magma body, than any incipient renewal of eruptive activity. The need for vigilance in both monitoring and crisis management are clearly underscored. It seems necessary to continue dedicated, full-scale monitoring of this type of dome-building eruption for some years after magma extrusion has ceased. It also seems that observed activity has to be substantially diminished before an eruption can be deemed to be over. The visible evidence of ongoing eruptive activity, despite the apparent absence of new magma at the surface, posed several problems for crisis managers. The civilian population, who continued to observe dome collapses, ash-venting and explosions at the volcano similar to those seen during the height of the emergency, remained concerned about the activity. Even when there was no magma extrusion, dome-collapse activity still produced tephrafallout deposits rich in crystalline silica (10-24% in the <10 m fraction; Baxter et al. 1999), which was of great concern for people living on the island (Moore et al. 2002). Ash plumes continued to threaten aviation in the region and tephra fallout caused occasional problems for new agricultural developments in the north of the island, as well as adding to the volume of loose material on the flanks of the volcano available for lahar generation. Dome collapses and explosions remained very dangerous, with their potential to take lives and impact property on the flanks of the volcano. The Montserrat experience emphasizes the necessity to maintain close monitoring for a considerable time after such a dome-building eruption appears to have peaked and declined.

Conclusions Volcanic activity at Soufriere Hills Volcano continued at a high level during a period of virtually no magma extrusion from March 1998 to November 1999. Dome-collapse pyroclastic flows and Vulcanian explosions, often without apparent precursors, continued to endanger life and property on the flanks of the volcano, and provided serious hazards to aviation in the eastern Caribbean. There was no evidence for ascent of magma to the surface between March 1998 and July 1998, although CGPS results showed that there was gradual inflation of the volcanic edifice, probably as a result of pressurization in the magma chamber and conduit. The 3 July 1998 dome collapse involved dome lava from the dome-building phase from November 1995 to March 1998 and it may have been caused by gas pressurization from the buried conduit, possibly enhanced by rainwater infiltration and weakening of the dome rock by hydrothermal activity. Following this collapse, the vent was exposed, and sluggish ascent of volatile-depleted magma may have begun in a conduit narrowed by crystallization at the wall. Over the next six months, activity became increasingly explosive. The limited evidence suggests that, from July 1998, magma ascent may have occurred in the conduit, although with limited driving pressure such that any dome growth was not sufficiently voluminous to be observed. The main signals of renewed vigorous magma extrusion were the onset of hybrid earthquakes and pumiceous fallout from explosions in early November 1999. It is proposed that much of the activity throughout the period of virtually no magma extrusion was due to cooling and degassing of the dome, conduit and deeper magma body, leading to pressurization of the crystallizing magma. Slow magma ascent after

July 1998 may have renewed the volatile and energy supply to higher conduit levels. Weak peaks in the intensity of the activity every five to six weeks were similar to the cyclicity observed during active dome growth. The cyclicity in both cases may have been linked to degassing and crystallization of the magma body and conduit. The activity at Soufriere Hills Volcano during this period underscores the need to continue to monitor dome-building eruptions for several years after magma extrusion has apparently ceased. The authors would like to thank all colleagues at the Montserrat Volcano Observatory, particularly R. Robertson and all the local staff who have contributed to the continued detailed documentation of the ongoing eruption at Soufriere Hills Volcano. The Government of Montserrat and the British Government (through DFID) supported the authors' work at the MVO. This paper is published by permission of the Director. British Geological Survey (NERC). Funding for UPRM equipment and GSM was provided by NASA grants NCC5-252 and NAG-6031.

References ASPINALL. W. P., MILLER. A. D., LYNCH. L. L., LATCHMAN. J. L.. STEWART, R. C., WHITE. R. A. & POWER, J. A. 1998. Soufriere Hills eruption. Montserrat 1995-1997: Volcanic earthquake locations and fault plane solutions. Geophysical Research Letters. 25. 3397-3400. ASPINALL. W. P., LOUGHLIN. S. C., MICHAEL. F. V. ET AL. 2002. The Montserrat Volcano Observatory: its evolution, organization, role and activities. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat. from 1995 to 1999. Geological Society, London. Memoirs. 21. 71-91. BAXTER. P. J., BONADONNA. C., DUPREE. R. ET AL. 1999. Cristobalite in volcanic ash of the Soufriere Hills Volcano. Montserrat: Hazards implications. Science. 283. 1143-1146. CALDER, E. S., COLE. P. D., DADE. W. B. ET AL. 1999. Mobility of pyroclastic flows and surges at the Soufriere Hills volcano. Montserrat. Geophysical Research Letters. 26. 537-540. CALDER. E. S., LUCKETT. R., SPARKS. R. S. J. & VOIGHT. B. 2002. Mechanisms of lava dome instability and generation of rockfalls and pyroclastic flows at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21. 173-190. CHOUET. B. A., PAGE. R. A., STEPHENS. C. D., LAHR. J. C. & POWER, J. A. 1994. Precursory swarms of long-period events at Redoubt Volcano (1989-1990). Alaska: Their origin and use as a forecasting tool. Journal of Volcanology and Geothennal Research. 62. 95-135. CHRISTIANSEN. R. L. & PETERSON. D. W. 1981. Chronology of the 1980 eruptive activity. In: LIPMAN. P. W. & MULLINEAUX. D. R. (eds) The 1980 Eruptions of Mount St Helens, Washington. United States Geological Survey. Professional Papers. 1250, 17-30. COLE. P. D., CALDER. E. S., DRUITT. T. H., HOBLITT. R., ROBERTSON. R., SPARKS, R. S. J. & YOUNG. S. R. 1998. Pyroclastic flows generated by gravitational instability of the 1996-97 lava dome of Soufriere Hills Volcano. Montserrat. Geophysical Research Letters. 25. 3425-3428. DAAG. A. S., DOLAN. M. T., LAGUERTA. E. P.. MEEKER. G. P., NEWHALL. C. G., PALLISTER. J. S. & SOLIDUM. R. U. 1996. Growth of a postclimactic lava dome at Mount Pinatubo. July-October 1992. In: NEWHALL, C. G. & PUNONGBAYAN. R. S. (eds) Fire and Mud: Eruptions and Lahars of Mount Pinatubo, Philippines. University of Washington Press. 647-664. DENLINGER. R. P. & HOBLITT. R. P. 1999. Cyclic eruptive behavior of silicic volcanoes. Geology. 27. 459-462. DEVINE, J. D., MURPHY. M. D., RUTHERFORD. M. J., SPARKS. R. S. J., CARROLL, M. R., YOUNG. S. R. & GARDNER. J. E. 1998a. Petrologic evidence for pre-eruptive pressure-temperature conditions, and recent reheating, of andesitic magma eruption at the Soufriere Hills Volcano, Montserrat. W.I. Geophysical Research Letters. 25. 3669-3672. DEVINE, J. D., RUTHERFORD. M. J. & GARDNER. J. E. 1998b. Petrologic determination of magma ascent rates for the 1995-1997 Soufriere Hills Volcano andesite magma. Geophysical Research Letters. 25, 3673-3676. DRUITT, T. H., YOUNG. S. R., BAPTIE. B. ET AL. 2002. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The

ACTIVITY DURING A PERIOD OF VIRTUALLY NO MAGMA EXTRUSION Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306. JACKSON, P., SHEPHERD, J. B., ROBERTSON, R. E. A. & SKERRITT, G. 1998. Ground deformation studies at Soufriere Hills volcano, Montserrat I: Electronic distance meter studies. Geophysical Research Letters, 25, 3409-3412. KRUSHENSKY, R. D. & ESCALANTE, G. 1967. Activity at Irazu and Poas Volcanoes, Costa Rica, November 1964-July 1965. Bulletin Volcanologique, 31, 75-84. LOUGHLIN, S. C., CALDER, E. S., CLARKE, A. B. ET AL. 2002. Pyroclastic flows and surges generated by the 25 June 1997 dome collapse, Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 191-209. MATTHEWS, S. J., GARDEWEG, M. C. & SPARKS, R. S. J. 1997. The 19841996 cyclic activity of Lascar volcano, Northern Chile: cycles of dome growth, dome subsidence, degassing and explosive eruptions. Bulletin of Volcanology, 59, 72-82. MATTIOLI, G. S., DIXON, T. H., FARINA, F., HOWELL, E., JANSMA, P. & SMITH, A. L. 1998. GPS measurement of surface deformation around Soufriere Hills volcano, Montserrat from October 1995 to July 1996, Geophysical Research Letters, 25, 3417-3420. MATTIOLI, G. S., RODRIGUEZ, L., SMITH, A. L., JANSMA, P. E. & HERD, R. A. 2000. Changes in magmatic flux recorded by campaign and continuous GPS geodesy of Soufriere Hills, Montserrat, BWI (abstract). Geosciences 2000. Geological Society, London. MILLER, A. D., STEWART, R. C., WHITE, R. A. ET AL. 1998. Seismicity associated with dome growth and collapse at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3401-3404. MOGI, K. 1958. Relations between the eruptions of various volcanoes and the deformations of the ground surfaces around them. Bulletin of the Earthquake Research Institute, 36, 99-134. MOORE, K. R., DUFFELL, H., NICHOLL, A. & SEARL, A. 2002. Monitoring of airborne particulate matter during the eruption of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 557-566. MORI, J., WHITE, R. A., HARLOW, D. H. ET AL. 1996. Volcanic earthquakes following the 1991 climactic eruption of Mount Pinatubo: Strong seismicity during a waning eruption. In: NEWHALL, C. G. & Punongbayan,, R. S. (eds) Fire and Mud: Eruptions and Lahars of Mount Pinatubo, Philippines. University of Washington Press, 339-350. MURPHY, M. D., SPARKS, R. S. J., BARCLAY, J. ET AL. 1998. The role of magma mixing in triggering the current eruption at the Soufriere Hills Volcano, Montserrat, West Indies. Geophysical Research Letters, 25, 3433-3436. NAKADA. S., SHIMIZU, H. & OHTA, K. 1999. Overview of the 1990-1995 eruption at Unzen Volcano. Journal of Volcanology and Geothermal Research, 89, 1-22. NORTON, G. E. 1997. Summary of the environmental effects of volcano emissions, Soufriere Hills Volcano, Montserrat. MVO Open File Report, 97/15. OPPENHEIMER, C., EDMONDS, M., FRANCIS, P. & BURTON, M. 2002. Variation in HC1/SO2 gas ratios observed by Fourier transform spectroscopy at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 621-639. RATDOMOPURBO, A & POUPINET, G. 2000. An overview of the seismicity of Merapi volcano (Java, Indonesia) 1983-1994. Journal of Volcanology and Geothermal Research, 100, 193-214.

481

ROBERTSON, R., COLE, P., SPARKS, R. S. J. ET AL. 1998. The explosive eruption of Soufriere Hills Volcano, Montserrat, West Indies, 17 September, 1996. Geophysical Research Letters, 25, 3681-3684. SPARKS, R. S. J., BARCLAY, J., CALDER, E. S. ET AL. 2002. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 409-434. STIX, J., TORRES, C. R., NARVAEZ, M. L., CORTES, J. G. P., RAIGOSA, A. J., GOMEZ, M. D. & CASTONQUAY, R. 1997. A model of Vulcanian eruptions at Galeras volcano, Columbia. Journal of Volcanology and Geothermal Research, 77, 285-303. SWANSON, S. E. & KIENLE, J. 1988. The 1986 eruption of Mount St Augustine: Field test of a hazard evaluation. Journal of Geophysical Research, 93(B5), 4500-4520. TAYLOR, G. A. 1958. The 1951 Eruption of Mount Lamington, Papua. Department of National Development, Bureau of Mineral Resources, Geology and Geophysics. Bulletin No. 38. VOIGHT, B. & ELSWORTH, D. 2000. Stability and collapse of gas-pressurized lava domes. Geophysical Research Letters, 48, 1-4. VOIGHT, B., HOBLITT, R. P., CLARKE, A., LOCKHART, A. B., MILLER, A. D., LYNCH, L. & McMAHON, J. 1998. Remarkable cyclic ground deformation monitored in real-time on Montserrat, and its use in eruption forecasting. Geophysical Research Letters, 25, 3405-3408. VOIGHT, B., SPARKS, R. S. J., MILLER, A. D. ET AL. 1999. Magma flow instability and cyclic activity at Soufriere Hills Volcano, Montserrat, B.W.I. Science, 283, 1138-1142. VOIGHT, B., CONSTANTINE, E. K., SISWOWIDJOYO, S. & TORLEY, R. 2000. Historical eruptions of Merapi Volcano, Central Java, Indonesia, 1768-1998. Journal of Volcanology and Geothermal Research, 100, 69-138. VOIGHT, B., KOMOROWSKI, J.-C, NORTON, G. E. ET AL. 2002. The 26 December (Boxing Day) 1997 sector collapse and debris avalanche at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKE-LAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 363-407. WATSON, I. M., OPPENHEIMER, C., VOIGHT, B. ET AL. 2000. The relationship between degassing and ground deformation at Soufriere Hills Volcano, Montserrat. Journal of Volcanology and Geothermal Research, 98, 117-126. WATTS, R. B., HERD, R. A., SPARKS, R. S. J. & YOUNG, S. R. 2002. Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 115-152. YOUNG, S. R., SPARKS, R. S. J., ASPINALL, W. P., LYNCH, L. L., MILLER, A. D., ROBERTSON, R. E. A. & SHEPHERD, J. B. 19980. Overview of the eruption of Soufriere Hills Volcano, Montserrat, 18 July 1995 to December 1997. Geophysical Research Letters, 25, 3389-3392. YOUNG, S. R, FRANCIS, P. W., BARCLAY, J. ET AL. 1998b. Monitoring SO2 emission at the Soufriere Hills Volcano: Implications for changes in eruptive conditions. Geophysical Research Letters, 25, 3681-3684. YOUNG, S. R., DUFFELL, H. J., VOIGHT, B. ET AL. 2000. Cyclic gas exhalations at Soufriere Hills Volcano, Montserrat and implications for real-time hazard assessment. Proceedings of IAVCEI General Assembly, Bali, Indonesia, July 2000.

Tephra fallout in the eruption of Soufriere Hills Volcano, Montserrat C. BONADONNA1, G. C. MAYBERRY 2 , E. S. CALDER 1 , R. S. J. SPARKS1, C. CHOUX3, P. JACKSON4, A. M. LEJEUNE 1 , S. C. LOUGHLIN5, G. E. NORTON6, W. I. ROSE2, G. RYAN7 & S. R. YOUNG4 1 Department of Earth Sciences, University of Bristol, Bristol BS8 1RJ, UK (e-mail: [email protected]) 2 Department of Geological Engineering and Sciences, Michigan Technological University, Houghton MI 49931, USA 3 Laboratoire Magmas et Volcans, Universite Blaise Pascal et CNRS, 63038 Clermont Ferrand, France 4 Montserrat Volcano Observatory, Mongo Hill, Montserrat, West Indies 5 British Geological Survey, Edinburgh EH9 3LA, UK 6 British Geological Survey, Keyworth, Nottingham, UK 7 Institute of Environmental and Natural Sciences, Lancaster University, Lancaster LA1 4YQ, UK

Abstract: Four mechanisms caused tephra fallout at Soufriere Hills Volcano, Montserrat, during the 1995-1999 period: explosive activity (mainly of Vulcanian type), dome collapses, ash-venting and phreatic explosions. The first two mechanisms contributed most of the tephra-fallout deposits (minimum total dense-rock equivalent volume of 23 x 106 m 3 ), which vary from massive to layered and represent the amalgamation of the deposits from a large numbers of events. The volume of co-pyroclasticflow fallout tephra is in the range 4-16% of the associated pyroclastic flow deposits. Dome-collapse fallout tephra is characterized by ash particles generated by fragmentation in the pyroclastic flows and by elutriation of fines. Vulcanian fallout tephra is coarser grained, as it is formed by magma fragmentation in the conduit and by elutriation from the fountain-collapse flows and initial surges. Vulcanian fallout tephra is typically polymodal, whereas dome-collapse fallout tephra is predominantly unimodal. Polymodality is attributed to: overlapping of fallout tephra of different types, premature fallout of fine particles, multiple tephrafallout sources, and differences in density and grain-size distribution of different components. During both dome collapses and explosions, ash fell as aggregates of various sizes and types. Accretionary lapilli grain size is independent of their diameter and is characterized by multiple subpopulations with a main mode at 50. Satellite data indicate that very fine ash can stay in a volcanic cloud for several hours and show that exponential thinning rates observed in proximal areas cannot apply in distal areas.

Tephra fallout was a significant phenomenon in the eruption of Soufriere Hills Volcano, Montserrat, during the 1995-1999 period. After a first explosive phreatic phase, tephra fallout was mainly associated with the extrusion of an andesitic lava dome, with generation of rockfalls and block-and-ash pyroclastic flows (Young et al. 1998). Magmatic explosive eruptions of sub-Plinian and Vulcanian type occurred in late 1996 (Robertson et al. 1998) and in 1997 (Druitt et al. 2002). Dome growth stopped in late March 1998, but collapses continued to occur in late 1998 and 1999 due to instability of the dome itself, and there were also episodes of vigorous ashventing (Norton et al. 2002). Volcanic plumes that led to tephra fallout on Montserrat were produced by four mechanisms: (i) magmatic explosive eruptions; (ii) elutriation of fines from dome-collapse pyroclastic flows and rockfalls; (iii) ash-venting; and (iv) phreatic explosions. In this paper tephra fallout is used to indicate fallout processes, whereas tephra is a collective term used to describe all particles ejected from volcanoes irrespective of size, shape and composition (Thorarinsson 1944). Ash plumes from both dome-collapse and fountain-collapse pyroclastic flows are here defined as co-pyrodastic-flow plumes (abbreviated to co-PF plumes), as even if the associated pyroclastic flows are fundamentally different, the process of plume formation by elutriation is similar. Finally, the central vent-derived plume in Vulcanian explosions is termed the vent plume. In the eastern Caribbean, low-level winds (1-5 km) typically blow to the west and intermediate-level winds (8-18 km) blow to the east, with a strong wind shear between 5 and 8 km. Wind profiles were analysed for the period 1992-1997 using data from the National Oceanic and Atmospheric Administration Climate Diagnostic Center (Bonadonna et al. 2002). Therefore most of the fallout tephra was deposited on the west coast of Montserrat. However, the standard deviation for the direction values varies between 30 and 162° (Bonadonna et al. 2002). Low-level wind, therefore, could sometimes disperse tephra to the NW, SW or south of the island, and intermediate-level winds to the north, south, SW, SE, NW and NE of the island. Wind velocity typically ranges between 3 and

10ms –1 for low-level winds and between 4 and 2 3 m s – l for intermediate-level winds (Bonadonna et al. 2002). Due to the small size of the island, a significant amount of tephra fell into the sea. Fallout tephra became an important hazard due to the frequency of tephra-fallout events and to its grain size and compositional characteristics, which adversely affected the quality of life of the people of Montserrat (Baxter et al. 1999). This paper describes the dispersal of the tephra-laden plumes and investigates the formation of tephra-fallout deposits. Grainsize and chemical analyses were carried out. Direct observations of tephra-fallout processes were made, and satellite data provided information on tephra dispersal over the sea and neighbouring islands. The limited size of the island only allowed the proximal tephra-fallout deposits (up to c. 10 km from the dome) to be studied in detail. Very proximal tephra-fallout deposits (within 3 km from the dome) were rarely accessible during the two series of Vulcanian explosions (August-October 1997) for obvious hazard and logistical reasons, and therefore the coarsest Vulcanian fallout tephra fraction (>2cm) could not be analysed as thoroughly as the 1 m to 2cm fraction.

Methods Fallout-tephra collection Starting from June 1996, trays to collect fallout tephra were placed at up to 15 localities around the volcano and measurements were made on a weekly basis through the eruption (Fig. 1). Additional collections were sometimes carried out after large tephra-fallout events. This sampling network was abandoned after the large dome collapse of 25 June 1997 because many localities became too hazardous for routine monitoring. During the August 1997 Vulcanian explosions a permanent network of sampling trays was not in use. However, fallout tephra from an individual Vulcanian explosion (i.e. 5 August 1997, 04:45 local time (LT)) was collected and used

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 483-516. 0435-4052/02/$15 C The Geological Society of London 2002.

483

484

C. BONADONNA ET AL.

1150 to 1370 kg m – 3 , and for fallout tephra produced by Vulcanian explosions from 1000 to 1 2 0 0 k g m – 3 . Therefore mean values of 1200 kg m –3 and 1 l 0 0 k g m – 3 were used for dome-collapse and Vulcanian fallout tephra respectively. From these isopach maps the volumes of tephra-fallout deposits were calculated using the method of Pyle (1989). Volumes of dome-collapse pyroclastic flows, and dome-collapse and Vulcanian tephra-fallout deposits were also converted into dense-rock equivalent (DRE) volumes (as in Table 1). assuming a dense-rock density of 2600 kg m–3 and dome-collapse pyroclastic-flow deposit density of 2000 kg m–3 (Sparks et al. 1998; Calder et al. 2002). The elutriation factor (i.e. volume percentage of a co-PF tephra-fallout deposit compared to the volume of the parent pyroclastic-flow deposit) was calculated on the basis of DRE volumes. Fallout tephra was also collected on adhesive carbon-paper tabs supported on a metallic substrate to enable later study by scanning electron microscope (SEM). Carbon tabs were exposed for 1 minute each in order to preserve individual ash clusters. During two Vulcanian explosions (28 September 1997, 23:03 LT. and 1 October 1997, 11:34 LT) plastic sheets (2m 2 ) were laid out for periods of five to ten minutes during the associated tephra fallout to investigate how accumulation varied with time. These experiments were carried out at 6.8km and 7.4km NW from the dome respectively (sites D and H on Fig. 1).

Analytical methods Petrology. A combination of X-ray diffraction (XRD). X-ray fluorescence (XRF), wet chemistry, electron microprobe and nuclear magnetic resonance (NMR) spectroscopy has been used to determine the mineralogy and chemistry of the fallout tephra. Toxic phases, such as cristobalite, tridymite and quartz, were identified, and mineral abundance was quantified in the <10 m fraction. Some results are described in Baxter et al. (1999) and Murphy et al. (2000). Fig. 1. Distribution of fallout-tephra collection sites on Montserrat for the period 3 June 1996 to 25 June 1997 (solid diamonds) and 10 September to 21 October 1997 (white triangles). Collection sites 1-3 (black numbers) have the most complete collection record of fallout tephra from the period 3 June 1996 to 18 June 1997 (described in Fig. 10). Grey circles with white numbers indicate the eight sections of tephra-fallout deposit described in Figure 15. White circles indicate the sites where some accretionary lapilli from the 26 December 1997 dome collapse were collected for analysis (AL1 and AL2). Dashed lines highlight the position of the original rivers inside the main valleys where pyroclastic flows generated by both dome collapses and fountain collapses have travelled. WR, White River valley; FG, Fort Ghaut: TyG, Tyre's Ghaut; MG, Mosquito Ghaut; TG, Tuitt's Ghaut; WG, White's Ghaut; TR, Tar River valley.

here as a comparison with fallout tephra from the SeptemberOctober Vulcanian explosions. A new network of 17 sampling trays was established in September 1997, covering the central and northern parts of the island (Fig. 1). Between 22 September and 21 October 1997, while there was an average of three Vulcanian explosions per day (Druitt et al. 2002), fallout tephra was collected daily, often within hours of individual explosions. The mass of dry material collected in each tray was weighed in order to calculate the dry mass per unit area. From these data, isomass contour maps were drawn. Isomass contour maps were produced both for individual events and for the accumulation of fallout tephra over extended periods of time resulting from large numbers of events. For tephra fallout occurring after particularly large events, thicknesses of the associated deposits could be measured. However, for smaller events for which it was impractical to measure the small thicknesses, isopach maps were generally drawn by converting data of mass per unit area to thickness using a representative density for the dry deposits. Measured dry densities for tephra-fallout deposits produced by dome collapses ranged from

Grain-size distribution. Samples of fallout tephra were dried and sieved at \o intervals (o = — log2d, where d is the particle diameter, in mm). Grain size of the fine ash (<63 m fraction) was analysed by the electrosensing zone method with an Elzone Celloscope (Muerdter et al. 1981) down to 8 m (7o). Ash samples (<125 m fraction) from the 5 August 1997 (04:45 LT) Vulcanian explosion were analysed using the laser diffraction technique with a Mastersizer Hydro 2000M (equipped with blue light), which can determine the volume percentage of particles from 2000 ( –lo) down to 0.02 m (150). Reflective index, refraction index and density used were 1.63. 0.1 and 2650 kg m–3 respectively. The Sequential Fragmentation Transport Analysis Windows application described in Wohletz et al. (1989) was used to identify lognormal subpopulations and to determine corresponding lognormal size distribution parameters (e.g. mode, weight fraction) and Inman (1952) parameters of median diameter (Md 0 ) and sorting coefficient ( 0 ) from particle-size data. Scanning electron microscopy (SEM). Selected fallout-tephra samples were analysed by SEM to investigate particle morphology and aggregation processes. Energy dispersive spectral (EDS) analyses were done to identify mineral phases. The SEM was also used to study accretionary lapilli. which were first isolated and impregnated with araldite resin. Finally, the SEM. together with Scion Image software, were used to investigate the mineralogical differences between the <63 m fractions of tephra that fell at different times during and following the 1 October 1997 (11:34 LT) Vulcanian explosion. Satellite image analysis. Thermal infrared channels 4 and 5 from the Geostationary Observational Environmental Satellite (GOES8) were used to investigate volcanic clouds and very fine ash. These

TEPHRA FALLOUT AT SOUFRIERE HILLS VOLCANO

channels have a spatial resolution of 4km and their respective wavelength ranges are 10.2-11.2 m (band 4) and 11.5-12.5 m (band 5). In order to distinguish volcanic clouds from surrounding meteorological clouds, the brightness temperature difference (BTD) between bands 4 and 5 was calculated by subtracting the brightness temperature for band 5 from that of band 4. Due to the differential absorption of upwelling radiation, meteorological clouds exhibit a positive BTD, whereas volcanic clouds show a negative BTD (Schneider et al. 1999). The magnitude of the negative brightness value of a volcanic cloud depends on the optical depth, particle size, particle refractive index, and particle size distribution in the cloud, and on the temperature difference between the volcanic cloud and the atmosphere (Prata 1989). In order to determine the mass of silicate particles, the average of the effective radius of the particles, and the optical depth of the cloud from the remote sensing data, a radiative transfer model developed by Wen & Rose (1994) was used. The model uses the GOES-8 band-4-minus-band-5 retrievals by combining radiative transfer calculations with a semi-transparent cloud model. The model assumes that the cloud is c. 1 km thick, homogenous, semi-transparent, parallel to the surface below it, and that the atmosphere above and below the cloud is clear. It is assumed that the particles in the cloud are all andesitic, spherical with a lognormal size distribution, and experience Mie scattering and absorption (Wen & Rose 1994). The Wen & Rose (1994) method was used for the processing of GOES-8 data. However, to apply the method to GOES-8 data, which have thermal bands in a slightly different position from the AVHRR (Advanced Very High Resolution Radiometer), the wavelength-dependent refractive index for andesite determined by Pollack et al. (1973) was also used. The results of mass estimates apply only to particles in the Mie range (typically c. 1-10 m radius). However, retrievals from GOES-8 satellite images can be minimum values for several reasons, such as interference with the infrared radiation to the sensor from lowertroposphere humid conditions and high water/ice content in the ash cloud.

Generation of tephra fallout Phreatic activity The first phase of activity, from July to November 1995, was characterized by numerous phreatic explosions, ranging from discrete

Fig. 2. (a) Co-PF plume generated by the 4 November 1997 dome collapse (1.5 x 10 6 m 3 DRE) (photo by E. S. Calder). (b) Early stage of co-PF plumes just forming on top of a small domecollapse pyroclastic flow down Mosquito Ghaut (photo by E. S. Calder). (c) Co-PF plumes forming on top of a dome-collapse pyroclastic flow down Mosquito Ghaut (25 June 1997, 4.9 x 106m3 DRE) (photo by R. B. Watts). The initial discrete plumes tend to amalgamate as they rise and then detach from the flow itself.

485

explosions to pulsating, dark plumes. Most of this activity produced plumes only a few hundred metres high and little ash was dispersed beyond the confines of English's Crater (old sector collapse scar on the east side of the dome; Fig. 1). However, significantly larger phreatic explosions occurred on 21 August, 31 October, 4 November and 9 November 1995 and generated plumes up to 2.5 km above sea level (a.s.l.). Each ash-laden plume was dark, with a white, steam-rich upper part. The phreatic explosions on 21 August and 31 October 1995 were accompanied by cold pyroclastic surges, which swept into Plymouth causing darkness for 15 to 20 minutes. Each of the largest phreatic explosions laid down thin (a few millimetres) deposits of ash across the Plymouth area. Dome-collapse activity Pyroclastic flows (>0.5 km runout, Calder et al. 2002) and rockfalls (typically <0.5-0.8 km runout) generated by instability of the growing andesitic dome produced co-PF plumes (Fig. 2). Drifting of co-PF plumes across the island was a persistent and major source of tephra fallout throughout the 1995-1999 period. The largest domecollapse pyroclastic flows had runout distances on land up to 6.7 km (Calder et al. 1999; Cole et al. 2002), and some entered the sea. Large dome collapses occurred mostly in 1996 and 1997 (Table 1). Up to March 1997, dome-collapse rockfalls and pyroclastic flows occurred in the Tar River valley (Fig. 1). In April 1997 pyroclastic flows started occurring to the south, down the White River valley, and in May and June to the north down Mosquito and Tuitt's Ghauts. In August 1997, pyroclastic flows also went down Fort Ghaut, reaching Plymouth. The largest dome collapse occurred on 26 December 1997, generating a co-PF plume up to c. 15 km a.s.l.. Most of the fallout tephra produced on 26 December 1997 was dispersed to the SW and fell in the sea (Ritchie et al. 2002). Large collapses also occurred after the lava dome stopped growing in March 1998 (e.g., 3 July and 12 November 1998 dome collapses; Norton et al. 2002). Co-PF plumes ascended to heights ranging from a few hundred metres up to 15 km a.s.l. The plumes formed as the parent pyroclastic flows travelled down the valley. In detail, for a given pyroclastic flow, there were initially multiple plumes, which formed at discrete distances from one another (Fig. 2b). As they rose, the discrete plumes tended to coalesce and formed a single cloud, which often detached completely from the generating pyroclastic flow and was transported downwind (Fig. 2c).

486

C. BONADONNA ET AL.

Fig. 2. (continued)

The volcanic cloud produced by the 6 November 1997 domecollapse pyroclastic flow was observed on GOES-8 satellite images and illustrates some general features of plume dispersal. Figure 3a shows the view of the volcanic cloud and meteorological clouds using GOES-8 channel 4 enhanced so that the volcanic clouds (most negative BTD values) are white and the surrounding meteorological clouds are grey to black. Figure 3b (channel 4 minus channel 5) images the silicate particles in the cloud. Due to interference by lower-tropospheric water vapour in bands 4 and 5, Figure 3b underestimates the area of the ash cloud. The 14:39:14 LT image shows the volcanic cloud early in the event, before it rose to its maximum height (10.8km a.s.l., Table 2). In the 15:09:08 LT and 15:39:05 LT images the cloud splits in two due to the different directions and speeds of wind at different altitudes. At this time intermediate-level winds to the SE were stronger than the low-level winds (Fig. 4a). Therefore the higher cloud was dispersed to the SE and travelled faster than the lower cloud. Both clouds were mostly dispersed over

the sea and did not produce any significant tephra fallout on Montserrat. However, at c. 17:00 LT tephra started to fall at Pointe a Pitre on Guadeloupe (c. 84km SE from Montserrat, indicated in the 14:39:19 LT image of Fig. 3b) and activity at the international airport on Guadeloupe was interrupted for a few hours (J.-C. Komorowski, pers. comm.). Table 2 shows information retrieved from GOES-8 satellite images corresponding to the 6 November 1997 dome collapse, as well as 25 June 1997 and 26 December 1997 dome collapses. The highest co-PF plume was typically generated next to the dome and the plume height tended to decrease along the valley (Fig. 2b and c). However, generation of co-PF plumes is also controlled by topography and vigorous plumes were also produced at breaks in slope, which enhance air entrainment (Cole et al. 1998). When pyroclastic flows entered the sea, co-PF plumes (typically grey and ash-rich) mixed together with plumes from secondary explosions (typically white and steam-rich).

Fig. 3. GOES-8 satellite images of four scenes from the 6 November 1997 (14:29 LT) dome collapse (in Table 2). Montserrat, Guadeloupe, Dominica and the southern part of Antigua are shown. Images for every 30 minutes are available, but those shown highlight the movement of the cloud. (a) Images of channel 4 (wavelength range: 10.3-11.3 m) showing the view of both volcanic (labelled with a red V) and meteorological clouds, which are imaged because they are colder than the subsurface. 14:39:14 LT: the volcanic cloud is located directly over Montserrat and has still not reached its maximum height, as indicated by a warm temperature (>0°C). 15:09:08 LT: the volcanic cloud is larger and moving predominantly to the east. 15:39:05 LT: the volcanic cloud is split into two parts due to the wind shear; the more easterly portion of the cloud is colder and higher. 17:39:05 LT: the volcanic cloud is barely visible, with the largest concentration closest to the volcano. (b) Mass images (particle mass in tonnes pixel –1 ; 1 tonne = 103 kg) constrained by negative brightness temperature difference (BTD) (channel 4 minus channel 5). The mass images show the mass of fine silicate particles (1-10 m radius) determined by their negative BTD, and their two-dimensional position in the drifting clouds. The largest mass of the cloud occurs at 15:09:08 LT. White dot on the 14:39:19 LT image locates Pointe-a-Pitre (in Guadeloupe), where tephra fallout started at about 17:00 LT.

488

C. BONADONNA ET AL. Table 1. Large dome collapses (DRE volume > 2 x 106 m3) during the 1995-1999 period of the Soufriere Hills Volcano eruption Event

PF deposit volume (DRE)* (x10 6 m 3 )

Volume (DRE) of fallout tephra (x10 6 m 3 )

Collapse valley

PF runout§ (km) (observed)

Plume max. height (km)

Tephra dispersal (observed)

SE W n/a W NW W

29 Jul. 96 4 Aug. 96 1 1 Aug. 96 21 Aug. 96 3 Sep. 96 17 Sep. 96 19 Dec. 96 16 Jan. 97

2.3 2.3 2.7 22 2.3 9.5 3.1 2.3

n/a n/a n/a n/a n/a 0.9 n/a n/a

TR TR TR TR TR TR TR TR

>3 3 >3 3 3 >3 3 >3

n/a 3 9 2.5 6 n/a 3 6

20 Jan. 97

2.3

n/a

TR

>3

9

0.8 2.3 0.4 4.9 7.0 11.0 4.6 35-45 19.2 2.3 3.8

0.02 n/a 0.02 n/a n/a n/a n/a 4.5 n/a n/a n/a

WR WR TR MG FG TG, WG WR WR TR TR TR

3.6 4.1 3 6.7 >5.6 >6.0 >5 >5 >3 >5 >3

31 Mar. 971 1 1 Apr. 97 27 May 97 25 Jun. 97 3 Aug. 97 21 Sep. 97 06 Nov. 97 26 Dec. 97 03 Jul. 98 12 Nov. 98 20 Jul. 99

sw

Low SW High SE Low SW High NE NW n/a n/a W WNW NW na SW ENE NW Low W High NW

4 n/a n/a 11.8 7 n/a 10.8 15 14 8 10

* Pyroclastic-flow deposit volumes include surge deposits and were surveyed throughout the eruption by MVO staff. They were estimated using a combination of GPS surveys, maps (calculated areas and thickness estimates), visual observations and simple geometry of the collapse scar (Calder et al. 2002; see text for details on DRE-volume calculations). Tephra-fallout deposit volumes were calculated using the method of Pyle (1989) apart from that of 26 December 1997, which was obtained from analytical calculations (Sparks et al. 2002; see text for details on DRE-volume calculations). \ Collapse valleys are shown on Figure 1. § Values of pyroclastic-flow runout preceded by '>' indicate that the flow entered the sea. The runout for 31 March 1997 refers to flows shed down the White River valley in the morning of that day. || Plume maximum heights were estimated from direct observations, with the exception of those of 25 June, 6 November and 26 December 1997. which were calculated from satellite data. Dome collapses with DRE volume <2 x 10 6 m 3 , used for elutriation-factor calculations. n/a, Data not available.

Table 2. Results of retrievals from satellite images of three dome collapses Dome collapse (h:min LT)

First image* (h:min:s LT)

25.06.97 (12:55)

13:09:06

06.11.97 (14:29)

15:09:08

26.12.97 (03:00)

03:09:05

Max. image (h:min:s LT)

14:09:05

15:39:07

10:39:06

Cloud top/bottom temp. (K)

Height (km)

Ash clouds total pixels

Ash cloud area§ (km 2 )

Mean effective radius ( m)

Pixel mean optical depth

Pixel mean mass (x10 6 kg)

213/280

11.8

123

1968

8.3

2.41

0.4

5.1

217/280 213/288

10.8

314 93

5024 1488

8.4 7.6

2.04 1.85

0.3 0.3

10.3 2.7

217/290 200/290

14.9

125 35

2000 560

7.4 7.4

1.20 2.31

0.2 0.3

2.3 1.2

479

7664

7.7

0.40

0.1

3.1

200/290

Total ash mass** (x107kg)

LT, local time. * First image available for the eruptive plume. Image with the greatest volcanic-cloud area. Cloud height is determined using the cloud-top temperature of the first image, as this is the most opaque, and comparing this value with radiosonde data from Guadeloupe at the nearest available time (within usually 12h). § Ash-cloud area is determined by multiplying number of pixels observed in each cloud by the resolution for each squared pixel (16km 2 ). || Size parameter that reflects the integral of the value of a size distribution of particles divided by its area (Wen & Rose 1994). Pixel mean optical depth represents logio(opacity) (Campbell 1987), and pixel mean mass is the silicate particle mass associated with each pixel. Both parameters are calculated using refractive index data from Pollack et al. (1973) for andesite. ** Total ash mass is obtained by multiplying pixel mean mass by the total number of pixels in the cloud. Values of total ash mass are minimum estimates because the humid conditions in the tropics interfere with the radiative transfer of infrared wavelengths to the sensor. Images for the 6 November 1997 dome collapse are shown on Figure 3.

TEPHRA FALLOUT AT SOUFRIERE HILLS VOLCANO

Fig. 4. Wind profiles for (a) 6 November 1997, showing a significant wind shear at about 3 km a.s.l., and (b) for 26 September 1997, showing constant wind direction to the north above 6km a.s.l. and uniform wind velocity. Data from Bonadonna et al. (2002).

Magmatic explosive activity Magmatic explosive activity resulted in a sub-Plinian phase in September 1996 (Robertson et al. 1998) and in two series of repetitive Vulcanian explosions during August to October 1997

Fig. 5. Vulcanian explosion of 5 August 1997 (16:57 LT). Photo taken about 5.5km NW from the volcano a few minutes after the beginning of the explosion. It shows the rising convective plume, fountain-collapse pyroclastic flows and initial elutriation from pyroclastic flows (photo by C. Bonadonna).

489

(Druitt et al 2002) (Fig. 5). The sub-Plinian explosion (column height up to c. 15km a.s.l.) occurred at 23:42 LT on 17 September 1996, following 9 hours of dome-collapse activity. Sub-Plinian discharge lasted about 40 minutes and caused tephra fallout mainly in the south of the island, with the ash reaching Guadeloupe, where the international airport had to close because of significant ash covering the runway markings. Thirteen Vulcanian explosions occurred between 4 and 12 August 1997, with a frequency of about two per day (Druitt et al. 2002). Five of these produced significant tephra fallout on the north of the island. Seventy-five explosions occurred between 22 September and 21 October 1997 (Druitt et al. 2002), of which 27 resulted in tephra fallout over the north of the island. All but one of these 88 explosions generated pyroclastic flows by fountain collapse. Each series of vertically directed explosive activity followed unloading of the conduit after major dome collapses: near-continuous dome-collapse flow generation during 17 September 1996 (9.5 x 10 6 m 3 ORE), a 7 x 10 6 m 3 DRE dome collapse down Fort Ghaut into Plymouth on 3 August 1997, and a 11 x 10 6 m 3 DRE dome collapse down Tuitt's Ghaut on 21 September 1997 (Calder et al. 2002). However, not every large dome collapse was followed by significant explosive activity (e.g. 26 December 1997 dome collapse: 46 x 10 6 m 3 DRE, Calder et al. 2002). The two series of Vulcanian explosions in 1997 were characterized by formation of a vertical eruption column and, in all but one case, by a collapsing fountain in the first 10-20 seconds of the event, which generated pyroclastic flows down all or some of the six major valleys around the volcano (Tar River, White River, Fort Ghaut, Tyre's Ghaut, Mosquito Ghaut and Tuitt's Ghaut; Fig. 1). The eruption columns were typically several kilometres high, reaching up to 15km a.s.l. in the strongest explosions. Fountaincollapse runout distances were observed up to 6 km from the dome (Cole et al. 1998; Calder et al. 1999; Druitt et al. 2002). Sequential pulses would coalesce together, forming a rising convecting plume above the vent. Co-PF plumes would also form above the associated fountain-collapse pyroclastic flows. Wind shear between 5 and 8 km was often observed, manifested by a splitting in the high volcanic cloud. However, during some explosions wind direction was constant enough to disperse the volcanic cloud uniformly. Figure 6 shows the GOES-8 image of the volcanic cloud produced by the explosion on 26 September 1997 (14:56 LT) dispersed NW by the wind, which that day did not vary much in either direction or velocity (Fig. 4b). This explosion resulted in heavy tephra fallout on Montserrat (Fig. 7). However, the volcanic cloud took at least 5 hours to lose most of the ash (Fig. 6). Table 3 shows the result

490

C. BONADONNA ET AL.

Fig. 7. Isopach map (in mm) for fallout tephra produced by the 26 September 1997 (14:56 LT) Vulcanian explosion (described in Table 3). The position of Soufriere Hills Volcano is also shown (black triangle). from the retrievals from GOES-8 satellite images of the 26 September 1997 (14:56 LT) explosion, as well as of the 7 August 1997 (12:05 LT) and 27 September 1997 (09:46 LT) explosions. The Vulcanian explosions involved three main types of tephra fallout (Druitt et al. 2002): (i) ballistic ejecta, (ii) tephra fallout from the vent plume and umbrella cloud and (iii) ash from co-PF plumes above fountain-collapse pyroclastic flows. The Vulcanian co-PF Fig. 6. GOES-8 satellite images of six scenes from the 26 September 1997 (14:56 LT) Vulcanian explosion (in Table 3). Montserrat. the northern part of Guadeloupe. Antigua. Barbuda. Nevis. St Kitts. St Eustatius. St Barts and the southern part of Anguilla are shown. Images for every 30 minutes are available, but those shown highlight the movement of the cloud, (a) Images of channel 4 (wavelength range: 10.3-11.3 m) showing the view of both volcanic (labelled with a red V) and meteorological clouds, which are imaged because they are colder than the subsurface. 15:09:05 LT: the volcanic cloud is over Montserrat and the meteorological clouds to the north. 15:38:59 LT: the volcanic cloud has spread over Montserrat. travelling NW. 16:39:04 and 17:39:02 LT: the volcanic cloud is gradually spreading over Nevis and St Kitts. 18:39:07 and 19:39:07 LT: the cloud dissipates over time and travels to the NW. 19:39:07 LT: The volcanic cloud is barely visible. (b) Mass images (particle mass in tonnes pixel - 1 : 1 tonne = 103 kg) constrained by negative brightness temperature difference (BTD) (channel 4 minus channel 5). The mass images show the mass of fine silicate particles (1-10/ m radius) determined by their negative BTD, and their two-dimensional position in the drifting clouds.

TEPHRA FALLOUT AT SOUFRIERE HILLS VOLCANO

491

Table 3. Results from retrievals from satellite images of three Vulcanian explosions Explosion (h:min LT)

First image (h:min:s LT)

07.08.97 (12:05)

15:09:04

26.09.97 (14:56)

15:09:05

27.09.97 (09:46)

10:09:06

Max. image (h:min:s LT)

16:09:05

15:38:09

13:09:03

Cloud top/bottom temp. (K)

Height (km)

Ash clouds total pixels

Ash cloud area (km 2 )

Mean effective radius (m)

Pixel mean optical depth

Pixel mean mass (x10 6 kg)

Total ash mass (x10 7 kg)

213/283

9.3

85

1360

6.5

0.66

0.1

0.8

213/290 217/280

11.3

319 82

5104 1312

6.7 7.6

0.56 2.58

0.1 0.4

2.5 3.3

217/280 213/284

10.8

104 75

1664 1200

8.6 6.4

2.89 1.98

0.5 0.3

5.4 2.0

211

3376

7.5

0.66

0.1

2.2

213/288

Parameters are explained in caption of Table 2. Images for the 26 September 1997 (14:56 LT) Vulcanian explosion are shown on Figure 6.

plumes reached heights up to c. 3 km a.s.l. Ash from these plumes was dispersed by low-level winds, tephra fallout occurring contemporaneously with, or preceding, that from the vent plume. The transition from one type of tephra fallout to the other did not vary systematically with distance from vent. At a given location tephra fallout was sometimes observed to stop for a few seconds and start again. To give an example of tephra-fallout duration, tephra fallout observed at MVO South (c. 1 km WNW of the crater; site D on Fig. 1) started between 10 and 45 minutes after each explosion and lasted 10 to 40 minutes. Sometimes the falling tephra was also observed to change colour with time, from pinkish in the early tephra fallout to grey. Heavy rain often followed tephra fallout.

Ash-venting Direct discharge of ash from fractures and vents in the dome has occurred throughout the 1995-1999 eruptive period (Fig. 8). Ash-venting plumes were typically small (<3km) in the first few

Fig. 8. Vigorous ash-venting occurring on 19 August 1996. Photograph taken about 5 km NE from the volcano (photo by R. B. Watts).

years of the eruption (up to 1998), and the volumetric contribution of this activity was small in comparison with the production of fallout tephra by dome collapses and Vulcanian explosions. However, the volumetric contribution of ash-venting became much more significant after the Vulcanian explosions stopped in October 1997 and the dome started growing again at a gradually accelerating rate (up to 9m3 s - 1 ; Watts et al. 2002). Ash-venting temporarily ceased when the dome growth stopped in March 1998. After a major collapse of the dome on 3 July 1998, a system of vents developed within the resulting collapse depression, and provided a source of considerable ash-venting. During December 1998 to October 1999, vigorous ash-venting episodes, with column heights up to 6km a.s.l., were commonly initiated with small explosions and could last for several hours (Norton et al. 2002). These small explosions produced convective plumes up to 6km a.s.l., and a few of the largest explosions generated pyroclastic flows by fountain collapse (Norton et al. 2002). The associated tephra-fallout deposits were unsubstantial and were mostly confined in English's Crater (Fig. 1), where a small ash cone built around the main vent.

492

C. BONADONNA ET AL.

Fig. 9. Examples of tephra-fallout structures from volcanic plumes, (a) Ash cloud from a fountain-collapse pyroclastic flow dispersed towards the west of the island, showing wavy, ash-rich vertical filaments, (b and c) Tephra fallout from the umbrella cloud produced by Vulcanian explosions in September 1997 (photos by C. Bonadonna): (b) a typical triangular cascade of material from the umbrella cloud is visible: (c) lateral outbursts can be seen which often develop when vertical protrusions touch the ground (e.g. protrusion A has formed a horizontal gravity current upon reaching the ground). A mid-air protrusion is also evident in the middle of the photograph (protrusion B). Three finger-like protrusions are also shown (protrusions C D and E) which have coalesced as they touch the ground.

Structures in plumes Tephra fallout was typically not uniform from the volcanic plumes. Two distinct types of structures were recognized. One was associated with low co-PF plumes (from both dome collapses and fountain collapses), where the base remained close to the ground (Fig. 9a). The other was associated with clouds from high plumes, which rose well clear of the ground (Fig. 9b and c). In the first case, downward-moving particle-rich streaks were observed within 2-3 minutes of the onset of tephra fallout (Fig. 9a). These structures are reminiscent of the fibrous streaks described in ice-rich cirriform clouds (Houze 1993). In the second case, the lower margin of the drifting plume developed a series of discrete and spaced vertical protrusions (Fig. 9b), which evolved into laterally directed gravity

flows upon reaching the ground. Sometimes the heads of these vertical protrusions coalesced (protrusions C D and E in Fig. 9c). These are reminiscent of fingering convective instabilities observed in experiments where vertical protrusions form at the fluid density interface (Carey 1997; Hoyal et ai 1999). Convective motions at the base of a sediment-laden cloud transport particles to the ground at a higher rate than their individual fall velocities (Hoyal et al. 1999). These structures (e.g. protrusion A in Fig. 9c) resemble meteorological microbursts characterized by a strong downdraft at their centre and strong divergent winds when they hit the ground (Houze 1993). Meteorological bursts are caused by evaporation and melting (Houze 1993). In volcanic plumes on Montserrat the principal causes of instability are likely to be (i) the presence of particles and (ii) evaporation.

TEPHRA FALLOUT AT SOUFRIERE HILLS VOLCANO

Fallout-tephra distribution

493

uced by co-PF plumes associated with the largest dome collapses (Table 1). Numerous rockfalls and small pyroclastic flows made

Fallout tephrafrom dome collapses (June 1996-June 1997) Figure l0a shows cumulative mass of fallout tephra for three proximal localities using the most complete collection record for the period from 3 June 1996 to 18 June 1997. The three-month gap in the data set, following the sub-Plinian activity of 17 September 1996, represents a period of sustained dome growth with minor rockfall activity and no major pyroclastic flows (Watts et al. 2002). The main contribution to tephra-fallout deposits in this period was prod-

only minor contributions to the total mass of fallout tephra. Figure 1 la shows the total accumulation of fallout tephra from 3 June 1996 to 25 June 1997 (main collapses in this period in Table 1). The map indicates the dominance of low-level winds to the west in controlling dispersal of tephra fallout from dome-collapse pyroclastic flows. Examples of tephra-fallout deposits produced by single dome collapses of different sizes during this period (Table 1), together with distributions of the corresponding pyroclastic-flow deposits, are given in Figure 12. The isopach map for the 17 September 1996

Fig. 10. Cumulative mass of fallout tephra (kgm - 2) for three localities with the most complete collection record, during the periods (a) 3 June 1996 to 18 June 1997 (sites 1, 2 and 3 on Fig. 1), and (b) 21 September to 10 October 1997 (sites D Q and O on Fig. 1). Grey boxes indicate starting and finishing dates of fallouttephra collection. White boxes indicate: (a) the dome collapses that contributed the most to accumulation of fallout tephra; (b) some tephra-collection dates. Most of the dome-collapse events indicated in (a) are described in Table 1, apart from those of March-April and May 1997, which represent a semi-continuous dome-collapse activity of small flows. During the period 30 March-2 April 1997 several dome-collapse pyroclastic flows travelled down White River valley with a maximum runout of c. 4km and a total volume of c. 4 x 10 6 m 3 (Calder et al. 2002). On 16 May 1997 a few moderate dome-collapse pyroclastic flows travelled down Tar River valley with a maximum runout of c. 3 km (volume estimates for these flows are not available).

494

C. BONADONNA ET AL.

Fig. 11. Isopach maps (in cm) of the total accumulated fallout tephra: (a) between 3 June 1996 and 25 June 1997 (mainly fallout tephra from dome-collapse activity, with main dome collapses during this period being 29 July 1996, 11 August 1996, 17 September 1996, 16 January 1997, 20 January 1997. 25 June 1997, in Table 1); and (b) between 4 August and 21 October 1997 (mainly fallout tephra from Vulcanian explosions). The position of Soufriere Hills Volcano is also shown (black triangle). Between 3 June 1996 and 25 June 1997 there was no significant tephra fallout in the very north and west of the island. and therefore fallout-tephra collection could not be carried out in this area (Fig. 1 la).

event (Fig. 12a) mainly represents fallout tephra from the 9-hour continuous dome-collapse activity, with a subordinate contribution from the sub-Plinian explosive phase. The low part of the volcanic cloud from the sub-Plinian phase travelled west at around 7 m s - 1 , but the highest part was transported ESE at around 18 m s - 1 , reaching Guadeloupe. Isopach maps for single dome collapses (Fig. 12) show two interesting features. First, the maximum thickness is centred neither on the dome nor on the valley of the collapse, but is shifted away from the source in the downwind direction (Fig. 12a and b). Second, for those cases where the wind direction was approximately normal to the valley axis (Fig. 12b and d), the isopach contours are focused on the dome source rather than the whole valley. This indicates that plumes generated near the dome give the dominant contribution to the whole tephra fallout, even though

co-PF plumes are distributed along the valley. This interpretation is consistent with direct observations of the most vigorous co-PF plume being close to the dome.

Vulcanian explosions (August-October

1997)

Figure l0b shows the accumulation of fallout tephra throughout most of the September-October 1997 explosion series (21 September-10 October 1997). During this period of activity the plume height ranged between 3 and 15km a.s.1, with an average of 6km a.s.l. Sites WNW, NNW and NNE of the dome (D, Q and O on Fig. 1) are compared to show the effect of wind dispersal. Site D (WNW of the dome) accumulated fallout tephra with a near-linear

Fig. 12. Isopach maps (in mm) for fallout tephra produced by dome collapses of different sizes described in Table 1: (a) 17 September 1996 (9.5 x 106 m3 DRE): (b)31 March 1997 (0.8 x 10 6 m 3 DRE); (c) 27 May 1997 (0.4 x 10 6 m 3 DRE); (d) 25 June 1997 (4.9 x 10 6 m 3 DRE). Associated pyroclastic-flow deposits are also shown. Circles in (b) (indicated by a b, c and d) are locations of samples for grain-size distribution data on Figure 17. The position of Soufriere Hills Volcano is also shown (black triangle).

TEPHRA FALLOUT AT SOUFRIERE HILLS VOLCANO

495

496

C. BONADONNA ET AL.

trend for much of the period, whereas Q (NNW of the dome) and O (NNE of the dome) accumulated fallout tephra more episodically. The areas to the NNW of the dome were affected by significant tephra fallout only when wind conditions were unusual and plume heights were 3 km a.s.l., whereas the areas to the NE of the dome accumulated significant fallout tephra only when wind conditions were unusual and plume heights were 7km a.s.l. A cumulative isopach map for the period 4 August-21 October 1997 was drawn by integrating data from the August-October 1997 collection and from field sampling carried out in late November 1997 (Fig. 1 1b). The map takes into account fallout tephra from the first Vulcanian series (4-12 August 1997). from a large dome collapse (21 September 1997; Table 1), from some small dome collapses (30 August, 10, 12, 15 and 17 September 1997; volume 10 6 m 3 ) and from the second Vulcanian explosion series (22 September21 October 1997). The main dispersal towards the west is evident. In contrast to the dome-collapse maps (Figs 1 la and 12), significant dispersal of Vulcanian tephra occurred to the east and to the north (Figs 7 and l1b). An example of the deposit from a single Vulcanian explosion is also given in Figure 7 for the 26 September 1997 (14:56 LT) Vulcanian explosion.

Volume estimates The Pyle (1989) method for the volume calculations of tephrafallout deposits was applied to both individual and cumulative isopach maps (Fig. 13). Semi-log plots of thickness versus square root of area for amalgamated deposits (Fig. 13a) and individual deposits (Fig. 13b) exhibit approximately an exponential thinning, although only three or four isopach contours can be reasonably drawn through the data. The map for the 17 September 1996 tephra-fallout deposit is the most complete and supports application of Pyle's (1989) method. The method can only give a minimum volume, as the exponential thinning can break down at large distances for small particles (Bonadonna et al. 1998), and there are no data for tephra that fell out to sea. Later we present and discuss

satellite data for the 26 September 1997 (14:56 LT) Vulcanian explosion and the 26 December 1997 dome collapse, which confirm that substantial amounts of very fine ash are transported well beyond the regions for tephra fallout predicted using Pyle's (1989) method, demonstrating that the exponential thinning law must break down. A minimum volume of tephra-fallout deposits for the period of 3 June 1996 to 25 June 1997 (i.e. mostly dome collapses) is 5 x 10 6 m 3 (2.3 x 10 6 m 3 DRE). The volume (DRE) of pyroclasticflow deposits for the same period is c. 44.3 x 10 6 m 3 (from data in Calder et al. 2002). Therefore the elutriated material represents at least 5% of the total dome-collapse volume for the same period. Individual dome collapses give minimum values ranging between 4 and 5% of the corresponding pyroclastic-flow deposit (Table 1). For the 17 September 1996 dome collapse. a value of 9% was obtained, but the isopach map used (Fig. 12a) includes fallout tephra from the 9-hour period of continuous dome collapse as well as fallout tephra related to the sub-Plinian explosive phase. Analytical estimates of fallout tephra from the 26 December 1997 dome collapse (Sparks et al. 2002) give a volume of 4.5 x 10 6 m 3 (DRE), which is 10-13% of the parent pyroclastic-density-current deposit (i.e. 35-45 x 10 6 m 3 DRE; Sparks et al. 2002). The minimum volume of tephra-fallout deposit produced during the period of 4 August to 21 October 1997 is c. 22 x 10 6 m 3 (9.3 x 10 6 m 3 DRE). Most of this tephra is related to Vulcanian explosions and less than 10 6 m 3 to dome collapses in August-September 1997 (30 August. 10, 12. 15. 17 and 21 September). Dividing 21 x 10 6 m 3 by 88 Vulcanian explosions gives an average volume of fallout tephra generated per explosion of 2.4 x 105 m3 (1.0 x 105 m3 DRE). Sequential sampling during Vulcanian explosions (discussed later) indicates that about 10% of the total fallout tephra is deposited by co-PF plumes. A minimum estimate of total co-PF fallout tephra produced during the August-October 1997 explosions is therefore 0.1 x 10 5 m 3 (DRE volume). This represents about 5% of the total volume of pumice-and-ash pyroclastic flow s produced during the August-October 1997 explosions (i.e. 1.9 x 105m3 DRE; Druitt et al. 2002). The minimum volume of fallout tephra calculated for the large Vulcanian explosion on 26 September 1997 (14:56 LT) is 5 x 10 5 m 3 (c. 2x 10 5 m 3 DRE) (Fig. 13b). Given that the volume of fallout tephra (vent-plume fallout + co-PF fallout) represents about 37% of the total volume involved in a Vulcanian explosion (Druitt et al. 2002). the total volume of the 26 September 1997 (14:56 LT) explosion is about 5 x l05 m3 DRE.

Characterization of fallout tephra Chemical and mineralogical compositions

Fig. 13. Thickness versus square root of area semi-log plots for (a) fallout tephra from 3 June 1996 to 25 June 1997 and 4 August to 21 October 1997 (cumulative isopach maps on Fig. 11) and (b) fallout tephra from 17 September 1996 dome collapse, 31 March 1997 dome collapse, 27 May 1997 dome collapse, and 26 September 1997 (14:56 LT) Vulcanian explosion (on Figs 7 and 12).

The petrological characterization of the andesitic lava and tephra from dome collapses and explosions is presented by Devine et al. (1998), Baxter et al. (1999) and Murphy et al. (2000). The Soufriere Hills lava dome consists of porphyritic andesitic lava with 35-45 wt% phenocrysts and about 15-20wt% microphenocrysts. Plagioclase, hornblende, orthopyroxene. Fe-Ti oxides and minor quartz constitute the phenocryst assemblage, and plagioclase. orthopyroxene. clinopyroxene. Fe-Ti oxides, rhyolitic glass and silica minerals make up the microphenocrysts and groundmass assemblage (Devine et al. 1998; Murphy et al. 2000). Fallout tephra generated by dome collapses and Vulcanian explosions differ in several aspects, reflecting the different fragmentation and eruptive histories of these two kinds of fallout tephra. For example, the <10 m ash generated by co-PF plumes from dome collapses contains 10-24% crystalline silica, whereas Vulcanian ash contains 3-6% (Baxter et al. 1999). The dominant silica polymorph is cristobalite. but there are also lesser amounts of tridymite and quartz in some samples. These differences are due to the fact that magma erupted explosively did not have time to form much cristobalite by devitrification of glass and precipitation of silica from hot gases. The cristobalite formed

TEPHRA FALLOUT AT SOUFRIERE HILLS VOLCANO

over several days to weeks within the dome and consequently the dome-collapse pyroclastic flow deposits and associated fallout tephra are enriched in cristobalite. This mineral is further concentrated in the respirable ash by selective crushing within the flows (Baxter et al 1999). Fallout tephra produced during Vulcanian explosions also contrasts in character with that generated from dome collapses, as it is composed both of fallout tephra from the vent plume and that elutriated from fountain-collapse pyroclastic flows. The <63 m fractions for six sequential 5-minute periods of tephra fallout from the 1 October 1997 (11:34 LT) Vulcanian explosion (collected at site H, Fig. 1) were analysed using the SEM (Fig. 14a) to investigate mineralogical variations with time during an individual tephra fallout. The <63 m fraction of a fallout-tephra sample from the 12 September 1997 dome collapse, collected at site I (on Fig. 1), is shown for comparison. Modal phase abundances are relatively constant in time for the Vulcanian tephra fallout. The dome-collapse fallout tephra shows a significantly higher content of crystalline silica compared to all the six periods of Vulcanian tephra fallout from the 1 October 1997 (11:34 LT) explosion (Fig. 14a). The same analysis was carried out on accretionary lapilli collected during the first 5 minutes of the 28 September 1997 (23:03 LT) explosion,

Fig. 14. (a) Variation of mineral and glass contents in the <63 m fraction for the 1 October 1997 (11:34 LT) Vulcanian explosion. Fallout tephra was sequentially collected during 5- or 10-minute periods at site H (on Fig. 1). Comparison with the <63 m fraction of a dome-collapse fallout tephra (12 September 1997) collected at site H (on Fig. 1) is also shown, (b) Averaged mineral and glass contents of accretionary lapilli from the first 5 minutes of tephra fallout produced by the 28 September 1997 (23:03 LT) explosion, collected at site D (on Fig. 1). Analysis was carried out on SEM backscattered images with Scion Image software. Proportion (%) is calculated as percentage of pixels over the total area (mt, magnetite; px, pyroxene; hbl, hornblende; pl, plagioclase).

497

yielding similar results: the most abundant phase is plagioclase, with subsidiary amounts of magnetite, pyroxene, hornblende, glass and silica (Fig. 14b). This observation indicates that the formation of accretionary lapilli was not a mineral-selective process.

Characteristics of the tephra-fallout deposits In May 1999, sections of tephra-fallout deposits in the west of the island were investigated. Preserved layers with thickness of >2cm were found only in a narrow arc between the Belham River valley and Plymouth (locations on Fig. 1). An abrupt transition zone occurred between Salem and the Belham River valley, where deposits increased in thickness from 1 to 15cm in less than 1 km distance. Between the Belham River valley and Plymouth, coinciding with the zone of maximum accumulation (Fig. 11), tephra-fallout deposits measured ranged from 12 to 43 cm thick. Tephra-fallout deposits from the 1995-1999 eruptive period are the amalgamated products of many individual events, most having been too small to generate discrete recognizable layers. The deposits are commonly massive and consist predominantly of fine ash with scattered pumice and

498

C. BONADONNA ET AL.

lithic lapilli. In well-preserved sections, however, layering is apparent (Figs 15 and 16), and major episodes of different styles of volcanic activity can be distinguished by variations in grain size and component characteristics. Four distinct lithofacies are identified (I-IV), representing an eruptive period dominated by a particular style of activity, as now described. (I) A pale grey, crystal-rich ash and fine lithic lapilli (maximum lithic clast: 6 mm) with no vesicular material are found at the tops of six sections (Fig. 16a). These layers, 0.5-1.5 cm in thickness, constitute the upper 3-12% of each section and are interpreted as being predominantly fallout tephra from 1998-1999 ash-venting and explosions (Norton el al 2002). (II) A lithofacies with a friable, porous texture comprises predominantly white to cream-coloured crystal-rich ash and pumice and lithic lapilli (maximum pumice clast: 5 cm; maximum lithic clast:

4 mm). All sections analysed contain one or two layers of this lithofacies. 1-1 0cm thick, constituting a total of 20-49% of each section. They are interpreted as tephra fallout during the two successive series of Vulcanian explosions in August and September October 1997. Both layers are similar and the parent explosive series could only be inferred from their respective positions within sections (Fig. 16a). Within these layers, sublayering relating to individual Vulcanian explosions could not be discerned by grain-size or colour variations. As an example, in section 4 (Fig. 16b). tephra-fallout deposits are intercalated with pyroclastic-flow deposits (3 August 1997 pyroclastic-surge and September October 1997 pumice-flow deposits). A 1-cm-thick layer of pumice-rich fallout tephra is preserved 9cm below the 3 August 1997 pyroclastic-surge deposit, and may represent the only preserved fallout tephra from the 17 September 1996 explosion.

Fig. 15. Eight sections through tephra-fallout deposit (locations marked on Fig. 1). Individual lithofacies types (I-IV) and subtypes (Illa-c) are indicated in the key. A lithofacies is defined by distinctive features representing an eruptive period dominated by a certain style of activity (i.e. ash-venting (I). Vulcanian explosions (II), and dome collapses (III)). With the exception of the 26 December 1997 collapse event (IIIa), these do not represent individual events in the eruption chronology. Lithofacies IV represents pervasively reworked, dominantly fine-grained ash inferred to be mainly dome collapse in origin. Section 4 is located within the Fort Ghaut drainage system (Fig. 1), and thus represents a thick accumulation rich in reworked fallout tephra and intercalated with pyroclastic-flow deposits. X marks the position of the 3 August 1997 pyroclastic-surge deposit (60cm). and Y marks the position of a pumice-flow deposit 1.4m thick. These two pyroclastic-flow horizons help to identify the origin of the overlying (August 1997) and underlying (September October 1997) Vulcanian tephra-fallout deposits respectively.

TEPHRA FALLOUT AT SOUFRIERE HILLS VOLCANO

Fig. 16. (a) Key section (section 3 on Figs 1 and 15) showing different lithofacies types (I-IV) and subtypes (IIIa-c) defined for tephra-fallout deposits accumulated during the 1995-1999 eruptive period of Soufriere Hills Volcano. This picture shows a surface rich in lithic lapilli from the ash-venting and explosions that occurred in 1998-1999 (ongoing at the time of fieldwork) (I), underlying fine-ash layers (IIIa, b and c), and a c. 10cm layer of friable crystal-rich ash and pumice lapilli from Vulcanian explosions (II). The basal portion comprises predominantly primary (IIIb) and reworked fine ash (IV) separated by a layer of Vulcanian fallout tephra (II) (c. 2cm thick), (b) Key section (section 4 on Figs 1 and 15) showing the intercalation of layers of fallout tephra with pyroclastic-flow deposits (3 August 1997 pyroclastic-surge (X) and September/October 1997 pumice-flow (Y) deposits). The section between the September/October 1997 pumiceflow deposit and the underlying 3 August 1997 pyroclastic-surge deposit comprises layers of fallout tephra from Vulcanian explosions (II) and reworked horizons (IV). See also sections in Fig. 15 for scale.

499

500

C. BONADONNA ET AL.

(III) Fine-grained lithofacies vary in colour from pale grey to pink, and are interpreted as accumulated fallout tephra from dome collapses. Based on textural characteristics, three subtypes could also be defined. IIIa is a relatively thick (0.2-7.2 cm) layer rich in accretionary lapilli, and compacted lapilli with irregular pore spaces. This subtype is identified as the individual tephra-fallout deposit of the 26 December 1997 dome collapse and forms a distinct marker horizon in each section. IIIb is a thinly bedded primary ash with subtle variations in colour between different millimetric layers. The variably coloured laminations are attributed to collapses of variably oxidized dome material, or to variations of post-emplacement oxidation. Coloured layers are not traceable laterally through the different sections and cannot be linked to specific events. These layers occur in greatest thickness in the lower parts of sections, but similar layers occur throughout. IIIc is compacted, partially reworked fine ash, often containing abundant round to irregular-shaped air bubbles and cavities. (IV) Pervasively reworked layers are common in upper portions of sections and are characterized by fine to medium-grained ash with thin beds, well-sorted horizons and examples of dune bedding. Occasionally pumice and/or lithic lapilli and other mixed-source exotic clasts occur. Two sections were capped by reworked finegrained ash (sections 1 and 4), probably derived from the underlying fallout of 26 December 1997 (IIIa). Discrete lithofacies mainly represent different styles of volcanic activity (i.e. ash-venting (I), Vulcanian explosions (II) and dome collapses (III)). The interplay between tephra accumulation rate and erosion at a given location determines the preserved thickness of each lithofacies type. Well-preserved lithofacies are primarily those of recent or large events or those that represent periods where the deposition rate was high (e.g. during the two series of Vulcanian explosions in 1997). With the exception of the 26 December 1997 collapse, individual dome collapses or explosive events cannot be identified. In the tropical climate, extensive erosion and reworking continuously affect the accumulating deposits, so that reworked horizons are common (i.e. IIIc and IV) and preserved thicknesses do not fully represent the original deposited fallout tephra.

Grain-size distribution Fallout tephra from dome collapses Grain-size distributions of fallout tephra from dome collapses were measured on samples collected for individual dome collapses only. Fallout tephra collected on a weekly basis represents accumulation of tephra deposited from various different wind profiles, plume heights and eruptive events. Grain-size data from deposits of several events are less easy to interpret in terms of the grain-size of individual plumes and for correlation of grain-size characteristics with transport conditions. Figure 17a-c shows the grain-size distributions for some of the samples collected at different distances from the vent after tephra fallout generated by two small dome collapses (31 March 1997. 0.8 x 106 m3 DRE; and 12 September 1997. 0.4 x 106 m3 DRE) and a large dome collapse (21 September 1997. 11 x 106 m3 DRE). Grainsize distributions on Figure 17b and c are predominantly unimodal with only minor subpopulations. Grain-size distributions for the 31 March 1997 dome collapse (Fig. 17a) are predominantly unimodal for samples beyond 5 km and bimodal for samples within 5 km. Figure 17d shows the total grain-size distributions of fallout tephra from the dome collapses on Figure 17a-c calculated by averaging all samples weighted by the thickness of the associated tephra-fallout deposit. The grain-size parameters are summarized in Table 4. We have identified three separate subpopulation modes in the dome-collapse samples (Table 4) using the Windows application described in Wohletz et al. (1989) (lognormal size distribution parameters). The fine subpopulation (Mode 1 in Table 4) dominates fallout tephra from the large dome collapse of 21 September 1997, whereas the smaller collapses are dominated by the intermediate subpopulation (Mode 2 in Table 4). Fine ash (<63 m) constitutes 40-45 wt% of the fallout tephra from small collapses and 70wt% of that from the large collapse of 21 September 1997. Fallout tephra from dome collapses, collected at individual sites 2-10 km from the vent, are characterized by Md0 between 2.9 and 5o and sorting between 0.9 and 2.2 (Fig. 18a). Md0 and sorting for

Fig. 17. Grain-size distributions of fallout tephra associated with dome collapses and collected at different distances from vent: (a) 31 March 1997 (0.6 x 10 6 m 3 DRE); (b) 12 September 1997 (0.4 x 10 6 m 3 DRE); (c) 21 September 1997 (11.0 x 10 6 m 3 DRE). (d) Total weighted average for the same events. Sample sites are located in Figure 1 for 12 September 1997 and 21 September 1997. and in Figure 12b for 31 March 1997. Distances of sample sites from the dome are indicated in brackets, o = — log2d. where d is the particle diameter (in mm).

TEPHRA FALLOUT AT SOUFRIERE HILLS VOLCANO Table 4. Main grain-size parameters of fallout tephrafrom three dome collapses

12 Sep. 97 Dome-collapse volume (DRE)(xl0 6 m 3 ) Collapse valley Number of sample sites Distance from vent (km) Wt% fine ash (<63 m) Median diameter, Md Sorting coefficient, Mode 1 0

Wt% Mode 2 0

Wt% Mode 3 0

Wt%

0.4

31 Mar. 97 0.8

501

The <63 m (>40) fraction (i.e. fine ash) typically varies between 30 and 80wt%, with only three samples having this fraction less than 30wt%.

21 Sep. 97

11.0

Fallout tephrafrom Vulcanian explosions Tuitt's Ghaut 10 2-10 42 3.6 1.6

White River 19 6-10 44 3.8 1.4

Tuitt's Ghaut 8 6-10 69 4.6 1.3

5.4 27

5.7 40

4.8 89

3.1 73

3.5 51

-

-

2.3 9

1.5 11

Grain-size distributions for these tephra-fallout deposits are also shown in Figure 17. Median diameter (Md ) and sorting coefficient ( ) are Inman (1952) parameters. Mode and wt% are the lognormal size distribution parameters calculated using the method described in Wohletz et al. (1989). = -log2d, where d is the particle diameter in mm.

the corresponding total tephra-fallout deposits vary between 3.6 and 4.60, and 1.3 and 1.6, respectively (Fig. 18c). Figure 18d illustrates the generally fine-grained character of dome-collapse fallout tephra, with the ash fraction >250 m (<20) always below 20wt%.

Grain-size analyses were also carried out on fallout tephra from individual Vulcanian explosions during the period August-October 1997. Figures 19a-e shows the grain-size distributions for samples collected at different distances from the vent after two small (28 September 1997, 23:03 LT, and 1 October 1997, 11:34 LT) and three large (5 August 1997, 04:45 LT, 26 September 1997, 14:56 LT, and 2 October 1997, 22:50 LT) Vulcanian explosions (Table 5). Due to the low plume heights (c. 3 and 5 km a.s.l. respectively) the first two explosions were characterized by uniform dispersal to the west by low-level winds (<5km a.s.1). During the 5 August 1997 04:45 explosion (vent plume up to 11 km a.s.1), the low (<5 km a.s.l.) and intermediate-level (8-18 km a.s.l.) winds were blowing uniformly to the NW. The low co-PF plumes from the 26 September 14:56 (Figs 4b and 6) and the 2 October 22:50 explosions were dispersed to the west, and the associated vent plumes (up to 12 and 9km a.s.l. respectively) were dispersed to the NW and NE respectively. These grain-size distributions are all strongly bimodal, with only the 1 October fallout tephra showing a predominant unimodality. The principal contrast between dome-collapse and Vulcanian fallout tephra is the prominence of coarse ash and lapilli in the latter (Fig. 18d). Most samples contain a coarse subpopulation of coarse ash and lapilli, with modes typically between 2mm and 63 m (-1 and 40), and a fine subpopulation with modes typically in the 63-8 m (4-7 ) range. Due to safety and logistical problems,

Fig. 18. Md versus sorting plots for individual grain-size distributions at different localities for: (a) dome-collapse fallout tephra; (b) Vulcanian-explosion fallout tephra; and (c) total weighted grain-size distributions of fallout tephra generated by dome collapses (black symbols) and Vulcanian explosions (white symbols). (d) Ternary plot for grain-size distributions of individual fallout-tephra samples from both dome collapses (black symbols) and Vulcanian explosions (white and grey symbols). The horizontal line highlights the dome-collapse field.

502

C. BONADONNA ET AL.

Fig. 19. Grain-size distributions of fallout tephra collected at different distances from the vent immediately after five Vulcanian explosions: (a) 5 August 1997 (04:45 LT); (b) 26 September 1997 (14:56 LT); (c) 28 September 1997 (23:03 LT); (d) 1 October 1997 (11:34 LT): (e) 2 October 1997 (22:50 LT). (f) Total weighted average for the last four events. Size interval: —3 to 130 for (a) and —3 to 10 for (b) to ( f ) . The <—125 m (< —3 ) fraction of ash samples from the 5 August 1997 (04:45 LT) Vulcanian explosion was analysed using the laser diffraction technique with a Mastersizer Hydro 2000M. which can determine the volume percentage of particles from 2000 down to 0.02 m.

TEPHRA FALLOUT AT SOUFRIERE HILLS VOLCANO

503

Table 5. Main grain-size parameters of fallout tephra from four Vulcanian explosions in 1997

Observed column height (km) Fountain-collapse valleys Number of sample sites Distance from vent (km) Wt% fine ash (<63 m) Median diameter, Md Sorting coefficient, Mode 1 0

Wt% Mode 2 Wt% Mode 3

26 Sep. 97 (14:56 LT)

28 Sep. 97 (23:03 LT)

1 Oct. 97 (11:34LT)

2 Oct. 97 (22:50 LT)

11 FG, TyG, TG, MG 12 5.7-9.5 46 3.7 2.6

3 FG, TyG, TG 10 5.7-7 63 4.6 2.9

5 FG, TyG, TG, MG, TR 7 7-7.6 56 4.2 2.0

9 n/a 13 7-9.5 46 3.7 2.6

4.9 58

5.2 78

4.9 71

4.5 74

1.0 42

0.2 22

1.7 29

0.3 24

-1.7 Wt%

Parameters are explained in footnote to Table 4. Valleys are shown in Figure 1. Grain-size distributions for these tephra-fallout deposits are shown in Figure 19. n/a, Data not available. we were unable to collect samples very close to the volcano (within 3 km) on a regular basis during the two series of Vulcanian explosions in 1997. However, qualitative field observations indicated that these proximal localities contained some pumice lapilli and blocks (Druitt et al. 2002). Had we been able to collect samples in those areas then the contrast between the coarse-grained Vulcanian tephra-fallout deposits and fine-grained co-PF deposits would be even more pronounced than that described with the available data. The proportions of the two subpopulations described above vary considerably with location (Figs 19 and 20). One factor appears to be distance from the volcano and is best illustrated by the fallout tephra from the 5 August 1997 (04:45 LT) explosion, where the most proximal samples are dominated by the fine subpopulation, and the coarse subpopulation increases with distance from the dome (Fig. 19a). The fine subpopulation is thought to be fallout tephra from the low co-PF plumes that is mainly deposited in proximal areas, whereas the coarse subpopulation is thought to be from the vent plume that, reaching higher levels, mainly sediments in more distal areas. However, distance from the volcano is not the only cause for grain-size variations, as the data do not show a systematic trend (Fig. 20c). Two additional factors are wind profile and position of individual sites compared to the dispersal axis. These influences can be observed in the data from the 26 September 1997 (14:56 LT) and 2 October 1997 (22:50 LT) explosions (Fig. 19b and e). In these two examples ash from the co-PF plumes was transported by lowlevel winds to the west, dominating deposition at localities P during the 26 September explosion, and M and D during the 2 October explosion (Fig. 19b and e; sites on Fig. 1). Coarse ash and lapilli from the high vent plume were transported towards the NW in the 26 September explosion and towards the NE in the 2 October explosion, so that the coarse subpopulation dominates at localities Q (NNW of the dome) and N (NNE of the dome) respectively in the two examples (Fig. 19b and e; sites on Fig. 1). Figure 19f shows the total grain-size distribution for Vulcanian explosions calculated as weighted average of sample grain-size distribution over the whole tephra-fallout deposit. All the distributions show polymodality. Had we been able to sample coarse-grained proximal deposits rich in pumice lapilli then the polymodality would have been even more pronounced: the proportion of the coarse mode is clearly underrepresented here due to absence of proximal samples. Table 5 summarizes the subpopulation parameters, and Figure 21 gives results of the distribution analysis for the tephrafallout deposit from the 26 September 1997 (14:56 LT) Vulcanian explosion. The two main modes (fine and coarse modes) vary between 47 and 31 m (4.4 and 5.0 ) and 1.7 and 0.4mm (-0.8 and

1.2 ) respectively for individual sites, and are at 33 and 50 m (4.9 and 1.00) respectively for the total weighted grain-size distribution (Table 5). Several minor subpopulations are also observed (e.g. Fig. 21). Md and sorting coefficients of Vulcanian tephra-fallout deposits at individual sites within 6-10 km from the vent, and corresponding total weighted grain-size distributions, vary between 1.5 mm and 22 m (—0.6 and 5.50) and 1 and 3 respectively (Fig. 18a-c). However, since these grain-size distributions are strongly bimodal, Md and sorting are of limited validity. Fallout tephra from Vulcanian explosions is best distinguished from that from dome collapses on a ternary diagram (Fig. 18d), where the dominance of lapilli and coarse ash is well shown. Analyses done with the laser-diffraction technique on the 5 August 1997 (04:45 LT) fallout tephra show that fine ash-rich samples (i.e. mainly ash from co-PF plumes) are characterized by 5-7 wt% of particles <8 m and the finest particles detected had an equivalent diameter of 0.2 m.

Production of very fine ash Here we compare the mass of very fine ash by two different methods: from satellite observations and from grain-size analysis of tephrafallout deposits on land. Very fine ash is here defined as that of 2-20 m in diameter, as this is the range that can be resolved by satellite-based methods. The two events studied were the 26 September 1997 (14:56 LT) Vulcanian explosion and the 26 December 1997 dome collapse. Masses of very fine ash retrieved from the GOES-8 images are given in Tables 2 and 3. For the Vulcanian explosion of 26 September 1997 (14:56 LT), for which we have good control of on-land data, the total mass of very fine ash is 5.4 x 10 7 kg (retrieved from the image with the maximum area). This represents c. 10% of the total deposit mass (5.5 x 108 kg) calculated from the deposit on land using Pyle's (1989) method. Extrapolation of the onland data using Pyle's (1989) method predicts that the deposit thins to negligible thickness by 25km from the volcano, so the cloud should contain no ash beyond 25km if this extrapolation is valid. The satellite data show that substantial amounts of very fine ash are in fact dispersed well beyond 25 km (Figs 6b and 22), and confirm that extrapolation of Pyle's (1989) method underestimates the volume of tephra-fallout deposits. In this example such a volume is underestimated by at least 10%. The total mass of fallout tephra produced by the 26 December 1997 dome collapse (11.7 x 10 9 kg) was estimated by Sparks et al. (2002) from plume height and collapse duration. Retrievals from

504

C. BONADONNA ET AL.

Fig. 21. Grain-size distributions of sites (a) B and (b) P (on Fig. 1). and (c) total weighted average of the fallout tephra generated by the 26 September (14:56 LT) Vulcanian explosion (see also Fig. 19b and f) plotted using the method described in Wohletz et al. (1989) to separate the lognormal subpopulations. Four subpopulations are evident for each sample and for the weighted average. Fig. 20. Variation of the weight fraction of (a) the fine and (b) the coarse modes observed in individual samples of fallout tephra produced by six Vulcanian explosions: 5 August 1997 (04:45 LT), 26 September 1997 (14:56 LT), 28 September 1997 (23:03 LT), 1 October 1997 (11:34 LT). 2 October 1997 (22:50 LT) and 10 October 1997 (18:40 LT). (c) Variation of the fine-mode fraction with distance from vent for the same samples. Size interval: 3 to 7 for the fine mode and -2 to 4 for the coarse mode. Wt% of fine and coarse populations was calculated using the method described in Wohletz et al (1989).

GOES-8 for the 26 December 1997 dome collapse (Table 2) show a surprisingly small amount of very fine ash in the cloud (3.1 x 10 7 kg), which represents c. 0.3% of the total mass. This is smaller than, for example, those retrieved for volcanic clouds produced by the 25 June 1997 dome collapse and by the 26 September 1997 (14:56 LT) Vulcanian explosion (Tables 2 and 3), even though the volume of the 26 December 1997 dome collapse was several times larger. We interpret this result as the consequence of the very high content of water in the cloud that formed when the pyroclastic density current of 26 December 1997 entered the sea. The water or ice from the water-rich cloud can mix with, or encase, the very fine ash particles (Rose et al. 1995), leading to rapid deposition. The

tephra-fallout deposit on land is dominated by large accretionary lapilli (up to 11 mm in diameter), providing independent evidence that ash had been scavenged from the ash plume. However, water or ice in the cloud can also mask the characteristic spectral signature of volcanic ash from the infrared sensor (IR), resulting in an underestimate of fine-ash content (Mayberry et al. 2002). Both these effects can drastically reduce the retrieved apparent values for the ash mass. The 25 June 1997 dome collapse and 26 September 1997 (14:56 LT) Vulcanian explosion did not form large, water-rich clouds. Therefore the water ice masking effect was not significant in these two cases.

Accumulation rates Accumulation-rate experiments were carried out during two Vulcanian explosions (Tables 6 and 7): 28 September 1997 (23:03 LT) and 1 October 1997 (11:34 LT). On 28 September 1997 an experiment was carried out at site D (6.8km from vent; Fig. 1). A Vulcanian explosion (Table 5) with fountain collapse had occurred at 23:03 LT, producing tephra fallout at site D between

TEPHRA FALLOUT AT SOUFRIERE HILLS VOLCANO

505

Table 6. Main parameters of sequential samples collected during tephra fallout from the 28 September 1997 (23:03 LT) Vulcanian explosion (at site D on Fig. 1) Time elapsed since explosion (min)

Mass/area (10 - 3 kgm - 2 ) Fallout rate (l0-3kgm-2s-1) Accretionary lapilli (max. diameter, mm) Median diameter, Md Sorting coefficient, Mode 1

Fig. 22. Diagram showing the extrapolation of the proximal exponential decay of deposit thickness to a thickness close to 0mm (using 10 - 5 mm as minimum thickness value) for the tephra-fallout deposit produced by the 26 September 1997 (14:56 LT) Vulcanian explosion. Small black diamonds represent isopachs drawn from field data (Fig. 7) and the solid line represents distal trend extrapolated from proximal data. Some positions of the associated volcanic cloud are also shown (determined from GOES-8 satellite images, Fig. 6).

23:14 LT and 23:27 LT. After the first minute of fallout, tephra started to fall as accretionary lapilli up to 4 mm in diameter. After the first 5 minutes of tephra fallout, sedimentation rate increased (Fig. 23a) and tephra fell as accretionary lapilli up to 3 mm in diameter. Some of these accretionary lapilli aggregated together forming clusters in the air up to 6 mm in diameter. Tephra fallout slowed down after 10 minutes and stopped after 13 minutes. At site D heavy rain started about one hour after the end of fallout. This fallout tephra is rich in fine ash (87wt%, weighted average). Samples collected between 0 and 10 minutes and between 15 and 20 minutes are coarse-tail skewed, suggesting bimodality. The sample collected between 10 and 15 minutes is unimodal (Fig. 23c). All four samples are characterized by a dominant fine subpopulation with a mode between 4.1 and 5.5 (mode 1 in Table 6). On 1 October 1997 a similar experiment was carried out at a site 7.4 km from the vent (H on Fig. 1). A Vulcanian explosion (Table 5) with fountain collapse had occurred at 11:34 LT, producing tephra fallout at site H between 11:49 and 12:26 LT. After the first 10 minutes tephra fallout stopped, then started again a few minutes

Wt% Mode 2

0-5

5-10

10-15

15-20

158.3

344.3

28.3

0.7

0.5

1.2

0.2

0.02

4 (after 1 min)

3

4.9 1.0

5.3 1.0

5.1 1.0

3.7 1.5

5.2 89

5.5 93

5.1 100

4.1 76

3.0 11

3.0 7

Wt% Mode 3 -

Wt%

-

1.8 21

-

-0.8 3

Parameters are explained in footnote to Table 4. Grain-size distributions for these samples are shown in Figure 23c.

later, reaching a maximum accumulation rate 25 minutes after the beginning of tephra fallout (Fig. 23a). After the first 24 minutes a significant amount of tephra that had accumulated on trees was also blown around. After the first 28 minutes tephra started to fall as accretionary lapilli, and this was followed by heavy rain. The 1 October fallout tephra was characterized by 54wt% (weighted average) of fine ash. The first 5 minutes of tephra fallout were characterized by mostly fine ash (c. 84 wt%) (Fig. 23b) and a nearly unimodal size distribution (Fig. 23d). This early fallout tephra (which represents c. 10wt% of the total tephra-fallout deposit at site H) is attributed to co-PF plumes above the fountain-collapse pyroclastic flows. During the following 20 minutes fallout tephra became progressively coarser, with fine ash always below 60wt%, and bimodality became prominent (Fig. 23d). This later fallout tephra is interpreted as principally sourced from the vent plume. From 25 to 35 minutes after the beginning of tephra fallout at site H (when most of the ash fell as millimetric accretionary lapilli) bimodality is less pronounced and fine ash content increases again,

Table 7. Main parameters of sequential samples collected during tephra fallout from the 1 October 1997 (11:34 LT) Vulcanian explosion (at site H on Fig. 1) Time elapsed since explosion (min)

Mass/area (10- 3 k g m -2 ) Fallout rate (10 - 3 k g m - 2 s - 1 ) Accretionary lapilli (max. diameter, mm) Median diameter, Md Sorting coefficient, Mode 1

Wt% Mode 2 0 Wt%

0-5

5-10

10-15

15-20

20-25

25-35

70.9 0.2 -

1.8 0.01 -

21.6 0.1 -

160.7 0.5 -

190.0 0.6 -

228.5 0.4 2

5.0 1.0

4.2 1.9

3.9 2.0

4.0 2.0

3.1 1.9

4.2 1.7

4.6 73

4.6 72

4.7 56

4.7 80

1.5 27

1.4 28

1.8 44

2.0 20

5.1 99 1.4 1

5.0 63 1.8 37

Parameters are explained in footnote to Table 4. Grain-size distributions for these samples are shown in Figure 23d.

506

C. BONADONNA ET AL.

Fig. 23. (a) Accumulation rate versus time and (b)wt% fine ash versus time for sequential samplings carried out during tephra fallout from two Vulcanian explosions (28 September 1997, 23:03 LT, and 1 October 1997. 11:34 LT) at 6.8km and 7.4km NW from the dome respectively (sites D and H on Fig. 1). (c and d) Plots showing the variation of grain-size distribution with time during the (c) 28 September 1997 (23:03 LT) and (d) 1 October 1997 (11:34 LT) Vulcanian explosions.

with 55wt% fine ash. All six samples are characterized by a large fine subpopulation with mode between 4.6 and 5.1 (mode 1 in Table 7) and a smaller coarse subpopulation (mode 2 in Table 7) with mode between 1.4 and 26.

Particle aggregation Aggregation processes were investigated by analysing samples from both Vulcanian explosions and co-PF plumes from dome collapses. SEM studies showed that fine-ash particles up to 50 m fell mainly as aggregates. Aggregates from both dome collapses and Vulcanian explosions show similar characteristics and could be separated into four categories: irregular aggregates (Fig. 24a), dust-coated crystals (Fig. 24b), accretionary lapilli (Fig. 24c-e) and mud rain (Fig. 24f). Both irregular aggregates and accretionary lapilli show a large range of dimensions, from 10 m to 100 m and from 150 m to 11 mm in diameter respectively. Accretionary lapilli are characterized by their typical spherical shape, and their external surfaces appear to be compact, with micrometric particles sticking together

and no evident secondary phases (Fig. 24d). Broken accretionary lapilli often show the presence of a large central particle (Fig. 24e). Dust-coated crystals consist of a phenocryst (>63 m) covered in fine particles (<20 m). and are characterized by their elongated shape and dimensions of up to 1.5 mm (Fig. 24b). Adhering dust is often considered to be produced by alteration processes (Heiken & Wohletz 1985). However, the samples analysed by SEM were collected during tephra fallout and typical alteration minerals (e.g. clay and zeolite) were not observed. Adhering dust is therefore better interpreted as an effect of electrostatic forces, which tend to make micrometric particles (<45 m ) stick to the surface of coarser particles (>63 m) (Schumacher 1994). Finally, mud rain appears as evaporated drops up to 6 mm in diameter. These have irregular shapes and some are characterized by coarser material on the rim (up to c. 1 mm thick), and by the presence of big. angular particles (up to c. 1 mm in diameter) throughout (Fig. 24f). Tephra fallout sampled during rain shows thinner or no dust coatings on crystals and no accretionary lapilli. Ritchie et al. (2002) show that the diameter of accretionary lapilli produced by the 26 December 1997 dome collapse increases with

TEPHRA FALLOUT AT SOUFRIERE HILLS VOLCANO

507

Fig. 24. SEM images of different kinds of aggregates collected on carbon tabs during tephra fallout. (a) Irregular aggregate from the 9 September 1997 dome collapse, (b) Dust-coated crystal aggregate (on a plagioclase crystal) from the 10 September 1997 dome collapse, (c) Accretionary lapillus from the 25 September 1997 (11:09 LT) Vulcanian explosion and (d) close-up of the surface showing the wide size range of aggregated particles. (e) Broken accretionary lapillus showing a plagioclase crystal as nucleus, 24 September 1997 (10:54 LT) Vulcanian explosion. (f) Mud-rain aggregate evaporated on a carbon tab during the 24 September 1997 (17:16 LT) Vulcanian explosion, showing the presence of large particles (0.5 to 1 mm in diameter) and a coarse rim up to about 1 mm wide.

C. BONADONNA ET AL.

508

of the whole fallout tephra. The dominant fine mode (mode 1 in Table 8) varies in the restricted range 27-22 m (5.2-5.50) and represents 87-93 wt% of the samples. Accretionary lapilli from the 28 September 1997 (23:03 LT) Vulcanian explosion studied with the SEM are loosely packed and show no apparent internal structure (Fig. 26a and b). Some accretionary lapilli (collected at site AL2 on Fig. 1) from the 26 December 1997 dome collapse show concentric structures (Fig. 26c-f). The external rim (c. 1.5-2 mm) is characterized by very fine and loose material, which easily detaches from the rest of the aggregate. The core is much coarser, and is characterized by glass-coated plagioclase and pyroxene crystals up to c. 1-2 mm minimum diameter. The core is also much more compact, with large single crystals embedded in a dominantly vitric matrix (Fig. 26e). Secondary minerals were not observed in the analysed accretionary lapilli, in contrast to observations elsewhere (e.g. Gilbert & Lane 1994). Fig. 25. Grain-size distributions for accretionary lapilli (AL1, up to 5mm; AL2, up to 7mm) that fell during the 26 December 1997 dome collapse (in Tables 1 and 2). They were collected at 2.5km (AL1) and 4.2km (AL2) from the dome (on Fig. 1). Parameters are summarized in Table 8.

distance from the pyroclastic-density-current deposit (from 3 mm up to 11 mm), reaching a maximum diameter where the tephra layer is thickest (6cm), and that accretionary lapilli are normally graded within the layers. Some of these accretionary lapilli (collected at sites AL1 and AL2 on Fig. 1; Fig. 25), together with some accretionary lapilli collected during the first 10 minutes of tephra fallout from the 28 September 1997 (23:03 LT) Vulcanian explosion, and described in the accumulation-rate test section (Fig. 23c), could be investigated using standard grain-size analysis techniques (Table 8). All the grainsize distributions are predominantly unimodal, with minor coarse subpopulations. The coarse modes (mode 2 and mode 3 in Table 8) vary from 8.6 to 0.1 mm (-3.1 to 3.2 ) and represent 7-13wt%

Morphology of fallout tephra SEM observations were mainly made on tephra collected on carbon tabs during tephra fallout between 6.8 and 7.4km from the dome. Fallout tephra from both dome collapses and Vulcanian explosions consists of crystals (mainly plagioclase but also pyroxene, hornblende and silica), glass shards and aggregates. However, the proportions and the morphological characteristics of these components are different for dome-collapse and Vulcanian fallout tephra. In fallout tephra from dome collapses, crystals (in the size range 0.3-1.5 mm) are characterized by sharp edges, stepped fractures and shallow, dish-shaped concavities (Fig. 27a and b). Round-edged crystals are also common in fallout tephra produced by large dome collapses (Fig. 27c). Some plagioclase crystals show cavities up to 20 m, which can be empty or filled with micrometric plagioclase fragments (Fig. 27d). These cavities are always associated with cracks in the crystal. Crystals are free of glass coatings and ranee

Table 8. Main parameters for samples of accretionary lapilli collected during the first 10 minutes of tephra fallout from a I'ulcanian explosion and from the tephrafallout deposit generated by a large dome collapse 28 Sep. 97 (23:03 LT)

26 Dec. 97

26 Dec. 97

Dome collapse (down White River valley) 15 AL1 2.5 5

Dome collapse (down White River valley) 15 AL2 4.2 7

0-5 min

5-10min

Type of event

Vulcanian explosion

Vulcanian explosion

Max. column height (km) Sample site (on Fig. 1) Distance from vent (km) Accretionary lapilli (max. diameter, mm) Wt% fine ash (<63/ m) Median diameter, Md Sorting coefficient, Mode 1

3 D 6.8 4

3 D 6.8 3

82.3 5.0 1.0

89.6 5.4 1.0

79.2 5.0 1.4

77.4 4.9 1.2

5.2 89

5.5 93

5.2 87

5.3 93

3.0 11

3.0 7

2.0 8

3.2 7

Wt% Mode 2 Wt% Mode 3 0

Wt%

-

-

-3.1 5

-

Parameters are explained in footnote to Table 4. The 26 December 1997 dome collapse is described in Table 1. Grain-size distributions for these samples are shown in Figures 23c and 25.

Fig. 26. SEM backscattered images of: (a) section of an accretionary lapillus collected at site D (Fig. 1) during the first 5 minutes of tephra fallout produced by the 28 September 1997 (23:03 LT) Vulcanian explosion; and (b) close-up of the same section on the bottom-left quadrant, (c) Section of an accretionary lapillus produced by the 26 December 1997 dome collapse and collected at site AL2 (on Fig. 1), and close-ups showing: (d) the contact between the external rim and the core; (e) the internal core; (f) the external rim. Note in (0 the presence of glass-coated crystals immersed in a matrix formed of very fine particles.

TEPHRA FALLOUT AT SOUFRIERE HILLS VOLCANO

28 September 1997 (23:03 LT) Vulcanian explosion

26 December 1997 dome collapse

509

510

C. BONADONNA ET AL.

Fig. 27. SEM images of dome-collapse ash collected on carbon tabs during tephra fallout, (a) Plagioclase crystal from a small dome collapse (10 September 1997) showing fresh fractures and adhering dust, (b) Fractured plagioclase showing adhering dust and dish-shaped fractures typical of transport abrasion (10 September 1997). (c) Rounded plagioclase crystal from tephra fallout generated by a large dome collapse (25 June 1997). (d) Crack and holes filled with micrometre-sized plagioclase fragments on a plagioclase crystal from the 10 September 1997 dome-collapse tephra fallout. (e) Non-vesicular glass fragment from the tephra fallout from a small dome collapse (10 September 1997) showing a rough surface, (f) Highly vesicular glass pyroclast from a large dome collapse (25 June 1997).

TEPHRA FALLOUT AT SOUFRIERE HILLS VOLCANO

from those with minor adhering dust (Fig. 27a-c) to dust-coated crystals (Fig. 24b). Adhering dust consists of crystals and glass shards with diameter <20 m. Glass shards are typically poorly vesicular to non-vesicular and they are mostly <50 m (Fig. 27e). They are characterized by a blocky appearance and the surface can be either rough or smooth. However, samples of ash from large dome collapses have also shown a very few highly vesicular glass shards with vesicles from spherical to ovoid (Fig. 27f). In fallout tephra from Vulcanian explosions, crystals (up to 3 mm) are mostly fractured, with pronounced cracks, holes, and stepped and dish-shaped fractures, and are characterized by rough surfaces (e.g. Fig. 28a and b). Most crystals show round edges (e.g. Fig. 28b). They are often glass-coated (e.g. Fig. 28a and c-f) and covered with adhering dust. Glass coatings are common in all Vulcanian fallout tephra and were observed on plagioclase and pyroxene crystals from 100 m to a few millimetres in size. The glass-crystal boundary is typically sharp (Fig. 28d), but can also be less evident (e.g. Fig. 28e). Generally cracks affect the glass coating (Fig. 28f). Glass pyroclasts vary from poorly vesicular (Fig. 28h) to highly vesicular, with spherical to ovoid vesicles up to 40 m, and thin vesicle walls (Fig. 28g). Vesicles are often filled with micrometric particles (200 nm to 4 m) and some vesicles have coalesced.

Discussion The tephra-fallout deposits generated during the eruption of Soufriere Hills Volcano from 1995 to 1999 result from a complex interplay of several different eruptive processes and transport conditions. Fallout tephra was principally deposited from co-PF plumes generated above dome-collapse and fountain-collapse pyroclastic flows, and from the vent plumes of numerous Vulcanian explosions. Ash-venting and early phreatic explosions also contributed minor amounts of fallout tephra. The largest individual events are recognizable as discrete layers in the final amalgamated deposits. Different eruptive processes are also largely distinguishable in the final tephra-fallout deposits because of lithology and grain size characteristics. The dome-collapse pyroclastic flows generated very fine-grained fallout tephra, usually dominated by grain-size below 63 m (90-96 wt% <250 m). Their fine-grained character is a consequence of elutriation of fine particles from pyroclastic flows to form co-PF plumes. In contrast, the Vulcanian explosions typically generated significantly coarser-grained fallout tephra, with substantial mass proportions of pumice lapilli and coarse ash (>250 m) in most samples. Vulcanian tephra-fallout deposits are somewhat complicated by the combination of tephra fallout from the vent plume, which typically supplies block-, lapilliand ash-sized particles, and tephra fallout from co-PF plumes, which only supplies ash-sized particles. However, tephra-fallout deposits from weak Vulcanian explosions commonly show grainsize characteristics similar to those from dome collapses. Grain-size distribution is strongly dependent on column height, wind profile and distance from the volcano. As a consequence, weak Vulcanian explosions can only transport fine particles at low heights, so there is less contrast in grain size with fine-grained deposits related to domecollapse pyroclastic flows. Several different transport processes significantly affect tephrafallout deposits: mainly wind shear, simultaneous tephra fallout from multiple sources, particle aggregation, and convective instabilities. The Caribbean is a region of complex wind shear, with lowlevel winds (<5km) predominantly moving to the west and intermediate-level winds (8-18 km) to the east. In detail, wind directions are quite variable so that winds can disperse tephra occasionally to the north and south too. The effects of wind shear were often obvious in direct observations of the rising plume, in the splitting of some plumes on satellite images, and in complex distributions of the resulting tephra-fallout deposits. Most dome-collapse pyroclastic flows generated co-PF plumes below 5km a.s.1., except for the largest collapses that generated plumes up to 15km a.s.1., and consequently the associated fallout tephra is distributed predomi-

511

nantly to the west. Many of the Vulcanian explosions generated vent plumes above 5 km a.s.l., and so fallout tephra could also be dispersed to the east, and occasionally to the north and south too. Most Vulcanian explosions occurred with simultaneous co-PF plumes and wind shear, which contributed to wide variations in fallout-tephra characteristics over the island. For example low-level winds could transport co-PF ash in a quite different direction to the coarser-grained fallout tephra from the associated vent plume. Aggregation is a major influence on tephra fallout and on the characteristics of the associated tephra-fallout deposits, and is manifested in irregular clusters, dust-coated crystals, accretionary lapilli, and mud rain. Aggregates commonly disintegrate on deposition, so that aggregation structures are not preserved in deposits. However, large accretionary lapilli are generally well preserved, forming distinctive fallout-tephra layers (e.g. tephra-fallout deposit from the 26 December 1997 dome collapse). Aggregation can cause premature fallout of fine particles (e.g. Carey & Sigurdsson 1982; Brazier el al. 1983). Convective structures were commonly observed in volcanic clouds too, and they can also contribute to early fallout of fine ash (Carey 1997). Our investigations enabled calculations of volumes of the tephrafallout deposits, but these could only be studied on the island at distances <10km from source. We have applied the method of Pyle (1989) to calculate volumes of fallout tephra from isopach maps, as this is the most flexible and reliable calculation method available to date. Pyle's (1989) method fits the on-land data well. However, preliminary comparison between field data for a Vulcanian explosion (i.e. 26 September 1997, 14:56 LT) and retrievals from corresponding satellite images shows that at least 10wt% of the very fine ash fraction (2-20 m) is not included in the volume of the resulting tephra-fallout deposit calculated by extrapolating the proximal exponential trend to infinity. Analyses of satellite images of some of the dome-collapse and Vulcanian explosion clouds show that a significant part of this fraction can stay in a volcanic cloud for several hours and can be transported large distances. These observations support the idea that volumes of fallout tephra need to be calculated by using a method that takes into account the amount of fine ash typically transported to distances far beyond the area of tephra fallout predicted by the assumption of exponential decay (Bonadonna et al, 1998; Rose 1998). Our investigations also showed that the underestimation of volumes given by these methods is very difficult to quantify and can vary from case to case. On the basis of our minimum estimates of volumes of fallout tephra, obtained by extrapolating on-land data to infinity using the exponential method of Pyle (1989), Vulcanian-explosion and domecollapse elutriation factors were calculated. The elutriation factor estimated for Vulcanian explosions by comparing the total volume of co-PF fallout tephra for the 1997 Vulcanian explosions and that of the associated fountain-collapse pyroclastic-flow deposits is 5%. The elutriation factors estimated for dome-collapse pyroclastic flows are in the range 4-5%. Dome-collapse elutriation factors were also estimated using a geochemical method and a method based on the plume dynamics for the 26 December 1997 dome collapse. In the geochemical method, the chemistry of the pyroclastic-flow deposits and the associated fallout tephra were used to estimate the proportions of elutriated ash from dome collapses by a mass balance and gave values of 7-16% (Horwell et al. 2001). In the plume dynamics method, the comparison between volume of fallout tephra and the associated pyroclastic-density-current deposit by plume-dynamics calculations gave values of 10-13% (Sparks et al. 2002). Elutriation factors based on extrapolated volumes of on-land fallout tephra are lower than those calculated by using the geochemical and the plumedynamics methods, probably due to the underestimation of volume of fallout tephra calculated by using Pyle's (1989) method. However, also given the uncertainties in the volume estimates of pyroclasticflow deposits and in the geochemical and plume-dynamics calculations, such an underestimation is difficult to quantify. We conclude that elutriation factors both for dome- and fountain-collapse pyroclastic flows on Montserrat are in the range 4-16%. Dome-collapse and Vulcanian elutriation factors estimated here are significantly lower than those derived from crystal concentration

512

C. BONADONNA ET AL.

Fig. 28. SEM images of Vulcanian-explosion ash collected on carbon tabs during tephra fallout, (a) Fractured plagioclase crystal showing cracks, holes and glass coating (27 September 1997. 17:15 LT). Note the cracks filled with particles, (b) Round and fractured plagioclase (25 September 1997. 11:09 L T ) . (c) Glass-coated pyroxene (27 September 1997. 17:15 LT). (d) Sharp boundary between crystal and glass coating on a plagioclase (24 September 1997. 19:00 LT). (e) Less evident boundary between plagioclase and glass (24 September 1997. 19:00 LT). (f) Crack on a glass-coated plagioclase crystal (27 September 1997, 17:15 LT). Note that the crack affects the glass coating too. suggesting that it formed after the crystal was ejected from the conduit and was entrained within the fountain collapse, (g) Poorly vesicular glass shard (8 October 1997. 15:10 LT). (h) Highly vesicular pumice fragment (24 September 1997. 12:30 LT) showing thin vesicle walls.

studies of fallout tephra from large fountain-collapse pyroclastic flows, which range between 30 and 50% of the mass of the associated ignimbrite deposits for the Minoan Ignimbrite (Greece) and ignimbrites of Vulsini Volcano (central Italy) (Walker 1971; Sparks & Walker 1977). This is probably due to the much larger magnitude of these eruptions compared to that of Soufriere Hills Volcano. Our estimates of elutriation factors are also significantly lower than the dome-collapse elutriation factor derived for the 1991 eruption of Unzen Volcano by Watanabe et al. (1999), who obtained an estimate of 30%. This significant difference is mainly due to the different methods used to calculate the volume of tephra-fallout deposits. The calculations by Watanabe et al. (1999) are based on the average mass of fallout tephra in the area of each isopach. However, if the calculations are done using the Pyle (1989) method, an elutriation factor of 10-20% is obtained (Y. Miyabuchi & K. Watanabe, pers. comm.), which is closer to our results. Therefore, considering an average elutriation factor of 10%, and given that c. 140 x 10 6 m 3 (DRE) of pyroclastic-flow deposit was produced by dome collapses (with volume >10 6 m 3 ; from data in Calder et al. 2002) during the 1995-1999 eruptive period of Soufriere Hill Volcano, we estimate a total volume of the associated tephra-fallout deposit of > 14 x 106 m3 (DRE). This is a much larger volume compared to the volume of total co-PF fallout tephra estimated for the AugustOctober 1997 explosions (>0.1 x 10 5 m 3 DRE).

Fallout tephra produced by both dome collapses and Vulcanian explosions generally show multiple grain-size subpopulations. Polymodality (i.e. two to four modes) is much more pronounced in Vulcanian fallout tephra. There are several possible causes of polymodality: (i) multiple types of fallout tephra (i.e. fallout tephra from vent plume and fallout tephra from associated fountain-collapse co-PF plumes; e.g. Sparks & Huang 1980); (ii) premature fallout of fine particles caused by aggregation and convective instabilities (e.g. Carey & Sigurdsson 1982; Brazier et al. 1983; Carey 1997; Hoyal et al. 1999); (iii) multiple, simultaneous sources of tephra fallout (i.e. tephra fallout from several plumes of different height); and (iv) differences in density and grain-size distribution of different mineral components during fragmentation of the andesite (Walker 1971). Polymodality is attributed to all these different factors taking place simultaneously during tephra fallout, and with the importance of each factor varying from case to case. We consider the combination of multiple fallout-tephra types as major factors in causing polymodal tephra-fallout deposits from Vulcanian explosions, at least in proximal areas. In particular, strong polymodality was always shown by tephra-fallout deposits from large Vulcanian explosions (i.e. explosions in which there is a significant difference between vent-plume and co-PF-plume heights, and then between the two different types of associated fallout tephra), whereas tephra-fallout deposits from small explosions and

TEPHRA FALLOUT AT SOUFRIERE HILLS VOLCANO

513

Fig. 28. (continued)

from dome collapses are mostly unimodal (Tables 4 and 5, Figs 17 and 19). This is consistent with computer simulations that show how Vulcanian tephra-fallout deposits result from the combination of co-PF-plume and vent-plume fallout tephra that, even when characterized by different dispersal, can partially overlap (Bonadonna et al. 2002). This overlap typically occurs in proximal areas and, due to the grain-size differences between fallout tephra from coPF and vent plumes, results in multiple populations in the grain-size distribution observed at proximal deposits from large Vulcanian explosions (e.g. Table 5, Fig. 19). Finally, most individual polymodal samples have a fine subpopulation with a mode typically in the 4 to 60 range (Table 5, Fig. 19). This mode overlaps with the characteristic grain-size mode of both fallout tephra from co-PF plumes (4.2-5.70; Tables 4 and 7, Figs 17 and 23d) and accretionary lapilli (4.9-5.40; Table 8, Figs 23c and 25). This fine mode is, therefore, also attributed to aggregation, which can cause premature fallout of fine particles, resulting in a composite deposit of both fine ash and coarse grains deposited at the same site. This interpretation is supported by modelling studies, which show that particles in the fine subpopulation could not deposit on the island unless they fall out as aggregates (Bonadonna et al. 2002). Aggregation is thought to be the main cause of bimodality in the case of multiple populations observed in some of the individual fallout-tephra samples from dome collapses for which grain-size characteristics of fallout tephra from different co-PF plumes is not significant (e.g. 31 March 1997, Table 4). The rapid sedimentation of fine particles due to the convective instabilities in the plumes (as in Fig. 9) may also contribute to polymodality. Plumes of different heights can transport and then sediment particles of different

sizes, and eruptive events generating plumes of different heights (e.g. dome collapses and Vulcanian explosions) produce tephrafallout deposits that are the result of this sedimentation from multiple plumes. If the difference in plume height is significant, these resulting deposits are likely to show minor subpopulations even when the types of fallout tephra involved are not significantly different. This is the case of fallout tephra from multiple plumes generated during large dome collapses (e.g. 21 September 1997 dome collapse; Table 4, Fig. 17). Differences in density and grain-size distribution of different mineral components during fragmentation can also produce bimodality (Walker 1971), which could explain the slightly bimodal distal samples of fallout tephra from the 31 March 1997 dome collapse. In fact, computer simulations support the idea that most of the aggregates fell in proximal areas for that dome collapse (Bonadonna et al. 2002), and thus cannot explain bimodality of distal samples. However, a detailed component analysis of fallout tephra is required to confirm such a hypothesis. Morphologies and component abundances of fallout tephra from dome collapses and Vulcanian explosions show contrasting features, reflecting different generation, fragmentation and transport processes. SEM analyses on the ash fraction of dome-collapse and Vulcanian tephra were made. Ash from Vulcanian explosions is characterized by poorly to highly vesiculated glass shards. Vesicular glass coatings on crystals were only observed in Vulcanian ash. Glass shards observed in ash from dome-collapse pyroclastic flows are mostly blocky and poorly vesicular, consistent with the nonexplosive nature of this process. However, vesicular glassy particles in ash from large dome collapses suggest an explosive component, or that collapse excavated vesicular material from the interior of the

C. BONADONNA ET AL.

514

dome. Abrasion features (e.g. rounding, superficial dish-shaped and stepped fractures) are common in both dome-collapse and Vulcanian ash. In contrast, fracturing of large crystals is a more significant feature in Vulcanian ash in comparison to crystals from dome collapses. Fractures are mostly a result of fragmentation processes in the conduit. However, some fractures and cracks affect the glass coating of crystals in Vulcanian ash too, indicating that they must have formed after the crystals were ejected and were entrained within the fountain collapse. This supports the evidence of secondary fragmentation within the flows also suggested by geochemical analyses (e.g. Horwell et al. 2001).

Conclusions Generation of tephra fallout Tephra fallout from Soufriere Hills Volcano during the 1995-1999 period was mainly produced by dome-collapse pyroclastic flows and Vulcanian explosions (with plumes of up to 15km a.s.l. for both processes). Phreatic explosions and ash-venting also contributed minor fallout tephra. Fallout tephra from low plumes (i.e. small dome-collapse plumes and weak Vulcanian explosions) was mostly deposited in the west of the island, whereas fallout tephra from higher plumes could also be deposited in the east, north and south of the island. These effects are related to changes in the dominant wind directions with altitude in the Caribbean. Volcanic clouds generated by both dome-collapse pyroclastic flows and Vulcanian explosions showed convective-instability structures, which cause the particles to reach the ground at a faster velocity than their individual fall velocities. Premature tephra fallout (which is attributed to aggregation and tephra-fallout convective instabilities) and tephra fallout from multiple sources cause the accumulation rate and fallouttephra grain-size distribution to vary significantly in time during individual events. Tephra-fallout

for the total fallout tephra produced by dome-collapse activity during the 1995-1999 eruptive period of Soufriere Hills Volcano has been calculated.

Grain-size distribution Grain-size distributions of fallout tephra produced by both dome collapses and Vulcanian explosions show polymodality. Polymodality is much more pronounced in Vulcanian fallout tephra than in that from dome collapses, and is typically characterized by a main fine mode and a main coarse mode between 4 and 70 (2-100 wt%) and -1 and 4 (0-98 wt%) respectively. Sometimes fallout tephra from dome collapses shows similar characteristics to those from weak Vulcanian explosions, as grain-size distribution is strongly dependent on column height, wind profile and distance from the volcano. Polymodality is the result of a combination of factors: (i) multiple types of fallout tephra (i.e. fallout tephra from vent plume and from associated fountain-collapse co-PF plumes); (ii) premature fallout of fine particles caused by aggregation processes and tephra-fallout convective instabilities: (iii) multiple, simultaneous sources of tephra fallout (i.e. tephra fallout from several plumes of different height): (iv) differences in density and grain-size distribution of different components during fragmentation.

Production of very fine ash Satellite images of some dome-collapse and Vulcanian explosion clouds show that a significant part of the very fine ash (2-20 m) can stay in a volcanic cloud and be transported to large distances from the source for up to several hours after initiation. A large amount of this very fine ash is not included in the volume of the resulting tephra-fallout deposits calculated by extrapolating the proximal exponential trend to infinity.

deposits

Tephra-fallout deposits from Soufriere Hills Volcano are an amalgamation of the deposits from many individual volcanic events, most being too small to generate discrete, recognizable layers. The deposits vary from massive to layered, reflecting accumulation of deposits from numerous events. Fallout-tephra layers from dome collapses are distinguishable from those generated by Vulcanian explosions, mainly because of the very fine-grained character of the former and the high contents of coarse ash and pumice lapilli of the latter. Four different lithofacies were identified: (I) primary ashventing fallout tephra; (II) primary Vulcanian-explosion fallout tephra; (III) primary dome-collapse fallout tephra; and (IV) pervasively reworked, dominantly fine-grained ash inferred to be mainly of dome-collapse origin. The interplay between tephra accumulation rate and erosion at a given location determines the preserved thickness of each lithofacies type.

Particle aggregation Most of the ash-sized particles in the tephra-fallout deposit fell as aggregates ranging from a few micrometres to several millimetres in diameter. Four types of aggregates could be distinguished: irregular clusters, dust-coated crystals, accretionary lapilli and mud rain. Accretionary lapilli grain-size distributions are independent of their size and are typically characterized by multiple sub-populations, with the finest subpopulation (mode = 5.2-5.5 ). representing c. 87-93 wt% of particles. Of all the accretionary lapilli studied, the small ones (up to 4 mm) do not show internal structures, whereas the large ones (up to 11 mm) are characterized by a loose, fine-grained external rim and a coarser and very compact core.

Morphology of fallout tephra Volumes of tephra-fallout deposits Tephra-fallout deposits from both dome collapses and Vulcanian explosions show an exponential thinning with distance from vent, within the distance range that could be investigated (2-10 km). A minimum volume of 2.3 x 10 6 m 3 (DRE) of fallout tephra was produced by intermittent dome-collapse activity over one year (3 June 1996 to 25 June 1997), and a minimum volume of 9.3 x 106 m3 (DRE) by two series of Vulcanian explosions (4 August to 21 October 1997). The minimum volume of fallout tephra associated with dome-collapse and fountain-collapse pyroclastic flows from Soufriere Hills Volcano in 1995-1999 is in the range 4-16% of the volume of the associated pyroclastic flow deposits, with an average value of 10%. A minimum volume of 14.0 x 106 m3 (DRE)

SEM studies have shown that ash-sized fallout tephra from both dome collapses and Vulcanian explosions consists of crystals (mainly plagioclase but also pyroxene, hornblende and silica), glass shards and particle aggregates. Ash from dome collapses is characterized by poorly vesicular to non-vesicular glass shards, with a blocky appearance. Ash from Vulcanian explosions is characterized by poorly to highly vesicular glass shards, with spherical to ovoid vesicles up to 40 m and thin vesicle walls. A glass coating on crystals can be used as a distinctive characteristic, as it is prominent in ash generated by explosive activity, but is absent in ash related to dome collapses. SEM studies suggest that fragmentation mainly occurred in the conduit during Vulcanian explosions, but also that secondary fragmentation (already shown by geochemical investigations) occurred in both dome-collapse and fountain-collapse pyroclastic flows.

TEPHRA FALLOUT AT SOUFRIERE HILLS VOLCANO We thank the staff of the MVO, who have contributed to the documentation of the eruption. Many thanks to S. Kearns for his invaluable assistance, expertise and daring in the use of the SEM. M. Bursik, G. Ernst and Y. Myabuchi are also thanked for fruitful and constructive discussions. T. Druitt, R. Cas and M. Rosi provided helpful reviews. Monitoring work on Montserrat was supported by the Department for International Development. C. Bonadonna is supported by an EC Marie Curie PhD Fellowship. R. S. J. Sparks is supported by a NERC Professorship.

References BAXTER, P. J., BONADONNA, C., DUPREE, R. ET AL. 1999. Cristobalite in volcanic ash of Soufriere Hills Volcano, Montserrat, British West Indies. Science, 283, 1142-1145. BONADONNA, C., ERNST, G. G. J. & SPARKS, R. S. J. 1998. Thickness variations and volume estimates of tephra fall deposits: the importance of particle Reynolds number. Journal of Volcanology and Geothermal Research, 81, 173-187. BONADONNA, C., MACEDONIO, G. & SPARKS, R. S. J. 2002. Numerical modelling of tephra fallout associated with dome collapses and Vulcanian explosions: application to hazard assessment on Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 517-537. BRAZIER, S., SPARKS, R. S. J., CAREY, S. N., SIGURDSSON, H. & WESTGATE, J. A. 1983. Bimodal grain-size distribution and secondary thickening in air-fall ash layers. Nature, 301, 115-119. CALDER, E. S., COLE, P. D., DADE, W. B. ET AL. 1999. Mobility of pyroclastic flows and surges at Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 26, 537-540. CALDER, E. S., LUCKETT, R., SPARKS, R. S. J. & VOIGHT, B. 2002. Mechanisms of lava dome instability and generation of rockfalls and pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 173-190. CAMPBELL, J. B. 1987. Introduction to Remote Sensing. Guildford Press, New York. CAREY, S. N. 1997. Influence of convective sedimentation on the formation of widespread tephra fall layers in the deep sea. Geology, 25, 839-842. CAREY, S. N. & SIGURDSSON, H. 1982. Influence of particle aggregation on deposition of distal tephra from the May 18 1980, eruption of Mount St-Helens volcano. Journal of Geophysical Research, 87, 7061-7072. COLE, P. D., CALDER, E. S., DRUITT, T. H., HOBLITT, R., ROBERTSON, R. E. A., SPARKS, R. S. J. & YOUNG, S. R. 1998. Pyroclastic flows generated by gravitational instability of the 1996-97 lava dome of Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3425-3428. COLE, P. D., CALDER, E. S., SPARKS, R. S. J. ET AL. 2002. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 231-262. DEVINE, J. D., MURPHY, M. D., RUTHERFORD, M. J. ET AL. 1998. Petrologic evidence for pre-eruptive pressure-temperature conditions, and recent reheating, of andesitic magma erupting at Soufriere Hills Volcano, Montserrat, WI. Geophysical Research Letters, 25, 3669-3672. DRUITT, T. H., YOUNG, S. R., BAPTIE, B. ET AL. 2002. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306. GILBERT, J. S. & LANE, S. J. 1994. The origin of accretionary lapilli. Bulletin of Volcanology, 56, 398-411. HEIKEN, G. & WOHLETZ, K. 1985. Volcanic Ash. University of California Press, Berkley. HORWELL, C. J., BRANA, L. P., SPARKS, R. S. J., MURPHY, M. D. & HARDS, V. L. 2001. A geochemical investigation of fragmentation and physical fractionation in pyroclastic flows from Soufriere Hills Volcano, Montserrat. Journal of Volcanology and Geothermal Research, 109, 247-262. HOUZE, R. A. 1993. Cloud Dynamics. Academic Press, San Diego. HOYAL, D., BURSIK, M. I. & ATKINSON, J. F. 1999. The influence of diffusive convection on sedimentation from buoyant plumes. Marine Geology, 159, 205-220.

515

INMAN, D. L. 1952. Measures for describing the size distribution of sediments. Journal of Sedimentary Petrology, 22, 125-145. MAYBERRY, G. C., ROSE, W. I. & BLUTH, G. J. S. 2002. Dynamics of volcanic and meteorological clouds produced on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKE-LAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 539-555. MUERDTER, D. R., DAUPHIN, J. P. & STEELE, G. 1981. An interactive computerized system for grain-size analysis of silt using electro-resistance. Journal of Sedimentary Petrology, 51, 647-650. MURPHY, M. D., SPARKS, R. S. J., BARCLAY, J., CARROLL, M. R. & BREWER, T. S. 2000. Remobilization of andesite magma by intrusion of mafic magma at Soufriere Hills Volcano, Montserrat, West Indies. Journal of Petrology, 41, 21-42. NORTON, G. E., WATTS, R. B., VOIGHT, B. ET AL. 2002. Pyroclastic flow and explosive activity at Soufriere Hills Volcano, Montserrat, during a period of virtually no magma extrusion (March 1998 to November 1999). In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 467-481. POLLACK, J. B., TOON, O. B. & KHANE, B. N. 1973. Optical properties of some terrestrial rocks and glasses. Icarus, 19, 372-389. PRATA, A. J. 1989. Infrared Radiative-Transfer calculations for volcanic ash clouds. Geophysical Research Letters, 16, 1293-1296. PYLE, D. M. 1989. The thickness, volume and grainsize of tephra fall deposits. Bulletin of Volcanology, 51, 1-15. RITCHIE, L. J., COLE, P. D. & SPARKS, R. S. J. 2002. Sedimentology of deposits from the pyroclastic density current of 26 December 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 435-456. ROBERTSON, R., COLE, P., SPARKS, R. S. J. ET AL. 1998. The explosive eruption of Soufriere Hills Volcano, Montserrat, West Indies, 17 September 1996. Geophysical Research Letters, 25, 3429-3432. ROSE, W. I. 1993. Comment on another look at the calculation of fallout tephra volumes. Bulletin of Volcanology, 55, 372-374. ROSE, W. I, DELENE, D. J., SCHNELDER, D. J. ET AL. 1995. Ice in the 1994 Rabaul eruption cloud - implications for volcano hazard and atmospheric effects. Nature, 375, 477-479. SCHNEIDER, D. J., ROSE, W. I, COKE, L. R., BLUTH, G. J. S., SPROD, I. E. & KRUEGER, A. J. 1999. Early evolution of a stratospheric volcanic eruption cloud as observed with TOMS and AVHRR. Journal of Geophysical Research-Atmospheres, 104, 4037-4050. SCHUMACHER, R. 1994. A reappraisal of Mount St-Helens ash clustersdepositional model from experimental-observation. Journal of Volcanology and Geothermal Research, 59, 253-260. SPARKS, R. S. J. & HUANG, T. C. 1980. The volcanological significance of deep-sea ash layers associated with ignimbrites. Geology Magazine, 117, 425-436. SPARKS, R. S. J. & WALKER, G. P. L. 1977. The significance of vitricenriched air-fall ashes associated with crystal-enriched ignimbrites. Journal of Volcanology and Geothermal Research, 2, 329-341. SPARKS, R. S. J., YOUNG, S. R., BARCLAY, J. ET AL. 1998. Magma production and growth of the lava dome of Soufriere Hills Volcano, Montserrat, West Indies: November 1995 to December 1997. Geophysical Research Letters, 25, 3421-3424. SPARKS, R. S. J., BARCLAY, J., CALDER, E. S. ET AL. 2002. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 409-434. THORARINSSON, S. 1944. Petrokronologista Studier pa Island. Geographes Annuales Stockholm, 26, 1-217. WALKER, G. P. L. 1971. Grainsize characteristics of pyroclastic deposits. Journal of Geology, 79, 696-714. WATANABE, K., ONO, K., SAKAGUCHI, K., TAKADA, A. & HOSHIZUMI, H. 1999. Co-ignimbrite ash-fall deposits of the 1991 eruptions of Fugendake, Unzen Volcano, Japan. Journal of Volcanology and Geothermal Research, 89, 95-112. WATTS, R. B., HERD, R. A., SPARKS, R. S. J. & YOUNG, S. R. 2002. Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills

516

C. BONADONNA ET AL.

Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 115-152. WEN, S. & ROSE, W. I. 1994. Retrieval of sizes and total masses of particles in volcanic clouds using AVHRR bands 4 and 5. Journal of Geophysical Research, 99, 5421-5431.

WOHLETZ, K. H.. SHERIDAN. M. F. & BROWN. W. K. 1989. Particle-size distributions and the sequential fragmentation transport-theory applied to volcanic ash. Journal of Geophysical Research-Solid Earth and Planets, 94. 15703-15721. YOUNG, S. R., SPARKS. R. S. J.. ASPINALL, W. P.. LYNCH. L. L.. MILLER. A. D., ROBERTSON, R. E. A. & SHEPHERD. J. B. 1998. Overview of the eruption of Soufriere Hills volcano. Montserrat, 18 July 1995 to December 1997. Geophysical Research Letters. 25. 3389-3392.

Numerical modelling of tephra fallout associated with dome collapses and Vulcanian explosions: application to hazard assessment on Montserrat C. BONADONNA1, G. MACEDONIO2 & R. S. J. SPARKS1 1

Department of Earth Sciences, University of Bristol, Bristol BS8 1RJ, UK (e-mail: [email protected]) 2 Osservatorio Vesuviano, Via Diocleziano 328, 80124 Napoli, Italy

Abstract: Hazardous effects of tephra fallout on Montserrat include roof collapse, aviation threats, health hazards from respirable crystalline silica, crop pollution, road safety and lahar generation. An advection-diffusion model was developed to investigate tephra dispersal from dome collapses and Vulcanian explosions, which generated most of the fallout tephra during the 1995-1999 eruptive period of Soufriere Hills Volcano. Wind field, atmospheric diffusion, gravity settling, aggregation and elutriation processes are considered. Computed isomass maps compare well with field observations and require aggregation of fine ash for good agreement. Probability maps were also compiled. Individual probability maps (for individual dome collapses and Vulcanian explosions) are based on the statistics of wind profiles and show that fallout tephra generated by individual eruptive events on a Montserrat scale do not cause serious damage in any area on Montserrat. Cumulative probability maps (for a given scenario of activity) are generated by sampling statistical distributions of wind profiles and eruptive events over an extended period of time. They show that persistent tephra fallout can accumulate enough material to cause roof collapses and serious damage to vegetation in the SW part of the island, and minor damage to vegetation in the north, as also confirmed by field data.

The study of tephra fallout from Soufriere Hills Volcano, Montserrat, became a priority for hazards assessment when tephra fallout started to have a substantial effect on the quality of life of people living and working close to the volcano. There are several hazardous effects of fallout tephra: roof collapse, aviation threats, health hazards from respirable crystalline silica, crop pollution, road safety and lahar generation (Blong 1984; Baxter et al. 1999). In this paper the term tephra is used in the original sense of Thorarinsson (1944), as a collective term for all particles ejected from volcanoes irrespective of size, shape and composition, whereas tephra fallout indicates the process of particle fallout. Processes leading to significant tephra fallout on Montserrat during the 1995-1999 period were mainly Vulcanian explosions and elutriation of fines from dome-collapse pyroclastic flows. Vulcanian explosions produced more fallout tephra in about five weeks in 1997 than did dome collapses in about one year from June 1996 to June 1997 (Bonadonna et al. 2002). However, the ash associated with dome-collapse pyroclastic flows seems to be more hazardous to health than Vulcanian tephra, as it contains more crystalline silica and is very fine-grained (Baxter et al. 1999; Moore et al. 2002). Furthermore, large dome collapses also occurred once dome growth had stopped in March 1998, being purely gravitational and not needing any magmatic input (Norton et al. 2002). In this paper an advection-diffusion model for dispersal of tephra from discrete sources is presented. This model is aimed at improving understanding of particle fallout from multiple plumes generated by dome collapses and Vulcanian explosions. Diffusion, advection by wind transport and particle sedimentation are described using a physical model, which is a two-dimensional modification of that presented by Armienti et al. (1988). Previous models (e.g. Suzuki 1983; Armienti et al. 1988; Glaze & Self 1991) considered dispersal of tephra from point-source plumes, in particular Plinian eruptions. Here we adapt these models to consider also weaker, fine-grained ash plumes from distributed sources in the areas inundated by pyroclastic flows. The results are compared with data gathered at Montserrat Volcano Observatory (MVO) throughout the Soufriere Hills Volcano 1995-1999 eruptive period. This model (HAZMAP) was developed as part of the emergency response programme on the effects of volcanic ash, which started in July 1997.

Elutriation from pyroclastic flows Tephra-fallout deposits from pyroclastic flows were first recognized and described by Lacroix (1904) and Hay (1959) in studies of the

eruption of Mont Pelee on Martinique and Soufriere of St Vincent in 1902. Sparks & Walker (1977) recognized that many deposits are commonly enriched in glass particles, and are complementary to the deposits of coevally emplaced pyroclastic flows, which are often enriched in crystals. They called this kind of tephra co-ignimbrite ash. Co-ignimbrite ash is very fine grained (typically <1 mm), resulting from the combination of progressive fragmentation of material within the pyroclastic flows and elutriation of fines by expanding gases. However, pyroclastic-flow deposits produced by dome collapses are not strictly ignimbrites, if ignimbrite is defined as a deposit formed from pumiceous pyroclastic flows (Sparks et al. 1973). In this paper, ash plumes from dome-collapse and fountaincollapse pyroclastic flows will be termed co-pyroclastic-flow plumes, or co-PF plumes. Experimental studies (Huppert et al. 1986; Carey et al. 1988; Woods & Caulfield 1992; Sparks et al. 1993; Woods & Bursik 1994), combined with observations and modelling (Sparks et al. 1986; Dobran et al. 1993; Hoblitt 1986; Woods & Kienle 1994; Calder et al. 1997), have helped in understanding the mechanisms leading to the formation of co-PF plumes. Such plumes form by release and expansion of gases as the juvenile fragments disintegrate, and by expansion of air entrained into the flow. Large flows can generate very high ash plumes by buoyant lift-off, where the whole upper part of the flow ascends buoyantly due to entrainment and heating of air and sedimentation (Sparks et al. 1993). Studies at Mount St Helens show that topography plays an important role in the formation of such co-PF plumes (Hoblitt 1986; Levine & Kieffer 1991; Calder et al. 1997). Breaks in slope, bends and jumps cause enhanced mixing with air and produce more vigorous pulses of plume buoyancy, aiding the formation of discrete plumes, which merge as they ascend. Calder et al. (1997) showed that the behaviour of ash plumes from pyroclastic flows at Mount St Helens (1980) were intermediate between that of a discrete thermal and a steady flux source. Eruption summary Soufriere Hills Volcano is an andesitic dome complex, which started to erupt on 18 July 1995 (Robertson et al. 2000) after being dormant for at least 350 years (Wadge & Isaacs 1988; Harford et al. 2002). From November 1995, a new lava dome started to grow inside English's Crater (Fig. 1) and the first substantial pyroclastic flow was produced on 31 March 1996 by collapse of the new dome. Pyroclastic flows were initially confined to the Tar River valley, progressively building a delta in the sea. In March 1997 pyroclastic flows started to move down the White River valley on the southern flanks (Fig. 1). In May and June 1997, pyroclastic flows started

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 517-537. 0435-4052/02/$15 © The Geological Society of London 2002.

517

518

C. BONADONNA ET AL.

Modelling HAZMAP is a Fortran code for the solution of the equations of particle diffusion, transport and sedimentation used to model the dispersion of a volcanic plume in the atmosphere and the deposition of fallout tephra. The model is an application of the theory of Armienti et al. (1988) and Macedonio et al. (1988). with a simplification of the equations from three to two dimensions. A similar application has been used to compile fallout hazard maps in the Vesuvius area (Italy), with the volcanic plume simulated as a vertical line and the vent located at the origin of the co-ordinate system (e.g. Barberi et al. 1990). The concentration of particles in the plume Cj is described by the following mass-conservation equation: (1)

Fig. 1. Topographic map of Montserrat showing the main valleys followed by both dome-collapse and fountain-collapse pyroclastic flows during the eruptive period 1995-1999 (WR, White River valley; FG, Fort Ghaut; TyG, Tyre's Ghaut; MG, Mosquito Ghaut; TG, Tuitt's Ghaut; WG, White's Ghaut; TR: Tar River valley). These valleys are highlighted by the positions of original rivers (dashed lines). Key localities are also shown: localities a, b and c (white triangles) are used for comparisons of computed with observed grain-size distributions (Fig. 18); locality c is also used to assess the accuracy of Monte-Carlo simulations (Fig. 23). Some of the most important populated areas are shown for the final discussion on hazard assessment.

travelling also down Tuitt's Ghaut and Mosquito Ghaut on the northern side of the volcano and down Fort Ghaut on the western side (Fig. 1). On 25 June 1997 a large dome-collapse pyroclastic flow nearly reached the sea on the eastern coast, and caused the deaths of 19 people (Loughlin et al 2002). On 3 August 1997 a pyroclastic flow reached the capital Plymouth via Fort Ghaut. A series of 13 repetitive Vulcanian explosions with associated fountain collapse then occurred from 4 to 12 August 1997 (Druitt et al. 2002). On 21 September 1997 another large dome collapse pyroclastic flow travelled down Tuitt's Ghaut and White's Ghaut (Fig. 1) and entered the sea, triggering another series of 15 Vulcanian explosions, which ended on 21 October 1997 (Druitt et al. 2002). After this second series of Vulcanian explosions, the dome continued to grow and produced further dome-collapse pyroclastic flows in early November 1997. On 26 December 1997 a flank failure accompanied the largest dome collapse to date (46 x 10 6 m 3 dense rock equivalent (DRE), Calder et al. 2002). This occurred down the White River valley, causing widespread damage to the south of the island. Three other large dome collapses occurred after the lava dome stopped growing in March 1998: 3 July 1998, 12 November 1998 and 20 July 1999 (Norton et al. 2002). Tephra fallout has continued since dome growth resumed in November 1999 and has largely been related to rockfall activity and small pyroclastic flows.

where x, y, z are the spatial co-ordinates, Cj is the mass concentration of particles ( k g m - 3 ) corresponding to the settlingvelocity class j, t is the time, wx and wy are the x and y components of the wind velocity, Kx and Ky are the horizontal diffusion coefficients, and vj is the settling velocity of the particles of the size class j. The vertical wind component w= is assumed to be negligible. Vertical diffusion coefficient Kz is small above 500m of altitude (Pasquill 1974), and therefore is assumed to be negligible. Horizontal diffusion is considered constant and isotropic (K = Kx = Ky). K depends on the typical area scale of the event, varying over the ranges 0-250 m2s-1 and 1000-10 000 m2 s-1 at the scales of a few tens of kilometres and a few hundreds of kilometres respectively (Pasquill 1974). K is not strictly a diffusion coefficient in HAZMAP. but an empirical parameter that takes into account the total factors affecting the horizontal expansion of the volcanic cloud (i.e. atmospheric eddy diffusion and gravity flow at the height of neutral buoyancy). Therefore values of K in HAZMAP are expected to be slightly greater than the values of diffusion coefficient presented by Pasquill (1974). HAZMAP investigates tephra dispersal from a system of discrete volcanic plumes, each of which is simulated as a vertical line and located at a point source i. Mass distribution along these lines is uniform unless stated otherwise. An analytical solution exists for Equation 1 for each instantaneous point source i. Each point source is processed independently due to the linearity of Equation 1, and the mass accumulated on the ground is obtained by summing that from each point source. Particle fallout is controlled by turbulent diffusion and wind transport on the horizontal axis, and by settling velocity on the vertical axis. Volcanic plumes are represented as instantaneous sources, with all the particles erupted at t = 0. Furthermore, every discrete plume and every pyroclastic flow is emplaced at the initial instant (t = 0). Particles are assumed to be spherical, and the whole grain-size distribution is transformed into a settling-velocity spectrum. Wind velocity and direction typically vary with height. In HAZMAP the atmosphere is divided into horizontal layers characterized by a uniform horizontal wind velocity and direction specific for each layer. Each point source i is located in a horizontal layer, and particles released from that point source are initially transported by the wind specific for that layer, until they fall into a lower layer, where they are affected by a different wind. This process continues until the particles reach the ground. Inside each layer, particle transport is described by the solution of Equation 1 in the .x — y plane. For emission from an instantaneous point source, the solution to Equation 1 is a Gaussian distribution of concentration in both the x and v directions. Particles spread horizontally due to the combined effects of turbulent eddy diffusion and gravity spreading of the plume, and are transported by the wind for the time - spent in each layer. tj is a function of the settling velocity vj of the particles and the layer thickness ( tj = = v j ). After the time the centre of the Gaussian distribution is shifted in the x — y plane by distances of xj = and along the axes .Y and v respectively, where wx and wy are the horizontal components of the wind speed in that layer. Particles falling from a point source i

MODELLING OF DOME-COLLAPSE AND VULCANIAN FALLOUT

located at (x i ,y i ,z i ) reach the ground at the time where ti,j = layers t j = z i/ v j- Therefore, the mass accumulation on the ground, Mi,j, of the particles j, released from the point source i is: (2)

where M0i,j is the mass of particles released from the source point i and corresponding to the settling-velocity class j, xi,j and yi,j are the co-ordinates of the centre of the Gaussian distribution (xi,j = Xi

+

layers

Xj,y i , j

2

= >yi + layers yj) and

i,j

is

the

width

°f

Parameters used in the modelling The simulation of tephra dispersal from dome-collapse and Vulcanian plumes is based on the data gathered during the eruptive period of Soufriere Hills Volcano from 1995 to 1999. Tephra dispersal depends mainly on plume height, grain-size distribution and wind profile. In the case of co-PF plumes the direction and runout of the parent pyroclastic flows are also important.

Plume height Vulcanian explosions generated two types of plume: a vent plume (main plume centred on the vent) and co-PF plumes (ash plumes from pyroclastic flows). During the eruptive period 1995-1999 of Soufriere Hills Volcano, Vulcanian vent plumes and co-PF plumes reached heights of up to 15 and 3 km above sea level (a.s.l.) respectively. Co-PF plumes from dome-collapse pyroclastic flows typically ascended to several thousands of metres. The highest co-PF plume was observed during the 26 December 1997 dome collapse, and reached about 15km a.s.l. (Mayberry el al. 2002). Given their very short periods of peak discharge, both co-PF plumes and Vulcanian vent plumes are modelled here as the instantaneous injection of a thermal of a given initial mass (Woods & Kienle 1994; Druitt et al. 2002). The height H (m) of a volcanic thermal rising into the troposphere can be expressed as: H= 1.89Q0

(3)

where Q=fxMxCx T is the excess thermal mass of the thermal injection (f is the mass fraction of the solids capable of losing their heat to the plume, M(kg) is the plume mass, C ( J k g - 1 K - 1 ) is the solids specific heat, and T(K) is the initial temperature contrast between the erupted mixture and the surrounding air). From Equation 3, a specific equation for the ventplume height Hyp (m a.s.l.) of a Vulcanian explosion was derived: Hyp = 55M0.25 + Hv -1

-1

erupted materials (Druitt et al. 2002). Given an elutriation temperature of 600 K for Vulcanian and dome-collapse pyroclastic flows (based on observed pyroclastic-flow temperatures; Cole et al. 1998; Calder et al. 1999), and 100% efficiency of thermal transfer between solids and gas in the plumes (due to the fine character of co-PF plume tephra), the equation of height of plume rise (a.s.l.) for both dome-collapse and fountain-collapse pyroclastic flows, HcoPF(m), is derived from Equation 3 as: = 45M 0.25

the

Gaussian distribution ( 2i,j = 2Kt i,j ). Definition of the physical system (total erupted mass, grain-size distribution, wind velocity profile, diffusion coefficient) and the computational grid are needed as input parameters. HAZMAP consists of four modules: the first module modifies the original grain-size distribution according to an aggregation model; the second module reads the modified grain-size distribution and redistributes the particles into settling-velocity categories according to their size and density; the third module calculates the vertical mass distribution along the column and generates the point-source distribution; the fourth module generates the output file (isomass/ probability maps) based on the input wind data, together with the input file produced by the previous module. All these modules and assumptions are discussed further below.

(4) v

where/is taken as 0.8, Cis 1100 J k g K , Tis 800K and H is the vent height (c. 1000m) (see Druitt et al. (2002) for a full description of these parameters). In Equation 4 the plume mass M is 33% of the total explosion mass, based on estimates on Montserrat that the collapsing fountain contains about two-thirds of the

519

(5)

w h e r e / i s taken as 1, C is H O O J k g ^ K r 1 and AT is 300K. Observations of co-PF plumes indicated that plume height is not constant for the whole length of the pyroclastic flow (Calder et al. 1997; Bonadonna et al. 2002). The most vigorous plume is the closest to the vent, and the height decreases away from vent (Bonadonna et al. 2002).

Grain-size distribution Grain-size distributions of fallout tephra generated by dome collapses vary with the volume of the parent pyroclastic flow (Bonadonna et aL 2002). Grain-size analyses were done on the onland fallout tephra deposited between 2 and 10km from the dome. This fraction (1 /mi to 2 cm) covers all particle sizes of fallout tephra from dome collapses, but does not include the coarsest fraction of Vulcanian fallout tephra (2-1 Ocm) in very proximal localities (i.e. within 2km from the dome; see Bonadonna et al. (2002) for full details). Fallout tephra from a small dome collapse (31 March 1997, 0.8 x 10 6 m 3 DRE; Fig. 2a) has all particle sizes below 1 mm (00), with about 45 wt% fine ash (<63 m, >4 ) ( = — Iog2d, where d is the particle diameter in mm). The grain-size distribution of fallout tephra from a large dome collapse (21 September 1997, 11 x 106 m3 DRE; Fig. 2b) shows particles below 8mm (-30), with about 70 wt% of fine ash. Fallout tephra from Vulcanian explosions is typically coarser than that from dome collapses. Most Vulcanian samples (e.g. 26 September 1997, 14:56 local time (LT); Fig. 2c) contain a coarse population with mode typically between 2 mm and 63 /on (—1 and 40), and a fine population with mode typically in the 63 to 4 //m range (4 to 70) Total grain-size distributions were calculated by averaging all samples weighted by the thickness of tephra-fallout deposits (Bonadonna et al. 2002). Scanning electron microscopy (SEM) studies of ash collected on Montserrat, and direct observations of tephra fallout, have shown that fine particles (< 100 m, >c. 3 ) mostly fell as aggregates (Bonadonna et al. 2002). The density of particles involved in the co-PF plumes from dome collapses is assumed to be 2600kgm -3 as these particles consist mainly of dense crystals and dense glass (Fig. 2d). Vulcanian ash particles with diameter of 2 mm to 8 m are modelled with a linear increase of density from 1000 kg m-3 to 2600 kg m-3 (Fig. 2d).

Mass distribution For simplicity, tephra mass is assumed to be uniformly distributed with height in all co-PF plumes (Distribution 4 Fig. 3d), whereas the mass distribution in the vent plume of large Vulcanian explosions is modelled using three other different models (Fig. 3a-c). In Figure 3, Ht is the top of the laterally spreading plume and Hb is the base, usually assumed to be the height of neutral buoyancy. For Distribution 1, all the mass is centred at a point half-way between Ht and Hb (Fig. 3a). For Distribution 2, the mass is evenly distributed between Ht and Hb (Fig. 3b). For Distribution 3 the mass is distributed according to the Suzuki (1983) model (Fig. 3c). The spacing ( zsp) along the plumes at which particles are released was studied using sensitivity tests and a suitable value of zsp = 100m achieved a good compromise between computing efficiency and accuracy (discussed further below).

520

C. BONADONNA ET AL.

Fig. 3. Diagram showing models of vent-plume mass distribution used in the simulations of Vulcanian explosions: (a) Distribution 1: (b) Distribution 2: (c) Distribution 3; and (d) Distribution 4. This diagram shows the mass distribution for a 5-km-high plume as an example (total explosion volume: 0.3 x 10 5 m 3 DRE). with an input mass in the plume of 2.8 x 10 7 kg of 100m. (c. 10 4 m 3 DRE). and a source-point vertical spacing The length of solid horizontal lines is proportional to the mass released at every plume step . The vertical dashed line indicates the plume centreline. The horizontal dashed lines indicate the top ( H t ) and the base (Hb) of the laterally spreading cloud.

density (assumed constant in this model). The drag coefficient C D is expressed as (Arastoopour et al. 1982): (7)

(8) Fig. 2. Observed grain-size distributions (weighted average; Bonadonna el al. 2002) for: (a) a small dome collapse (31 March 1997, 0.8 x 10 6 m 3 D RE); (b) a large dome collapse (21 September 1997. 11 x 10 6 m 3 DRE); and (c) a large Vulcanian explosion (26 September 1997, 14:56 LT, 0.5 x 10 6 m 3 DRE). (d) Model of particle density used to compute tephra dispersal from dome-collapse and Vulcanian plumes. Dome-collapse particle density is assumed to be constant with size (black triangle). For Vulcanian particles (grey circles) a linear increase of density with is assumed (for — l < diameter < 7 ). o — —log 2 d, where d is the particle diameter (in mm).

where Re is the Reynolds number of the particles: (9)

where // is the dynamic viscosity of the air. Grain-size distributions (Fig. 2a-c) are converted into distributions of settling velocity v (Fig. 4). where particles are represented by settling-velocity classes. Grain-size distributions are assumed to be continuous functions of o:

(10)

Settling velocity

(6)

with o = -log2d/, where d is the particle diameter (mm) and N is the weight fraction of the particles. Settling-velocity distributions are assumed to be continuous and monotonic functions of v. so that a biunique relationship between f and v is defined. The settlingvelocity function is defined as:

where g is the gravitational acceleration, d is the particle diameter, p is the particle density, CD is the drag coefficient and p a , is the air

(I1)

The settling velocities of the particles are obtained by the balance between gravity and air drag:

MODELLING OF DOME-COLLAPSE AND VULCANIAN FALLOUT

521

Fig. 4. Settling-velocity distribution of (a) 'non-aggregated' and (b) 'aggregated' fallout tephra from a small dome collapse (31 March 1997), and (c) 'non-aggregated' and (d) 'aggregated' fallout tephra from a large Vulcanian explosion (26 September 1997, 14:56 LT). See text for details of settling-velocity classes. Corresponding 'non-aggregated' and 'aggregated' grain-size distributions are shown in Fig. 2a and c, and Figs 8c (Model 3) and 9d (Model 4) respectively.

Combining Equations 10 and 11, the particle fraction Nn, for each settling-velocity class n, is given by:

(12)

Values of and , corresponding to each vn and vn-1, are calculated using Equation 6. is not a continuous distribution, as its values are known at sieving steps. To perform the integration (Equation 12), a uniform distribution of particles within each sieving step is assumed. The variation of particle density with grain size used in the calculations is shown in Figure 2d. When aggregation processes are taken into account, "nonaggregated' grain-size distributions (i.e. original grain-size distributions, consisting of individual free particles; Fig. 2a-c) are first converted into "aggregated' grain-size distributions (using aggregation models discussed further below), and then converted into settling-velocity classes (Fig. 4b and d). For typical grain-size distributions of fallout tephra from dome collapses and Vulcanian explosions on Montserrat, the whole population of particles can be represented by 16 classes. The first class (Class 1: 1 m s - 1 ) represents particles with settling velocity from 0 to 1 . 5 m s - 1 , the second class (Class 2: 2 m s - 1 ) represents particles with settling velocity from 1.5 to 2 . 5 m s - 1 , the third class (Class 3: 3 m s - 1 ) represents particles with settling velocity from 2.5 to 3 . 5 m s - 1 , and so on until Class 16 (i.e. particles with settling velocity from 15.5 to 16.5ms - 1 ).

5-8 km and 18-20 km, typical for the Eastern Caribbean subtropical climate. The high-level shear corresponds to the tropopause position (Fig. 5). Low-level (1-5 km) and high-level (20-30 km) winds typically blow west. Intermediate-level winds (8-18 km) typically blow east. The standard deviation for the direction values varies between about 30° and 162°. Therefore, low-level winds can also blow to the NW, SW or south, and intermediate-level winds to the north, NW, NE, SW, SE or south. Wind velocities range between 3 and 1 0 m s - 1 for low-level winds, between 4 and 2 3 m s - 1 for intermediate-level winds, and between 4 and 2 0 m s - 1 for highlevel winds (Fig. 5b).

Wind profile HAZMAP reads wind data as north and east vectors for any wind vertical spacing , then interpolates the data to fit them to the source-point vertical spacing that characterizes the computed plumes from which the particles are released. North and east vectors give direction and velocity of wind at a specific height. Figure 5 shows vertical profiles of wind velocity and direction (with corresponding standard deviation) above Montserrat for the period 1992-1997, based on data from the National Oceanic and Atmospheric Administration (NOAA) Climate Diagnostic Center (http:// www.cdc.noaa.gov/). In this study wind vectors are given in 30 levels starting from 1 km a.s.l. and spaced at 1 km intervals (i.e. = 1 km). Figure 5a shows two evident wind shears at about

Fig. 5. Average (a) wind direction (provenance + 180°) and (b) wind speed over the period 1992-1997, with corresponding standard deviations. Archived daily averages of wind direction and speed for 17 pressure levels were provided through the NOAA Climate Diagnostic Center. The average wind profile for 30 levels (i.e. wind-profile vertical spacing, = 1 km) with corresponding standard deviation was calculated. Prevailing easterly lowand high-level wind direction and westerly intermediate-level wind with two wind shears (5-8 km and 18-20 km) are evident. The shaded area shows the fluctuation of the tropopause above Montserrat (W. I. Rose, pers. comm.).

C. BONADONNA ET AL.

522

Diffusion

coefficient

A value of 3000m2 s-1 was found as a best fit of observed data for the Mount St Helens 1980 eruption (Armienti et al. 1988). A value of 2700m 2 s-1 is used here for simulations of both dome collapses and Vulcanian explosions. Sensitivity tests are discussed further below.

Directions of pyroclastic flows During the eruptive period 1995-1999 of Soufriere Hills Volcano, pyroclastic-flow activity was confined to seven valleys around the volcano: Fort Ghaut (west). White River valley (south), Tar River valley (east). White's Ghaut (NE), Mosquito Ghaut and Tuitt's Ghaut (NNE), Tyre's Ghaut (north) (Fig. 1).

Runouts of pyroclastic flows The runout of a pyroclastic flow is the length over which co-PF plumes are sourced. Data collected on Montserrat (Calder et al. 1999) yield the following empirical relationship between the runout of a dome-collapse pyroclastic flow and its volume: RDC = 21851og 10 6

6

V-9645

(13) 2

(0.2 x 10 < V < 14.3 x 10 ; number of points = 10; R = 0.85) where RDc is the dome-collapse pyroclastic-flow runout (m) and V is the pyroclastic-flow volume (m 3 ). Fountain-collapse pyroclastic flows are more mobile than dome-collapse pyroclastic flows; therefore the runout distance is determined using the empirical equation (from data in Calder et al. 1999): RFC= 1950log 10 F- 5519

(14)

(0.02 x 106 < V < 0.14 x 106; number of points = 4; R2 = 0.77) where R F c is the fountain-collapse flow runout (m) and Kis the pyroclastic flow volume (m 3 ).

Particle source horizontal co-ordinates A system of point-source co-ordinates r = (Xi, Y) is generated based on the following assumptions. First, discrete plumes are modelled 1 km apart down the pyroclastic-flow valleys (Fig. 6). This assumption is based on observations of typical spacing between discrete plumes observed in pyroclastic flows of Soufriere Hills Volcano and from the 7 August 1980 Mount St Helens pyroclastic flow (Hoblitt 1986; Levine & Kieffer 1991; Calder et al. 1997). Second, an exponential decay of elutriated mass along the pyroclastic-flow valley is assumed, starting from the co-PF plume located 1 km from the dome. The heights of co-PF plumes consequently decreases with distance (from Equation 5). This distribution of plume mass is supported by observations. The highest plume is formed close to the dome base, as it forms at the base of the dome talus where a significant break in slope enhances elutriation (Calder et al. 1997). Plume heights are observed to decline with distance (Bonadonna et al. 2002). For dome-collapse simulations 10% of the total elutriated mass is assigned to the first plume centred on the dome (Fig. 7a). The remaining mass decreases exponentially from the second plume at 1 km distance (Fig. 7a) according to the empirical equation: (15a)

with (15b)

Fig. 6. Map of Montserrat showing the source distribution, used in modelling, of co-PF plumes along each of the main valleys around the volcano. Source points are 1 km apart (black diamonds). The position of Soufriere Hills Volcano is also shown (black triangle). For key to names of the valleys, see caption to Figure 1.

where is the distance from the dome, R is the pyroclastic-flow runout, and 0.9M T O T is the total elutriated mass from all the co-PF plumes, excluding the co-PF plume centred on the dome, a is an empirical factor for decrease of elutriated mass down the valley of collapse. Sensitivity tests, discussed further below, show that the best fit to observed data is given for a =0.1. Corresponding plume heights (Fig. 7b) are calculated using Equation 5. For simulations of Vulcanian explosions, the vent-plume height is calculated using Equation 4. The elutriated mass in each co-PF plume is distributed assuming an exponential decrease starting from the first co-PF plume (positioned 1 km away from the vent plume) (Fig. 7c). Therefore A/(r i ) is calculated using Equation 15a with: (15c)

where MTOT is the total elutriated mass. Co-PF plume height is then calculated from the elutriated mass using Equation 5 (Fig. 7d). In the case where the computed runout (calculated using Equations 13 and 14) is longer than the valley concerned, the calculation is truncated at the coastline (i.e. pyroclastic-flow runout = valley length), and the remaining mass is assumed to be lost when the pyroclastic flow enters the sea.

MODELLING OF DOME-COLLAPSE AND VULCANIAN FALLOUT

523

Fig. 7. Computed (a) elutriated masses and (b) heights of co-PF plumes versus distance from the dome for a 107 m3 (DRE) dome collapse (computed runout: 6 km < valley length). Ten per cent of the total elutriated mass is assigned to the plume located on the dome (x = Okm); elutriated mass is then assumed to decrease exponentially, starting from the plume located 1 km from the dome (x = 1 km). Computed (c) masses and (d) heights of vent plume (white bars) and co-PF plumes (black bars) versus distance from vent for a 5 x 105 m3 (DRE) Vulcanian explosion. Elutriated masses and heights of Vulcanian co-PF plumes are simulated for a fountain collapse down four valleys as an example, but for simplicity only co-PF plumes from one valley are shown (computed runout: 4km < valley). In (c) 33% of the total mass is assigned to the vent plume (x =0 km); elutriated mass in each valley is then assumed to decrease exponentially starting from the first co-PF plume (x = 1 km). Co-PF plume heights in (b) and (d) are calculated using Equation 5; vent-plume height in (d) is calculated using Equation 4.

Volumes of dome collapses and Vulcanian explosions

HAZMAP outputs

The eruption of Soufriere Hills Volcano from 1995 to 1999 produced dome collapses of various sizes, up to a maximum of about 45 x 10 6 m 3 DRE (26 December 1997 dome collapse; Calder et al. 2002). Vulcanian explosions in the period August-October 1997 each discharged an average of 0.3 x 10 6 m 3 DRE of magma (Druitt et al. 2002). Peak discharge lasted 45-70 seconds and an average of 67% of the rising material would collapse back, generating fountain-collapse pyroclastic flows down all or some of the valleys around the volcano (Druitt et al. 2002). DRE volumes were converted into mass assuming a density of 2600 kg m-3 for dense rock.

HAZMAP produces two different output results: isomass and probability maps. Isomass maps show the accumulation of mass per unit area described by Equation 2 for one specific wind profile, and are mainly used to test the model and to assess the agreement with field data. Probability maps give the probability distribution of a particular mass loading around the volcano based on the statistical distribution of wind profiles, and are used for assessment of tephrafallout hazard. Mass-loading values (i.e. deposit thresholds) are based on certain threshold values of tephra mass per unit area that produce specific types of damage (e.g. collapse of building roofs, different degrees of damage to vegetation and crops). Probability maps and deposit thresholds are described further below.

Elutriated mass Bonadonna et al. (2002) show that 4-16% of the pyroclastic flow mass was elutriated forming co-PF plumes from both dome collapses and Vulcanian explosions in Montserrat. Here the volume percentage of ash elutriated from the pyroclastic flow is taken as a typical value of 10% of the DRE volume of the associated pyroclastic-flow deposit.

Computational grid HAZMAP uses geographical co-ordinates (in m) in order to overlap the output file on a digital elevation model (DEM) map of Montserrat. Therefore co-ordinates of plume sources (in the input file for module 4) are given in geographic co-ordinates. The output grid is 20 x 20 km, with a spacing between nodes on the x and y axes of 100m.

Aggregation models Direct observations and experimental studies have shown that most particles <100 m in diameter fall as aggregates of three main types, according to the amount of liquid involved (Sparks et al. 1997): dry aggregates, accretionary lapilli and mud rain. Aggregation processes cause premature fallout of tephra, can generate anomalous deposit thicknesses (e.g. Carey & Sigurdsson 1982; Hildreth & Drake 1992), and can also be responsible for polymodal grain-size distributions of tephra-fallout deposits (e.g. Carey & Sigurdsson 1982; Brazier et al. 1983). Fallout tephra from dome collapses and Vulcanian explosions from Soufriere Hills Volcano is rich in fine ash, with typically 4070wt% of particles with diameters <63/im. Sedimentation of particles during both types of eruptive activity was significantly affected by aggregation, as shown by direct observations and SEM studies (Bonadonna et al. 2002). Therefore, incorporation of

524

C. BONADONNA ET AL.

particle aggregation into the simulations of tephra fallout from both dome collapses and Vulcanian explosions on Montserrat is crucial. However, the types, sizes and amounts of aggregates formed during tephra fallout varied significantly from case to case (Bonadonna el al. 2002). Here we combine results from previous studies on particle aggregation and observations made on Montserrat. Four different aggregation models are presented and compared with field data (Table 1). Model 1 is the result of the best fit of computer simulations with the Campanian Y-5 ash field data (Cornell el al. 1983). It assumes aggregation of 50wt% of particles in the size range 125-63 m, 75wt% of particles in the size range 63-31 m, and 100 wt% of particles <31 m. All aggregates (of unspecified type) have a diameter of 250 m ( = 2) and a density of 2000 k g m - 3 . Models 2, 3 and 4 are based on observations (Bonadonna el al 2002) and are specific to the formation of accretionary lapilli due to the aggregation of particles < 125 m in the presence of water in the volcanic plume. The diameter (m) of the accretionary lapilli is determined by the equation (Gilbert & Lane 1994): (16)

where w is the plume mass loading (kgm - 3 ), E the aggregation coefficient, x the thickness of the umbrella cloud (m), p the particle the aggregate porosity and the pore density density (kgm (kgm - 3 ). Values used for and are 2600 kg m - 3 , 0.4 and 1000 kg m-3 respectively. The aggregate density used is 1840kgm - 3 , the weighted average between particle and pore density. The plume mass loading is calculated for dome-collapse and Vulcanian plumes considered as discrete volcanic thermals using the model of Sparks el al (1994). In these calculations, temperatures are assumed to be around 600 K (from measurements of pyroclasticflow temperatures; Cole el al. 1998, 2002; Calder el al 1999) and 1100K (andesitic magma temperature) for (i) dome-collapse and Vulcanian co-PF plumes, and (ii) Vulcanian vent plumes respetively. The aggregation coefficient E depends on the particle size, and for Models 2, 3 and 4 three different values are used (from Gilbert & Lane 1994): 0.3 (for particles in the size range 125-31 m ), 0.75 (for particles in the size range 31-8 m ) and 1.0 (for particles in the size range 8-1 m). Therefore, using these three values of E, three different size categories of accretionary lapilli are determined from Equation 16 for each dome collapse based on the highest co-PF plume. In simulations of Vulcanian explosions, the heights of the vent plume and of the associated co-PF plumes can be significantly different. Therefore, three size categories of accretionary lapilli are determined for both vent plume and the highest co-PF plume, resulting in a total of six size categories for each explosion simulated. The proportions of ash of different size fractions involved in the formation of the accretionary lapilli (weighted over the whole fallout tephra) vary significantly from case to case (Bonadonna el al. 2002). Models 2, 3 and 4 differ in the fractions of different particle

sizes involved in the aggregation process (Table 1). Model 2 is used as the ash-rich end-member and assumes aggregation of 100 wt% of particles with diameters < 125 m (Table 1). Models 3 and 4 are based on a sequence of samples collected during 5- or 10-minute periods during the Vulcanian explosions of 28 September 1997 (23:03 LT) and 1 October 1997 (11:34 LT) respectively (Bonadonna el al. 2002). Both explosions produced tephra fallout from co-PF plumes and from the vent plume. During the 28 September explosion, about 95wt% of tephra fell in Old Towne (in Fig. 1) as accretionary lapilli up to 4mm in diameter over the first 10 minutes of tephra fallout. During the 1 October explosion, about 35 wt% of tephra fell as accretionary lapilli up to 2mm in diameter over the last 10 minutes of tephra fallout at the same location. Proportions of particles involved in the aggregation process are given for four size ranges of <125 m and weighted over the whole tephrafallout deposit: 125-63 m. 63-8 m, 8-4 m , 4-2 m (Table 1). In Models 2, 3 and 4 the amount of aggregated ash is then partitioned equally amongst the three categories of accretionary lapilli calculated using Equation 16. As an example, Figures 8 and 9 show the results of applying the four aggregation models to the grain-size distributions for the 31 March 1997 dome collapse and 26 September 1997 (14:56 LT) Vulcanian explosion respectively. When 'non-aggregated' grain-size distributions (Figs 2a-c and 4a. c) are converted into 'aggregated' grain-size distributions (Fig. 4b and d). a significant bimodality in the corresponding settling-velocity spectrum is shown.

Results We have run simulations to investigate the sensitivity of results to the main parameters and assumptions, the main factors being diffusion coefficient, aggregation model, grain-size distribution, plume mass distribution of Vulcanian vent plume, source-point vertical spacing , factor of elutriated-mass decrease in co-PF plumes Q, and volume of the erupted material. All these parameters have been cross-checked to establish the absolute best fit to field data. Simulations presented here (Table 2) were carried out by varying the investigated parameter and by using the values that give the best fit to field data for the other parameters. Best-fit results were obtained by minimization of the misfit function mf: (17)

where N is the number of data and xobs and xcomp are the observed and computed mass accumulation per unit area respectively. The misfit function is an estimate of the global agreement between observed and computed data, and so observations and model results were also compared at individual locations.

Table 1. Description of the aggregation models used in the simulations

Aggregate type Aggregate diameter ( m) Aggregate density ( k g m - 3 ) Wt% of particles in aggregates: 125-63 m 63-3 1 m <31 m

Model 1

Model 2

Model 3

Model 4

Not specified 250 2000

Accretionary lapilli 180-11500* 1840

Accretionary lapilli 180-11500* 1840

Accretionary lapilli 180-11500* 1840

50 75 100

Wt% of particles in accretionary lapilli: 125-63 Aim 63-8 m 8-4 m 4-2 m

100 100 100 100

91 95 91 63

42 35 48 67

Model 1 represents a slightly modified version of an aggregation model presented by Cornell el al. (1983). Models 2, 3 and 4 are based on observations made on Montserrat. * Accretionary lapilli diameters are calculated using Gilbert & Lane (1994) theory, and computed for volcanic thermals with heights ranging from 2 to 15km and temperatures ranging from 600 to 1100K.

MODELLING OF DOME-COLLAPSE AND VULCANIAN FALLOUT

525

Fig. 8. Application of four aggregation models (Table 1) to the original grain-size distribution (i.e. individual 'non-aggregated' particles) of fallout tephra from the 31 March 1997 dome collapse (black circles): (a) Model 1 (Cornell et al. 1983); (b) Model 2; (c) Model 3; (d) Model 4. The four models differ for the fraction of particles involved in the aggregation process. A certain fraction of free particles (i.e. non-aggregated particles) (grey bars) is aggregated in clusters of unspecified type (white bars) in Model 1 and in accretionary lapilli (black bars) in Models 2-4.

Fig. 9. Application of four aggregation models (Table 1) to the original grain-size distribution (i.e. individual 'non-aggregated' particles) of fallout tephra from the 26 September 1997 (14:56 LT) Vulcanian explosion (black circles): (a) Model 1 (Cornell et al. 1983); (b) Model 2; (c) Model 3; (d) Model 4. The four models differ for the fraction of particles involved in the aggregation process. A certain fraction of free particles (i.e. nonaggregated particles) (grey bars) is aggregated in clusters of unspecified type (white bars) in Model 1 and in accretionary lapilli (black bars) in Models 2-4.

Effect

of diffusion

coefficient

(Runs 1 in Table 2)

The effective horizontal diffusion depends on the typical area scale of the event (Pasquill 1974), and, near source, also on gravity spreading of the plume. Dome collapses and Vulcanian explosions generated by Soufriere Hills Volcano are of comparable scale, and so it was anticipated that best-fit values of diffusion coefficient would be similar for both kinds of activity. Sensitivity tests for diffusion coefficient were made on the fallout tephra from the 31 March 1997 dome collapse, as this has the most complete data set (Runs 1). The best fit to the data is 2700m 2 s-1 (Fig. 10), which is very similar to the best-fit value found for the Mount St Helens 1980 eruption (3000m 2 s - 1 ; Armienti et al. 1988). Effect

of aggregation processes (Runs 2 in Table 2)

The effect of particle aggregation was investigated for the cases of the 31 March 1997 dome collapse (Runs 2a) and the 26 September

Fig. 10. Diffusion coefficient versus misfit function (mf in Equation 17) for the tephra-fallout deposit from the 31 March 1997 dome collapse. The lowest (i.e. best fit) is obtained for a diffusion coefficient of 2700m 2 s-1 (Runs 1 in Table 2).

C. BONADONNA ET AL.

526

Table 2. Parameters used in the simulations of dome collapses and Vulcanian explosions Runs

Test

Sensitivity 1 Diff. coeff. 2a Aggreg. model 2b Aggreg. model 3a Grain size 3b Grain size 4a Vent-plume mass distr. 4b Vent-plume mass distr. 5 6 7a 7b 8a 8b

zsp a

Volume Volume Best fit Best fit

General results Grain size Diff. fallout Diff. fallout Diff. fallout Diff. fallout

9 l0a l0b l0c l0d

Individual probability maps 11 Thresholds: 12, 120, 180 k g m - 2 12 Thresholds: 12, 120 kgm- 2

Figure

Event

Diffusion coefficient (m2s-1)

Grain-size Vent-plume distribution mass distribution

10 11 12 13a 13b 14 15 16a 16b 16c 16c 17 17

DC DC

1000-10000 2700 2700 2700 2700 2700 2700 2700 2700 2700 2700 2700 2700

A A B

18 19a 19a 19b 19b

VE (EC) DC

VE (FC) VE (FC) VE (NFC) DC DC DC VE DC

VE (FC) DC

VE (vent) VE (co-PF) VE (vent) VE (co-PF)

20a-e, 21 DC 20f

VE (FC)

Aggregation model

Volume DRE (x106m3)

Wind z sp profile (m)

B B A A A B A B

— 4 1-4 1-4 4 4

3 0-4 0-4 0-4 0-4 4 4 3 3 3 4 3 4

0.8 0.8 0.5 0.8 0.5 0.5 0.5 0.8 0.8

I I II I II II II I I I I I II

2700 2700 2700 2700 2700

A B B B B

4 4 4 4

3 4 4 4 4

0.8 0.2

2700

C

-

2700

B

4

A, B A, B

100 100 100 100 100 100 100

5-1500

0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1

100 100 100 100 100

0.1 0.1 0.1 0.1

0.03* 0.02f 0.005 *

I II II III III

100 100 100 100 100

0.1 0.1 0.1 0.1 0.1

3

10

TOT

100

0.1

4

1

TOT

100

0.1

0.2-1.5 0.2-1 0.8 0.5

0.04-100

Values in bold are the parameters varied to test the sensitivity of the model. Simulations: DC, dome collapse: VE. Vulcanian explosion: VE (FC). Vulcanian explosion computed with fountain collapse; VE (NFC), Vulcanian explosion computed without fountain collapse: VE (vent): contribution of vent-plume tephra fallout only; VE (co-PF), contribution of co-PF plume tephra fallout only. Grain-size distributions are shown in Fig. 2 (A. 31 March 1997 dome collapse; B, 26 September 1997 (14:56 LT) Vulcanian explosion; C, 21 September 1997 dome collapse). Vent-plume mass distributions 1. 2. 3 and 4 are described in Fig. 3. Aggregation models 1, 2, 3 and 4 are described in Table 1. Model 0 indicates the case of no aggregation. Volumes used in these simulations are from field data (Bonadonna et al. 2002), apart from: * total co-PF tephra volume (i.e. 10% of pyroclastic-flow DRE volume); total volume of the 10 October 1997 (18:40 LT) Vulcanian explosion was calculated using Equation 4 and the plume height provided by NOAA (in Druitt et al. 2002) reduced by 20% as suggested by Druitt et al. (2002). Wind profiles used in simulations are interpolated according to the characteristic Zsp (i.e. source-point vertical spacing), and determined for each specific day when the event simulated occurred (I, 31 March 1997; II, 26 September 1997; III, 10 October 1997). TOT. total data (i.e. 2192 daily wind profiles for the period 1992-1997; see text for details). zsp is the vertical spacing of point sources along the plumes, a is the coefficient of elutriated-mass decrease along the pyroclastic flow (in Equation 15).

1997 (14:56 LT) Vulcanian explosion (Runs 2b) separately. The four aggregation models used give significantly different results for dome-collapse and Vulcanian fallout tephra (Figs 11 and 12). The computed deposit is thicker in proximal areas for models in which more particles are involved in aggregation (Models 2 and 3), as might be expected. When particle deposition is simulated with aggregation, proximal mass loadings are up to ten and four times those for no aggregation (Figs 1 1b and 12b), in the cases of the dome collapse and the Vulcanian explosion respectively. The grain-size distribution in the Vulcanian simulations is coarser than that of dome-collapse simulations. Therefore, a significant fraction of fall- out tephra in Vulcanian simulations is expected to accumulate in proximal areas, even when aggregation does not take place. This also results in a smaller variation of the misfit function ( mf ) for Vulcanian simulations compared to dome-collapse simulations ( mf= 0.3 and 1.0kgm - 2 respectively; Figs 12g and llg). Model 3 gives the best fit for the 31 March 1997 dome collapse (Fig. llg), whereas Model 4 gives the best fit for the 26 September 1997 explosion (Fig. 12g), even though Vulcanian simulations are less sensitive to aggregation processes than those of dome collapses. Effect

of grainsize distribution (Runs 3 in Table 2)

Tephra-fallout deposits from the 31 March 1997 dome collapse and the 26 September 1997 (14:56 LT) Vulcanian explosion were com-

puted using the four aggregation models in Figures 8 and 9 applied to the two respective total grain-size distributions (Fig. 2a and c). The results do not vary significantly when the grain-size distribution is varied (Fig. 13). This can be explained by the aerodynamic similarities of coarse ash and lapilli, which characterize Vulcanian tephra, and fine-ash aggregates, formed during both Vulcanian and dome-collapse tephra fallout. Effects of premature fallout of fine ash caused by aggregation during fallout of finegrained tephra can be reproduced with simulations run with coarser grain-size distributions and vice versa. Effect

of plume mass distribution (Runs 4 in Table 2)

The effect of vertical mass distribution within the Vulcanian vent plumes was investigated using the four different distribution models in Figure 3. The mass distribution in the co-PF plumes is maintained uniform (Distribution 4 in Fig. 3), as some of the co-PF plumes are too low to be sensitive to differences in the mass distribution. These models were tested on the 26 September 1997 (14:56 LT) Vulcanian explosion. Simulation results are only weakly sensitive to different ventplume mass distributions (Fig. 14). However, the best agreement with observations was obtained using the uniform mass distribution (Distribution 4) (Fig. 14e). The double accumulation maximum in Figure 14a and b is due to the simultaneous release of particles from two tephra-fallout sources with a narrow range of source

MODELLING OF DOME-COLLAPSE AND VULCANIAN FALLOUT

527

Fig. 11. Isomass maps (kgm - 2) of the 31 March 1997 (morning) dome collapse showing (a) field data, and the same deposit computed using (b) the original ('non-aggregated') grain-size distribution (in Fig. 2a), and the original distribution "aggregated' using: (c) Model 1; (d) Model 2; (e) Model 3; (f) Model 4 (in Fig. 8). (g) Misfit function ( m f i n Equation 17) calculated for computed deposits (b)-(f) in comparison to field data in (a) (Runs 2a in Table 2). TF, tephra-fallout deposit; PF, pyroclastic-flow deposit; NA, non-aggregated grain-size distribution. In (a) localities a, b and c (also in Fig. 1), used for investigations of computed grain size (Fig. 18), are shown as circles. Contours of 0.05, 0.1, 1 and 5 k g m - 2 are shown in all maps. Contours of 0.01, 10, 15 and 2 0 k g m - 2 are also shown for computed deposits.

points (i.e. vent plume with Distributions 1 and 2 and low co-PF plumes). The effect of vent-plume mass distribution was also investigated using the same Vulcanian explosion, but without fountain collapse (i.e. all erupted mass concentrated in the vent plume, Runs 4b) (Fig. 15). Distributions 3 and 4 give slightly better fits compared to Distributions 1 and 2. Tephra dispersal computed using Distributions 1 and 2 (Fig. 15a and b) is dominated by intermediate-level winds, as all the particles are released high in the column.

Effect of source-point vertical spacing ( zsp) (Runs 5 in Table 2) The spacing of vertical source points up the simulated eruptive plumes ( zsp) can create numerical artefacts, such as a shift downwind of the mass accumulation maximum and artificial mass accumulation double maxima. These numerical artefacts occur when the vertical spacing ( zsp) is too large and the accumulation of particles from different point sources does not overlap. Sensitivity tests carried out using fallout tephra from the 31 March 1997 dome collapse show that decreasing zsp improves the numerical accuracy (Runs 5, Fig. 16a). A zsp value of 100m was chosen as a compromise between minimizing the computing time and maximizing the computing accuracy.

Effect of decrease of elutriated mass along the pyroclastic flow (mass-decrease coefficient a in Equation 15; Runs 6 in Table 2) Co-PF plume mass is assumed to decrease exponentially with distance from the dome. In order to determine the influence of the empirical mass-decrease coefficient a in Equation 15, sensitivity tests were carried out on the 31 March 1997 dome-collapse fallout tephra (computed PF runout = 3 km, Table 3). Only values of a in the range 0.04-10 give results consistent with observations. Smaller values result in the concentration of elutriated mass in the co-PF plumes closest to the vent, and would require a grid and a physical model of higher resolution. Larger values result in a constant height of co-PF plumes. The best fit to observed data is given by a = 0.1 (Fig. 16b). Variations of a do not significantly affect the computed deposit for the 26 September 1997 (14:56 LT) Vulcanian explosion. We also use a = 0.1 for Vulcanian explosions.

Effect of volume (Runs 7 in Table 2) Results are sensitive to the variation of the input volume, both for dome collapses and Vulcanian explosions (Fig. 16c). The best fit to the data is given by a total DRE volume of 0.8 x 106 and

528

C. BONADONNA ET AL.

Fig. 12. Isomass maps ( k g m - 2 ) of the 26 September 1997 (14:56 LT) Vulcanian explosion showing (a) field data, and the same deposit computed using (b) the original ('non-aggregated') grain-size distribution (in Fig. 2c), and the original distribution 'aggregated" using: (c) Model 1: (d) Model 2: (e) Model 3: (f) Model 4 (in Fig. 9). (g) Misfit function ( m f i n Equation 17) calculated for computed deposits (b)-(f) in comparison to field data in (a) (Runs 2b in Table 2). TF, tephra-fallout deposit; NA, non-aggregated grain-size distribution. Contours of 1 2. 10 and 1 5 k g m - 2 are shown in all maps. Contours of 0.01. 0.1. 5. 20. 30 and 4 0 k g m - 2 are also shown for computed deposits.

0 . 5 x l 0 6 m 3 for the 31 March 1997 dome collapse and the 26 September 1997 (14:56 LT) Vulcanian explosion respectively, in good agreement with field data (Bonadonna et al. 2002). Best fit to field data and local differences (Table 3 and Runs 8 in Table 2) All the above sensitivity tests were carried out on the two tephrafallout deposits with the most complete observations: 31 March 1997 dome collapse (number of sample points, 18; distances from the dome, 2-9 km; mass-accumulation range, 0.02-7.3 k g m - 2 ; mf = 0.4kgm- 2 ) and 26 September (14:56 LT) 1997 Vulcanian explosion (number of sample points, 12; distances from the dome, 6-9 km; mass-accumulation range, 0.02-16.6 kg m - 2 ; mf= 3 . 9 k g m - 2 ) . Local differences between the best-fit computed model and observations were also investigated (Fig. 17). The greatest differences were recorded for very low values of mass accumulation for both deposits (local difference, 70-500%; mass range, 0.02-1.0 k g m - 2 ) . This may be a result of the high measurement error involved in the sampling of very thin deposits. The local difference of the largest values of mass accumulation available (mass range, 1.0-16.6kgm -2 ), varies between 2 and 50%. Table 3 shows that there is good agreement between computed and observed runout (from Equation 13) and column height (from Equations 4 and 5) for both cases studied.

Variation of grain-size distribution with distance from vent (Runs 9 in Table 2) Tephra-fallout deposits computed with HAZMAP. using one of the aggregation models, consist of both aggregates and free individual particles (e.g. Fig. 18). Aggregates are generally not well preserved in the observed tephra-fallout deposits, and only the 'disaggregated' distribution is available from field data. In order to compare computed and observed grain-size distributions, computed aggregates were numerically 'disaggregated' and the resulting 'disaggregated' particles were added to the computed free particles fallen at one locality (Fig. 18). The comparison between computed grainsize distribution and observed grain-size distribution at three different localities for the 31 March 1997 dome collapse (localities in Figs 1 and 1 la) shows good agreement. Computed data are consistent with about 95 wt% of tephra falling as accretionary lapilli up to 2 mm in diameter at locality a (Fig. 18a). The model also indicates a significant decrease of the proportions of accretionary lapilli with distance from the dome, being absent at the most distal locality (locality c, 9 km from the dome). As in our simulations accretionary lapilli are released from the plume at t = 0 together with all the other particles, we conclude that during the 31 March 1997 collapse most of the accretionary lapilli must have formed in the proximal area of the volcanic cloud and also accumulated in the proximal deposit. Bimodality, however, is predicted even when accretionary lapilli are absent (Fig. 18c).

MODELLING OF DOME-COLLAPSE AND VULCANIAN FALLOUT

Fig. 13. Comparison of the misfit function (mf/in Equation 17) obtained for: (a) the computed tephra-fallout deposit from the 31 March 1997 dome collapse, simulated with the grain-size distribution from the 31 March 1997 dome collapse (DC1, black bars, Fig. 2a), from the 21 September 1997 dome collapse (DC2, grey bars, Fig. 2b), and from the 26 September 1997 (14:56 LT) Vulcanian explosion (VE, white bars, Fig. 2c); (b) the computed tephra-fallout deposit from the 26 September 1997 (14:56 LT) Vulcanian explosion, simulated with the three grain-size distributions as above. NA, non-aggregated grain-size distribution; Models 1-4, aggregation models in Table 1 (Runs 3 in Table 2).

Differential fallout from Vulcanian explosions (Runs 10 in Table 2) Fallout tephra generated by Vulcanian explosions with fountain collapse is a combination of contributions from the vent plume and co-PF plumes. The heights of vent and co-PF plumes typically differ significantly. Particularly when the wind direction varies with altitude, these two types of fallout tephra are characterized by different dispersal (here called differential fallout). Simulations of tephra fallout from the vent plume and from co-PF plumes were compared for two Vulcanian explosions (26 September 1997, 14:56 LT, and 10 October 1997, 18:40 LT; Fig. 19). Although there is considerable overlap in dispersal, there is no significant deposition of computed co-PF ash in areas of northern Montserrat from the earlier Vulcanian explosion (Fig. 19a), in agreement with observations (Bonadonna et al 2002). Computed fallout tephra from the vent plume generated by the later Vulcanian explosion has a much wider dispersal than that from co-PF plumes (Fig. 19b). Differential fallout can strongly affect general grain-size characteristics of tephra-fallout deposits, as high vent plumes are typically coarser grained than the associated co-PF plumes.

Probability maps Assessment of fallout tephra hazard is here based on the probability of reaching certain mass-accumulation thresholds in a particular

529

Fig. 14. Computed isomass maps (kgm - 2) of the tephra-fallout deposit from the 26 September 1997 (14:56 LT) Vulcanian explosion, simulated with different mass distributions in the vent plume (Fig. 3): (a) Distribution 1; (b) Distribution 2; (c) Distribution 3; (d) Distribution 4. (e) Misfit function (mfin Equation 17). TF, computed tephra-fallout deposit. Contour interval is 1 k g m - 2 for (a) and (b), and 3 k g m - 2 for (c) and (d) (Runs 4a in Table 2).

area. Hazardous deposit thresholds (Blong 1984) are estimated for a 1200 k g m - 3 deposit density and from observations made on hazardous effects on Montserrat. Experience on Montserrat indicates that tephra accumulation of 1 2 k g m - 2 (equivalent to approximately 1 cm) causes minor damage to agriculture (also suggested by Blong 1984). Accumulation of fallout tephra of 12 k g m - 2 is also useful for assessing health effects, as deposits thicker than about 1 cm were observed to persist on the ground for many weeks and months, and can be remobilized, causing health hazards (Moore et al. 2002). A threshold of 120kgm - 2 is approximately the value for collapse of wooden and corrugated roofs on Montserrat. Accumulations of 250 k g m - 2 or more are required for failure of concrete roofs. These thresholds reduce when tephra is wet. Other deposit thresholds are (Blong 1984): 1 8 0 k g m - 2 (partial survival of vegetation: zone 2); 600kgm - 2 (partial survival of vegetation: zone 1); 1800 kg m-2 (zone of near-total vegetation kill); and 2400 kg m-2 (zone of total vegetation kill). Only the deposit thresholds of 12, 120, 180 and 600kgm - 2 are considered in this study, as they represent the most significant values for the assessment of tephra fallout hazard from eruptive events on a Montserrat scale. Probabilities of reaching significant thresholds were calculated for fallout tephra produced both by dome collapses and Vulcanian explosions, based on wind statistics. Unless stated otherwise, six

530

C. BONADONNA ET AL.

Fig. 15. Computed isomass maps ( k g m - 2 ) of the tephra-fallout deposit from the 26 September 1997 (14:56 LT) Vulcanian explosion, simulated with no fountain collapse and with different mass distributions in the vent plume (Fig. 3): (a) Distribution 1; (b) Distribution 2; (c) Distribution 3; (d) Distribution 4 (in Fig. 3). (e) Misfit function (mfin Equation 17). TF, computed tephra-fallout deposit. Contour interval is 2 k g m - 2 for (a) and (b), and 5 k g m - 2 for (c) and (d) (Runs 4b in Table 2).

years of daily wind data were used (1992-1997; Fig. 5). Therefore, 2192 isomass maps (i.e. daily isomass maps for six years) were calculated using daily wind profiles, then compared with the hazardous deposit thresholds. Contributions from daily isomass maps were calculated and summed in each node of the grid. Finally, all the values were renormalized to 100% probability of exceeding the deposit threshold. Probability maps were compiled for individual eruptive events (individual probability maps) and for a given scenario of activity resulting from the occurrence of several eruptive events during a certain period of time (cumulative probability maps).

Probability maps for individual events A DRE volume of 107 m3 was used in the compilation of individual probability maps for dome collapses, as this volume could have a significant impact on populated areas. Such a volume is comparable to large dome collapses on Montserrat, such as 3 August and 21 September 1997 (Calder et al 1999). Dome collapses were

Fig. 16. Sensitivity tests on: (a) vertical spacing of source points zsp; (b) decrease-factor of elutriated mass along the pyroclastic flow a: (c) input volume DRE (mfis from Equation 17) (Runs 5-7 in Table 2).

studied for the five main valleys around the volcano (Fig. 1). Individual probability maps were also compiled for a relatively large (10 6 m 3 DRE) 1997 Vulcanian explosion with fountain collapse down the same five valleys (Fig. 1). Figure 20 shows individual probability maps computed for a 1 2 k g m - 2 deposit threshold (Runs 11 and 12 in Table 2). Significant probabilities (>10%) of reaching a deposit threshold of 120 k g m - 2 for a 10 7 m 3 DRE dome collapse are confined to a narrow area around the valleys of collapse (Fig. 21). The probability of reaching a deposit threshold of 180kgm - 2 or higher is <0.1% in any area. For a 10 6 m 3 DRE Vulcanian explosion the probability of reaching a deposit threshold of 1 2 0 k g m - 2 or higher is also
Table 3. Results of the best fit to field data for a dome collapse and a Vulcanian explosion Dome collapse

Vent-plume maximum height (m) Co-PF plume height range (m) Runout (m) Accretionary-lapilli diameter (mm): vent plume co-PF plumes Wt% aggregated particles: vent plume co-PF plume Plume mass loading (kgm- 3 ) vent plume co-PF plumes

Vulcanian explosion

Observed*

Computed

Observed*

Computed

_ 4000 (max) 4000

_ 1000-4400 3311

1 1 300

11 395 300-2400 3644

-

0.8, 2.0, 2.7

0.8,1.9,2.6 0.2, 0.6, 0.8

82

23 23

0.07

0.04 0.03

-

-

* Data from Bonadonna et al. (2002). Calculated for volcanic thermals with temperature of 1100 K and 600 K for Vulcanian vent plumes, and for Vulcanian and dome-collapse co-PF plumes respectively. See also Table 2 (Runs 8) for input parameters. Dates: dome collapse, 31 March 1997; Vulcanian explosion, 26 September 1997 (14:56 LT).

Cumulative probability maps Cumulative probability maps were compiled for a given eruption scenario, which involves many individual events occurring over a certain period of time (e.g. the scenario in Table 4). These maps are more complex than probability maps compiled for individual events, as they need to include individual probabilities of individual eruptive events, combinations of wind profiles and accumulation/erosion of tephra-fallout deposits over a certain period of time. As the process of erosion cannot be easily predicted, minimum-deposit and maximum-deposit probabilities are investigated. Minimum-deposit probabilities represent the probability of reaching certain thresholds in a case where fallout tephra is completely eroded away between each separate event. Therefore, tephra accumulation is assessed separately for each dome collapse and Vulcanian explosion in the given activity scenario. Maximum-deposit probabilities represent the probability of reaching certain thresholds in a case where fallout tephra continuously accumulates with no erosion between separate events. In this case tephra accumulation is assessed as the final cumulative deposit produced by all dome collapses and Vulcanian explosions assumed in the given scenario. The eruptive scenario used to compile cumulative probability maps for Montserrat is described in Table 4 and is an approximation of the July 1995-March 1998 activity of

Fig. 17. Comparison between field data and best-fit computed data of mass accumulation (kgm - 2 ) for the fallout tephra from 31 March 1997 dome collapse (black triangles) and 26 September 1997 (14:56 LT) Vulcanian explosion (grey diamonds). The equiline (perfect agreement) is also shown, as a solid line, for reference (Runs 8 in Table 2).

Fig. 18. Comparison between computed grain-size distribution (white diamonds, 'disaggregated' distribution; black bars, accretionary lapilli; grey bars, free particles) and observed grain-size distribution (black triangles) for tephra-fallout deposit in three localities at different distances from the dome for the 31 March 1997 dome collapse: (a) locality a 2km from the dome; (b) locality b 4 km from the dome; and (c) locality c, 9 km from the dome. Localities are shown in Fig. 1 and mass-accumulation values are shown in Figure 11. The numerically 'disaggregated' computed distribution (white diamonds) is used as a comparison with the field data (black triangles). See text for details (Runs 9 in Table 2).

532

C. BONADONNA ET AL.

Fig. 19. Tephra-fallout deposits computed separately for the vent plume (solid lines) and the co-PF plumes (dashed lines) produced by the two Vulcanian explosions of: (a) 26 September 1997 (14:56 LT) (computed vent-plume height = 11.3km, computed co-PF-plume height range = 0.3-2.4 km): and (b) 10 October 1997 (18:40 LT) (computed vent-plume height = 7.6km. computed co-PF-plume height range = 0.4-1.4 km) (Runs 10 in Table 2). Contours shown for the ventplume tephra-fallout deposits are 0.1. 0.5. 1.5. 10 and 2 0 k g m - 2 (black numbers). Contours shown for the co-PF tephra-fallout deposits are 0.1. 0.5. 1. 2 and 3 k g m - 2 (grey numbers). Computed co-PF plume sources are also shown (black diamonds) (valleys followed by fountain-collapse pyroclastic flows from Druitt el al. 2002).

Soufriere Hills Volcano. Three years of wind data, and dome and fountain collapses down the four main valleys around the volcano are used in these computations (Tar River valley. White River valley. Fort Ghaut and Tuitfs Ghaut; see Fig. 1). Figure 22 shows the maps of the minimum-deposit probability of reaching 12, 120, 180 and 600 kg m-2 of tephra accumulation for the scenario of Table 4. First, the probability of reaching a deposit threshold was calculated separately for each individual dome collapse and Vulcanian explosion in Table 4. Cumulative probabilities for the minimum deposit were then calculated by the union of individual probabilities of each eruptive event. Over a certain period of activity, dome collapses and Vulcanian explosions represent independent, but not mutually exclusive, events. Therefore, if A\ and A 2 are the events fallout-tephra accumulation deposit threshold' for the dome collapse (or Vulcanian explosion) 1 and 2 respectively, the probability of the union of A\ and A2 is:

(18) and as A\ and A2 are independent: Resolving Equation 18 for n events:

The summation is taken over all of the possible subsets of size r of the set {1.2 n} (Ross 1989). An equivalent solution for the same problem is obtained by calculating the probability of the intersection of all the complements in order to analyse the probability of 'never

Fig. 20. Individual probability maps for a 10 7 m 3 (DRE) dome collapse (DC) down (a) Tar River valley, (b) White River valley, (c) Fort Ghaut, (d) Tyre's Ghaut and (e) Tuitt's Ghaut: (f) a 10 6 m 3 (DRE) Vulcanian explosion (VE) with a symmetrical fountain collapse down the same valleys. Co-PF plume sources used in the modelling are also shown (black diamonds). Contour interval is 10% probability. The threshold used is 1 2 k g m - 2 (minor damage to agriculture), and wind data cover a six-year period starting from 1992 (Runs 11 and 12 in Table 2).

MODELLING OF DOME-COLLAPSE AND VULCANIAN FALLOUT

Fig. 21. Individual probability maps for a threshold of 120 kg m- 2 (minimum threshold for roof collapse), and computed separately for a 107 m3 (DRE) dome collapse down the five main valleys around the volcano (Tar River valley, White River valley, Fort Ghaut, Tyre's Ghaut and Tuitt's Ghaut; see Fig. 1). Computed co-PF plume sources are also shown (black diamonds). Contours shown are the 10% and 90% probability for each tephra-fallout contribution. The probability of reaching a threshold of 120kgm~ 2 for a 10 6 m 3 (DRE) Vulcanian explosion is
(20)

533

Fig. 22. Cumulative probability maps for the minimum tephra-fallout deposit corresponding to the activity scenario described in Table 4. Dome and fountain collapses are simulated down four valleys around the volcano (Tar River valley, White River valley, Fort Ghaut and Tuitt's Ghaut; see Fig. 1). These maps are based on the assumption of complete erosion between separate events (computed in terms of union and intersection of individual probabilities, see text for details). Thresholds used: (a) 1 2 k g m - 2 (minor damage to agriculture); (b) 120kgm - 2 (minimum accumulation for roof collapse); (c) 180kgm - 2 (partial survival of vegetation: zone 2); (d) 600kgm - 2 (partial survival of vegetation: zone 1). Computed co-PF plume sources are also shown (black diamonds). Contour interval is 10% probability for (a), (b) and (c); in (d) only contours of 1%, 10% and 20% probability are shown. Wind data cover a three-year period starting from 1995. The maximum-deposit probability is given by adding all the probable contributions of tephra-fallout deposition produced by individual dome collapses and Vulcanian explosions in the scenario of activity in Table 4 and considering erosion to be negligible. However, the number of combinations of tephra-fallout deposit produced by individual dome collapses and Vulcanian explosions,

Table 4. Parameters used in the simulations for cumulative-probability maps (approximation of the July 1995-March 1998 activity) Event

Volume DRE (x10 6 m 3 )

Number of events

Diffusion coefficient (m 2 s - 1 )

Grain-size distribution

DC DC DC DC DC DC VE

50 30 10 3 1 0.5 0.4

1 1 2 7 16 30 90

2700 2700 2700 2700 2700 2700 2700

C C C B B B A

Vent-plume mass distribution

Aggregation model

4

3 3 3 3 3 3 4

Events: DC, dome collapse; VE, Vulcanian explosion (computed with fountain collapse). Volumes and numbers of events represent an approximation for three-year period of activity of Soufriere Hills Volcano (July 1995-March 1998). Grain-size distributions are shown in Figure 2 (A, 31 March 1997 dome collapse; B, 26 September 1997 (14:56 LT) Vulcanian explosion; C, 21 September 1997 dome collapse). Vent-plume mass Distribution 4 is described in Figure 3d. Aggregation Models 3 and 4 are described in Table 1.

C. BONADONNA ET AL

534

valleys of collapse and wind profiles is impracticable to compute. Therefore, maps for the maximum-deposit probability were computed using a Monte-Carlo technique in which the computer generates pseudo-random numbers (random numbers that are the result of an algorithm). Pseudo-random numbers were then used to simulate naturally random parameters such as wind profile and valley of collapse. The wind profile is naturally random and does not depend on the eruptive event. The valley of collapse depends on the direction of preferential dome growth, which is assumed to be naturally random (Watts et al. 2002). Therefore, the valley of collapse is also assumed to be random. The accuracy of the Monte-Carlo simulations was studied by calculating several times the probability of reaching a certain deposit threshold (12kg m - 2 ) at a particular locality (site c in Fig. 1) for different numbers of runs and for different numbers of wind profiles (one, three and six years) (Fig. 23). These simulations (50, 100, 200,

500, 1000 runs) were repeated 20 times for the case of one year and ten times for the cases of three and six years, using different pseudorandom sequences. The standard deviation for the three wind-data groups decreases with the number of runs, but does not vary significantly with the number of wind profiles used (Fig. 23b). The variation of this standard deviation is fitted well by a power law (Fig. 23b). The mean probability also does not vary significantly for the three ranges of wind profiles used (Fig. 23c) (standard deviation of the mean probability of the three populations of wind profiles: 0.7, 0.6, 0.8 for one year, three years and six years of wind profile respectively). Maps in Figure 24 are computed using three-year wind profiles (1995-1997) over 200 runs.

Discussion HAZMAP results show good agreement with field data, in terms of both mass accumulation of fallout tephra, and variations of grainsize distribution with distance from vent. This agreement has some implications for tephra-fallout processes and allows probability maps to be compiled. Simulations of co-PF and Vulcanian vent plumes as thermal injections give good agreement with field data (Table 3). This is supported by the comparison of theory with observations for coignimbrite plumes (Woods & Kienle 1994). and by field observations

0

200

400 600 800 1000 Number of runs

1200

Fig. 23. Accuracy of the Monte-Carlo technique used to compile cumulative probability maps for the maximum tephra-fallout deposit produced by the activity scenario in Table 4 (Fig. 24). The number of runs and the number of wind profiles are investigated on the basis of: (a) variability of probability values; (b) variability of standard deviation (a) of probability values (corresponding power-law fit is shown); (c) variability of the mean of probability values. Probability values are computed for locality c in Fig. 1, a threshold of 1 2 k g m - 2 , wind profiles of one-, two- and three-year periods, and 50, 100, 200, 500 and 1000 runs. Each simulation was run 20 times for the case of one-year period and ten times for the cases of two- and three-year periods.

Fig. 24. Cumulative probability maps for the maximum tephra-fallout deposit corresponding to the activity scenario described in Table 4. Dome and fountain collapses are simulated down four valleys around the volcano (Tar River valley, White River valley. Fort Ghaut and Tuitt's Ghaut; see Fig. 1). These maps are based on the assumption of no erosion between separate events (i.e. computed using a Monte-Carlo technique, see text for details). Thresholds used: (a) 1 2 k g m - 2 . minor damage to agriculture: (b) 1 2 0 k g m - 2 , minimum accumulation for roof collapse; (c) 180kgm - 2 . partial survival of vegetation, zone 2; (d) 6 0 0 k g m - 2 partial survival of vegetation, zone 1. Computed co-PF plume sources are also shown. Isomass contour interval is 20% probability. Maps are computed using a three-year period of wind profiles (1995-1997) and 200 runs.

MODELLING OF DOME-COLLAPSE AND VULCANIAN FALLOUT

for Vulcanian explosions described in Druitt et al. (2002). Modelling dome-collapse and Vulcanian plumes as vertical lines, neglecting particle re-entrainment, still give adequate agreement near the source. The model agrees well with observed data only if particle aggregation is incorporated into the simulations of tephra fallout from dome collapses and Vulcanian explosions (Figs 11 and 12). Dispersal of co-PF plumes from dome collapses is more sensitive to different aggregation models than that of Vulcanian plumes, since co-PF plumes are composed almost entirely of fine ash. Aggregation models presented in this paper represent a first attempt to simulate numerical aggregation mechanisms that generate different types of aggregates, such as irregular aggregates and accretionary lapilli (Bonadonna et al. 2002). As a first approximation, the aggregation model of Cornell et al. (1983) (Model 1) can be used to simulate the formation of irregular aggregates, as it assumes one fixed aggregate size of 250 m ( = 2), in agreement with the typical sizes of irregular aggregates observed on Montserrat (10-100 m). However, observations of fallout tephra from both dome collapses and Vulcanian explosions on Montserrat have shown that aggregates vary in size, and that accretionary lapilli are also characterized by larger diameters (150 m to llmm). Models 2, 3 and 4 better describe the formation of accretionary lapilli and their size variations. In these three models, the size of accretionary lapilli is specific for each event, varying from about 200 m for small co-PF plumes (Ht = 2 km) to about 11 mm for the largest co-PF plumes (Ht = 15 km) (Table 1). These sizes are consistent with observations (Bonadonna et al. 2002). However, Models 2, 3 and 4 can only describe three size categories for each plume height (based on three different aggregation coefficients), whereas the whole distribution of accretionary lapilli sizes can be characterized by several more sizes, ranging from a few hundred micrometres to a few millimetres (Bonadonna et al. 2002). Too little is known about the dynamics of aggregation to allow the percentage of particles incorporated into the accretionary lapilli to be predicted a priori. To achieve better results, further studies on the interaction between atmospheric conditions, plume height and plume-mass loading are needed. The empirical Models 3 and 4 give adequate results for hazardassessment purposes. Differential fallout (combination of wind shear and simultaneous tephra dispersal from Vulcanian vent and co-PF plumes) is mainly due to differences in plume heights (Fig. 19). Wind direction and speed typically vary with height; therefore, the (higher) vent plume can be dispersed in very different ways from the co-PF plumes. Field data have shown that differential fallout can have significant effects on the resulting grain-size distribution pattern around the volcano, as the vent plume is typically characterized by coarser particles and is affected by different aggregation processes, mainly due to a thicker umbrella cloud and to different plume mass loadings. The tephra-fallout deposit is predicted to vary from bimodal and fine-grained in the area of overlap, to unimodal and coarsegrained in the areas where only the vent-plume tephra deposits are present. This feature is commonly observed in tephra-fallout deposits from large Vulcanian explosions (Bonadonna et al. 2002). Variations with distance from vent of the computed grain-size distributions agree with field data. Computed tephra-fallout deposits show variations of the computed accretionary-lapilli size distributions with distance from vent, with most of the accretionary lapilli falling in proximal areas. Aggregation is the most obvious interpretation for strong bimodality in grain-size distributions of fallout tephra from dome collapses, which in the tephra-fallout deposit from the 31 March 1997 dome collapse was observed within 5 km from the dome (Bonadonna et al. 2002). However, both computed results and field data from some distal samples are slightly bimodal (Fig. 18c), even though the computed deposit consists of non-aggregated particles only. Bimodality shown by field data in distal areas might be due to very small aggregates and crystal aggregates, which are often observed with SEM analysis (Bonadonna et al. 2002). However, bimodality shown by computed data is due to the fact that the assumed input grain size is already bimodal (Fig. 2a).

535

Individual probability maps show that tephra fallout from dome collapses and Vulcanian explosions on a Montserrat scale can accumulate a mass of fallout tephra above the threshold of 'minor damage to agriculture' (12 kg m - 2 ) over large areas of the island, but cannot cause significant widespread damage to buildings and vegetation (>120kgm - 2 ). However, persistent dome-collapse activity (1996-1999) associated with Vulcanian explosions (AugustOctober 1997) accumulated enough material on Montserrat to reach values of over 600kgm - 2 (partial survival of vegetation) in the southwestern part of the island, and up to 12kgm - 2 (minor damage to vegetation) in the northern part (Bonadonna et al. 2002). We conclude that the use of cumulative probability maps, taking into account the cumulative effects of multiple eruptive events, is necessary in hazards assessment of fallout tephra generated during long-lasting eruptions. There are large differences between the minimum-deposit and maximum-deposit cumulative probability maps computed on the basis of the July 1995-March 1998 period. For example, were fallout tephra washed away after every individual dome collapse and Vulcanian explosion (Fig. 22), the probability of reaching the first threshold for partial survival of vegetation (600kgm - 2 ) would be <10% everywhere on the island. However, a continuous accumulation of fallout tephra with negligible erosion extends the area of 100% probability for such a threshold to a consistent area on the SW of the island (Fig. 24), in agreement with field data (Bonadonna et al. 2002). Which map is actually relevant depends on the effectiveness of erosion between events. During 1995-1998, erosion processes were more effective in the north, where deposits were thinner, whereas localities in the SW preserved much of the tephra-fallout deposits, even after several years. This shows that in long-lasting eruptions with many tephra-fallout events, an understanding of erosion is required to produce accurate hazard maps. Both probability maps of individual dome collapses and Vulcanian explosions, and cumulative probability maps show that the area on Montserrat that is most prone to tephra-fallout hazards is the WSW sector, with only minor damage to vegetation occurring up to St John's (Fig. 1). The north of the island (north of Woodlands; Fig. 1) receives little fallout tephra. Only combinations of unfortunate wind directions, high eruptive columns and vigorous activity can transport significant fallout tephra into northern Montserrat. Due to the wind direction and the decreasing distance from the dome, the currently populated areas in the NW (Woodlands; Fig. 1) and centre-west (Old Towne; Fig. 1) of the island are those where the chances of significant tephra fallout increase markedly. It is, however, unlikely that accumulation great enough to be life-threatening (e.g. roof collapse) will occur. Due to the prevailing wind directions, the Bramble Airport area (NNE of the island; Fig. 1) is very unlikely to be affected by tephra from plumes lower than 5km a.s.l., which are typical of small dome collapses (<2 x 106m3 DRE). Due to the rarer occurrence of higher plumes from large dome collapses and Vulcanian explosions, the probability of accumulating up to 1 cm of fallout tephra at the airport in three years of activity (in Table 4) is still high (20-90%), but the probability of accumulation above 10cm is low (0.5-5%).

Conclusions The HAZMAP model simulates tephra dispersal by solving for the diffusion, transport and sedimentation of particles in two dimensions from a distribution of source points. The model tested on the eruption of Soufriere Hills Volcano from 1995 to 1999 shows good agreement with observed data, allowing realistic probability maps for tephra-fallout hazard to be compiled. Good agreement between observations and model results is obtained when: (1)

Vulcanian vent plumes and Vulcanian and dome-collapse co-PF plumes are simulated as instantaneous injections of thermals;

536

C. BONADONNA ET AL.

(2)

runouts of Vulcanian and dome-collapse pyroclastic flows are calculated as logarithmic functions of the pyroclasticflow volume; (3) co-PF plumes are simulated with a 1-km spacing along the parent pyroclastic flow; (4) elutriated masses in Vulcanian and dome-collapse co-PF plumes decrease exponentially with distance from the dome base.

Comparison of field data with simulated tephra-fallout deposits from dome collapses and Vulcanian explosions shows best-fit results when: (5) mass distribution in volcanic thermals is uniform; (6) decrease of elutriation masses in successive co-PF plumes follows a factor of exp(10/PF runout); (7) diffusion coefficient is 2700 m 2 s - l ; (8) particle aggregation is described considering a variation in the aggregate-size range. Some general features of tephra-fallout deposits from dome collapses and Vulcanian explosions also emerge from our modelling results. (9)

(10)

(11) (12)

Simulation of tephra dispersal from dome collapses is significantly sensitive to inputs of total volumes and aggregation models used, whereas simulations of Vulcanian explosions are mainly sensitive to volume inputs. When particle aggregation is considered, simulations of dome collapses and Vulcanian explosions are not very sensitive to grain-size distributions. Aggregation processes are a major factor in the modelling of tephra dispersal from dome-collapse co-PF plumes, which are rich in fine ash, representing a major cause of grain-size distribution bimodality in dome-collapse fallout tephra. Aggregate size and percentage of particles incorporated into aggregates exert an important control on tephra accumulation. Computed isomass maps for large Vulcanian explosions predict a differential fallout dispersal. This is due to the combination of plume height and wind shear when tephra fallout from the vent plume occurs simultaneously with that from co-PF plumes from fountain-collapse pyroclastic flows of large Vulcanian explosions. This has significant implications for the grain-size distributions of tephra-fallout deposits, as also confirmed by field data.

From the perspective of tephra-fallout hazards assessment, the following points emerge. (13)

(14)

(15)

Tephra fallout from individual dome collapses and individual Vulcanian explosions on a Montserrat scale does not produce deposits thick enough to cause widespread damage to vegetation and human activities. The amalgamation of tephra from many events of tephra fallout generated during a long-lasting eruption represents a significant hazard in terms of damage to agriculture, roof collapses, polluting effects and exposure to particulate pollution. Hazards of fallout tephra produced during long-lasting eruptions need to be assessed using individual-probability maps together with cumulative-probability maps compiled for the minimum and maximum deposits.

In terms of hazards assessment on Montserrat, we finally make the following conclusions. (16)

Tephra fallout from dome collapses is the most hazardous of tephra-generation processes observed on Montserrat because: (i) dome-collapse activity was the most persistent processes during the whole 1995-1999 eruptive period; (ii) dome col-

(17)

lapses can also occur after the lava dome stops growing; (iii) fallout tephra from dome collapses contains more crystalline silica in the <10 m fraction than fallout tephra from other processes. The SW sector of Montserrat is the most prone to fallout tephra, experiencing damage to vegetation for fallout tephra from individual events and also serious damage to vegetation and buildings as a consequence of three years of activity (of the 1995-1998 type). The north, NE and NW sectors do receive some fallout tephra from individual dome collapses and Vulcanian explosions, but not enough to cause serious damage to vegetation and buildings. Three years of activity can also cause minor damage to vegetation in these sectors.

The authors would like to thank A. Longo. who initiated the model presented here, for her work on fallout hazard assessment in the Vesuvius area. Computer modelling was made easier by the helpful advice of R. Herd and H. Dixon. Many thanks also to A. Hogg for helpful discussions concerning parts of this study. T. Druitt and A. Woods are gratefully acknowledged for thorough and constructive reviews. Studies on Montserrat were supported by the Department for International Development. C. Bonadonna is supported by an EC Marie Curie PhD Fellowship. R. S. J. Sparks is supported by a NERC Professorship.

References ARASTOOPOUR, H.. WANG. C. H. & WEIL. S. A. 1982. Particle-particle interaction force in a dilute gas-solid system. Chemical Engineering Sciences, 37, 1379-1386. ARMIENTI, P., MACEDONIO, G. & PARESCHI, M. T. 1988. A numerical-model for simulation of tephra transport and deposition - applications to May 18 1980, Mount-St-Helens eruption. Journal of Geophysical Research, 93, 6463-6476. BARBERI, F., MACEDONIO, G., PARESCHI. M. T. & SANTACROCE, R. 1990. Mapping the tephra fallout risk: an example from Vesuvius. Italy. Nature. 344, 142-144. BAXTER, P. J., BONADONNA, C., DUPREE. R. ET AL. 1999. Cristobalite in volcanic ash of the Soufriere Hills Volcano. Montserrat. British West Indies. Science. 283, 1142-1145. BLONG, R. J. 1984. Volcanic Hazards. A Sourcebook on the Effects of Eruptions. Academic Press. Sidney. BONADONNA. C., MAYBERRY. G. C.. CALDER. E. S. ET AL. 2002. Tephra fallout in the eruption of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs, 21. 483-516. BRAZIER, S., SPARKS, R. S. J., CAREY, S. N.. SIGURDSSON. H. & WESTGATE. J. A. 1983. Bimodal grain-size distribution and secondary thickening in air-fall ash layers. Nature. 301. 115-119. CALDER, E. S., SPARKS, R. S. J. & WOODS. A. W. 1997. Dynamics of coignimbrite plumes generated from pyroclastic flows of Mount St. Helens (7 August 1980). Bulletin of Volcanology. 58, 432-440. CALDER, E. S.. COLE, P. D., DADE. W. B. ET AL. 1999. Mobility of pyroclastic flows and surges at the Soufriere Hills Volcano. Montserrat. Geophysical Research Letters, 26. 537-540. CALDER, E. S., LUCKETT. R.. SPARKS. R. S. J. & VOIGHT, B. 2002. Mechanisms of lava dome instability and generation of rockfalls and pyroclastic flows at Soufriere Hills Volcano. Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21, 173-190. CAREY, S. N. & SIGURDSSON. H. 1982. Influence of particle aggregation on deposition of distal tephra from the May 18 1980. eruption of Mount St-Helens volcano. Journal of Geophysical Research, 87. 7061-7072. CAREY. S. N., SIGURDSSON. H. & SPARKS. R. S. J. 1988. Experimental studies of particle-laden plumes. Journal of Geophysical Research. 93. 15314-15328. COLE, P. D., CALDER, E. S., DRUITT, T. H.. HOBLITT. R.. ROBERTSON, R. E. A., SPARKS, R. S. J. & YOUNG. S. R. 1998. Pyroclastic flows generated by gravitational instability of the 1996-97 lava dome of Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3425-3428.

MODELLING OF DOME-COLLAPSE AND VULCANIAN FALLOUT COLE, P. D., CALDER, E. S., SPARKS, R. S. J. ETAL. 2002. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 231-262. CORNELL, W., CAREY, S. & SIGURDSSON, H. 1983. Computer-simulation of transport and deposition of the Campanian Y-5 Ash. Journal of Volcanology and Geothermal Research, 17, 89-109. DOBRAN, F., NERI, A. & MACEDONIO, G. 1993. Numerical-simulation of collapsing volcanic columns. Journal of Geophysical Research, 98, 4231-4259. DRUITT, T. H., YOUNG, S. R., BAPTIE, B. ETAL. 2002. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306. GILBERT, J. S. & LANE, S. J. 1994. The origin of accretionary lapilli. Bulletin of Volcanology, 56, 398-411. GLAZE, L. S. & SELF, S. 1991. Ashfall dispersal for the 16 September 1986, eruption of Lascar, Chile, calculated by a turbulent-diffusion model. Geophysical Research Letters, 18, 1237-1240. HARFORD, C. L., PRINGLE, M. S., SPARKS, R. S. J. & YOUNG, S. R. 2002. The volcanic evolution of Montserrat using 40 Ar/ 39 Ar geochronology. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 93-113. HAY, R. L. 1959. Formation of the crystal-rich glowing avalanche deposits of the St. VINCENT, B. W. I. Journal of Geology, 67, 540-562. HILDRETH, W. & DRAKE, R. E. 1992. Volcano Quizapu, Chilean Andes. Bulletin of Volcanology, 54, 93-125. HOBLITT, R. P. 1986. Observations of the eruption of July 22 and August 7 1980, at Mount St. Helens, Washington. US Geological Survey Professional Papers, 1335. HUPPERT, H. E., TURNER, J. S., CAREY, S. N., SPARKS, R. S. J. & HALLWORTH, M. A. 1986. A laboratory simulation of pyroclastic flows down slopes. Journal of Volcanology and Geothermal Research, 30, 179-199. LACROIX, A. 1904. La Montagne Pelee et ses eruptions. Masson, Paris. LEVINE, A. H. & KIEFFER, S. W. 1991. Hydraulics of the August 7, 1980, pyroclastic flow at Mount St. Helens, Washington. Geology, 19, 1121-1124. LOUGHLIN, S. C., CALDER, E. S., CLARKE, A. B. ET AL. 2002. Pyroclastic flows and surges generated by the 25 June 1997 dome collapse, Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 191-209. MACEDONIO, G., PARESCHI, M. T. & SANTACROCE, R. 1988. A numerical simulation of the Plinian fall phase of 79 AD eruption of Vesuvius. Journal of Geophysical Research - Solid Earth and Planets, 93, 14817-14827. MAYBERRY, G. C., ROSE, W. I. & BLUTH, G. J. S. 2002. Dynamics of volcanic and meteorological clouds produced on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 539-555. MOORE, K. R., DUFFELL, H., NICHOLL, A. & SEARL, A. 2002. Monitoring of airborne particulate matter during the eruption of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 557-566.

537

NORTON, G. E., WATTS, R. B., VOIGHT, B. ETAL. 2002. Pyroclastic flow and explosive activity at Soufriere Hills Volcano, Montserrat, during a period of virtually no magma extrusion (March 1998 to November 1999). In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 467-481. PASQUILL, F. 1974. Atmospheric Diffusion. John Wiley, Chichester. ROBERTSON, R. E. A., ASPINALL, W. P., HERD, R. A., NORTON, G. E., SPARKS, R. S. J. & YOUNG, S. R. 2000. The 1995-1998 eruption of the Soufriere Hills volcano, Montserrat, WI. Philosophical Transactions of the Royal Society of London Series a - Mathematical Physical and Engineering Sciences, 358, 1619-1637. Ross, S. M. 1989. A First Course in Probability. Macmillan, New York. SPARKS, R. S. J. & WALKER, G. P. L. 1977. The significance of vitricenriched air-fall ashes associated with crystal-enriched ignimbrites. Journal of Volcanology and Geothermal Research, 2, 329-341. SPARKS, R. S. J., SELF, S. & WALKER, G. P. L. 1973. Products of ignimbrite eruptions. Geology, 1, 115-118. SPARKS, R. S. J., MOORE, J. G. & RICE, C. J. 1986. The initial giant umbrella cloud of the May 18th 1980, explosive eruption of Mount St. Helens. Journal of Volcanology and Geothermal Research, 28, 257-274. SPARKS, R. S. J., BONNECAZE, R. T., HUPPERT, H. E., LISTER, J. R., HALLWORTH, M. A., MADER, H. & PHILLIPS, J. 1993. Sediment-laden gravity currents with reversing buoyancy. Earth and Planetary Science Letters, 114, 243-257. SPARKS, R. S. J., BURSIK, M. L, CAREY, S. N., WOODS, A. W. & GILBERT, J. S. 1994. The controls of eruption-column dynamics on the injection and mass loading of ash into the atmosphere. In: First International Symposium on Volcanic Ash and Aviation Safety, Seattle, Washington. US Government Printing Office, Washington, 81-86. SPARKS, R. S. J., BURSIK, M. L, CAREY, S. N., GILBERT, J. S., GLAZE, L. S., SIGURDSSON, H. & WOODS, A. W. 1997. Volcanic Plumes. John Wiley, Chichester. SPARKS, R. S. J., BARCLAY, J., CALDER, E. S. ETAL. 2002. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 409-434. SUZUKI, T. 1983. A theoretical model for dispersion of tephra. In: SHIMOZURU, D. & YOKOYAMA, I. (eds) Arc Volcanism, Physics and Tectonics. Terra Scientific, Tokyo, 95-113. THORARINSSON, S. 1944. Petrokronologista Studier pa Island. Geographes Annuales Stockholm, 26, 1-217. WADGE, G. & ISAACS, M. C. 1988. Mapping the volcanic hazards from Soufriere Hills Volcano, Montserrat, West-Indies using an imageprocessor. Journal of the Geological Society, 145, 541-552. WATTS, R. B., HERD, R. A., SPARKS, R. S. J. & YOUNG, S. R. 2001. Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 115-152. WOODS, A. W. & BURSIK, M. I. 1994. A laboratory study of ash flows. Journal of Geophysical Research - Solid Earth, 99, 4375-4394. WOODS, A. W. & CAULFIELD, C. C. P. 1992. A laboratory study of explosive volcanic-eruptions. Journal of Geophysical Research - Solid Earth, 97, 6699-6712. WOODS, A. W. & KIENLE, J. 1994. The dynamics and thermodynamics of volcanic clouds - Theory and observations from the April 15 and April 21, 1990 eruptions of Redoubt volcano, Alaska. Journal of Volcanology and Geothermal Research, 62, 273-299.

Dynamics of volcanic and meteorological clouds produced on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat G. C. MAYBERRY, W. I. ROSE & G. J. S. BLUTH Department of Geological Engineering and Sciences, Michigan Technological University, Houghton, MI 49931, USA

Abstract: The 26 December 1997 explosive activity of Soufriere Hills Volcano, Montserrat, provided an opportunity to study the evolution of a volcanic cloud by merging data from various satellites with wind-trajectory data. The activity involved a debris avalanche that descended SSW from the lava dome, to the coast, and a pyroclastic density current that traversed the coast and entered the sea. The slope failure and subsequent dome collapse occurred at c. 07:01 universal time (UT; 03:01 local time), lasted 15.2 minutes, and produced an upwardly convecting volcanic ash cloud that cloud temperatures suggest rose to c. 15 km. The volcanic ash cloud was unusual because the pyroclastic density current transported hot fine ash to the sea, where it rapidly transferred its heat to the sea water. The evaporation of large volumes of water produced a volcanogenic meteorological (VM) cloud that convected along with the volcanic ash cloud. The evolution of the volcanic and VM clouds was studied using an isentropic wind trajectory model and data from three satellite sensors: Geostationary Observational Environmental Satellite 8 (GOES 8), Advanced Very High Resolution Radiometer (AVHRR), and Total Ozone Mapping Spectrometer (TOMS). The high temporal resolution of the GOES 8 images filled many of the time gaps the other satellites left, and allowed quantitative retrievals to be performed using a two-band infrared retrieval method. The three-dimensional morphology of the volcanic cloud was reconstructed using GOES 8 data and by determining the heights of air parcels from wind-trajectory data. The volcanic cloud was estimated to contain up to 4.5 x 107 kg of silicate ash. Between c. 07:39 UT and 13:39 UT the ash signal of the volcanic cloud was masked by the VM cloud, which had a mass of up to 1.5 x 108 kg of ice. Ice forms when moist air is convected upwards to temperatures of less than -40°C and becomes saturated. Ice formation in volcanic clouds is especially likely when hot volcanic material is cooled by seawater rather than the atmosphere. The efficiency of evaporation of the seawater was calculated to be c. 5%, based on physical and GOES 8 data. TOMS data showed the SO2 in the volcanic cloud rose higher than the ash in the volcanic cloud, as has occurred in several other eruptions. A comparison between GOES 8 and AVHRR data showed that AVHRR data retrieved higher fine-ash silicate masses and higher cloud areas than GOES 8 due to the finer spatial resolution of AVHRR images. The effect on retrieval data of the high water vapour content in the lower troposphere of the tropical atmosphere was quantified; the high humidity in the Montserrat region caused the characteristic ash signal to the infrared sensors to be depressed by up to 80%. This signal depression caused a corresponding underestimation of the mass and area of the volcanic cloud when the infrared brightness temperature difference retrieval technique was used.

Over the last two decades scientists have pursued progressively more sophisticated methods of using satellite data for the study of volcanic clouds (e.g. Krueger 1983; Prata 1989; Wen & Rose 1994; Krotkov el al. 1999). Most recently, the merging of satellite and wind-trajectory data has provided a more holistic look at volcanic cloud behavior. Studies of the 1994 Rabaul eruption demonstrated that the eruption cloud contained large amounts of ice (Rose et al. 1995). Data showed that the June 1992 Spurr eruption cloud was sheared and thinned from below as it travelled away from the volcano (Shannon 1996). The 1982 El Chichon eruption cloud was found to have separated into SO2 and ash-rich portions at different altitudes (Schneider et al. 1999). In addition, the high temporal resolution of geostationary infrared data has provided an opportunity to perform more detailed analysis of volcanic clouds by filling in the time gaps left by polar orbiting satellites (Rose & Schneider 1996). The development of quantitative algorithms for retrieving the characteristics of volcanic clouds has made it possible to evaluate reactions and chemical/physical processes in volcanic clouds as they move (Wen & Rose 1994; Krueger et al. 1995; Krotkov et al. 1999). While preliminary work has been conducted on the use of geostationary satellites in quantitative volcanic cloud studies (Rose & Schneider 1996; Davies & Rose 1998), a need exists for multisensor studies to ensure that the high temporal resolution of geostationary satellite data and the high spatial resolution of polar orbiting satellite images are used to their full potential. Over 800 scenes of volcanic clouds from Soufriere Hills Volcano are available for analysis (Davies & Rose 1998). Davies & Rose (1998) briefly describe some details of the evolution of the volcanic cloud from the 6 November 1997 dome collapse of Soufriere Hills Volcano. They discuss the presence of a volcanogenic meteorological (VM) cloud, and the effect of using the brightness temperature difference (BTD) technique to delineate the volcanic cloud in a region with high humidity. Several volcanic cloud studies using polar-orbiting satellites have shown that separation of SO2 from the ash in volcanic clouds occurred (Shocker 1996; Schneider et al. 1999; Constantine et al. 2000), but the cause of the separation has not been explained. The

frequent data provided by geostationary satellites are needed to study the physical and chemical processes in short-lived volcanic clouds. The goal of this paper is to describe and analyse the dynamics of volcanic and VM clouds utilizing satellite and wind-trajectory data. This work applies diverse satellite and wind-trajectory model methodology to the largest Soufriere Hills eruptive event of 1997, the 26 December dome collapse, which generated a pyroclastic density current that travelled to the coast and entered the sea, producing a VM cloud from evaporation of seawater and a volcanic cloud that rose to c. 15 km. To help describe and analyse the clouds, data from Geostationary Observational Environmental Satellite 8 (GOES 8), the Advanced Very High Resolution Radiometer (AVHRR), the Total Ozone Mapping Spectrometer (TOMS), an isentropic wind trajectory model, and ground observations are used. The main objectives of this study are as follows. (a) Determine the three-dimensional morphology of the volcanic cloud using wind-trajectory data and GOES 8 imagery. (b) Compare the imagery and ash cloud retrieval data from AVHRR and GOES 8 which have different spatial resolutions. (c) Analyse the masking effect of the VM cloud when it overlapped the volcanic cloud from the satellite perspective. The evolution of the VM cloud is reconstructed, the volume of seawater evaporated to form it is estimated, and the effect of masking by the VM cloud on retrieved volcanic-cloud data is discussed. (d) Investigate the effect of the humid lower troposphere in the Montserrat region on retrieving cloud properties using the infrared brightness temperature difference technique. (e) Investigate the cause of the separation of SO2 and volcanic ash in the Soufriere Hills volcanic cloud. The 26 December 1997 collapse The largest dome collapse of the Soufriere Hills eruption occurred on 26 December 1997. The SW flank gave way and moved down the White River valley as a debris avalanche (Sparks et al. 2002). This

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 539-555. 0435-4052/02/$15 © The Geological Society of London 2002.

539

540

G. C. MAYBERRY ET AL.

ticles <5 m (Baxter et al. 1999). Although the ash was fine-grained and would be expected to have a long atmospheric residence, much of it fell out very quickly as accretionary lapilli SW of the volcano (Bonadonna et al. 2002). The high humidity in the tropical atmosphere, and or the excess water in the volcanic cloud from interaction with the VM cloud, may have been factors in the creation of the large quantity of accretionary lapilli. Meteorology of the Montserrat region

Fig. 1. Ash masses in volcanic clouds. The total mass of ash in measured and mapped fallout deposits is plotted against the maximum mass of fine (diameter 1-25 m) ash measured in volcanic clouds using two-band infrared satellite data (Rose et al 2000).

was followed by a fast-moving pyroclastic density current which generated an important volcanic cloud (Bonadonna et al. 2002). While the 26 December 1997 volcanic ash cloud is large relative to other Soufriere Hills volcanic clouds, it is an order of magnitude smaller than those of 1992 from Crater Peak of Mount Spurr, Alaska, and more than two orders of magnitude smaller than the 15 August 1991 eruption of Cerro Hudson, Chile (Fig. 1). Details of the 26 December 1997 event are given by Sparks et al. (2002). As the pyroclastic density current travelled down White River Valley to the sea, it quickly overtook the debris avalanche. Sparks et al. (2002) used estimates of the volume of the dome before and after the eruption to estimate the volume of dome material in the pyroclastic density current at 35-45 x 106 m3 (c. 40% of the dome). The density current had a peak velocity of 80-90 m s-1, and a minimum flux of 10 8 kgs - 1 . It devastated 10km 2 of southern Montserrat and extended the former coastline a distance of up to 200 m. The dispersal of the material from the pyroclastic density current is described by Sparks et al. (2002) and Ritchie et al. (2002). The fine ash (0.05-63.5 m radius; Baxter et al. 1999) component of the pyroclastic density current convected, due to thermal differences between air heated by the hot ash and ambient air, forming a column that rose several kilometres into the atmosphere (Cole et al. 1998). Light ash fell over southwestern Montserrat, as well as SE of Montserrat on the island of Guadeloupe, but most of the fine ash was deposited in the sea (Bonadonna et al. 2002). Approximately 1.8-3.2 x 10 6 m 3 of the 35-45 x10 6 m 3 of dome material was deposited on land, therefore about 30-43 x 10 6 m 3 of the material is not accounted for (Sparks et al. 2002). This missing volume is thought to have partly entered the sea, and partly convected into the buoyant ash plume. Though the lateral extent of the submarine deposit is unknown, areas with deposits up to 50m thick were found, suggesting that the bulk of the missing material was deposited in the sea (Sparks et al. 2002). The activity occurred at night and could not be observed directly, so we know little about the interaction between the pyroclastic density current and the ocean. However, it is probable that the entry of the pyroclastic density current into the sea caused evaporation of seawater and the formation of a VM cloud.

The weather in the Caribbean is often monotonous. with patterns that can last for weeks. Conditions near sea level are hot and dry, and mountains are often draped with orographic clouds. Beneath the temperature inversion (1-2 km), the air is moist with a layer of cumulus clouds. The trade winds. or tropical easterlies, are pronounced in the tropics between 0 and 6km. especially during the winter. The trade winds produce predictable weather patterns because they travel consistently towards the SW at c. 7 m s - 1 . while upper level winds (>6 km) often travel in a different direction (Barry & Chorley 1992). The tropopause in late December. based on radiosonde data from Guadaloupe. was about 16-17 km above sea level. Data and data analysis Geostationary Observational Environmental Satellite (GOES) GOES sensors utilize five bands ranging from visible to thermalinfrared (Table 1) and operate at the geosynchronous altitude of 35800km above the Earth's surface. Geostationary data are advantageous when studying volcanic clouds because the data have a fine temporal resolution. Images are collected approximately every 30 minutes compared to two to four times daily for polarorbiting satellites. The 26 December 1997 activity was imaged by Table 1. GOES 8 and A VHRR spectral hand information Band

GOES A(m)

Resolution (km)

AVHRR A (m)

1

0.55-0.75 3.8-4.0 6.5-7.0 10.2-11.2 11.5-12.5

1 4 8 4 4

0.58-0.68 0.725-1.10 3.55-3.93 10.3-11.3 11.5-12.5

2 3 4 5

* AVHRR subpoint resolution is 1.1 km in all bands.

Fallout deposits of 26 December 1997 Only a small fraction of the airborne ash of 26 December 1997 was sensed by infrared satellite detectors because much of it was coarser than the satellite could detect (1.5-12 m radius) and fell out very quickly after the eruption. Andesitic ash produced by fragmentation of the Montserrat lava dome during various dome collapses contained 60-70 wt% of 5-62.5 m particles and 13-20wt% par-

Fig. 2. Area of the volcanic cloud of Soufriere Hills Volcano. 26 December 1997, as determined by GOES band 4 and BTD. The BTD area matches the band 4 area only early in the event and generally is about 10% of the band 4 area, decreasing with time.

DYNAMICS OF VOLCANIC AND METEOROLOGICAL CLOUDS

GOES 8 which scans bidirectionally, alternating west to east and east to west (Kidder & Vonder Haar 1995). Advanced Very High Resolution Radiometer (A VHRR) The AVHRR is a multispectral, infrared radiometer that operates on polar-orbiting meteorological satellites operated by the National Oceanic and Atmospheric Administration (NOAA) with five spectral bands (Table 1). The 26 December 1997 volcanic cloud was imaged by the AVHRR sensor on NOAA-14, which is a sunsynchronous satellite, with a daytime north-to-south equatorial crossing (Lillesand & Kiefer 1987). AVHRR has a swath width of approximately 2800 km, and images the same area twice daily with daytime and night-time passes. Local area coverage (LAC) data (pixel scale l . l k m at nadir) were used in this study to analyse physical details of the volcanic cloud.

541

Retrieval methods for GOES and A VHRR In order to discriminate the volcanic cloud from the background and from meteorological clouds, the brightness temperature difference (BTD) between bands 4 and 5 was used (Prata 1989). The effective radius and optical depth of the volcanic cloud at infrared wavelengths was then retrieved using a radiative transfer model developed by Wen & Rose (1994). The model has simplifying casedependent assumptions. The cloud is assumed to be c. 1 km thick, homogeneous, semi-transparent, parallel to the surface below it, and the atmosphere above and below the cloud is assumed to be cloudfree. It is also assumed that particles in the cloud are all andesitic, spherical in shape with a log- normal size distribution. The scheme detects those particles with radii of 1.5-12 m, sizes that experience Mie scattering and absorption at infrared wavelengths (Wen & Rose 1994). Band 4 (10.2-11.2 m) of the GOES 8 sensor is slightly offset from band 4 (10.3-11.3 m) of the AVHRR sensor (Table 1), and

Fig. 3. Selected GOES band 4 volcanic cloud images, 26 December 1997. The volcanic cloud starts out as a bright spot, reflecting a very low brightness temperature and high opacity; with time the cloud grows in size but dims in contrast as it becomes more transparent.

542

G. C. M A Y B E R R Y ET AL.

this was corrected using wavelength-dependent data collected by Pollack et al. (1973). We retrieved volcanic cloud information in two parts, to allow for differing cloud heights. Cloud heights were constrained by wind trajectory data. The area of the volcanic cloud in band 4 images was measured to compare them with areas of the ash in the volcanic cloud shown in the BTD images (Fig. 2). This procedure is subjective and was done to estimate any underestimation of ash mass in the volcanic cloud due to the tropospheric moisture.

GOES and A VHRR results Band 4 brightness temperatures. GOES 8 collected 33 images of the volcanic cloud on 26 December 1997, approximately every 30 minutes from 07:09 universal time (UT; 03:09 local time) (c. 8 minutes after the eruption began), until 23:39 UT. The pattern of movement of the volcanic cloud is shown by six selected band 4 images in Figure 3. The volcanic cloud cannot be distinguished from meteorological clouds in band 4, but this type of representation does show the overall pattern of dispersal well, and it displays the full size of the cloud, even after the infrared ash signal is too weak to be detected. Throughout the c. 17 hours shown in Figure 3, the volcanic cloud was elongated hundreds of kilometres to the SE, while simultaneously migrating 120 km to the SW (from the volcano to the cloud centroid). The first image (Fig. 3a) shows a sharply defined volcanic cloud, with a high optical depth reflected in the cold temperature, located near the volcano with an area of 770km 2 , and elongated 40km to the SE. The cloud ultimately rose to c. 15km, based on comparing radiosonde data from the Guadeloupe weather station (Fig. 4) and the cloud brightness temperatures (200 K) derived from band 4 data of high-optical-depth pixels of the cloud in the two earliest images (07:09 UT and 07:39 UT; the latter is not shown in Fig. 3). The radiosonde data (Fig. 4) show that the temperature was 200 K at two altitudes; in the troposphere at c. 15km, and in

the stratosphere at c. 20km. Wind-trajectory analyses show that the volcanic cloud did not rise to 20km. therefore c. 15 km was the maximum altitude that the volcanic cloud reached. BTD retrieval results. Figure 5 shows the volcanic cloud as discriminated by BTD. Figures 2 and 3 show how the area of the volcanic cloud changed with time during 26 December. It increased in size rapidly at first, slowing and reaching a maximum at about 16:00 UT and then slowly decreased in size. The band 4 areas of the cloud were larger by about an order of magnitude during much of the day. The difference in areas occurred in part because the 26 December 1997 dome collapse was smaller than other eruptions shown in Figure 1, so the concentration of fine ash in much of the volcanic cloud was not high enough for the sensor to detect. The volcanic ash signal for each pixel must be strong enough to override the effects of any hydrometeors in the volcanic cloud, or the effects of high humidity depressing infrared radiation transmission in the tropical lower troposphere (Coll & Caselles 1997). Figure 6 summarizes the time trends of the 26 December 1997 volcanic cloud, including mass, area and mean optical depth. This represents the BTD-determined ash-rich core of a larger cloud. The two-dimensional shapes of these cloud cores are shown in Figure 5. Retrieval results from GOES 8 are given in Table 2. Between 07:39 and 09:39 UT the volcanic cloud area decreased by 40%, and total ash mass of the volcanic cloud decreased by 75% (Fig. 6). These decreases were probably caused by interference of the transmission of the ash signal to the sensor by a VM cloud. The times when the volcanic cloud ash signal was masked by the VM cloud is shown in Figure 6. After 09:39 UT the total area and silicate mass increased until 16:39 UT and 13:09 UT, respectively, with a general decrease thereafter. The optical depth decreased rapidly after 09:39 UT until 13:39 UT, when it began to decrease at a slower rate. The maximum mass of fine ash (45 x 106 kg) was detected at 07:39 UT, c. 38 minutes after the explosive activity began. The volcanic cloud core

Fig. 4. Upper atmosphere temperature profile at 11:00 UT at Guadaloupe. Asterisks indicate the two altitudes where the temperature was 200 K. Pressure (P) in millibars.

DYNAMICS OF VOLCANIC AND METEOROLOGICAL CLOUDS

543

Fig. 5. Selected GOES brightness temperature difference (BTD) images, 26 December 1997. The size of the volcanic cloud as outlined by BTD is much smaller than that outlined by band 4 (Fig. 3), as only the dense cores of the cloud, richer in volcanic ash, are detected by cutoff values less than -0.5 K for BTD.

as seen with BTD data expanded to 16 times its size in the first image (Fig. 5a) over 11.6 hours to a maximum area of 8000km2 (Fig. 5d). Figures 7 and 8 show the spatial distribution of the cloud optical depth, fine ash mass and mean effective radius (Table 2). The algorithm used in the retrieval model produces masses derived from the optical depth; therefore they are directly related (Fig. 7). The cloud's overall highest ash mass and optical depth occurred in the first hour after the onset of the explosive activity (07:39 UT), when the cloud had an average optical depth of 2.6 and was partially opaque (optical depth >4). The mass and optical depth generally decreased until 15:39 UT although they increased slightly in the northern section of the volcanic cloud beginning at 13:09 UT.

AVHRR results. A single NOAA 14 AVHRR image from 17:47 UT on 26 December 1997 was also analysed (Fig. 6) by the same algorithm and tabulated with the GOES data in Table 2.

Total Ozone Mapping Spectrometer (TOMS) TOMS is an ultraviolet spectrophotometer that operates in a polar sun-synchronous orbit on board NASA's Earth Probe (EP) satellite. TOMS compares spectral radiance of the sunlit atmosphere with radiance of a sunlit calibrated diffuser plate from a satellite platform (Krueger et al 1995). EP/ TOMS has a 2100km swath

544

Fig. 6. Changes in fine ash mass, and optical depth (unitless) of the volcanic cloud on 26 December 1997. Data from Table 2. The two AVHRR points are labelled; all of the others are from GOES. The points between about 8 and 12 UT (shaded area) are during the period of masking effects of the volcanogenic meteorological cloud. (Krueger et al. 1995). The algorithm used in the program compares radiance values collected by TOMS with pre-computed radiances from a look-up table. The look-up table includes the 26 standard ozone profiles and a layer of SO2 with amounts from 1 DU to 600 DU at 20km (N. Krotkov, pers. comm.). The total concentration of SO2 in the volcanic cloud was corrected for background levels (i.e. sensor-dependent levels) of SO2 (after Constantine et al. 2000). The EP/TOMS data are not rectilinear. so the SO2 located between the pixels scanned is calculated by assuming that the area of SO2 in the cloud missed by the scan is a function of the ratio between the scan area and the total cloud area (S. Schaefer, pers. comm.). The final SO2 mass value represents the SO2 in the cloud. including SO2

and 90% Earth coverage; it images the same area once daily, with 39km pixel size at nadir (McPeters et al. 1998). Four bands within the 312-380 nm region are used to determine total-column SO2 loadings (Krueger et al. 1995). SO2 concentrations are measured in Dobson units (DU), which represent the thickness of gas at standard temperature and pressure that affects the reflection of ultraviolet radiation in a column that extends from the satellite to the reflective surface. A data visualization software package (IDL) program developed by NASA (Krueger et al. 1995) calculates the total concentration of SO2 by multiplying the DU of SC2 in each pixel collected by TOMS by the area of the pixel and a conversion factor, then summing all of the pixels in the area of interest

Table 2. GOES and A VHRR infrared volcanic ash retrieval results Time (UT)

No. of ash pixels

Cloud area (km 2 )

Mean effective radius ( m)

Mean optical depth

Fine ash mass (x 10 6 kg)

07:09 07:39 08:39 09:39 10:39 11:39 13:09 13:39 14:39 15:39 16:39 17:39 17:47*

35 79 48 48 113 291 403 354 460 451 503 372 451 (7217)* 222 318 284 257 243 126

560 1300 770 770 1800 4700 6500 5700 7400 7200 8000 5900 8700

7.0 8.3 8.6 5.6 5.7 7.7 7.6 6.7 6.7 6.2 6.0 5.2 5.0

2.6 3.3 3.4 2.0 1.0 0.7 0.6 0.7 0.6 0.5 0.4 0.4 0.6

13 45 29 11 13 32 38 36.5 37 29 27 18 31

3600 5100 4500 4100 3900 2000

4.8 5.0 4.3 4.7 4.2 4.0

0.4 0.3 0.3 0.3 0.2 0.2

13 8.5 6.4 6.5 4.4 22

18:39 19:39 20:39 21:39 22:39 23:39

* AVHRR data have smaller pixels (higher spatial resolution) so an equivalent number is calculated to compare. Data from Wen & Rose (1994). Mean optical depth is unitless. Local time = universal time (UT) minus 4 hours.

DYNAMICS OF VOLCANIC AND METEOROLOGICAL CLOUDS

between the scanned pixels but excluding background values, with an approximate error of % (Krueger et al. 1995). In addition, the aerosol index (AI) is mapped from TOMS data. AI is a unitless, relative scale that identifies relative amounts of aerosols and ash in volcanic clouds (Seftor et al. 1998). AI is the difference in the logarithms of backscattered radiance at two near-

545

ultraviolet wavelengths compared with what would be expected for a pure Rayleigh atmosphere over a Lambertian surface (Krotkov et al. 1999). Meteorological clouds have AI values of approximately 0 while absorbing media (ash clouds, smoke and dust) have positive values that increase with higher optical depth (a maximum value of 6 was used in this study). The practical application of the AI

Fig. 7. Spatial distribution of optical depth values (unitless) within the volcanic cloud cores for selected data, 26 December 1997. The first two images are enlarged in scale by a factor of four compared to the others.

546

G. C. MAYBERRY ET AL.

Fig. 8. Spatial distribution of effective radius retrievals of volcanic ash (in m) within the volcanic cloud cores for selected data 26 December 1997. The first two images are enlarged in scale by a factor of four compared to the others.

scale to cloud analysis can be complex, but it is similar to the infrared retrievals of Wen & Rose (1994). As shown by Krotkov et al (1999, fig. 2), an AI value for a given particle size distribution can be considered linearly related to ash optical depth (hence mass) under typical cloud conditions. Thus, for a mean effective radius of 1 m, an AI value of 10 would represent approximately twice the mass of an AI value of 5. For other mean effective radii, the relationship is still roughly linear, but not necessarily one-to-one. However, if

the particle size distribution is multimodal the AI/mass relationship also becomes a function of multiple curves (one for each size distribution). One TOMS observation of the volcanic cloud is available at 15:43 UT (Figs 9 and 10), nearly coincident with the areal maxima for the volcanic cloud and with one of the GOES data sets. The swath that captured the volcanic cloud terminated at the SE edge of the cloud (Fig. 9), so the SO2 mass of 340 x 10 6 kg is a

DYNAMICS OF VOLCANIC AND METEOROLOGICAL CLOUDS

547

Fig. 9. TOMS SO2 cloud at 15:43 UT, 26 December 1997. In this plot the footprints of individual pixels are plotted with their SO2 burden in Dobson units (equivalent to milli-atm cm and shown by colour scale). Note that the east edge of the SO2 cloud is undefined because of the edge of the orbital swath.

Fig. 10. TOMS aerosol index (AI; unitless) map at 15:43 UT, 26 December 1997. Note that the AI image shows an elongate volcanic cloud, with a very different shape from the SO2 (Fig. 9).

minimum value. The TOMS AI map (Fig. 10) located ash south of Montserrat with an area of c. 150000km2 (about 30% larger than the band 4 area in Fig. 3) and elongated towards the SSE. The ash extended from west of Montserrat to the SSE until c. 60° west and then continued east creating a hook shape that is also con-

spicuous in the 10:39 UT and 13:09 UT GOES 8 band 4 images (Fig. 3b and c). This comparison shows clearly that the ideal TOMS AI data, collected at solar noon, can outline the volcanic ash much more fully than GOES or AVHRR data. Note that the SO2 was located in the easternmost portion of the volcanic cloud. Separation

548

G. C. MAYBERRY ET AL.

of SO2 and ash in volcanic clouds has been observed in a number of other cases (Rose et al 1995; Holasek et al. 1996; Shocker 1996; Schneider et al. 1999) and will be discussed below.

Isentropic wind trajectory model An isentropic wind trajectory model (Allen et al. 1999) was used to recreate the meteorological conditions during the 26 December, 1997 collapse in an effort to simulate the three-dimensional dispersal of ash in the volcanic cloud. Meteorological data are poor in the tropics because the weaker Coriolis effect at low latitudes causes wind dynamics to be more complex than at higher latitudes. A qualitative reconstruction of ash dispersal over time was produced using wind data. Five models of the wind data by three meteorological centres were analysed in this study: the National Meteorological Center (NMC), the United Kingdom Meteorology (UKM) and the American Society of Meteorologists (ASM). NMC data with a 1° x 1° grid produced the model of the volcanic cloud that visually correlated most accurately with the position of the volcanic cloud in GOES 8 imagery. Using NMC wind field data for 26 December 1997, the model was run by starting a set of 72 air parcels distributed in a 25 km diameter ring centred on the volcano when the explosive activity began (Schoeberl et al. 1992). In the model the air parcels are advected by wind, temperature and pressure values derived from NMC's balanced wind equations and then interpolated to isentropic (constant potential temperature) levels over the length of time the volcanic cloud was visible by GOES 8 (Schneider et al. 1999). Data are available every six hours starting at 06:00 UT on 26 December 1997 (c. 1 hour prior to the collapse) and they were averaged between the six-hour intervals. The positions of the air parcels at various heights were compared to the position of the volcanic cloud in band 4 and BTD GOES 8 images to determine the vertical extent of the volcanic cloud. The isentropic wind trajectory model was run for a range of altitudes starting at 07:00 UT on 26 December 1997 with all points initially located at the volcano. Trajectory points were computed for air parcels at different alti-tudes at four-hour intervals (Fig. 11). The trajectory points were compared to satellite images of the volcanic and VM clouds to determine the matching altitude. From comparing positions of the parcels in the trajectory results and the orientation of the clouds in satellite imagery, we discovered that the volcanic cloud extended from about 2 to 15km altitude, while

the VM cloud rose to c. 17 km. Wind trajectory data (Fig. 11) show that by 12:00 UT on 26 December ash at 2km altitude had travelled at 24kmh - 1 to the west, ash at 4km altitude had travelled at 15kmh - 1 to the SW, and ash at 6km altitude had travelled at 26 km h-1 southward. At altitude of 8 km and above the ash moved SE; the wind speed was 41kmh - 1 at 8km and increased to 7 2 k m h - 1 , 104kmh -1 and 109 km h-1, respectively at 10km, 12km and 14km altitudes. At 16km the wind speed slowed to 7 0 k m h - 1 . By 00:00 UT on 27 December 1997 wind speeds at 12km and 14km were similar (98kmh -1 and 85kmh - 1 , respec- tively) and in the same direction (SE), causing co-location of ash at these heights. Overall wind speeds increased from 2 to 14km altitude at 12:00 UT and from 2 to 12 km altitude at 24:00 UT. Thus in general the higher parts of the cloud travelled farther from the volcano.

Discussion Determination of the three-dimensional morphology of the volcanic cloud Three-dimensional visualization of the volcanic cloud was derived from combining our observations. The conspicuous SSE-pointing hook shape in the 10:39 UT and 13:09 UT band 4 images (Fig. 3b and c) can be explained by the trajectory of the wind. The hook shape formed because wind speeds at an altitude of 14km (109kmh - 1 ), and at 16km ( 7 0 k m h - 1 ) . were slower than speeds below them (13km altitude, l l 9 k m h - 1 ) , and because before 13:09 UT winds at altitudes above 14km blew in a slightly (15 0 ) more easterly direction. A cross-sectional view of the volcanic cloud trajectory showing the latitude of ash at various heights was created from wind trajectory data (Fig. 12). The plot shows the position of the volcanic cloud 24 hours after the eruption began, with the wind field data averaged between 00:00 UT on 26 December 1997 and 00:00 UT on 27 December 1997. By c. 07:00 UT on 26 December the wind speed gradually increased up to c. 9 km altitude. then decreased up to c. 12 km, and then increased again causing the volcanic cloud to have an S-shaped front. This is significant because, in the absence of wind shear, the satellite would see overlapping layers of fine ash at the same latitude. The wind direction on 26 December 1997 was constantly to the SE (with azimuth varying <150 ) between

Fig. 11. Isentropic trajectory model results for 26 December 1997. Each line connects points of parcels moving at equal altitudes of 2, 4, 6, 8, 10, 12, and 14 km. Points represent 4 hours of movement. All sets begin at the volcano at 06:39 UT and include 10:39, 14:39, 18:39 and 22:39 UT, with each point being farther from the volcano's position reflecting continued movement by winds.

DYNAMICS OF VOLCANIC AND METEOROLOGICAL

c. 8km and 16km altitude, with little wind shear, thus overlapping layers were seen by the satellite sensors. This three-dimensional visualization of the cloud helps us evaluate whether the assumption of cloud shape parallel to the Earth's surface is violated seriously.

CLOUDS

549

Because the cloud's slope towards the upwind direction is slight (rise of c. 10 km in a run of c. 800km), it is very nearly parallel.

Comparison of satellite retrieval data We compared similar volcanic cloud data from two satellites. Figure 13 shows the correlation of the AVHRR and GOES 8 images. The GOES 8 image was collected about 8 minutes before the AVHRR image and as a result the contours showing the ash position determined from GOES 8 data are shifted about 7 km NE of the position of the AVHRR ash cloud. Wind speeds reached a maximum of 1 1 9 k m h - 1 at 13km altitude and the average speed between 2 and 15km was 5 1 k m h - 1 , so the ash in the AVHRR image had been moved by wind during the time between data collections (7km, 8 minutes, c. 52kmh - 1 ). Retrievals for both data sets are compared in Table 2. Correcting for differing pixel resolutions, the area of volcanic cloud detected by AVHRR is 47% higher than GOES. The mean optical depth retrieved by the AVHRR is also higher by 47%, and these differences result in the ash mass being 81% higher. These differences are most likely caused by the higher spatial resolution of the LAC AVHRR data, which results in a larger footprint of ash cloud, resulting from the addition of pixels at the optically thin fringes (i.e. BTD close to —0.5°C).

Fig. 12. A cross-sectional north-south view of dispersed volcanic cloud parcels at 07:00 UT on 27 December 1997, 24 hours after the start of the eruption. The plot is meant to show how the cloud's trajectory at different altitudes led to an overlapping relationship. The vertical axis is plotted in a meteorological format, with potential temperature (left) and actual height (right).

Fig. 13. Comparison of the position of brightness temperature difference (BTD) volcanic cloud anomalies from GOES 8 (17:39 UT) and AVHRR (17:47 UT). The GOES data are contoured with black lines at BTD= -0.5, -1.5, -2.5 K. The slight positional offset of the cloud position is explained by the time differences and the prevailing winds.

The volcanogenic meteorological cloud associated with the 26 December event The pattern of movement of the VM cloud, which we believe formed from the evaporation of sea water by the pyroclastic density current with sea water (see discussion below), was revealed by the

550

G. C. MAYBERRY ET AL.

GOES 8 data. Part of the 26 December volcanic cloud had high (>5 0 C) positive BTD values, unlike most other studied volcanic clouds (Shannon 1996; Constantine et aL 2000; Schneider el al, 1999). The positive BTD values were observed from 07:09 UT to 11:39 UT (Fig. 14). While volcanic clouds typically have negative BTD values (red and orange in Fig. 14), and the lowest meteorological clouds have positive BTD values between c. 0 to 5°C (green and

yellow), in this case a portion of the volcanic cloud had BTD values >50C (blue). These blue regions show the regions where the signal of the VM cloud dominates. The high BTD values reflect the presence of ice and liquid water, rather than ash. As Figure 14 illustrates, at 07:09 UT pixels >5 C were concentrated west of the volcanic cloud, and at 07:39 UT they fringed most of the volcanic cloud and increased in area. The VM cloud then masked portions of the volcanic

Fig. 14. Selected GOES images, highlighting the position of the volcanic cloud and the volcanogenic meteorological (VM) cloud, 26 December 1997. Red and brown pixels depict the volcanic cloud (BTD < —0.5) while blue pixels depict the VM cloud (BTD>0).

DYNAMICS OF VOLCANIC AND METEOROLOGICAL

cloud until 11:39 UT. The VM cloud is generally located along the northern fringe of the volcanic cloud near the leading edge (Fig. 14). Modelling of the microphysics in volcanic clouds shows that ice is the dominant phase at c. 6 km in altitude where temperatures are generally less than -40 °C (Herzog et al 1998). Thus the VM cloud was most likely composed primarily of water below 6 km and ice above 6km. The volcanic cloud from the eruption of Rabaul in 1994 was shown to have abundant ice (Rose et al. 1995) and this was derived from seawater entering the vent rather than external seawater/density current interactions. The ice in the Rabaul cloud produced highly positive BTD values in AVHRR images, which masked the characteristic spectral signature of the ash in a similar manner. The VM cloud masks the characteristic spectral signature of ash in the volcanic cloud by dampening the negative BTD value of the ash pixels with the highly positive BTD meteorological cloud pixels. The satellite detects upwelling thermal radiation that is scattered and absorbed by clouds, and in this case detects effects of both hydrometeors (snow, hail, graupel and raindrops) and volcanic ash. One assumption of the Wen & Rose (1994) retrieval method - that the atmosphere above and below the volcanic cloud is clear - is violated in this case. The masking effect of the VM cloud is evident on plots of the volcanic cloud's physical characteristics (Fig. 6). The masking effect was most prominent between 07:39 UT, when the meteorological cloud rose high enough to disturb the characteristic spectral signature of the ash, and 13:09 UT, when it began to dissipate. Although the 26 December 1997 collapse was not directly observed, the idea that the VM cloud resulted from interaction between the pyroclastic density current and seawater is supported by observations and photographs of smaller Soufriere Hills volcanic events, particularly the 12 May 1996 dome collapse. The pyroclastic flow on 12 May 1996 involved 0.4 x 106 m3 of dome material that descended Tar River valley (Cole et al 2002). It travelled to the sea where the upper surge component detached and travelled 200 m across the sea surface, while the dense basal avalanche entered the

Fig 15. Trajectory data (from Fig. 11) at 10:39 UT (black dots with numbers represent altitude in km) superimposed over the 10:39 UT brightness temperature difference map from Fig. 13.

CLOUDS

551

sea (Cole et al. 2002). The flow front velocity increased from 8-15ms - 1 to a maximum of 2 5 m s - 1 . Lofting ash plumes were seen above the distal end of the flow and boiling water near the pyroclastic fan produced dense steam that rose several hundred metres (Cole et al 2002). Information about the hydrometeors in the VM cloud was retrieved from the 09:39 UT GOES 8 image (when the VM cloud had its highest optical depth) using the model of Wen & Rose (1994) and following the same methodology as Rose et al (1995) used in the analysis of the ice in the Rabaul volcanic cloud. The VM cloud was 10000km 2 and the mean effective radius of its ice was 24 m. The total mass of ice in the VM cloud was thus estimated from the remote sensing data at 1.5 x 108 kg, which is c. 3.5 times the mass of the silicate ash in the volcanic cloud (4.5 x 107 kg). Without other data the imagery cannot resolve whether the VM cloud underlies, overlaps or overlies the volcanic cloud. The VM cloud had its largest area at 10:39 UT (12000km 2 ) and, based on wind trajectory data, one part of the VM cloud was higher than the volcanic cloud, at a height of 16 km or more (Fig. 15). The range of trajectory-based heights for the VM cloud go from about 4 to perhaps 17km. At 10:39 UT there appears to have been a slowermoving, lower (4-8 km altitude) part of the VM cloud south of Guadaloupe and west of Dominica, and a higher (8-17 km altitude) part located to the east of Martinique and St Lucia. By 11:39 UT the northern section of the VM cloud dissipated and the southern portion increased in area towards the east. After 11:39 UT the remaining portion of the VM cloud continued to dissipate until it was below satellite detection limits by 13:39 UT, while the volcanic cloud remained detectable. The evolution of the VM cloud is shown schematically in Figure 16. Dissipation of the VM cloud coincided with local noon (16:00 UT) on Montserrat, so its disappearance may be associated with increasing solar radiation of a surrounding undersaturated atmosphere. Also, as Rose et al (1995) proposed for Rabaul, the presence of ice in the volcanic cloud may have reduced the residence time of

552

G. C. MAYBERRY ET AL.

This is a maximum mass of water, assuming all of the ash that reached the sea was used to evaporate the water with 100% efficiency. The efficiency of evaporation would be far less than 100% for several reasons. Larger fragments in the pyroclastic density current probably entered the sea, sank, and heated large quantities of water without reaching the boiling point, so that evaporation did not occur. It is likely that much of the heat exchange may have heated the water only slightly, not enough to boil. The VM cloud was estimated above to have a maximum mass of 1.5x 10 8 kg (equivalent to 4.7 x 10 14 J) of water vapour at 09:39 UT on 26 December 1997. This amounts to an efficiency of c.5% if we assume that all of the H2O in the cloud came from evaporation induced by the heat exchange between the pyroclastic density current and the sea. If the entire mass of andesitic ash (1.0 x 10 11 kg) was fed into a bouyant 'thermal' plume model such as outlined by Sparks et al. (1997, fig. 7.23, p. 206), a cloud height of c. 15 km would result. If the entire steam mass estimated by the evaporation at 100% efficiency (3.19 x 10 10 kg) was fed into a similar thermal model the height would be similar. This is close to the observed height of both the ash and VM clouds (15-17 km, see above). Exact agreement would be unlikely because the thermal energy of the andesite was partly transferred to seawater evaporation and part of the thermal energy was dissipated into heating of seawater without evaporation. It is likely that the rise of both the ash and VM cloud were nearly simultaneous and reinforced each other, so the height of both clouds is the result of their collective thermal energy. The effects of high humidity on volcanic cloud detection

Fig. 16. Schematic cross-sectional diagram of the clouds of 26 December 1997, as seen looking north at various stages in the first hour beginning at 07:00 UT (VC, volcanic cloud; VM, volcanogenic meteorological cloud).

the ash by coating the particles with ice, increasing their mass and diameter and causing accelerated fallout. The accretionary lapilli deposits SW of the volcano may have formed due to hydrometeor/ ash interactions soon after the eruption, but before the volcanic cloud drifted SSE of the volcano. The generation of meteorological clouds from pyroclastic density currents that reach the sea is likely to have occurred often in the geological record, for example in the 1883 Krakatoa eruption (Carey et al. 1996), and could represent an important means of atmospheric perturbation by large eruptions. Estimation of the mass of water in the volcanogenic meteorological cloud The maximum mass of water evaporated by the pyroclastic density current upon entering the sea can be estimated using heat-balance equations and constrained using information on the physical characteristics of the VM cloud retrieved from GOES 8. The temperature of the andesitic magma is known to be 850°C (Barclay et al. 1998), and an estimated 85% of the total mass of the pyroclastic density current reached the sea (R. S. J. Sparks, pers. comm.). The density of dense andesite is 2600 kg/m 3 (McBirney 1993), and the temperature of the ambient surface seawater derived from band 4 GOES 8 data was 22°C. The heat balance equation is: maCa(Ta - 373) = m w C w (373 - 295) + mwHv (1) where Ca is the specific heat of andesite, 1200Jkg -1 K-1 (McBirney 1993); ma is the mass of andesite, 1.0 x 1011 kg; Ta is the temperature of the Montserrat andesite = 850°C=1123K; Cw is the specific heat of water, 4128 J k g - 1 K - 1 ; Hv is the latent heat of vaporization, 2.5 x 106 Jkg -1 . We can then solve for mw , the water mass = 3.19 x 10 10 kg.

Figure 2 shows the comparison of areas of the volcanic cloud using band 4 alone and the BTD method. The differences in area are largely due to the effects of high tropospheric humidity on a weak volcanic cloud signal. Coll & Caselles (1997) used real data to validate split-window algorithms for sea and land surface temperatures and showed that under humid conditions atmospheric transmission decreases with increased humidity. Rose & Prata (1997) found that high tropospheric water vapour in the tropics dramatically affected the BTD values by shifting volcanic cloud pixels with low optical depth to higher BTD values by up to about 3 K. This resulted in fewer pixels being identified as volcanic ash. Use of a variable BTD cutoff to distinguish the volcanic cloud is investigated by Yu (2000), who used a new method of discriminating the volcanic cloud, based on atmospheric corrections for both the band 4 and band 5 brightness temperatures and the BTD, which also shifts the retrieval grid significantly. These corrections allow more accurate estimates of ash masses and areas to be used in tropical regions where humidity is high. Based on Yu's (2000) preliminary work, the masses given in Table 2 are probably too low by factors of about three to five. SO2 and ash separation in the 26 December 1997 volcanic cloud TOMS images at 15:43 UT of the positions of the SO2 gas (Figs 9 and 17) and the volcanic ash (Fig. 10) show that the SO2 was concentrated in the easternmost portion of the volcanic cloud. Since only one TOMS image was collected on 26 December 1997, it is not possible to determine if the SO2 gradually separated from the volcanic cloud, or if it was leading the volcanic cloud to the east. The eastern edge of the SO2 mass is not well known because it was at the edge of the TOMS swath. The SO2 pattern seen by TOMS is circular and there was no tail of SO2 into the volcanic cloud to the west, suggesting that the SO 2 was separated and was not interacting with the volcanic ash by 15:43 UT. Since the SO2 and the highest portion of the volcanic cloud overlapped, it is most likely that the SO2 rose to at least the maximum height reached by the ash (c. 15km). It is clear that SO2 did not rise above 17km, which was the height of the tropopause, because the morphology of the plume does not reflect the effects of the change in wind direction and speed

DYNAMICS OF VOLCANIC AND METEOROLOGICAL CLOUDS

553

Fig. 17. TOMS SO2 burden contours at 15:43 UT (white lines showing contours every two Dobson Units from 15 to 23), superimposed over the GOES band 4 image (band 4 brightness temperatures in °C) at 15:39 UT.

above 17km. The wind changed speed and direction dramatically from 75kmh - 1 to the SE at 17km to 21kmrh - 1 to the SSW at 18km. Table 3 lists volcanic clouds in which SO2 separation from volcanic ash have been observed. Four explanations for the separation have been suggested, as listed below. Gravitational sedimentation in the volcanic cloud. According to Holasek et al. (1996) SO2/ash separation may occur due to gravitational sedimentation of ash within the volcanic cloud. They performed experiments that illustrate the vertical segregation within a volcanic cloud where a relatively light, gas-rich portion of the volcanic cloud quickly rises, and the denser, ash-laden portion rises more slowly until neutral buoyancy is reached. In their study of the April 1982 eruption of El Chichon, Schneider el al (1999) propose that gravitational separation is more likely to occur if the eruption intensity is high enough to emplace the volcanic cloud a few kilometres higher than the level of strong wind shear, where separation is aided by disparate wind directions; for example at the troposphere/stratosphere interface. While the volcanic clouds in Table 3 that experienced separation were emplaced several kilometres above the tropopause, the 26 December 1997 volcanic cloud was not. However, wind speeds were much slower at 10km (72kmh - 1 ) than at 12km (104kmh - 1 ) and 14km (109kmh - 1 ), and the direction of the wind varied by only c. 5°, so gravitational separation could have been accelerated by disparate wind speeds between 10 and 12km altitude. Table 3. Examples of gas/ash separation of volcanic clouds Eruption (date)

SO2 height Ash height Ref. (km) (km)

El Chichon (4/82) Hudson (8/91) Lascar (4/93) Soufriere Hills (12/97)

22-26 14-18 >18 14

19-21 10-14 12-18 4-12

Schneider et al. (1999) Constantine et al. (1999) Shocker (1996) Rose & Mayberry (2000)

Accelerated sedimentation caused by interaction between the VM and volcanic clouds. Separation of the SO2 from ash in volcanic clouds may be promoted by the presence of hydrometeors (Herzog et al. 1998). According to Rose et al. (2000) the excess water from hydrometeors may interact with the ash in volcanic clouds, causing accelerated sedimentation of ash. Sedimentation in the 26 December 1997 volcanic cloud may have been accelerated by hydrometeors produced by interaction between the pyroclastic density current and seawater. The abundant accretionary lapilli from the eruption suggest that there was interaction between the ash in the volcanic cloud and the water in the VM cloud resulting in aggregation and accelerated sedimentation.

Formation of a gas-rich cap in the conduit. In this case, SO2/ash separation occurs before the volcano erupts, when a gas-rich cap forms in the magma conduit due to degassing of the magma body and movement of gas through a foamy, vesiculated magma body (Mader 1998). When an eruption occurs, the gas-rich cap is released first, rising higher than the ensuing volcanic cloud. We do not know whether any evidence for this idea exists, but it would require a gasrich part of the explosion to have occurred in the first seconds of the event, perhaps many minutes before convection of the elutriated ash from the pyroclastic density current occurred.

Scavenging of SO2 by hydrometeors in the VM cloud. Rose et al. (1995) propose that low SO2 concentrations were observed in the Rabaul volcanic cloud because excess water in the cloud from hydrometeors scavenged or absorbed the SO2. They contend that scavenging occurred because the SO2 in the volcanic cloud was either adsorbed on ice hydrometeors or was dissolved in the supercooled water in the volcanic cloud, which then formed ice around ash nuclei at c. 6km altitude and fell out of the volcanic cloud. A similar process could have occurred in the 26 December 1997 volcanic cloud, but it is unclear why the highest part of the volcanic cloud would not have experienced it.

554

G. C. MAYBERRY ET AL.

Conclusions The abundance of satellite, meteorological and ground-based data available during the 26 December 1997 dome collapse and explosive decompression of Soufriere Hills Volcano provided an opportunity to analyse the dynamics of the volcanic and volcanogenic meteorological (VM) clouds produced, even though the early parts of the activity were in darkness. The pyroclastic density current generated from the collapse of the lava dome produced a volcanic ash cloud. When it reached the sea, the hot ash heated the sea surface and evaporated seawater, producing a VM cloud. The volcanic cloud was recorded by the GOES 8, AVHRR and TOMS sensors; the more ephemeral VM cloud was recorded only by GOES 8 due to its greater frequency of data collection. The volcanic cloud had a maximum mass of c. 4.5 x 107 kg of fine (1-25 m diameter) volcanic ash, and grew to a maximum area of about 100000km 2 at about 16:00 UT. Wind data were used to generate a trajectory model that helped define the three-dimensional geometry of the volcanic and VM clouds. The volcanic cloud rose to c. 15 km, growing in area to the SE while it simultaneously migrated SW, and the VM cloud was at similar altitudes but rose slightly higher to a maximum altitude of c. 17 km. The VM cloud had a maximum mass of 1.5 x 108 kg of ice at 09:39 UT and grew in area for several hours. It interacted with the volcanic cloud, adding excess water/ice to it. The VM cloud masked the volcanic cloud from c. 07:39 to 13:39 UT by interfering with the signal of the silicate ash to the GOES 8 sensor, causing underestimations of ash mass values. The VM cloud dissipated after c. 6 hours, possibly due to evaporation or ice particles falling out of the cloud. The higher spatial resolution of the AVHRR data showed a larger area of the fringe volcanic ash pixels than the GOES 8 data. The larger area of ash retrieved by the AVHRR sensor resulted in mean optical depth and total area values that were c. 45% higher, a total mass value that was 80% higher. These differences were higher than they would have been for larger eruptions. Even though the 26 December 1997 collapse was among the largest of the series of Soufriere Hills Volcano events of 1995-1999, it was one to three orders of magnitude smaller than eruptions of recent years elsewhere in the world. The 26 December 1997 data set also provided an important example of a small volcanic cloud with complex evolution in a humid region. Applying the -0.5°C BTD cutoff to the volcanic cloud outlines only the densest 20% of the volcanic cloud. Atmospheric corrections must be made in order to detect the thin edges of small volcanic clouds in a humid atmosphere. The problem is more important when the eruption is very small, such as the 26 December 1997 event, so that the ash signal is nearly lost at the optically thin edges of the volcanic cloud. TOMS data revealed that the SO2 rose higher than the ash-rich portion of the 26 December 1997 volcanic cloud. The SO2 may have been higher than the ash due to sedimentation causing accelerated fallout of the ash, and the shearing of the volcanic cloud by disparate wind speeds at different altitudes. Alternatively it could be the result of the early release of a gas-rich cap in the 26 December 1997 event or scavenging of the SO2 by excess water in the ash cloud below 6km altitude. This project was funded by the NASA TOMS Research Program, the National Science Foundation, a Michigan Space Grant, and the American Geological Institute. We obtained the GOES data for this study through the co-operation of D. Johnson at the Research Applications Program at NCAR who received the data and allowed us to copy it from their data storage. M. Davies made the initial survey of all the 1997 Montserrat GOES data and got it ready to study. We would like to thank NASA Goddard Space Flight Center for the use of their facilities and their knowledge, and especially P. Newman, S. Schaefer, A. Krueger and N. Krotkov for sharing their expertise. D. Schneider and the staff of the Alaska Volcano Observatory helped to make the processing easier. Thanks also go to G. Norton, S. Young and the staff of the Montserrat Volcano Observatory for sharing information about the volcano. S. Sparks and the staff and students at the University of Bristol have also been instrumental to the success of this pro-

ject. M. Bursik, L. Glaze and D. Rothery made several helpful suggestions. and T. Druitt did an extraordinarily attentive editing job. which clarified much of the text.

References ALLEN, D. R., SCHOEBERL, M. R. & HERMAN, J. R. 1999. Trajectory modelling of aerosol clouds observed by TOMS. Journal of Geophysical Research, 104, 27461-27471. BARCLAY, J., RUTHERFORD, M. J., CARROLL, M. R.. MURPHY, M. D., DEVINE, J. D., GARDNER, J. & SPARKS. R. S. J. 1998. Experimental phase equilibria constraints on pre-eruptive storage conditions of the Soufriere Hills magma. Geophysical Research Letters, 25. 3437-3440. BARRY, R. G. & CHORLEY, R. J. 1992. Atmosphere, Weather and Climate. Routledge, London. BAXTER, P. J., BONADONNA, C., DUPREE. R. ET AL. 1999. Cristobolite in volcanic ash of Soufriere Hills volcano, Montserrat, British West Indies. Science, 283, 1142-1145. BONADONNA, C., MAYBERRY. G. C., CALDER, E. S. ET AL. 2002. Tephra fallout in the eruption of Soufriere Hills Volcano. Montserrat. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society of London, Memoirs, 21. 483-516. CAREY, S., SIGURDSSON, H., MANDEVILLE, C. & BRONTO. S. 1996. Pyroclastic flows and surges over water: An example from the 1883 Krakatau eruption. Bulletin of Volcanolology, 57, 493-511. COLE, P. D.. CALDER, E. S., SPARKS. R. S. J. ET AL. 2002. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society of London, Memoirs, 21. 231-262. COLL, C. & CASELLES, V. 1997. A split-window algorithm for land surface temperature from Advanced Very High Resolution Radiometer data: Validation and algorithm comparison. Journal of Geophysical Research, 102, 16697-16713. CONSTANTINE, E. K., BLUTH. G. J. S. & ROSE. W. I. 2000. TOMS and AVHRR observations of drifting volcanic clouds from the August 1991 eruptions of Cerro Hudson. In: CRISP. J. & MOUGINIS-MARK. P. (eds) Volcano Remote Sensing, AGU Monograph. 116, 45-64. DAVIES, M. A. & ROSE, W. I. 1998. GOES imagery fills gaps in Montserrat volcanic cloud observations. Eos. Transactions American Geophysical Union, 79, 505-507. HERZOG, M., GRAF, H. F., TEXTOR, C. & OBERHUBER, J. M. 1998. The effect of phase changes of water on the development of volcanic plumes. Journal of Volcanology and Geothermal Research, 87. 55-74. HOLASEK, R. E., WOODS, A. & SELF, S. 1996. Experiments on gas separation processes in volcanic umbrella clouds. Journal of Volcanologv and Geothermal Research, 70. 169-181. KIDDER, S. Q. & VONDER HAAR. T. H. 1995. Satellite Meteorology. Academic Press. San Diego. KROTKOV, N. A., TORRES. O., SEFTOR. C. ET AL. 1999. Comparison of TOMS and AVHRR volcanic ash retrievals from the August 1992 eruptions of Mount Spurr. Geophysical Research Letters, 26, 455-458. KRUEGER. A. J. 1983. Sighting of El Chichon sulfur dioxide with the Nimbus 7 Total Ozone Mapping Spectrometer. Science, 220. 1377-1378. KRUEGER. A. J.. WALTER. L. S.. BHARTIA. P. K., SCHNETZLER. C. C., KROTKOV, N. A., SPROD. I. & BLUTH. G. J. S. 1995. Volcanic sulfur dioxide measurements from the Total Ozone Mapping Spectrometer instruments. Journal of Geophysical Research, 100. 14057-14076. LILLESAND. T. M. & KIEFER. R. W. 1987. Remote Sensing and Image Interpretation. John Wiley. New York. MADER, H. M. 1998. Conduit flow and fragmentation. In: GILBERT, J. S. & SPARKS, R. S. J. (eds) The Physics of Explosive Volcanic Eruptions. Geological Society, London. Special Publications. 151. 815-823. McBIRNEY, A. R. 1993. Igneous Petrology. Jones and Bartlett, Boston. MCPETERS, R., BHARTIA. P. K., KRUEGER, A. ET AL. 1998. Earth Probe Total Ozone Mapping Spectrometer ( TOMS) data product user's guide. NASA Reference Publication. TP-1998-206895 POLLACK, J. B., TOON, O. B. & KHARE, B. N. 1973. Optical properties of some terrestrial rocks and glasses. Icarus, 19. 372-389. PRATA, A. J. 1989. Infrared radiative transfer calculations for volcanic ash clouds. Geophysical Research Letters. 16. 1293-1296.

DYNAMICS OF VOLCANIC AND METEOROLOGICAL CLOUDS RITCHIE, L. J., COLE, P. D. & SPARKS, R. S. J. 2002. Sedimentology of deposits from the pyroclastic density current of 26 December 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 435-456. ROSE, W. I. & MAYBERRY, G. C. 2000. Use of GOES thermal infrared imagery for eruption scale measurements, Soufriere Hills, Montserrat. Geophysical Research Letters, 27, 3097-3100. ROSE, W. I. & PRATA, A. J. 1997. Atmospheric corrections for two band infrared volcanic cloud discriminations and retrievals. Eos. Transactions American Geophysical Union, F818. ROSE, W. I. & SCHNEIDER, D. J. 1996. Satellite images offer aircraft protection from volcanic ash clouds. Eos. Transactions American Geophysical Union, 77, 529-532. ROSE, W. I., DELENE, D. J., SCHNEIDER, D. J. ET AL. 1995. Ice in the 1994 Rabaul eruption cloud: Implications for volcanic hazard and atmospheric effects. Nature, 375, 477-479. ROSE, W. I., BLUTH, G. J. S. & ERNST, G. G. J. 2000. Integrating retrievals of volcanic cloud characteristics from satellite remote sensors - a summary. Philosophical Transactions of the Royal Society Series A, 358, 1585-1606. SCHNEIDER, D. J., ROSE, W. I., COKE, L. R., BLUTH, G. J. S., SPROD, I. E. & KRUEGER, A. J. 1999. Early evolution of a stratospheric volcanic eruption cloud as observed with TOMS and AVHRR. Journal of Geophysical Research, 104, 4037-4050.

555

SCHOEBERL, M. R., LAIT, L. R., NEWMAN, P. A. & ROSENFIELD, R. E. 1992.

The structure of the polar vortex. Journal of Geophysical Research, 97, 7859-7882. SEFTOR, C. J., Hsu, N. C., HERMAN, J. R. ETAL. 1998. Detection of volcanic ash clouds from Nimbus-7/TOMS, Journal of Geophysical Research, 102, 16749-16760. SHANNON, J. M. 1996. 3-D reconstruction of the Mount Spurr volcanic clouds using A VHRR, TOMS, and wind trajectory data. MS Thesis, Michigan Technological University. SHOCKER, H. L. 1996, Lascar volcanic clouds of 1993: Merging of satellitebased remote sensing from TOMS, A VHRR and ATSR during three days of atmospheric residence. MS Thesis, Michigan Technological University. SPARKS, R. S. J., BURSIK, M, L, CAREY, S. W. ET AL. 1997. Volcanic Plumes. John Wiley & Sons, Chichester. SPARKS, R. S. J., BARCLAY, J., CALDER, E. S. ET AL. 2002. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 409-434. WEN, S. & ROSE, W. I. 1994. Retrieval of sizes and total masses of particles in volcanic clouds using AVHRR bands 4 and 5. Journal of Geophysical Research, 99, 5421-5431. Yu, T. 2000. Improved algorithms for volcanic clouds using multispectral infrared remote sensing. PhD Dissertation, Michigan Technological University.

Monitoring of airborne particulate matter during the eruption of Soufriere Hills Volcano, Montserrat K. R. MOORE 1 , H. DUFFELL2, A. NICHOLL3 & A. SEARL3 1

Department of Geology, National University of Ireland Galway, Galway, Ireland (e-mail: [email protected]) 2 Department of Earth Sciences, University of Cambridge, Cambridge CB2 3EQ, UK 3 Institute of Occupational Medicine, 8 Roxburgh Place, Edinburgh EH8 9SU, UK

Abstract: A programme of air quality monitoring was devised to investigate concentrations and behaviour of respirable ash in the air on Montserrat. Sampling strategies used were short environmental tests, occupational tests, and continuous environmental monitoring in periods of both high and low volcanic activity. The results obtained between September 1996 and July 1999 allow comparison of the contributions made to airborne particles by different eruptive styles. Vulcanian explosions and ash-venting increased average background concentrations (c. 30 g m - 3 ) of airborne particles (<10 m in diameter) by up to a factor of ten. Large dome collapses released large amounts of ash into the atmosphere in an extremely short time and airborne particle concentrations reached more than 600 g m - 3 . Aggregates of ash particles broke up on impact with the ground and ash re-suspension by wind caused airborne particle concentrations of approximately 250 g m - 3 , which were sometimes significantly higher than those caused by tephra fallout. Airborne particle concentration decreased with distance from the ash source, and wind and rain effectively removed ash from the atmosphere. Generally, airborne particle concentrations in inhabited areas of Montserrat remained low, even during elevated volcanic activity, when volcanic products were carried away by winds.

Tephra fallout containing respirable ash has been a significant feature of the volcanic activity during the eruption of Soufriere Hills Volcano, Montserrat. Prior to March 1997, pyroclastic flows were largely restricted to the Tar River valley east of the volcano, but airborne ash was dispersed to the area west of the dome by wind (Fig. 1). Subsequently fallout affected the entire island, with central and northern parts of Montserrat being most affected between June and October 1997. The island of Montserrat is less

Fig. 1. Location map showing air quality monitoring sites (dots) and thickness of fallout from different styles of volcanic activity. The shaded area shows the extent of pyroclastic flow and surge deposits up to March 1998. Ash-mass loading contours (Bonadonna et. al. 2002) give ash deposited by dome collapses (solid lines; cm) and Vulcanian explosions (long dashed lines; cm) and by ash-venting (short dashed lines; kgm - 2 ).

than 16 km in length and up to 12 000 people lived on the island when the eruption started in July 1995. Although two-thirds of the population left the island, and the remainder were relocated to the north, a significant number of people have been exposed to airborne ash. The first measurements of airborne ash particles were made in March 1996 (Allen et al. 2000) to investigate the potential health hazards. Further surveys of exposure to respirable dust and silica were undertaken in September 1996, June 1997 and August 1997. In October 1997, a network of environmental testing sites was established to provide daily information about air quality for the island's population. Epidemiological studies on the effects of urban air pollution have shown that short-term exposure to elevated levels of respirable particulate matter is associated with an increased prevalence of the symptoms of respiratory conditions such as bronchitis and asthma (Committee on the Medical Effects of Air Pollutants 1998). Workplace exposure to crystalline silica is associated with silicosis (a progressive lung disease characterized by the development of fibrotic tissue) and, potentially, lung cancer (International Agency for Research on Cancer 1997). The criteria that should be considered when investigating hazards associated with inhalation of volcanic ash include: (i) the proximity of the population to the source of ash; (ii) the frequency of ash emissions and duration of exposure; (iii) the grain size of ash particles and whether they are of respirable size; and (iv) the mineral composition of the ash products and their toxicity. A substantial proportion (13-20 wt%) of the ash from Soufriere Hills Volcano was within the particle size that could be inhaled, and its composition had the potential to be highly toxic (Baxter et al. 1999). Thus, there was a need for airborne ash levels to be monitored in order to determine population exposure. There was also a need to monitor an individual's personal exposure, as people's activities disturbed deposited ash, giving rise to much higher exposures than might be anticipated from background air quality measurements. This paper synthesizes the results that have been amassed under the Air Quality Monitoring Programme. Its aim is to describe the effects on local concentration of airborne ash of different eruptive styles, eruptive intensities, location and meteorological conditions. A review is also given of some of the personal exposure data and of the factors affecting personal exposure to ash. Overview of the eruption The Soufriere Hills Volcano eruption has been characterized by different eruptive styles, commencing with phreatic activity (July

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 557-566. 0435-4052/02/$15 © The Geological Society of London 2002.

557

558

K. R. MOORE ET AL.

1995), which continued until magma reached the surface in November 1995. Dome collapses resulted in rockfalls and pyroclastic flows (Cole el al. 2002; Kokelaar 2002) with associated ash plumes (Bonadonna el al. 2002). The first major dome collapse occurred on 29 July 1996. Dome collapse activity was confined to the Tar River valley (Fig. 1) until after April 1997, when pyroclastic flows started travelling down the south, west and north flanks of the volcano. The dome stopped growing in March 1998, but dome collapses continued (Norton el al. 2002). Explosive activity has included subPlinian activity (17 September 1996) and two series of Vulcanian explosions (4-12 August 1997 and 22 September-21 October 1997). Vulcanian explosions typically resulted in vertical eruption columns and fountain-collapse pyroclastic flows with associated ash plumes (Druitt el al. 2002). Ash-venting (direct discharge of ash from fractures) has generated ash plumes throughout the eruption. Phreatic activity (July-November 1995) generated plume heights of up to 2.5km above sea level (a.s.l.), and ash-venting produced plumes up to 6 km. The dome-collapse activity and Vulcanian explosions generated most of the tephra fallout, as aggregates of ash particles, during the 1995-1999 period (Bonadonna el al. 2002). Both vertically directed explosions and dome collapses produced plumes up to 15km a.s.l. Tephra fallout accumulated predominantly on the west of the island, due to low-level (0-5 km) easterly prevailing winds with velocities of 3-10ms - 1 . Wind shear between 5 and 8 km was often observed due to a change in direction between low- and intermediate-level winds. Intermediate-level (8-18 km) winds typically blew towards the east but could also blow to the north, at speeds of 4-23 m s - 1 (Bonadonna el al. 2002). Thus, wind direction normally acted to minimise exposure of humans living in the north of the island to respirable volcanic particles.

Health hazards from airborne volcanic ash The health hazards posed by volcanic ash have been examined during eruption of a number of volcanoes, including Mount Spurr, 1992 (Gordian el al. 1996; Cloudhury el al 1997), Mount Sakurajima, 1985-1986 (Yano el al. 1990) and Mount St Helens, 18 May 1980 (Dollberg el al. 1986; Martin el al. 1986). Studies of respiratory disease in Anchorage, Alaska, in relation to the eruption of Mount Spurr, 95km away (Gordian el al. 1996; Cloudhury el al. 1997), found that volcanic ash combined with combustion-related fine particles had an acute, adverse effect on respiratory health, even at relatively low ambient concentrations (70 g m - 3 of particulate matter <10 m in diameter). In contrast, the main health effect of brief exposures to high concentrations of a coarser volcanic ash (up to 225 gm-3 of particulate matter >15 m in diameter) from Mount Sakurajima in Japan was a prevalence of eye symptoms (Yano el al. 1990). The composition of this volcanic ash was dominated by feldspar (albite), with negligible crystalline silica, and was found to be biologically inactive. Approximately one-third of tephra fallout from the 18 May 1980 Mount St Helens eruption had grain size less than 15 m (Carey & Sigurdsson 1982), contained little crystalline silica (Dollberg el al. 1986) and did not cause longterm damage to health (Martin el al. 1986).

Composition of Montserrat ash The lava dome at Soufriere Hills Volcano has andesitic composition, with interstitial glass of high-silica rhyolite composition (Devine el al. 1998; Baxter el al. 1999). Tephra fallout from dome collapses in the 1995-1999 period under consideration was enriched in crystalline silica, glass and feldspar due to preferential fragmentation of fine groundmass minerals to form the finest fractions, which were elutriated into ash plumes generated above pyroclastic flows (Baxter el al. 1999; Horwell el al. 2001). The contents of crystalline silica in the <10 m fraction of tephra fallout from dome

collapses and Vulcanian explosions were 10-24 wt% and 3-6 wt% respectively. Most crystalline silica was in the form of cristobalite (Baxter el al. 1999). Cristobalite and tridymite are polymorphs of quartz that form over time from vapour-phase crystallization and devitrification of volcanic glass within the dome (Baxter el al. 1999). Thus, tephra fallout from dome collapses was enriched in cristobalite and tridymite compared to that from Vulcanian explosions. The greater biological activity of cristobalite and tridymite than quartz may be related to their more open structures and metastable nature (Ross el al. 1993). Preliminary toxicity studies suggest that bulk ash was substantially more toxic than the relatively inert material rutile (Wilson el al. 2000) and, in some assays, had activities similar to that of quartz, a toxic material that causes silicosis (Searl & Nicholl 1997).

Methods Derivation of exposure guidelines Airborne particulate matter with mean aerodynamic diameter <10 m (termed PM10; concentration measured in g m - 3 ) is able to reach the lung, according to the Thoracic Sampling Convention (International Organisation for Standardisation (ISO) 1991). This includes a respirable fraction of particles that are small enough, (<4 m, termed PM4) to penetrate to the alveolar part of the lung (ISO 1991; Ashton & Gill 2000). A daily PM10 average background concentration of 50 g m - 3 is considered to pose no risk to the health of most people in the UK (Expert Panel on Air Quality Standards 1995). Occupational limits are much higher as exposure is not constant. The accepted occupational exposure limit for airborne particles (from all sources, including combustion particles and pollution) in 1996 was 5000 g m - 3 PM10 (Nicholl 1996), and the threshold limit value for cristobalite was 50 g m - 3 (American Conference of Governmental Industrial Hygienists 1996). The UK Department of Health used both background and workplace exposure limits to derive a series of Action Levels, above which precautions should be taken on Montserrat. The occupational limits are designed to protect workers exposed for 40 hours a week, 46 weeks a year. The eruption on Montserrat produced airborne particles that were potentially omnipresent, so environmental exposure lasted for approximately five times more hours per year. The occupational threshold limit value for respirable cristobalite was therefore equivalent to an environmental threshold on Montserrat of 10 g m - 3 (Searl & Nicholl 1997). Given that airborne ash particles contained up to 25% of <10 m crystalline silica that was mostly cristobalite (Baxter el al. 1999). this was equivalent to a bulk ash environmental threshold of 40 g m - 3 . In October 1997, the Department of Health in the UK proposed a general advisory banding system for a 1-hour average PM10 level of bulk airborne particles (including ash) for use on Montserrat: • •

• •

<50 g m - 3 : particle levels should be regarded as Low and nobody should need to wear a facemask; 51-100 g m - 3 : particle levels should be regarded as Raised and those who have had their health affected by dusty periods in the past, especially those who suffer from asthma, should ensure that they have masks available; 101-300 g m - 3 : particle levels should be regarded as Very High and masks should be worn; >300 g m - 3 : particle levels should be regarded as Alert Level. masks should be worn and efforts should be made to reduce exposure.

Analytical methods The TSI DUSTTRAK™ Aerosol Monitor Model 8520 is a portable laser-photometer that measures PM10. It uses light-scattering

559

AIRBORNE PARTICULATE MATTER

technology to determine mass concentration in real-time. The term 'dust' is used to encompass particulate matter of all origins, including soil erosion, volcanic eruption and pollution. The DUSTTRAK™ monitors are supplied calibrated using ISO standard dust ('Arizona road dust') and were programmed to sample PM10 every second (resolution % or 1 g m - 3 , whichever is greater), with a one-minute average recorded by the instrument. Equipment was adapted for two monitoring strategies. (1) Prior to October 1997, short-duration tests (normally between 15 minutes and one hour) using portable DUSTTRAK™ monitors located areas most affected by elevated volcanic and human activity. (2) A network of 16 stationary and continuously monitoring, or 'static', test sites (Fig. 1) was then established to provide measurements of PM10. Location of static test sites was classified according to human activity: quiet sites, with little to no human activity or traffic, provided environmental background PM10; busy sites were active thoroughfares close to major roads. A more accurate measure of an individual's exposure was achieved by the use of cyclone samplers. These separate the <4 m respirable fraction from the larger fractions of airborne dust drawn through the sampler by a pump, worn around the waist. Samples collected on Gelman 25mm GLA-5000 PVC filters were analysed gravimetrically (Health and Safety Executive 1993) to determine total mass of respirable dust, and by the direct on-filter method of X-ray diffraction (Health and Safety Executive 1994) to analyse for cristobalite. The errors in measurement of a <4 m dust mass are approximately 0 g, and many of the samples were close to the detection limit. Detection limits in the measurement of cristobalite are smaller, at approximately 10 g. Cyclone sampler tests on individuals were of two kinds. Occupational tests involved the subject wearing a sampler for the duration of their working day and included all relevant vocational activities. Activity-related tests lasted only for duration of a single activity, for example running or sweeping. A 36-hour experiment (20-30 June 1997; Table 1) confirmed that the mean concentration of <10 m airborne particles measured by the DUSTTRAK™ was only slightly higher than the concentration of <4 m (respirable) airborne particles determined by gravimetric methods. Thus, airborne particle concentration monitored by the DUSTTRAK™ was also a good indicator of the respirable fraction of airborne material. The cyclone samplers were also used intermittently for static site measurement.

Monitoring programme Monitoring of airborne particles during the eruption of Soufriere Hills Volcano from 1995 to 1999 can be summarized in seven main survey periods (Table 1). The intensive sampling strategy of the first survey, 21-28 September 1996, consisted of short-duration tests and monitoring of individual's exposure. This survey followed the sub-Plinian eruption of 17 September 1996 when, after nine hours of dome collapse, a sustained explosive eruption generated an ash plume up to 15 km a.s.l. (Robertson et al. 1998). A layer of ash 4cm thick was deposited on the capital Plymouth and ash deposits extended as far north as locality 14 (Fig. 1). Plymouth and villages to the south had been previously evacuated (April 1996), but daytime entry to Plymouth was still possible for essential work and access to the port. Of 38 successful tests in the first survey, nine were tests on individuals and 29 were static measurements at both 'quiet sites' and 'busy sites'. During the second survey period (29 September 1996-20 June 1997) intermittent samples were collected during elevated volcanic activity only. During the third survey (20-30 June 1997) a large dome collapse on 25 June 1997 generated pyroclastic flows and caused fallout as far north as locality 11 (Loughlin et al. 2002). Localities 13-16 had all been evacuated. The sampling strategy consisted mostly of short-duration environmental tests using both cyclone samplers and DUSTTRAK™ monitors. For the purposes of this survey, Montserrat was divided into three main areas (Table 2) to enable examination of exposure of populations displaced northwards: localities 14-16 (Fig. 1) in the southern area; localities 10-12 in the central area; localities 4, 6 and 8 in the northern area. Between 10 August and 13 October 1997 (the fourth survey; Table 1), short-duration tests using DUSTTRAK™ monitors were mostly unaffected by the first series of Vulcanian explosions (4-12 August 1997; Druitt et al. 2002). Twenty-seven of the 75 explosions between 22 September and 21 October, and dome collapses on 30 August, 10, 12, 15 and 17 September 1997, affected static test sites. Localities 10-12 were evacuated during the fourth survey. The fifth survey consisted of static test sites to monitor airborne particle concentrations in areas inhabited between 14 October 1997 and 22 August 1998 (localities 1-9, Fig. 1). Events that strongly affected the network were the ash-venting of 28 January to 6 February 1998 and the 3 July 1998 dome collapse. Dome growth ceased in

Table 1. Summary of major tests conducted under the Air Quality Monitoring Programme Survey periods

Monitoring equipment

Test sites occupied*

Additional monitoring

Major volcanic events

21-28 Sep 1996

Cyclone sampler, 6-1 Oh tests

10, 12, 14, 15, 16

Occupational: shop assistant, road workers, police

SP eruption (17 Sep 1996)

29 Sep 1996-20 Jun 1997

Cyclone sampler, intermittent 8 h tests

6, 10, 14, 16

DC (Dec 1996-Mar 1997)

20-30 Jun 1997

Cyclone sampler (24 h) and DT(1 x 36hand < l h )

4,6, 10, 11, 12, 14, Occupational: cleaner, waiter, gardener, office worker 15, 16

DC (25 Jun 1997)

10 Aug-13 Oct 1997

DT: short (15-30min) and long (30-120min) tests

2, 4, 7, 8, 9, 10, 11, 12

Activities: walking, driving, children's activities

VE (4-12 Aug 1997); DC (Aug-Sep 1997)

14 Oct 1997-22 Aug 1998

DT continual testing network

1 , 3 , 4 , 5,6, 7,9, 10, 12

Occupational: drivers, gardener, teacher, police, scientists, school children

VE (22 Sep-21 Oct 1997); DC (4, 6 Nov 1997, 26 Dec 1997, 3 July 1998); Ash-venting (28 Jan-6 Feb 1998); no dome growth from Mar 1998

8 Sep 1998-May 1999

DT 24 h tests during activity only

3, 10

8 May-9 Jul 1999

DT continual testing network

7, 10, 13, 14

* Test site locations are given in Figure 1. DC, dome collapse; VE, Vulcanian explosions; SP, sub-Plinian. DT, DUSTTRAK™ monitors.

DC (5 Nov 1998, 12 Nov 1998) Weather stations

DC (5 June 1999)

560

K. R. MOORE ET AL.

Table 2. Summary of DUSTTRAK™ and cyclone meter measurements of environmental concentrations of PM10 ( gm-3 ) between 20 and 30 June 1997

Little volcanic ash, wet Little volcanic ash Moderate volcanic ash High volcanic ash

Southern

Central

Northern

35-40 60-150 500+ 1000+

18-35 28-49 37-70 102-400

17-20 25-33 36-53 50-70

High volcanic ash, days on which all surfaces are covered with at least 1 mm of ash; moderate volcanic ash, days on which surfaces are covered with at least a visible layer of dust. For the purposes of this survey, Montserrat was divided into three main areas: localities 14-16 (Fig. 1) in the southern area; localities 10-12 in the central area; localities 4, 6 and 8 in the northern area. Weather conditions were dry, unless stated otherwise.

March 1998 before recommencing in November 1999 (Watts et al. 2002). During the sixth survey period (8 September 1998-May 1999) continuous testing was carried out intermittently during elevated volcanic activity only, most notably during the 5 November 1998 dome collapse. A network of four DUSTTRAK™ monitors (Table 1) was established during the seventh survey (8 May-9 July 1999) to record airborne particle concentrations in inhabited northern areas (localities 1-13, Fig. 1) and at localities 14 and 15. to which the public had access during the daytime. A comprehensive study of the effect of weather conditions on airborne ash was also carried out. Residual volcanic activity increased during this period, culminating in the 5 June 1999 dome collapse.

Results Static site measurements Cyclone sampler and DUSTTRAK™ results consistently showed that airborne particle concentrations were at least double, and up to 20 times greater, at southern localities (14-16; Fig. 1) than at northern localities (1-8; Fig. 1). since the former were closer to ash deposits and the ash-dispersal axis due to low-level (0-5 km) winds. Ambient PM10 levels of 0 g m - 3 (24-hour averages) were recorded at localities 1-12 during periods of negligible volcanic activity (Fig. 2). These were less than the 50 g m - 3 safety limit. Dome collapses, Vulcanian explosions and major ash-venting all produced airborne particle concentrations that reached alert level (>300 g m - 3 ) of the advisory banding system. For example, a 1-hour average PM10 level of 2110 g m - 3 was recorded during major ash-venting on 4 February 1998 (locality 9; Fig. 1). The maximum recorded 24-hour average PM10 level of approximately 600 g m - 3 (locality 12; Fig. 1) resulted from tephra fallout after the 5 June 1999 dome collapse. Although most airborne particles on Montserrat had a volcanic origin, a small proportion were from erosion of soil and from sea salt (Twomey 1977). High airborne particle concentrations were recorded near beaches on days when the sea was rough or strong winds were blowing. As Montserrat is a small island with few roads, very little of the airborne particles were attributable to combustion. An additional source of airborne particles was Saharan dust (Fig. 2) that was occasionally transported to the Caribbean on equatorial winds. During these periods. PM10 levels exceeding 100 g m - 3 were recorded across all parts of the island (in the

Fig. 2. Summary of data from all static testing sites and frequencies of volcanic events throughout the monitoring period (September 1996-July 1999) as: (a) an average of short-period tests and (b) a seven-day running average of all sites. This smoothes small fluctuations, such as those caused by localized rain showers. All results are measured concentration of particles <10 m in diameter using the DUSTTRAK™ apparatus, except those from the 21-28 September 1996 survey when only cyclone meters were used giving measured concentration of particles <4 m in diameter (labelled PM4). Hazard monitoring alert levels are shown for reference. The asterisks denote significant quantities of Saharan dust. (c) Volcanic activity is represented as the number of days in each month for which ash-producing phenomena were reported to the Montserrat Volcano Observatory and excludes small pyroclastic flows and rockfalls that failed to produce significant ash plumes.

AIRBORNE PARTICULATE MATTER

absence of volcanic activity), and a haze was observed between Montserrat and neighbouring islands. Cristobalite PM4 levels were lower than l0 gm-3 (21 September 1996-30 June 1997), except in the capital Plymouth (Fig. 1), following the 17 September 1996 sub-Plinian explosion and the 25 June 1997 dome collapse. Airborne concentrations of other silica polymorphs have been below the limit of detection for X-ray diffraction of on-filter samples. Effects of eruption intensity and ash-production mechanism. Islandwide static-site test results for the period 21 September 1996-9 July 1999 are summarized in Figure 2. Averages of short-duration tests (Fig. 2a) have a high airborne particle concentration bias as monitoring equipment was employed most extensively during high levels of volcanic activity and it is difficult to draw comparisons with 24-hour averages (Fig. 2b). Elevated airborne particle concentrations were recorded after six dome collapses (one followed by a sub-Plinian explosion on 17 September 1996), one major ashventing episode (3-5 February 1998) and the Vulcanian explosions of 22 September-21 October 1997. Figure 2 gives no indication of the relative magnitude of eruptive events but, with the exception of the ash-venting on 3-5 February 1998, high levels of PM10 were caused only by volcanic activity that generated extensive pyroclastic flows and ash plumes. The magnitude of an eruptive event, as represented by DRE dome-collapse volume (Calder et al. 2002) for 25 June 1997 (4.9 x 10 6 m 3 ), 3 July 1998 (15-19 x 10 6 m 3 ) and 5 November 1998 (0.8 x 10 6 m 3 ), does not correlate well with the average island-wide concentration of airborne particles. The highest local airborne particle concentrations were detected during the 5 June 1999 dome collapse: the maximum 1-minute average PM10 level was 128mgm - 3 and 24-hour average PM10 levels were between 184 and 602 g m - 3 (localities 10, 13 and 14; Fig. 1). The 25 June 1997 dome collapse generated similar volumes of <10 m ash, with airborne particle concentrations decreasing rapidly with distance from the volcano (from south to north; Table 2). For the 3 July 1998 and 5 November 1998 dome collapses at locality 10, 24-hour average PM10 levels were approximately 220 g m - 3 . The island-wide average of 59 short-duration tests (24-hour data unavailable; Table 1) between 22 September and 11 October 1997, during the Vulcanian explosions, recorded a PM10 value of 200 g m - 3 . The 24-hour average PM10 levels during ash-venting on 4 February 1998 were 105 and 190 g m - 3 (localities 8 and 9). Tephra fallout deposits created by three eruption mechanisms are illustrated in Figure 1. (1) Tephra fallout thickness locally in excess of 10cm accumulated to the west of the volcano, mainly from dome collapses, between 3 and 25 June 1997 (Bonadonna et al. 2002). Maximum local tephra thickness after any individual dome collapse was > 50 cm (17 September 1996) near locality 16 but included the ash produced by the subsequent sub-Plinian explosion. (2) Vulcanian explosions (August-October 1997) deposited more than 20cm of airfall tephra (Bonadonna et al. 2002), also to the west of the volcano. The thickness of ash deposits from dome collapses and Vulcanian explosions decreased rapidly to the north, with distance from the volcano. Tephra fallout after vigorous ashventing is represented as mass loading because smaller volumes of ash and deposits of highly variable thickness were produced. (3) Up to 5 k g m - 2 of tephra from ash-venting (1-25 February 1998) was equivalent to < l c m thickness. With reference to ash-deposit thickness, dome collapses and Vulcanian explosions were eruptive events of high intensity and major ash-venting was a much less significant ash-production mechanism. Dome collapses and vertically directed explosions produced similar orders of ash-deposit thickness (10-20 cm maximum) but much higher PM10 levels were recorded after some dome collapses (e.g. 600 g m - 3 on 5 June 1999) than after the Vulcanian explosions. Lower energy ash-venting produced much thinner ash deposits (< 1 cm) but a similar order concentration (c. 200 g m - 3 ) of airborne particles < 10 m in diameter to the Vulcanian explosions and some dome collapses (3 July 1998 and 5 November 1998).

561

Effect of ash plume height and wind profile. Volcanic activity produced plumes up to 15km a.s.l. Wind directions varied from generally easterly below 5 km a.s.l. to generally westerly or southerly between 8 and 18km, resulting in wind shear between 5 and 8km a.s.l. Therefore, many of the volcanic events that produced ash plumes between September 1996 and July 1999 did not elevate airborne particle concentrations in the monitored northern parts of the island. For example, the largest dome collapse of the eruption on 26 December 1997 (DRE collapse volume = 46 x 10 6 m 3 ) failed to elevate PM10 levels at any test sites (Fig. 2), since both the collapse and plume transport were directed towards the south (Sparks et al. 2002). The fact that only six of all dome collapses affected the monitoring network was a consequence of the directed nature of dome collapses and variable wind direction, which generally acted to distribute tephra in the unmonitored south of the island. Fallout following the 3 July 1998 collapse on the east flank of the dome produced a maximum local 24-hour PM10 level of 314 gm-3 (locality 10, 4 July 1998). The plume was detected by satellite moving ENE at 10-15 km a.s.l., but it deposited tephra in the inhabited NW due to low-level winds (0-5 km). In contrast, the collapse of lower-temperature andesite from the east flank of the dome on 5 June 1999 formed a low-level cloud only and produced a maximum local 24-hour PM10 level of 602 g m - 3 . As tephra was not carried away by intermediate-level winds (818km), much greater volumes of <10 m ash were transported in the direction of southeasterly prevailing low-level (0-5 km) wind than for other dome collapses. During the first cyclic Vulcanian explosions (4-12 August 1997), easterly low-level and westerly intermediate-level wind during tests (10-29 August 1997; Fig. 3) ensured that localities 1-12 were largely unaffected by airborne ash <10 m in diameter. Ash particles generated by the second series of Vulcanian explosions (22 September-21 October 1997) were transported to localities 1-12 (30 August-11 October 1997; Fig. 3) only by intermediate-level winds (8-18 km). Therefore, both low (<5km a.s.l.) and intermediate (8-18 km a.s.l.) level winds were able to disperse airborne particles to populated areas from plumes caused by major dome collapses and explosions.

Fig. 3. Plot comparing PM10 level ( g m - 3 ) between periods when volcanic ash was transported away from and towards monitored parts of the island by wind. Data are averages of between two and 24 tests, the mean duration of which was usually from 15 to 30 minutes. Localities are arranged from north to south. All localities were sites with little human activity between 10 and 29 August 1997 with the exceptions of localities l0a and l0b. Localities 10 and 12 were evacuated between August and September 1997, with populations displaced to the north (localities 8 and 9) as indicated by the arrow.

562

K. R. MOORE ET AL.

Fig. 4. Airborne particle concentrations (24-hour average of PM10) at static monitoring sites in northern Montserrat (localities 7-9; Fig. 1) from 27 January to 11 February 1998. All sites had little human activity except locality 9a, a busy site near a main road with a high level of human activity. All sites received ashfall from 28 January onwards, but the notable increase in airborne particle concentrations only occurred after the heavy, dry ashfalls of 3-5 February. Resuspension of the same ash by wind on 7 and 8 February caused higher airborne particle concentrations than initial ashfall.

Vigorous ash-venting that occurred during enhanced rockfall activity between 28 January and 6 February 1998 resulted in weak bent-over plumes (2-3 km a.s.l.), and affected inhabited northern Montserrat between 3 and 5 February because of unusual southerly and southeasterly winds (Fig. 4). Therefore, low-level winds only (<5km a.s.l.) are responsible for transporting fine ash from weak volcanic phenomena with small plumes.

Effects of resuspension by the wind. Resuspension of tephra by wind on 7-8 February 1998, after ash-venting on 4 February, generated 24-hour average PM10 levels of between 93 and 265 g m - 3 at localities 7-9 (7 February; Fig. 4). Local PM10 levels from the resuspension were approximately two times greater than during fallout from ash-venting (50-191 g m - 3 on 4 February; Fig. 4). Resuspended ash persisted as a thick, dusty haze over the few metres closest to the ground and clouds of visible dust blew off ashcovered surfaces. Resuspension by the wind generated raised (51-100 g m - 3 ) to very high (101-300 g m - 3 ) levels for several days following fallout of tephra from all three eruptive mechanisms (Fig. 5). Concentrations of resuspended airborne particles <10 m on 7 February 1998 (following ash-venting on 4 February) and 10 June 1999 (following the 5 June dome collapse) were of similar magnitude, with PM10 levels of approximately 250 g m - 3 . However, the PM10 level during tephra fallout directly over the monitoring network was considerably larger for the 5 June 1999 dome collapse than during tephra fallout directly over the monitoring network for the ashventing on 4 February 1998. Had it not been for extremely heavy rainfall (8 and 11 June 1999), resuspension of volcanic ash following the 5 June 1999 dome collapse could have been considerably higher. During negligible volcanic activity, resuspension and transport of ash from deposits (March 1998; Fig. 1) elevated PM10 levels in populated areas (localities 1-12), though rarely to more than l00 g m - 3 , if low-level (0-5 km) winds were southerly or southeasterly. Resuspension of airborne particles generated progressively smaller PM10 levels with time after an event: Average PM10 level at locality 3 between 6 and 8 March 1998 was 64 g m - 3 ; subsequently, peaks of 45, 31 and 21 gm-3 were recorded on 13, 21 and 30 April 1998 respectively (Fig. 6). The relationship between wind speed and airborne particle concentration was investigated over the period 15 May-9 July 1999 (Fig. 7). PM10 was elevated to very

Fig. 5. Decrease in the concentration of airborne particles (24-hour averages of PM10 at all sites) in the atmosphere following volcanic phenomena. Day 1 is the first elevated airborne particle concentration recorded in each case. The events are: residual ash from Vulcanian explosions on 14 October 1997 (this was the first day of 24-hour monitoring (Table 1), which post-dates the explosions that had a strong effect on testing sites); ash-venting on 4 February 1998: and dome collapses on 3 July 1998. and 5 June 1999.

high (101-300 g m - 3 ) levels at all static test sites (Table 1) between 26 May and 1 June 1999, and 5 June and 13 June 1999. due to ashventing and dome collapses respectively. With little or no volcanic activity, resuspension of ash was enhanced by increased wind speed, which varied from 2 to 12.5ms - 1 . correlating positively with average (across all test sites) 24-hour PM10 levels of 13 to 81 g m - 3 . During volcanic activity in dry conditions, wind was also required to transport airborne particles away from the island: following an explosion and ash-venting between 22 and 27 May 1999, wind speeds dropped to l.5-3ms - 1 (27-31 May 1999; Fig. 7) and dust levels exceeded 100 g m - 3 .

Fig. 6. PM10 levels ( g m-3) presented as 24-hour averages from 9 April to 3 May 1998. Locality 3 is a site of high human activity on a major road with dust readily resuspended by wind. The last heavy ash fall occurred on 5 February; there were some light ashfalls in February and March and airborne particle concentrations then gradually decreased.

AIRBORNE PARTICULATE MATTER

Fig. 7. Airborne particle concentrations (24-hour averages of PM10) and weather data between 15 May and 9 July 1999. Weather data (right-hand axis) were downloaded from weather station 1 in eastern Montserrat until it was struck by lightning on 9 June 1999. Weather data from Montserrat heliport (weather station 2) were used from then onwards. The graph shows that sites have recorded similar results except on 5 June 1999, during a large dome collapse, and on 17 June when ash fell on locality 14 only. Saharan dust is again identified in the atmosphere (asterisks).

Effect of rain and air humidity. Figure 7 shows that rain and airborne particle concentration had a strong inverse relationship, with rainfall peaks and PM10 level depressions on 22 June, 26 June and 4 July 1999. The short residence time of ash with particle size < 10 m in the atmosphere (generally less than three days; Figs 4-6) was the result of frequent rain showers. In the absence of volcanic activity and Saharan dust, as little as 5 mm of rain could reduce PM10 levels to less than 25 g m - 3 (Fig. 7) very rapidly. A rain shower at 22:50 local time (LT) on 23 March 1998, when there was negligible volcanic activity (Fig. 8), depressed PM10 levels by 40 g m - 3 in 40 minutes following a more than two-fold increase in PM10 from 18:00 to 22:30 LT, related to decreasing humidity. The frequency and duration of rain showers dictated the depth to which water percolated through a tephra deposit, hindering subsequent resuspension by increasing humidity. Additional effects of heavy rains were that the uppermost layers of accumulated ash deposits were compacted and mudflows were generated, forming a crust of consolidated ash that was more resistant to resuspension by wind than loose ash. Effect of topography on static data. Localized fallout elevated PM10 levels at specific localities regardless of topography, e.g. there was a very high (101-300 g m - 3 ) PM10 level at locality 14 only on 17 June 1999 (Fig. 7). However, some local variations in the 24-hour PM10 data between 14 October 1997 and 9 July 1999 (Table 1) could not be explained simply by volcanic activity, pre-

Fig. 8. DUSTTRAK™ profiles at three localities on 23 March 1998 (PM10 levels in g m - 3 ) . Results are averages of 60 1-second measurements. Broad trends affecting all sites typify gradually changing weather conditions.

563

vailing wind direction and humidity. Data from localities 10 and 14 are illustrated in Figure 7 as representative of parts of the island that were affected by fallout from the 5 June 1999 dome collapse. PM10 levels up to 20 gm-3 greater than those shown for locality 14 (Fig. 7) were recorded between 18 June and 4 July 1999 at locality 13, despite a greater distance from the volcano (Fig. 1). Locality 13 had close proximity to the Belham River valley, which contained a large reservoir of unconsolidated ash, from pyroclastic flows generated by the 5-17 June 1999 dome collapses, that could be resuspended. Locality 7 generally had 10-30 g m - 3 lower PM10 levels than all other sites (10 August 1997-9 July 1999). This site was usually unaffected by resuspension of ash deposits because the range of hills in the centre of Montserrat acted as a barrier to lowlevel winds from the south. In contrast, locality 5 (Fig. 1) had PM10 levels up to 25 g m - 3 higher than other localities at a similar distance from the volcano because the apparatus was situated on the windward side of a hill. Thus, topographic depressions partly controlled the distribution of ash deposits and the interaction of topography with low-level winds could act either to reduce or increase local airborne ash particle concentrations. Effects of human activity and shelter. Background concentrations of airborne particles were higher in populated areas because human activity aided resuspension of ash deposits. Figure 3 shows that the highest PM10 levels were recorded at sites near to busy roads or with a lot of pedestrian activity, which shifted northwards (arrow; Fig. 3) with the evacuation of people from localities 10-12 to localities 8 and 9 (August-September 1997). Irrespective of volcanic activity, airborne particle concentrations were elevated (140-180 g m - 3 ) where there was a supply of accumulated ash that could be reworked (localities 8, 9, l0a, 10b; Fig. 3). PM10 levels recorded indoors and outdoors simultaneously show that variations related to volcanic activity and climate were observed 1.5-2 minutes earlier outside (a in Fig. 9) and PM10 levels were up to 50 g m - 3 lower inside (b in Fig. 9). Peaks of short duration (1-2 minutes) could be observed out of doors due to resuspension of ash by wind or cars, which did not register indoors (d in Fig. 9). Likewise, sweeping and cleaning caused indoor peaks only (c in Fig. 9). PM10 levels of >2000 g m - 3 represent the highest transient concentrations of airborne particles recorded indoors, and were caused by cleaning during fallout on 24 September 1997 (locality 11). Elevated airborne particle concentrations due to gusts of wind generally had higher frequency but much lower magnitude than those due to indoor activities.

Measurements on individuals Measurements showed that occupation and human activity influenced exposure of individuals to airborne particles on Montserrat.

Fig. 9. DUSTTRAK™ profiles (PM10 levels in g m - 3 ) for both inside (dashed line) and outside (solid line) a building. Gas and ash emissions from the volcano were observed throughout the day (15 August 1997).

K. R. MOORE ET AL.

564

Table 3. Summary of personal exposure to PM10 ( g m - 3 ) associated with various activities

Background level Indoors activities Outdoor locomotion Dusty occupations

High ash

Moderate ash

No ash

1000 <4500 2000 5000

300 <600 600 1000

30 50 50 100

Background exposure levels are ambient airborne particle concentrations, which depend on the amount of unconsolidated ash lying on the ground. High and moderate ash are defined in Table 2.

Results of tests using DUSTTRAK™ monitors attached to individuals are summarized in Table 3. Indoor activities included sweeping and playing, and outdoor locomotion summarizes the activities of walking and driving. People in dusty occupations (Table 3) who were exposed to the highest PM10 levels were roadworkers, gardeners and cleaners (Searl el al in press). Most of the population lived and worked in areas where ash deposits were not normally present (localities 1-9 after September 1997) and background PM10 levels were approximately 30 g m - 3 . An individual's exposure to airborne particles approached the occupational limit of 5000 g m-3 for respirable dust only when there was a lot of ash present (background PM10 levels of 1000 gm - 3 ; Table 3). The occupational limit of 50 g m - 3 for cristobalite <4 m in diameter was exceeded on only one occasion, by a grader driver in Plymouth after the 25 June 1997 dome collapse (he wore a respirator during his working day). Climate had a significant effect on the activity-related exposure of an individual to airborne particles. Activity-related indi-

Fig. 10. Effect of climatic conditions on dust levels produced by different human activities with negligible volcanic activity. Airborne particles are measured as PM10 ( g m - 3 ). Weather conditions were frequent showers with very damp ground (3 April 1998), and warm, dry and windy (6 April 1998).

vidual exposures of 130-500 g m - 3 (6 April 1998; Fig. 10) were recorded on dry, breezy days (background PM10 level = 16 g m - 3 ) and 70-250 g m - 3 (3 April 1998; Fig. 10) in very damp, windless conditions (background PM10 level = 12 g m - 3 ), when there was no volcanic activity. Humidity effectively hindered resuspension of ash by any kind of human activity. High-energy activities indoors (activity 3; Fig. 10) elevated individual exposure more than highenergy activities outdoors (activity 2; Fig. 10). regardless of humidity. Airborne particles were dispersed outdoors from the site of resuspension by wind but were trapped indoors, so that an individual's exposure on windy days in the absence of volcanic activity could be greatest indoors.

Discussion Airborne particle concentrations (of PM4 and PM10) measured between September 1996 and July 1999 showed more than a ten-fold variation, with airborne ash from the eruption of Soufriere Hills Volcano transported to nearby populated areas by low (<5km a.s.l.) or intermediate (8-18 km a.s.l.) level winds, depending on plume height. Plume height is controlled by eruption mechanism, ash mass flux at source, total mass of material erupted and thermal energy (Woods 1998). Tephra dispersal is dominated by plume dynamics proximal to the volcano and distally by atmospheric dynamics (Bursik 1998). Large dome collapses generated a range of plume heights and the highest PM10 levels occurred during fallout when low-temperature tephra was transported solely in low-level winds (5 June 1999). The energy of release of comparatively hightemperature material during vertically directed explosions or collapse of a growing dome generated large volumes of tephra and high plumes, but the fallout had lower PM10 levels. Splitting of high plumes at 5-8 km a.s.l. increased the probability of plumes being transported to populated areas (localities 1-9; Fig. 1) but lowered the concentration of airborne particles for two reasons. (1) Splitting of the plume reduced the volume of tephra transported in any one direction. (2) Tephra was normally transported to the populated north of the island in that part of the plume affected by intermediate-level wind (8-18 km a.s.l.) and. as plumes erupted into a region of higher winds will be distorted into a more elongate shape (Bursik 1998), more fallout occurred off-island. Ash-venting produced less total tephra but raised PM10 levels when all erupted material was transported in a weak plume at very low altitudes (2-3 km a.s.l.) directly over monitoring sites. Ash was observed falling in aggregates with diameter of approximately 100 m (Bonadonna et al. 2002). Aggregation in the plume increased grain size (Bacon & Sarma 1991; Lane et al. 1993), particularly during explosive volcanic eruptions that generated fine particles with an electrostatic charge (Gilbert et al. 1991; Bonadonna et al. 2002), and aided sedimentation of tephra. The disintegration of aggregates on deposition (Sorem 1982; Bonadonna et al. 2002) released individual small particles and increased PM10 levels during resuspension of ash by wind. Levels of PM10 during resuspension, as a function of PM10 levels during fallout, were not noticeably higher for explosions than for dome collapses or ashventing. This could be because aggregation was no more extensive during the explosions, but it is more likely to be due to the small sample population available for different ash-producing mechanisms and prevailing weather conditions. The efficiency of resuspension of ash by wind was affected by humidity, as condensed water on particle surfaces enhanced adhesion of particles (Gilbert et al. 1991). Prediction of areas most exposed to airborne ash particles < 10 m in diameter on Montserrat due to either fallout or resuspension of ash by wind may be possible using comparisons of static-site 24-hour data with locality 7 sheltered by Centre Hills, as a reference. Between February and May 1998, with negligible volcanic activity and low-level winds towards the west or NNE. localities 9-12 (central Montserrat) generally had PM10 levels 2.5-3 times higher than at locality 7 (Fig. 11). The difference between locality 7 and central Montserrat was not as great when volcanic activity was

AIRBORNE PARTICULATE MATTER

565

toxic than from Vulcanian explosions (Baxter et al. 1999), where the abundance of fresh glass is greater. Given that it was most commonly dome collapses that affected the static-site monitoring network, and that cristobalite was concentrated into fine ash, the low frequency of tests where cristobalite <4 m in diameter exceeded the environmental threshold of 10 g m - 3 was unexpected. It is possible that most cristobalite was not in the fraction of ash smaller than 4 m in diameter and was not collected by cyclone samplers. However, effective removal of ash by rain and strong lowlevel winds would have ensured that the residence time of cristobalite, if it were in respirable ash in the atmosphere, was short.

Conclusions The air quality monitoring programme examined the origins of airborne particles in the atmosphere above Montserrat, dispersal of <10 m volcanic ash, consequences of aggregate disintegration and the relative PM10 levels produced by fallout and resuspension of ash by wind, and concluded the following. (1)

(2)

(3) Fig. 11. Relative concentrations of airborne particles at static monitoring sites. Consistently low PM10 levels were measured at locality 7 (Fig. 1), sheltered from ash-carrying winds by the Centre Hills, and were arbitrarily assigned a value of 1. Excess PM10 at other sites was then interpreted as the component of dust that was volcanic ash. Data used were gathered during low volcanic activity (dots) between 10 February and 30 April 1998 (localities 3-12) and elevated volcanic activity (squares) between 15 May and 8 July 1999 (localities 7, 10, 13 and 14).

(4) (5)

elevated (May-June 1999) and wind transported airborne particles over monitoring sites, with PM10 levels at localities 13 and 14 less than 1.5 times greater than at locality 7 (Fig. 11). The construction of airborne-particle risk maps must therefore incorporate different volcanic phenomena, extent of ash deposits, wind direction (lowand intermediate-level) and humidity. Irrespective of volcanic activity, background PM10 levels were highest during the day, particularly in warm, dry and windy conditions. Turbulent motions caused by friction of wind at the ground surface mix air rapidly and are exacerbated by the buoyancy of warm daytime air (Barry & Chorley 1998), while cooling of the surface at night forms an adjacent cold layer of air that inhibits turbulence. Moreover, the human population is most active during the day (Searl et al. in press) and certain activities increased individual exposure further, particularly high-energy activities on dry days, with a lack of adequate ventilation. The time spent in different activities varies for different sections of society (Searl et al. in press). For example, children will play a lot more than adults and may, therefore, have higher individual exposure levels. The potential health hazard of ash from Sakurajima Japan (1985-1986; Yano et al 1990) and Mount St Helens USA (18 May 1980; Dollberg et al. 1986) was considerably lower than that on Montserrat because the ash contained negligible crystalline silica. The cristobalite-rich tephra of dome collapses from Soufriere Hills Volcano, and fractionation of cristobalite into fine ash by selective crushing within pyroclastic flows (Horwell et al. 2001), implies that airborne paniculate matter from dome collapses is likely to be more

(6)

Dust with particle diameters <10 m contained material from volcanic ash, Saharan dust, soil erosion, sea salt and vehicle exhausts. The last three were always present in small quantities but only Saharan dust affected all monitoring sites equally. Volcanic ash caused the largest fluctuations in airborne particle concentration but, generally, PM10 levels in inhabited areas of Montserrat remained low, even during elevated volcanic activity, unless tephra was transported by southerly or southeasterly winds. With one exception, raised PM10 levels were caused by volcanic activity that generated extensive pyroclastic flows and eruption plumes, i.e. large dome collapses and vertically directed explosions. Plumes <5 km high from weak ash venting or lower-temperature dome collapses were transported by, and concentrated airborne particles in, low-level winds. High plumes (up to 15km a.s.l.) from vigorous, higher-temperature eruptions were split at 5-8 km a.s.l. and had greater probability of affecting the population in the north of the island. Resuspension of ash particles by wind was capable of generating PM10 levels considerably higher than fallout due to disintegration of ash particle aggregates. Residence times of airborne particles <10 m in the atmosphere, following either fallout or resuspension of ash by wind, was normally less than three days due to the frequent rain showers over Montserrat. Rain and mudflows caused compaction of the top layer of ash, which formed a hard crust that hindered resuspension by wind. Heavy rains transported ash in runoff and strong winds removed ash from over the island to the sea. Background airborne particle concentrations varied with location. PM10 levels, originating from both tephra fallout and resuspended ash deposits, decreased with distance from the ash source (to the north), with local perturbations occurring due to the interaction of low-level winds with topography. The highest aerosol concentrations were consistently recorded in localities with most human activity. It is possible for ash hazard maps to be produced, but they must include the type of volcanic activity, low-level (0-5 km a.s.l.) and intermediatelevel (8-18 km a.s.l.) wind directions, air humidity, and extent of ash deposits.

The authors would like to thank the staff of the Montserrat Volcano Observatory and all the residents who allowed their houses to be used as testing sites. C. Bonadonna, S. Dornan, C. Horwell, G. Ryan and C. Walker contributed to the data used in this paper. S. Loughlin and L. Ritchie are thanked for stepping in when there were no Air Quality Monitoring Officers on the island. The Department for International Development funded monitoring work on Montserrat. The authors would also like to thank P. Baxter, C. Bonadonna, T. Druitt, C. Horwell and J. Gilbert for their valuable and constructive reviews.

566

K. R. MOORE ET AL.

References ALLEN. A. G.. BAXTER. P. J. & OTTLEY. C. J. 2000. Gas and particle emissions from Soufriere Hills Volcano. Montserrat, West Indies: characterization and health hazard assessment. Bulletin of Volcanology, 62. 8-19. AMERICAN CONFERENCE OF GOVERNMENTAL INDUSTRIAL HYGIENISTS (ACGIH) 1996. TLVs ami BEIs (1996): Threshold Limit Values for Chemical Substances, Physical Agents and Biological Exposure Indices. World Wide Web Address: http: www.acgih.org.html. ASHTON. I. & GILL, F. S. 2000. Monitoring for Health Hazards at Work (third edition). Blackwell. Oxford. BACON. D. P. & SARMA. R. A. 1991. Agglomeration of dust in convective clouds initialised by nuclear bursts. Atmospheric Environment. 25A. 2627-2642. BARRY. R. G. & CHORLEY, R. J. 1998. Atmosphere, Weather and Climate (sixth edition). Routledge, London. BAXTER. P. J.. BONADONNA. C.. DUPREE. R. ET AL. 1999. Cristobalite in volcanic ash of the Soufriere Hills Volcano, Montserrat. British West Indies. Science. 283. 1142-1145. BONADONNA. C.. MAYBERRY. G. C.. CALDER. E. S. ET AL. 2002. Tephra fallout in the eruption of Soufriere Hills Volcano, Montserrat. In: DRLTTT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoir. BURSIK. M. 1998. Tephra dispersal. In: GILBERT, J. S. & SPARKS, R. S. J. (eds) The Physics of Explosive Volcanic Eruptions. Geological Society, London, Special Publications, 145, 115-144. CALDER, E. S., LUCKETT. R., SPARKS, R. S. J. & VOIGHT, B. 2002. Mechanisms of lava dome instability and generation of rockfalls and pyroclastic flows at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London. Memoirs. 21, 173-190 CAREY, S. & SIGURDSSON. H. 1982. Influence of particle aggregation on the deposition of distal tephra from the May 18 1980 eruption of Mount St. Helens. Journal of Geophysical Research . 87, 7061-7072. CLOUDHURY. A. H.. GORDIAN. M. E. & MORRIS. S. S. 1997. Associations between respiratory illness and PM10 air pollution. Archives of Environmental Health. 53, 113-117. COLE, P. D., CALDER. E. S.. SPARKS. R. S. J. ET AL. 2002. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 231-262. COMMITTEE ON THE MEDICAL EFFECTS OF AIR POLLUTANTS (COMEAP) 1998. Quantification of the Effects of Air Pollution on Health in the United Kingdom. Department of Health, HMSO, London. DEVINE. J. D., MURPHY, M. D., RUTHERFORD, M. J. ET AL. 1998. Petrologic evidence of pre-eruptive pressure-temperature conditions, and recent reheating, of andesitic magma erupting at the Soufriere Hills Volcano. Montserrat, W. I. Geophysical Research Letters, 25(19), 3669-3672. DOLLBERG. D. D.. BOLYARD. M. L. & SMITH, D. L. 1986. Evaluation of physical health effects due to volcanic hazards: crystalline silica in Mount St Helens volcanic ash. American Journal of Public Health. 76. 53-58. DRUITT. T. H.. YOUNG, S. R., BAPTIE, B. ET AL. 2002. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano. Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21, 281-306. EXPERT PANEL ON AIR QUALITY STANDARDS (EPAQS) 1995. Particles. Department of Environment. HMSO. London. GILBERT,J. S.. LANE. S. J., SPARKS. R. S. J. & KOYAGUCHI, T. 1991. Charge measurement on particle fallout from a volcanic plume. Nature, 349. 598-600. GORDIAN, M. E., OZKAYNAK, H., XUE, J., MORRIS, S. S. & SPENGLER. J. D. 1996. Particulate air pollution and respiratory disease in Anchorage. Alaska. Environmental Health Perspectives, 104(3), 290-297. HEALTH AND SAFETY EXECUTIVE 1993. General methods for the gravimetric determination of respirahle and total inhalable dust. Methods for the Determination of Hazardous Substances, 14. HEALTH AND SAFETY EXECUTIVE 1994. Cristobalite in respirahle airborne dusts. Methods for the Determination of Hazardous Substances, 76. HORWELL. C. J.. BRANA, L. P., SPARKS, R. S. J., MURPHY, M. D. & HARDS, V. L. (2001). A geochemical investigation of fragmentation and physical fractionation in pyroclastic flows from the Soufriere Hills

Volcano. Montserrat. Journal of Volcanology and Geothermal Research. 109. 247-262. INTERNATIONAL AGENCY FOR RESEARCH ON CANCER (IARC) 1997. Silica, some silicates, coal dust and para-aramid fibrils. Monographs on the Evalulation of Carcinogenic Risks to Humans. 66. INTERNATIONAL ORGANISATION FOR STANDARDISATION (ISO) 1991. Air Quality Particle Size Fraction Definitions for Health-Related Sampling. Approved for publication as CD 7708. ISO. Geneva. KOKELAAR. B. P. 2002. Setting, chronology and consequences of the eruption of Soufriere Hills Volcano. Montserrat (1995-1999). In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano. Montserrat. from 1995 to 1999. Geological Society. London. Memoirs. 21. 1-43. LANE. S. J.. GILBERT. J. S. & HILTON. M. 1993. The aerodynamic behaviour of volcanic aggregates. Bulletin of Volcanology. 55. 481-488 LOUGHLIN. S. C., CALDER. E. S.. CLARKE. A. B. ET AL. 2002. Pyroclastic flows and surges generated by the 25 June 1997 dome collapse. Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano. Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21. 191-209. MARTIN. T. R.. WEHNER. A. P. & BUTLER. J. B. 1986. Evaluation of physical health effects due to volcanic hazards: the use of experimental systems to estimate the pulmonary toxicity of volcanic ash. American Journal of Public Health. 76. 59-65. NICHOLL. A. 1996. Report on the sampling of airborne volcanic dust on the island of Montserrat. West Indies, 21st-28th September 1996. Report for the Department for International Development. Institute of Occupational Medicine Report No. OH 3070. NORTON. G. E.. WATTS. R. B.. VOIGHT. B. ET AL. 2002. Pyroclastic flow and explosive activity at Soufriere Hills Volcano. Montserrat. during a period of virtually no magma extrusion (March 1998 to November 1999). ///: DRUITT. T. H. & K O K E L A A R . B. P. (eds) The Eruption of Soufriere Hills Volcano. Montserrat. from 1995 to 1999. Geological Society. London. Memoirs. 21. 467-481. ROBERTSON. R.. COLE. P.. SPARKS. R. S. J. ET AL. 1998. The explosive eruption of Soufriere Hills Volcano. Montserrat. West Indies. 17 September 1996. Geophysical Research Letters. 25(18). 3429-3432. Ross. M.. NOLAN. R. P.. LANGER. A. M. & COOPER. W. C. 1993. Health effects of mineral dusts other than asbestos. In: GUTHRIE. G. D. & MOSSMAN. B. T. (eds) Health Effects of Mineral Dusts. Mineralogical Society of America. Reviews in Mineralogy. 361-365. SEARL. A. & NICHOLL. A. 1997. Assessment of the exposure of the population of Montserrat to volcanic ash containing cristobalite. Report for the Department for International Development. Institute of Occupational Medicine Report No. P752. SEARL. A.. NICHOLL. A. & BAXTER. P. J. (in press). Assessment of the exposure of islanders to ash from health risks associated with exposure to volcanic ash from the Soufriere Hills Volcano. Montserrat British West Indies. Occupational and Environmental Medicine. SOREM, R. K. 1982. Volcanic ash clusters: tephra rafts and scavengers. Journal of Volcanohgy and Geothermal Research 13: 63-71. SPARKS. R. S. J.. BARCLAY. J.. CALDER. E. S. ET AL. 2002. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano. Montserrat, from 1995 to 1999. Geological Society. London. Memoir. TWOMEY. S. 1977. Atmospheric Aerosols. Elsevier. Oxford. WATTS. R. B.. HERD. R. A.. SPARKS. R. S. J. & YOUNG. S. R. 2002. Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano. Montserrat. from 1995 to 1999. Geological Society. London. Memoirs. 21. 115-152. WILSON. M. R.. STONE. V.. CULLEN. R. T.. SEARL. A.. MAYNARD. R. L. & DONALDSON. K. 2000. In vitro toxicology of respirable Montserrat volcanic ash. Occupational and Environmental Medicine. 57. 727-733. WOODS. A. W. 1998. Observations and models of volcanic eruption columns. In: GILBERT. J. S. & SPARKS. R. S. J. (eds) The Physics of Explosive Volcanic Eruptions. Geological Society. London. Special Publications. 145.91-114. YANO. E., YOKOYAMA. Y.. HIGASHI. H.. NISHII. S.. MAEDA. K. & KOIZUMI. A. 1990. Health effects of volcanic ash: a repeat study. Archives of Environmental Health. 45. 367-373.

Seismicity, gas emission and deformation from 18 July to 25 September 1995 during the initial phreatic phase of the eruption of Soufriere Hills Volcano, Montserrat C. A. GARDNER 1 & R. A. WHITE 2 1

USGS, Cascades Volcano Observatory, 5400 MacArthur Blvd, Vancouver, WA 98661, USA 2 USGS, MS 910, 345 Middlefield Rd, Menlo Park, CA 94025, USA

Abstract: On 18 July 1995, after more than three years of irregularly increasing seismicity, phreatic explosions opened a new vent on Soufriere Hills Volcano, about 4km east of the capital city of Plymouth, Montserrat. By early August 1995, the volcano was monitored by a nine-station seismic network, three telemetered electronic tiltmeters, and daily correlation spectroscopy (COSPEC) flights to measure SO2 emission rates and to observe vent areas. Seismicity and SO2 emission rates implicated magma intrusion as the cause of the seismic unrest. Strong evidence of magma ascent to shallow levels, however, did not appear until late September 1995, when increasing numbers of hybrid events heralded the formation of a small dome and spine. We infer from the data that intrusion of a small volume of magma occurred in July, but stalled on ascent. Degassing of the stalled magma formed a carapace that thickened with time. We suggest that either volatile build-up beneath the degassed carapace, or aseismic intrusion of fresh material, finally forced the stalled magma to the surface in late September 1995. A similar cycle of activity appears to have occurred during the second phreatic phase between late September and mid-November 1995.

A major challenge during periods of volcanic unrest is to correctly interpret monitoring data. Part of the difficulty lies in the fact that, worldwide, there are few good data sets for periods of unrest, but there are many ways in which volcanic unrest can unfold. Further difficulty arises because many fundamental processes that drive eruptions and eruptive behaviour are still poorly understood. Regardless, correct interpretation of monitoring data and successful communication of that interpretation can help prevent disaster (e.g. Mount Pinatubo 1991; Newhall & Punongbayan 1996), whereas incorrect interpretation and (or) poor communication can lead to disaster (e.g. Nevado del Ruiz 1985; see Voight 1990). While it is obvious that monitoring data are critical for successfully forecasting volcanic activity, they are also critical for understanding how volcanoes work. Such data provide rare realtime information on subsurface processes that influence eruptive behaviour. No single monitoring tool - seismicity, deformation, visual observations, gas emissions, etc. - measures all parameters of a volcanic system, so multiple data sets are needed. Eventually, theoretical and experimental models of eruptive processes must be able to replicate all the timescales and patterns recorded by monitoring data in order to be considered successful. In this paper we present detailed visual, seismic, gas emission and deformation data from 18 July to 30 September 1995, which recorded the reawakening of a small-volume andesitic system after 350 years of quiescence (Young et al. 1996). We then use the data to constrain hypotheses regarding the nature of the pre-eruptive plumbing system. We focus on this narrow time period of the eruption because (a) it brackets the time between the opening of the first phreatic vent on 18 July 1995 and formation of a small dome and spine in late September 1995, and (b) it is the only pre-eruptive time period for which there are SO2 emission data. We broadly defined the term phreatic to imply that the activity was dominated by hydrothermal, meteoric and (or) magmatic volatile emissions with little or no juvenile material reaching the surface. Prior to 18 July 1995, precursory activity consisted solely of higher-thannormal seismicity that had begun three years earlier (Miller et al. 1997). After September 1995, the volcano experienced a second phreatic phase that lasted until the beginning of near-continuous dome extrusion in mid-November 1995 (see Young et al. 1998b; Sparks & Young 2002).

Monitoring Visual Airborne observations of vent areas and soufrieres were made almost daily from 28 July until 8 September 1995. After 8 September

until early October 1995, vent areas were visited on foot as time and weather permitted.

Seismic A seismic network of five short-period seismometers existed on the island when the first phreatic vent opened on 18 July 1995 at

Fig. 1. Map showing locations of seismic stations (black triangles and fourletter designations), electronic tiltmeters (white triangles and two-letter designations), and paths for COSPEC flights (dashed lines) for July to September 1995. The short-dashed line shows the normal COSPEC flight path, as winds were typically from the east. The longer-dashed line shows the flight path for 12-13 August 1995 when winds were from the south. MVO, location of the Montserrat Volcano Observatory after 21 August 1995; CPD, Castle Peak dome; MA, Montserrat Airport; stipple pattern shows extent of the city of Plymouth. Contour intervals 400 feet.

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 567-581. 0435-4052/02/$15 © The Geological Society of London 2002.

567

568

C. A. GARDNER & R. A. WHITE

Soufriere Hills Volcano. This network was quickly augmented, and, by the end of July, consisted of nine short-period seismometers (Fig. 1; Aspinall et al. 1998; Power et al 1998). Signals from each station were telemetered either by FM radio or telephone line to a central recording site in downtown Plymouth until 21 August 1995, after which time the Montserrat Volcano Observatory (MVO) was relocated to a site further north (Fig. 1). Signals were recorded on a PC-based digital acquisition system similar to that described by Murray et al. (1996). This system provides earthquake hypocentres and magnitudes, as well as real-time seismic amplitude measurements (RSAM; Endo & Murray 1991) and spectral seismicamplitude measurements (SSAM; Stephens et al. 1994). Earthquake hypocentres and magnitudes were calculated using the HYPO71PC program (Lee & Valdes 1985). Earthquakes were located using a four-layer velocity model that was derived empirically by modifying the velocity model used at Guadeloupe (C. Antenor, pers. comm. 1995) to minimize the root mean square of the residual error (RMS) from a test set of earthquakes from August 1995 at Montserrat. This model begins at sea level and has layer boundaries at 2, 3 and 15km depths corresponding to P-wave velocities of 2.5 (sea level), 3.5, 6, and T k m s - 1 respectively. A Vp/Vs ratio of 1.75 was assumed for the location process. Because the earthquake data set includes both poorly and well located events, only earthquakes that fit the following criteria were used for analyses: (a) horizontal and vertical errors <0.75 km; (b) number of P and S phases >7; and (c) RMS < 0.2s. Earthquake magnitudes were calculated by coda duration measurements derived for the eastern Caribbean (Shepherd & Aspinall 1983).

Montserrat airport located on the east side of the island (Fig. 1), as well as from onboard instruments. Four to seven traverses beneath the plume were made when SO2 was detected: the values from each traverse were then averaged for a daily emission rate. The similarity in SO2 emission rates collected at various times over a period of several weeks (Table 1) suggests that extrapolating a one-hour window of measurements to a daily emission rate was generally a reasonable assumption, except for periods during heightened or unusual seismicity. Then, measurements were often made twice daily, with one group of measurements often much higher than the other. The odour of SO2 was usually detected about the same time that the COSPEC registered SO2, indicating that we were within, rather than below, the plume during the traverse; thus daily averages should be treated as minimum values. The odour of H2S was detected during the first several COSPEC flights, suggesting that some SO2 was being scrubbed by groundwater. By 31 July, however, the odour of SO2 was dominant, indicating that the conduit had 'dried out' (Doukas & Gerlach 1995). The odour of SO2 remained dominant until emission rates dropped below detection level (<40 tonnes per day ( t d - 1 ) . or 0 . 5 k g s - 1 ) in midAugust 1995. During the first week of measurements, car traverses along the north-south road on the west side of the island were made as well. Data from these traverses corresponded well with those of the airborne traverses (Table 1). The largest uncertainty in SO2 measurements is in determining wind speed or the rate at which the plume is moving. For wind

SO2 emissions Airborne SO2 measurements using a correlative spectrometer (COSPEC) installed in a Lynx helicopter were made on almost a daily basis from 29 July until 2 September 1995 (Fig. 2). Light interference from the helicopter blades produced a noisy trace. Nonetheless, by choosing intermediate values of background, high and low calibration, and the SO2 trace (Fig. 3), consistent results (Table 1) were obtained even with three different operators collecting and reducing data. Airborne traverses were flown perpendicular to the plume path (Fig. 1). Winds were generally from the east, and wind speed and direction at altitude were obtained from balloon soundings at

Fig. 2. Photograph of COSPEC mounted in the Lynx helicopter. The instruments were fastened with bungy cords to the back of the pilot's seat and to an O-ring in the floor. SO2 measurements were made with the doors open and the optical sensor pointed upwards (photograph by T. J. Casadevall)

Fig. 3. (a) Photograph of steam plumes from the 18 July and 28 July vents taken on 13 August 1995. (b) Portion of the chart recorded data from the 12 August 1995 COSPEC flight. The trace is very noisy because of light interference from the helicopter blades. To calculate emission rates, intermediate values (solid lines) were chosen to determine the baseline, high and low calibrations, and the SO2 flux above baseline. On this day. both plumes contributed to the SO 2 emission rate.

569

SEISMICITY AND GAS EMISSION DURING PHREATIC PHASE Table 1. COSPEC data for the period 29 July to 2 September 1995 Date

Method

Time

N

SO2 (td- 1 )

SO2 (kgs - 1 )

Wind speed (ms- 1 )

29 July 29 July 30 July 30 July 31 July 1 Aug. 2 Aug. 4 Aug. 4 Aug. 5 Aug. 5 Aug. 6 Aug. 6 Aug. 7 Aug. 8 Aug. 9 Aug. 10 Aug. 1 1 Aug. 12 Aug. 13 Aug. 14 Aug. 15 Aug. 16 Aug. 17 Aug. 18 Aug. 19 Aug. 19 Aug. 20 Aug. 20 Aug. 21 Aug. 22 Aug. 23 Aug. 24 Aug. 25 Aug. 26 Aug. 28 Aug. 29 Aug. 30 Aug. 31 Aug. 2 Sep.

a a a a a a a a

14:30 15:41 10:48 15:32 09:25 09:30 09:44 10:14 14:41 09:33 10:32 09:41 15:12 10:00 09:16 09:27 09:13 09:38 09:32 16:36 10:40 10:02 09:13 09:28 09:14 09:25 15:17 09:29 15:30 16:20 16:30 12:02 09:22 09:12 09:54 09:38 17:07 17:10 09:40 10:03

9

300 ? 590 ?

3.5 6.8 0.9 3.0 3.1 3.7 3.1 9.4 2.5 4.2 4.7 13.9 1.8 3.0 2.2 2.0 2.0 2.8 2.2 0.9 1.4 .7 .9 1.9 .4 2.1 2.0 0.8 0.5 bd bd 0.5 bd bd bd 1.0 bd bd bd bd

nd 3.6 5.1 5.6 6.2 7.2 6.7 12.9 7.7 7.2 6.2 6.2 5.1 8.3 7.7 7.7 6.2 5.1 3.1 1.5 6.4 6.6 8.9 9.8 8.2 7.8 8.0 8.5 6.7 7.7 7.7 7.7 8.3 10.0 12.3 7.7 4.3 nd 4.1 7.7

' C

a c a a a a a a a a a a a a a a a a a a a a a a a a a a a a a

3? 4 5 5 5 3 5 4 6 3 4 3 5 3 6 4 5 5 4 5 7 6 6 5 6 4 6 4 4 4 4 5 6 5 5 2 3 3 4

260 270 320 270 8 10

5 80 70 40 5

0 5 365 0 405 5 1200 0 5 260 0 190 5 0 170 0 245 5 190 0 80 0 120 0 5 165 0 5 120 0 5 5 2 40 2 bd bd 40 6 bd bd 50 2 0 bd bd bd bd

a, airborne; c, car measurement; N, number of runs on which the calculations are based; starting time for first COSPEC measurement; bd, below detection limit (<40td - 1 ); nd, no data.

speeds of 3.6 m s-1 or more (Table 1), the total error is probably less than 30%, but for lighter winds of less than 2.5 m s-1, SO2 emission rates may be in error by as much as 50 to 100% (see discussion in Daag et al. 1996).

Deformation Three high-resolution electronic tiltmeters (see Voight et al. 1998 for specifications) were installed by early August 1995 (Fig. 1). The Amersham and Spring Hill tiltmeters (3.7 WSW and 2.5km SW of the 18 July vent, respectively) were transmitting data by 1 August, and the Long Ground tiltmeter (2km NE of the vent area) by 4 August. Tilt data were transmitted and recorded once every 10 minutes. In addition, dry-tilt lines, using an array established by K. Yamashita and R. Fiske in the 1970s, were re-established in midAugust with measurements continuing until early September 1995. Also, in late August to early September 1995, electronic distance meter (EDM) measurements were made on the west and east flanks of the volcano (J. Shepherd, pers. comm. 1995).

Observations Vent and fumarole observations Heightened seismicity that began in 1992 was the fourth seismic crisis centred on Soufriere Hills Volcano in the past 100 years: 1897-1898, 1933-1937, 1966-1967 and 1992 to present (Young et al. 1998b). Unlike during the previous three crises, nearby fumaroles, namely Gages Lower and Upper, Galway's and Tar River Soufrieres (Fig. 4), were largely unperturbed (T. J. Casadevall pers. comm. 1995; Hammouya et al. 1998). Although data regarding the soufrieres during past crises are sparse, reports document that heat flux, temperature and (or) composition variations occurred at some or all of the soufrieres (Perret 1939; MacGregor 1949; Shepherd et al. 1971). MacGregor (1949) noted that during both the 1897-1898 and 1934-1937 crises, the soufrieres increased in temperature, and that H2S output increased at Gages Lower Soufriere. He states that the lack of SO2 during the 1933-1937 crisis indicated to him that a local eruption was unlikely. However, Perret (1939) discusses a 'new ingredient' at

570

C. A. GARDNER & R. A. WHITE

Gages Soufriere in 1934 that was exceedingly irritating to the eyes and respiratory passages, and that the smell of H2S was absent. The appearance of this 'new ingredient' coincided with an increase in temperature of the Gages Lower Soufriere from 75 to 94°C, which later that year rose to 115-120°C. Ferret (1939) also noted that most of the activity occurred at Gages Soufriere, with Galway's Soufriere showing relatively little change. During the 1966-1967 crisis, however, heat flow and temperature both increased at Galway's Soufriere, but no changes occurred at Gages Upper Soufriere, and no mention was made of Gages Lower or the Tar River Soufrieres (Shepherd et al. 1971). Temperatures at all the soufrieres have been between 96 and 99°C since the 1967 crisis (Hammouya et al. 1998). Only Galway's Soufriere has shown any change during the present crisis. There, a slight change of slowly decreasing amounts of CO2 coincident with a slight increase in H2S began in early August 1995 and continued until at least April of 1996; however, significant changes were not observed in any of the other gas species, nor in gas pressure or temperature (Hammouya et al. 1998). During the previous three seismic crises no new vents had opened on the volcano, but at approximately 14:00 (all times given are local times) on 18 July 1995, residents in the southern third of Montserrat noted loud rumblings, sulphurous smells, and light ashfall coming from Soufriere Hills Volcano (Miller et al. 1997). The following day a field party reported minor explosions about every 20 minutes, with debris reaching 40m from a vent on the WSW flank of Castle Peak dome (Figs 4 and 5a). Light ashfall was again reported during the morning of 22 July, and a few minor tephra-producing events ensued over the next several days. On 28 July, a 'large' explosion and a volcanotectonic earthquake swarm (about 50 earthquakes of about Ml) occurred early in the morning. Later that day, a new vent, 1 to 2m in diameter, was noted on the SE side of English's Crater near the headwaters of the Hot River (Figs 4 and 5b). On 4 August, a vigorous steam-and-ash emission enlarged the 28 July vent to about 10m in diameter. From

its inception, the 28 July vent had a vigorous steam plume in contrast to the 18 July vent which was only intermittently active, Plume vigour at the 28 July vent decreased abruptly about 16:00 on 19 August, after which time only a wispy plume emanated from the vent. During late July 1995, water in English's Crater flowed around and into the 28 July vent, with water observed within the vent in early August (T. J. Casadevall, pers. comm. 1995). On about 6 or 7 August the water disappeared from the vent area and was not observed again. A new fissure with several small vents was observed on 17 August; none had existed on 16 August (Figs 4 and 5c). The fissure and new vents extended southward from the 18 July vent towards the 28 July vent for several tens of metres. These vents were not a source of continuous emission, but enlarged over the next several weeks, largely coalescing with the 18 July vent, On 27 August, new vents opened on the NNE side of Castle Peak between 10:00 and 14:00 (Figs 4 and 5d). Heavy rains from Hurricane Iris obscured views of the volcano, but field crews reported loud noises from the direction of Castle Peak dome, the odour of SO2 by early afternoon, and at 16:30 a light ashfall. On 28 August, two small vents, at altitudes of 500 and 550m were observed on the north flank of Castle Peak dome. Activity quickly diminished at these vents, and by 30 August only a few faint wisps of steam emanated from them.

Seismic activity Classification of seismic signals. Seismic events during the phreatic phase are classified into five main categories (Fig. 6) volcanotectonic (VT) earthquake, long-period (LP) event, hybrid event, explosion signal and tremor - on the basis of waveform, spectral content and temporal relationships to other earthquakes, This scheme closely follows that of Lahr et al. (1994) and Miller

Fig. 4. (a) Map showing locations of vents (labelled by date of initial activity) and other features in and around English's Crater. GLS, Gages Lower Soufriere; GUS, Gages Upper Soufriere; TRS, Tar River Soufriere; GS, Galway's Soufriere; CP, Chances Peak; GM, Galway's Mountain. The hachured line shows the outline of English's Crater and the dashed line the outline of Castle Peak dome in July to September 1995. Drainage shown in solid lines, (b) Chronology of vent activity. Strong activity, a vigorous steam plume was present; weak activity, only wisps emanated from the vent; moderate activity, a condition between the two; and no activity when not even wisps of steam emanated from the vent. Ash emissions are indicated when ash either fell on the far flanks of the volcano or the ash plume rose > 1 km over the vent. On days when more than one visual observation was made, there may be more than one symbol denoting activity.

SEISMICITY AND GAS EMISSION DURING PHREATIC PHASE

571

Fig. 5. Photographs of vents active during the initial phreatic phase from 18 July to 30 September 1995. (a) 18 July vent on 11 August looking SE. (b) 28 July vent on 30 July, looking west, (c) South moat in English's Crater with crack that began opening between the 18 July and 28 July vents on 16 August, looking NW. (d) Steam plumes from 28 July vent and 27 August vents from Montserrat Airport (photographs a and b by T. J. Casadevall).

et al. (1997, 1998), but subtle differences require explanation so that signals (Fig. 6d) were often of longer duration (several minutes), events described here can be compared with events observed at with the amplitude usually increasing slowly to a peak, then slowly other volcanoes. There is considerable overlap amongst the five decaying. LP events have been attributed to the resonance of fluidcategories: (a) rapidly occurring events of any of the first three types filled cracks or conduits induced by pressure transients in the fluids, (VT, LP or hybrid), may appear at times as types of tremor; and either magma or gas (Chouet 1996). (b) explosion signals occasionally have codas containing VTs, LPs Hybrid events are generally impulsive, low-frequency, monoor tremor. Owing to attenuation of high frequencies over distance, chromatic events with a superimposed high-frequency component the signal envelopes and spectra described below only apply to during the first few seconds (Fig. 6c). These events have narrowseismic stations within 3km of the vents. band spectra with dominant frequencies of 0.5-4 Hz (Miller et al. Volcanotectonic (VT) earthquakes generally are impulsive, have 1998). Hybrid event swarms at Montserrat generally showed a clear P and S arrivals, are of short duration, and have broad spectra progression from (a) initially more impulsive, more high-frequency, with peak frequencies above 5 Hz (Lahr et al. 1994; Fig. 6a). These irregularly spaced events, to (b) less impulsive, lower frequency, events are believed to result from brittle fracture in response to stress more regularly spaced events of similar magnitude, to (c) very changes associated with magmatic activity. We subdivided VTs into emergent, almost monochromatic events of very regular spacing two groups: (a) mainshock-aftershock sequences, in which the and almost identical magnitude (White et al. 1998). Unlike LP largest earthquake occurs at or near the beginning of the sequence events, hybrid events show a mix of first motions at different seismic and the number of subsequent smaller events decays approxistations. When the interval between small hybrid events became less mately as l/t; and (b) VT swarms, in which the largest earthquake than a few seconds, the signal appeared as tremor on analogue occurs roughly near the middle of the sequence and is not much stations more than several hundred metres from the vent area larger than several other earthquakes, and in which the rate of (Neuberg et al. 1998), but could be recognized as individual rapidevents may change through time, sometimes more than once. fire hybrid events by careful spectral analysis. When hundreds or thousands of small VTs occur in a short time Explosion signals last for tens of seconds to minutes and have frame, they may appear as high-frequency or spasmodic tremor an envelope that does not decay rapidly (Fig. 6d). They have (see below). broadband spectra and a great variety of onset and waveform Long-period (LP) events generally have peak frequencies of 1 to appearances. At Montserrat, explosion signals were often paired 2.5 Hz and are nearly monochromatic (Miller et al. 1998). During with LP events that occurred within the coda of explosions (Fig. 6d) the initial phreatic phase at Montserrat, LPs occurred both as or as discrete events following explosions by minutes to hours, isolated events and also within the coda of explosion signals. Interestingly, while many of the larger explosion signals accompanied ash plumes, many of the smaller ones apparently were not Isolated LP events were of generally short duration (<1 min) with emergent onsets and tectonic earthquake envelopes (Fig. 6b). LP accompanied by venting; thus, we infer that these were buried events that occurred within the codas of, or shortly after, explosion explosions.

572

C. A. G A R D N E R & R. A. WHITE

Fig. 6. Representative seismic signals from Montserrat recorded during August and September 1995. (a) Volcanotectonic (VT) signals from 13 August, (b) Long-period (LP) signal from 13 August, (c) Hybrid signals from 24 September, (d) Explosion signal with LP coda and two later explosion signals without LP codas, all from 1 September, (e) Spasmodic tremor from 7 August, (f) Harmonic tremor from 10 August. All seismograms from the Gages (MGAT) seismic station located 1 km NW of Castle Peak dome (Fig. 1).

Tremor is a continuous seismic disturbance that lasts from several minutes to days or weeks (Harlow et al 1996). Both highfrequency (HF) tremor and low-frequency (LF) tremor occurred. HF, or spasmodic tremor (Fig. 6e), occurred as impulsive, closely spaced events with highly irregular amplitudes. HF tremor has broadband spectra with considerable energy between 5 and 20 Hz and a dominant frequency above 10 Hz. LF, or harmonic tremor (Fig. 6f), generally occurred as a rather even-amplitude, narrowband, sometimes nearly monochromatic signal with a dominant frequency between 1 and 4 Hz. At times, harmonic tremor was composed of closely spaced LP or hybrid events (Neuberg et aL 1998). Seismic chronology. For this report we define the beginning of the seismic crisis as the time when anomalous seismicity was first detected in January 1992 (Ambeh et al. 1995; Miller et al 1997). We define the beginning of the phreatic phase as the opening of the first vent on 18 July 1995 and the duration of the initial phreatic phase as the period from 18 July until the appearance of a small dome and spine in late September 1995 (Young et aL 1998b). A compilation of seismic data summarizing many thousands of individual signals during the initial phreatic phase, based on seismograms

from the seismic station (MGAT; Fig. 1) 1 km NW of Castle Peak dome, is shown in Figure 7. Highlighted in grey on Figure 7 are the main events referred to below. January 1992 to 18 July 1995. The following summary is taken from Miller et al. (1997). Heightened seismicity associated with the 1995 volcanic activity can be traced back to January 1992. Sporadic activity occurred throughout 1992, with the most notable swarm occurring on 8 November, when seven events were located just north of Montserrat at depths ranging from 12 to 19km and magnitudes between 2.4 and 3.6. The year 1993 was relatively quiet, with just 37 events recorded. Activity increased by an order of magnitude in 1994 (338 events) and increased further during the first seven months of 1995 (306 events). Two small VT swarms off the island occurred prior to the phreatic explosion of 18 July, one on 30 June and the other 15 July 1995. 18 July-5 August 1995. The opening of the two main phreatic vents, the 18 July and 28 July vents (Fig. 4). occurred during this interval. The 18 July vent opened without notable accompanying seismicity, whereas the opening of the 28 July vent appears coincident with the largest explosion of the phreatic period (Fig. 7). No single type of seismicity dominated this interval. The opening of the 18 July vent occurred within an aftershock sequence, perhaps from the 15 July VT sequence. After 18 July, there were several lowamplitude tremor episodes, many of which were of too short a duration to be shown (Fig. 7). Most of these tremor episodes began with an explosion signal, and some ended with one. Explosion signals and LP events occurred sporadically throughout this period. Located events were distributed diffusely in time and space around the summit area and were concentrated between 0 and 2 km depths (all depths are given as below sea level, with 0 being at sea level). The summit of the volcano is approximately 1 km above sea level. 5-6 August 1995. This period is dominated by two VT swarms (Fig. 7). The first began around 07:00 on 5 August with an earthquake approximately 2.5km NE of Castle Peak dome at a depth of 2.5 km (Fig. 8a). Subsequent events migrated farther ENE and moved deeper, eventually reaching a depth just over 6km. There were virtually no earthquakes above 2 km. Thirteen VTs of M > 2 . 5 were recorded. The largest earthquake, M3.4, occurred about 14:00 on 5 August, after which seismicity decayed as l / t . A quiet period of about 2 hours preceded the 6 August VT sequence, which began at 06:52 with an explosion signal. Several large-magnitude events (M>2.5), at about 2-3 km depth and smaller shallower events, located from the summit to a couple of kilometres to the NE, initiated the sequence (Fig. 8b). These decayed as a mainshock-aftershock sequence, with aftershocks centred mostly under the summit at about 2 km depth. Five events of M2.5-3.0 were recorded. A small burst of low-magnitude VT seismicity occurred early on 7 August between 05:00 and 07:00. The entire sequence ended by 20:00 on 7 August. 6-12 August 1995. This interval is dominated by tremor (Fig. 7). Late on 6 August, HF tremor began, which appeared to be composed of numerous quasi-regularly spaced VT earthquakes. Early on 7 August, roughly at 03:00. 04:00 and 05:00, bands of LF tremor appear to be superimposed upon the HF tremor. About 08:00 on 9 August, tremor briefly increased in amplitude to 20mm and became longer-period, before dropping back to the average of 5mm around 11:00. Later that day, the spasmodic character of the HF tremor diminished and LF tremor dominated before quickly decaying in the early hours of 11 August. Two bands of HF tremor initiated by explosions occurred between 08:00 and 10:00 and also between 16:30 and 17:30 on 11 August. The period between these two bands was fairly quiet, with only a couple of explosion signals and a few VTs. Low-amplitude HF tremor occurred from 08:00 to 12:00 on 12 August. That and a few explosion signals preceded the 12 August VT sequence. 72-75 August 1995. An explosion at 16:10 on 12 August initiated another VT swarm that increased slowly in number and magnitude of events for several hours until a M4.0 event at 02:22 on 13 August (Figs 7 and 8c). Afterwards the swarm decayed as a mainshock-aftershock sequence. Hypocentres were located 3.5km WNW of Castle Peak dome. For the first hour, earthquakes were

Fig. 7. Rainfall, SO2, and seismic data for the period 18 July to 30 September 1995. Each vertical bin represents a 4-hour interval (0-4; 4-8; 8-12; 12-16; 16-20; 20-24). We have no rainfall data prior to 28 July or after 29 August 1995. Rainfall data were from Montserrat Airport (Fig. 1) and were recorded daily. We assigned each day's value to the 8-12 time bin. SO 2 data (td -1 ) are reported in the time bin in which the COSPEC flight or car traverse was made. There are no data for the time periods 18-29 July 1995 and 3-30 September 1995. Tremor data represent the average amplitude on seismograms from the Gages seismic station (MGAT) located 1 km NW of Castle Peak dome (Fig. 1): tremor had to occur for at least 2 hours in order to be represented on the plot. Explosions are given as duration of saturation on the MGAT record and are cumulative, i.e. if more than one explosion occurred during a given time bin, the durations were added together. VTs are the number of volcanotectonic earthquakes greater than 1 cm amplitude on the MGAT record. LPs are the number of long-period events greater than 1 cm amplitude on the MGAT record. The LP codas from explosions that met the 1 cm amplitude criterion were counted as a single LP event. Hybrid energy was calculated as the log amplitude multiplied by the number of events. The stipple line for the second hybrid swarm is a schematic of data from Miller et al. (1997). The vertical grey lines highlight time periods during which significant events, discussed in the paper, occurred (i.e. vent opening, seismic swarm, etc.). This database includes all earthquakes for the time period 18 July to 30 September and not just the subset of earthquakes used for analyses in Figure 8.

Fig 8. Map showing epicentral areas of VT and hybrid swarms during August and September 1995 and time depth plots for each swarm. Sea level is 0 and the dashed line is 2km below sea level (the summit of Soufriere Hills is about 1 km above sea level), (a) 5 August, off-summit VT swarm, (b) 6 August, near-summit VT swarm, (c) 12-13 August, off-summit VT swarm, (d) 21 August, near-summit VT swarm, (e) 30-31 August, near-summit VT swarm. (f) 8-9 September, off-summit VT swarm, (g) 17 September, near-summit VT swarm, (h) 23-26 September, near-summit hybrid swarm associated with formation of a small dome and spine on about 25 September. The size of symbols is keyed to magnitudes (smallest circle <M1.5: small intermediate Ml.5-2.0; large intermediate M2.0-2.5; largest circle >M2.5). The low number of hybrid events shown in this data set as compared to the same time period in Figure 7 is because only those events that fit the criteria discussed in the seismic method section were included here.

SEISMICITY AND GAS EMISSION DURING PHREATIC PHASE

located at depths of slightly less than 2.0 to 4.5km, with the main energy release occurring over the next four hours at depths between 3 and 4km. The swarm faded out late on 14 August; however, it was not until late on 16 August that the majority of VTs were located again under Soufriere Hills. Starting at 00:30 on 14 August, explosion signals occurred regularly, at intervals between 2.5 and 3.5 hours, for the next 32 hours. 15-19 August 1995. Tremor dominated this interval (Fig. 7), although intervals of quiescence, each lasting for several hours, occurred early on 15 August and again on 16 August. Tremor began as LF tremor for the first several hours, after which it diminished and HF tremor dominated. On 16 August, LF tremor again dominated the record and did so for most of the rest of the time interval. Shortly after noon on 19 August, an explosion occurred during the tremor that produced an ash column to about 1 km above Soufriere Hills. Tremor continued for several more hours, then abruptly ceased at 15:57. 19-23 August 1995. Several explosion signals accompanied by ash plumes (> 1 km above Soufriere Hills) and a shallow seismic swarm characterize this interval. Two ash-producing explosions occurred on 19 August (at 12:20 and 19:47), one on 20 August (16:55), one on 21 August (08:02) and one late afternoon on 22 August (15:55) (Fig. 4). The 21 August explosion sent an ash plume to 7 km and a cold pyroclastic density current over the west wall of English's Crater (Sparks & Young 2002). All events consisted of an explosion signal with an LP coda, but none triggered tremor episodes (Fig. 7). About 19:40 on 20 August a swarm of small-amplitude VT events began. These were located beneath the crater mostly at depths of 1 to 2 km (Fig. 8d) and diminished slowly, ending on 23 August. 23-27 August 1995. Starting at 16:00 on 23 August, explosion signals dominated the record and occurred at intervals of 2 to 5 hours. Some explosion signals had LP codas or were followed by discrete LPs, but others did not. No ash plumes were observed during this interval. 27-29 August 1995. This interval coincides with the passage of Hurricane Iris near Montserrat and the opening of the 27 August vent on the NNE side of English's Crater (Fig. 4). The interval is dominated by low-amplitude tremor and a sequence of discrete LP events (Fig. 7). The largest LP of the phreatic period (M2.5) occurred late in the afternoon on 29 August, after which LF tremor appeared in bands lasting a couple of hours. Tremor, which began on 27 August, ceased by 08:00 on 29 August. 29 August-4 September 1995. Explosion signals again dominated the record (Fig. 7) and a small VT swarm occurred late on 30 August to 1 September (Fig. 8e). Starting around 06:30 on 29 August, explosion signals occurred at intervals between 1 and 6 hours and then more regularly about one every hour. Most of these explosion signals had LP codas lasting up to 10 minutes, but no ash was produced during these events. Late on 30 August to early on 1 September, three small VT swarms (17:40 to 19:50, 23:55 to 03:30, and 07:00 to 07:30) occurred at depths of 1 to 2km under the crater. Each swarm started after an explosion signal or series of explosion signals. None of the explosion signals associated with the VT swarms had an LP coda. After the VT swarms ended, the explosion signals changed character in that LP codas again often followed explosions signals, some of which were associated with ash plumes and lightning. A vigorous explosion signal around 14:00 on 3 September sent an ash plume to about 4km, but no LP coda was associated with this event. 4-7 September 1995. Hurricane Luis passed between the islands of Antigua and Montserrat during this period and storm noise dominated the record between explosion signals at 18:30 on 4 September (ash plume to about 3.5km, but again no LP coda) and 13:15 on 7 September. Spectral analysis shows that during this period the volcano was very quiet. The hurricane activity had a dominant frequency of 0.5 Hz and rain >8-10Hz. There is very little on the SSAM record between those frequencies that might correspond to volcano-seismic events. 7-14 September 1995. Seismic activity returned after Hurricane Luis passed, but in general this interval was mostly quiet (Fig. 7).

575

A strongly felt VT swarm began around 22:10 on 8 September, but lasted only a few hours. The swarm was centred about 3.5 km north (Fig. 8F) of Soufriere Hills, and consisted of at least six events of M>2.5. All depths were between 3 and 4km. After the swarm, seismic activity was very low for the following several days. 14-30 September 1995. Only extremely brief periods (a few minutes) of low-energy hybrid activity occurred prior to midSeptember, but hybrid events dominated seismic activity during the latter half of the month. The first of three episodes began early on 14 September (Fig. 7). On 15 September, discrete LP events occurred at the highest rate seen during the initial phreatic phase and by mid-day, low-amplitude tremor began, the first continuous tremor (>4 hours) detected since 29 August. Tremor and LP activity tapered off just as a VT swarm began on the morning of 17 September. This swarm consisted of numerous low-magnitude events (M < 1.0), with hypocentres slightly NE of Castle Peak dome and at depths between 2 and 4km (Fig. 8g). Low-energy hybrid events again dominated the record on 20 and 21 September (Miller et al. 1997). The third and most energetic episode began late on 23 September and tapered off late on 27 September. The hybrid event rate peaked about every 6-10 hours. Depths varied but the majority of hybrids occurred between 1 and 2km (Fig. 8h). 30 September-mid-November 1995. This is the interval of the second phreatic phase which was dominated by one to five explosion signals per day, many of which were immediately followed by small VT swarms beneath the crater. From 11 to 31 October, 10-40 minute episodes of LF tremor occurred about once per day following small explosion signals. After 31 October the tremor was replaced by discrete LP events. On 4 November a significant ash plume was noted. After 6 November, explosion signals became smaller and less frequent, ending on 13 November. Hybrid events reappeared on 12 November for the first time since September and peaked during 14-20 November 1995 (see White et al. 1998, fig. 2).

Chronology of SO2 emissions COSPEC measurements began on 29 July, a day after phreatic explosions opened the 28 July vent. Early SO2 emission rates -1 averaged about (3.5kgs - ] ) (Fig. 7 and Table 1; Young et al. 1998a) until 4 August when an emission rate of about 810 200t d-1 (9.4 kg s-1) was measured during the morning flight (Table 1). This increase occurred after an explosion signal that was accompanied by an energetic steam-and-ash emission that enlarged the 28 July vent; by afternoon, however, SO2 emission rates returned to previous values. On 5 August, a vigorous VT swarm, felt throughout Plymouth, began about 2.5km NNE of Castle Peak dome (Figs 7 and 8a). SO2 emission rates varied little during this swarm (Table 1 and Fig. 7). The following morning, another VT sequence began, this time beneath Soufriere Hills Volcano. The SO2 emission -1 rate during this sequence was (13.9kgs - 1 ), the highest measured during the phreatic phase. By afternoon the swarm had tapered off and SO2 emission rates had also decreased (Fig. 7 and Table 1). After 6 August, average emission rates declined slightly -1 to an average of about (2.3kgs - 1 ). Throughout early August, a vigorous steam plume emanated from the 28 July vent, which was presumed to be the sole source of SO2 since only a wispy steam plume emanated from the 18 July vent. On 11 August, however, the steam plume from the 18 July vent became more vigorous and the following day was about as vigorous as that from the 28 July vent (Fig. 3a). Unusual southerly winds on 12 August required the COSPEC flight to be flown along an eastwest traverse (Fig. 1). The trace shows distinct peaks for each plume (Fig. 3b), indicating that both were contributing SO 2 (Table 1). A VT sequence, consisting of numerous felt earthquakes and centred approximately 3.5km WNW of Soufriere Hills Volcano (Figs 1, 6 and 7) occurred late in the afternoon of 12 August until early morning of 13 August. SO2 emission rates did not vary over this time period. The low value measured on 13 August (Table 1) was due to the extremely light winds that day.

576

C. A. GARDNER & R. A. WHITE

On 19 August, two noteworthy events occurred. The first, around noon, was an ash-producing explosion with a plume that reached about 2km. The ash came from the 18 July vent area, while the 28 July vent continued vigorously to jet a white steam plume. Gas emission data from before and after the explosion showed no change in SO2 emission rates from previous days (Table 1 and Fig. 7). The second event occurred around 16:00, when the vigorous steam plume at the 28 July vent abruptly decreased, which coincided with the abrupt cessation of tremor (Fig. 7). Two measurements on 20 August showed that SO2 emission rates had dramatically dropped to less than 65 t d - 1 (0.8 kg s - 1 ) . On 21 August, the SO 2 emission rate fell below detection level (<40td - 1 ( 0 . 5 k g s - 1 ) ; Table 1 and Fig. 7). Between 21 August and 2 September, the SO2 emission rate remained at or below detection level, except for a brief increase around 27 August. This increase coincided with the opening of the 27 August vents on the NNE flank of Castle Peak dome (Fig. 5d), and was the only post-19 August tremor episode in August. SO2 emission rate was 90 10 td-1 (1 k g s - 1 ) on 28 August (Table 1 and Fig. 7); however, activity at the vents decayed quickly and by the following day, gas emission rate was below the detection level, tremor had ceased, and only wisps of steam emanated from the 27 August vents. COSPEC measurements ceased on 2 September 1995 and were not restarted until the spring of 1996 (see Young et al. 1998a). Deformation observations During the initial phreatic phase no unambiguous tilt excursions were noted at the three electronic tiltmeters, from the dry-tilt array, nor from the EDM lines, except for a transient excursion during 1-2 September on the Long Ground (LG) tiltmeter (Figs 1 and 9). The excursion began between 22:50 and 23:00 on 1 September 1995 and returned to baseline between 00:10 and 00:20 on 2 September. No weather phenomena could account for the excursion, and no excursion was noted at either of the other two electronic tiltmeters. The end of the tilt event correlated with an energetic explosion signal at 00:06 with a LP coda that started about 00:09 and ended about 00:21. Discussion The seismic crisis that began in 1992 was the fourth one centred on Soufriere Hills Volcano in the past 100 years, but the only one to culminate in an eruption. During the initial phreatic phase, it

Fig. 9. Plot showing tilt excursion at the Long Ground tiltmeter (LG; Fig. 1), located about 2km NE of Castle Peak dome on 1-2 September for the X (S 540 W) and Y (S 380 E) axes.

differed from those of 1897-1898, 1934-1937 and 1966-1967 in other ways as well: (a) it was the first during which phreatic vents opened; (b) it was the first to have unambiguous SO2 emissions (although Perret's (1939) work suggests that SO 2 may have been present during the 1935-1937 crisis): (c) it did not affect nearby soufrieres (Hammouya et al. 1998: Boudon et al. 1998); and (d) unlike during the 1966-1967 crisis (Shepherd et al. 1971), no permanent deformation occurred. Below we discuss the evidence for magma as the source of the seismic unrest and data that suggest that the magma stalled on ascent. We examine relationships between tremor and vent observations. SO 2 degassing and other seismic signals, and we explore hypotheses regarding the possible causes of low SO 2 emission rates during the initial phreatic phase. Lastly, we discuss observations that bear on the magmatic plumbing system prior to the formation of the dome and spine in late September 1995. and for the period prior to the start of near-continuous dome formation between late September and mid-November 1995 (Young et al. 1998b; Sparks & Young 2002). Evidence for magma as the cause of the seismic unrest during the initial phreatic phase Whether or not magma is the driving force behind seismic unrest at a volcano, specifically in such a way as to lead to an eruption, is often not easy to determine. Although there is now no question that unrest at Montserrat was related to magma intrusion, there was considerable debate about this during the initial phreatic phase. We submit that SO 2 emission rates alone, but especially in concert with the varied seismic activity, strongly supported a magma-intrusion hypothesis by late July 1995. which suggested a much greater likelihood of eruption than during the previous three seismic crises. One issue at Montserrat was whether SO 2 emission rates measured between late July and late August 1995 (<40 to 1200t d -1 (<0.5 to 13.9kgs - 1 ); Table 1) were from a magmatic source or the result of boiling or oxidation of the hydrothermal system. We suggest that the hydrothermal system was an unlikely source at Soufriere Hills Volcano on the basis of the following argument. Data from geothermal wells worldwide indicate that gases in geothermal steam plumes are dominantly CO 2 and H 2 S with trace amounts of other gases, but rarely SO 2 (Henley & Ellis 1983). Upon boiling of the hydrothermal system. CO 2 and H 2 S are strongly partitioned into the vapour phase (Doukas & Gerlach 1995). The easiest place to oxidize H2S to form SO2 is in the atmosphere where there is more free oxygen than in groundwater. Because of oxidation, H 2 S has a residence time of only one to two days in the atmosphere (Stern et al. 1984) with a half-life of about one day (Graedel 1977). From July to September 1995. COSPEC measurements were made at distances of about 2 km from the vent. At these distances, plumes measured were in the atmosphere for at most an hour, and usually only minutes, which would have been insufficient time to convert all the H2S to SO2. One mole of SO2 is produced for every mole of H2S oxidized. If the SO2 were solely from oxidation of H 2 S, it would have required that significant quantities (more than several hundred tonnes) of H 2 S were still in the atmosphere at the time the SO 2 was measured. At such quantities. H 2 S should have been easily smelled by COSPEC operators, because the 'rotten egg' odour of H2S is extremely potent, requiring only a small amount for detection (3-20 ppb; Reiffenstein et al. 1992). A slight odour of H 2 S was detected during the first few gas flights (T. J. Casadevall, pers. comm. 1995). but by 31 July, and up to 19 August 1995, only the odour of SO2 was noted. The decline of SO2 emission rates to below detection level after 19 August 1995. however, points out that, although moderate to high SO2 emission rates may be diagnostic of magmatic involvement, low or no SO2 emissions should not necessarily be interpreted as a lack of magmatic involvement. Explanations for very low SO2 emission rates that are consistent with magma intrusion include scrubbing of SO2 by groundwater (Doukas & Gerlach 1995; Symonds & Gerlach 1998) or creation of a degassed plug that may prevent or reduce gas transport to the surface (Casadevall et al.

SEISMICITY AND GAS EMISSION DURING PHREATIC PHASE

1994; Daag et al 1996; Zapata et al. 1997). The low SO2 emission rates (15-40td -1 or 0.2-0.5 kg s - 1 )during the phreatic phase of the 1980 eruption of Mount St Helens (Casadevall et al. 1981) certainly could not outweigh the evidence for magma intrusion from seismic and deformation data.

Magma transport from mid-July to late-September 1995 As volcanic unrest unfolds, knowing at precisely what depth magma resides within the conduit, precisely when it ascends, and whether that ascent is continuous or episodic would be extremely useful. During later periods of dome growth at Montserrat, there was good correlation between deformation, seismicity and dome growth (Voight et al. 1998; Sparks & Young 2002) such that quite reliable forecasts of volcanic activity could be made, but rarely do such ideal data or conditions exist. The seismic data indicate that there was no appreciable magma ascent during the initial phreatic phase until formation of the dome and spine during vigorous hybrid activity in late September 1995. Hypocentres for summit swarms (Fig. 8b, d,e, g) between August and late-September were all clustered around depths of 2km, and similarities in earthquake waveforms during these swarms suggest that the earthquakes were being generated from the same source and were not propagating upwards along a crack. VT hypocentres were as deep as 7 km during the 5 August swarm (Fig. 8a), but the swarm started at a depth of around 2km and migrated deeper, not shallower, and away from the summit. The 6 August swarm that started several hours later (Fig. 8b) had events as shallow as 0 km (sea level), but this swarm started at about 2 km depth with the majority of subsequent events clustered around 2 km. Locatable explosion signals and HF tremor also had depths around 2 km, although LF tremor may have been slightly shallower at depths of less than 1 to 1.5km. Thus no seismicity during the initial phreatic phase provided any apparent evidence of upward magma movement. Swarms with hypocentres several kilometres from the volcano (Fig. 8a,c, f) had deeper hypocentres and b-values more consistent with tectonic than volcanic earthquakes (Power et al. 1998). The seismic data, coupled with the lack of an appreciable change in SO2 emission rate during the 5 and 12-13 August earthquake swarms (Fig. 7; no SO2 data exist for the 7-8 September swarm), suggest that the three off-edifice sequences were related to stress adjustments around the volcano, perhaps resulting from changes in magmatic pressure at depth, rather than directly from magma movement (see also Aspinall et al. 1998). Hypocenters clustered around 2 km during the 6 August swarm, but there were numerous events between 0 and 2km early in the swarm (Fig. 8b). A SO2 emission rate of 1200+l00td -1 (13.9kgs - 1 ) measured on the morning of 6 August was three to four times the average daily emission rate and the highest seen during the initial phreatic phase (Fig. 7 and Table 1). This coincidence suggests that the shallow seismicity may have been caused by exsolution of volatiles due to (a) intrusion of new, more gas-rich magma, (b) decompression- or crystallization-induced degassing, or (c) accumulated volatiles forcing a path to the surface. That SO2 emission rates returned to pre-swarm levels by that afternoon, and that hypocentres during subsequent edifice swarms clustered at depths about 2km, suggest that the cause of the increased gas emissions did not involve a large influx of new material or ascent of magma to appreciably shallower levels. Another possible explanation for the high SO2 emission rate and shallow earthquakes of the 6 August swarm may be that fault rupture during the 5 August swarm acted to decompress conduit magma, thereby accelerating degassing. This additional supply of volatiles may have initiated the phreatic explosion that started the 6 August swarm and provided the higher SO2 emission rate. Except for the transient tilt signal observed on 1-2 September, neither the electronic tiltmeters, dry tilt, nor EDM lines showed deformation prior to 25 September 1995. This suggests that, if deformation at depth had accompanied intrusion, it had occurred

577

prior to early August, the placement of the electronic tiltmeters would have picked up changes in volume mostly at depths >3 km, and that no permanent shallow deformation had occurred by early September 1995. The 1-2 September 1995 electronic tilt signal returned to baseline (Fig. 9), which indicated that some ephemeral process had occurred, perhaps a large vesiculation event, rather than a volume change due to intrusion of a significant volume of magma. Because of the timing of this tilt event, any gas emission related to a vesiculation event would have dissipated prior to the COSPEC flight on 2 September. The lack of observed deformation during the initial phreatic phase is in contrast with: (a) data from the 1966-1967 crisis when deformation rates were as high as 12 microradians per month (Shepherd et al. 1971); (b) deformation observed after formation of the dome and spine in late-September 1995 data (Jackson et al. 1998); and (c) deformation during the magmatic portion of the eruption (Voight et al. 1998; Sparks & Young 2002). Clear deformation did occur after (and possibly just before) formation of the late-September dome and spine (Jackson et al. 1998). EDM lines shortened 40 cm between 28 September and the start of continuous dome extrusion on about 15 November 1995 (Jackson et al. 1998). The electronic tiltmeters detected no deformation during this interval (28 September-15 November 1995), supporting Jackson et a/.'s (1998) interpretation that the deformation was shallow. Devine et al. (1998) postulated that early dome magmas (mid-November to December) had an ascent time of about 60 days on the basis of hornblende reaction rims observed in December 1995 deposits. We suggest that the petrologic data (Devine et al. 1998), deformation data (this paper and Jackson et al. 1998) and high groundmass crystallinity (including a pure silica phase) of dome rocks erupted in early December 1995 may indicate that a shallow cryptodome was emplaced or shallow intrusion occurred between 28 September and 15 November 1995. A note should be made about the nature of the late September dome and spine. First, the extrusion appears to have been shortlived. Second, it is unclear if either of these features represented juvenile material or older conduit fill. The spine was oxidized (Sparks & Young 2002), but the reason for the oxidation is unknown because no samples of spine or dome were collected. Regardless of whether they were juvenile or not, their presence strongly suggests that magma reached very near, if not actually to, the surface in late September 1995.

Observations of tremor in relation to other types of seismicity and SO2 degassing Tremor commonly occurs prior to eruptions (McNutt 1996). It falls into that family of seismic events in which fluid generally plays an active role in generating seismic waves (Chouet 1994). Correlation of tremor with volcanic activity, however, varies widely, and depends upon source mechanisms (Chouet 1988). Low-amplitude tremor was noted on seismic records by 20 July. It then dominated the seismic record during two periods in early to mid-August before stopping suddenly on 19 August (Fig. 7). This abrupt cessation coincided precisely in time with field observations of a dramatic decrease in the steam plume from the 28 July vent and, in general, with a diminished SO2 emission rate. From 20 August to 30 September only wisps of steam emanated from the 28 July vent and only three brief periods of low-amplitude tremor occurred. The coincidence of abrupt cessation of tremor, marked decrease in plume vigour from the 28 July vent and decreased SO2 emission rate suggests strong links between these three phenomena. The links, however, are not always clear. From 28 July until 19 August, a vigorous steam plume emanated from the 28 July vent and moderate SO2 emissions were detected, whether tremor occurred or not (see 1-5, 5-6 and 12-13 August, Fig. 7). The highest SO2 emission rate recorded during the phreatic phase occurred just prior to and during the 5-6 August VT swarms, when no tremor was recorded. It is

578

C. A. GARDNER & R. A. WHITE

important to note, however, that after 20 August, SO2 emission rates were above detection limits only during a brief period of tremor that coincided with activity at the 27 August vents. Because there were no gas measurements between 2 and 30 September 1995, we do not know whether tremor episodes on 15-16 and 22 September were also accompanied by SO2 degassing. During the initial phreatic phase, tremor and VT swarms were essentially mutually exclusive (Fig. 7). If tremor were only spasmodic HF tremor, then this mutual exclusion may be explained as a change from closely spaced (tremor) to more discrete VT earthquakes. During the two main tremor episodes in August 1995, however, both HF and LF tremor occurred, often simultaneously. Thus, this exclusivity may be real or perhaps the result of a limited database. No exclusivity was seen among hybrid events, long-period earthquakes, explosions and tremor (see 14-16 September, Fig. 7). Explosion signals occurred during periods of tremor (Fig. 7) with seemingly no effect on the tremor (amplitude or frequency), nor on the vigour of the 28 July steam plume. Were these processes decoupled and, if so, how? The ash-producing explosion around noon on 19 August may provide some insight. On that day tremor dominated the seismic records as it had done for many days, the 28 July vent had a vigorous steam plume, and there was no steam plume activity at the 18 July vent. The ash plume coincident with the explosion signal, however, clearly came from the 18 July vent (Fig. 4). Thus, for this event, the two seismic signals appear to be related to different pathways - tremor with the 28 July vent and the explosion with the 18 July vent - which may have been the case at other times as well. Unfortunately we cannot test this hypothesis, because the resolution of the seismic data is not sufficient to locate precisely explosions or tremor under one vent or the other. Ashproducing explosions from the 18 July vent continued after tremor and plume activity at the 28 July vent ceased and SO2 emission rates were below detection level. If the above hypothesis is correct, it would suggest that the 19 August explosion occurred at a depth above which the two conduits diverge because the explosion did not affect plume vigour at the 28 July vent (Fig. 10). The explosion signals themselves are enigmatic because it is difficult to tie them to any given physical effect at the volcano. Some explosion signals were associated with ash plumes that rose to several kilometres above the volcano, but many were not. We consider those explosion signals that had no surface expression to be buried explosions. The only explosion that produced an ash plume

from the 28 July vent occurred on 4 August (Fig. 4), which was coincident with an increase in the SO2 emission rate (Fig. 7), but most explosion signals seemed to have had no effect on SO2 emission rates. Until 19 August, some explosions appeared to trigger tremor, but after that date there was no such relationship during the rest of the initial phreatic phase. Some explosion signals that had LP codas were accompanied by ash plumes whereas others were not, and many explosion signals that were accompanied by ash plumes did not have LP codas.

Models to account for the low SO2 emission rates during the intitial phreatic phase SO2 measurements were made almost daily from 29 July to 2 September 1995 and except for 4 and 6 August. SO2 emission rates were generally low (<40-300td -1 (0.5-3.5kgs -1 ); Fig. 7 and Table 1). Were the low SO2 emission rates the consequence of SO2 removal by hydrolysis, i.e. scrubbing by water (Doukas & Gerlach 1995; Symonds & Gerlach 1998), or of a small volume of magma being degassed? It is difficult to test either hypothesis, but the implication of each scenario is quite different in regard to the volume of magma that may have been residing at shallow levels. There was certainly plenty of surface water available to scrub SO2; from late July to 7 August, water flowed into the 28 July vent, and standing water was sometimes seen in the vent (T. J. Casadevall. pers. comm. 1995). There were also long-lived ponds in the moat between Castle Peak and the rim of English's Crater and on Castle Peak dome itself. And there were periods of intense rainfall. The data, however, do not appear compatible with a scrubbing model, at least one in which surface water played a role. SO2 emission rates were highest during early gas monitoring, when water flowed into the 28 July vent, and lower when there was no visible water in the vent area. SO2 emission rates were fairly constant regardless of dry spells or rainy weather (Fig. 7). After 20 August, the only time SO 2 was detected during the remainder of the initial phreatic phase was during the opening of the 27 August vents, which coincided with high rains caused by Hurricane Iris. The lack of an H2S odour after late July also suggests that the conduit had 'dried out' and that SO 2 was not being scrubbed by groundwater. Unfortunately we do not have SO2 measurements for the period between early September 1995 and April 1996, when (a) extrusion of the dome and spine occurred in late September 1995. (b) nearly continuous dome extrusion began in mid-November 1995. and (c) the early stages of dome growth were taking place (Young et al. 1998a). However, during the period of fairly continuous lowvolume extrusion rates (2-3m 3 s - 1 : Sparks et al. 1998) from late (northern hemisphere) spring to mid-summer 1996, average SO2 emission rates were about 3 0 0 t d - 1 (3.5kgs - 1 ) (Younget al. 1998a), which were similar to SO 2 emission rates during the initial phreatic phase. Although similar emission rates do not substantiate instrusion of a small volume of magma to shallow levels, it supports the idea that low SO 2 emission rates during the initial phreatic phase are compatible with such a model.

Speculations on the nature of the plumbing system during the phreatic phase

Fig. 10. Speculative schematic diagram of the plumbing system at Soufriere Hills Volcano prior to late September 1995. The thick black line at depth signifies a conduit filled with magma, and the thinner lines represent gas conduits that were not filled with magma. The extent of the highly altered zones associated with the soufrieres is unknown as denoted by question marks. A zone of hydrothermal fluids probably exists beneath Castle Peak dome as suggested by Boudon et al. (1998), but is not included in order to better show other features (modified after Boudon et al. 1998).

Of the data presented, the location of the phreatic vents, response of the nearby soufrieres, pattern of SO2 emission rates, depths of VT hypocentres and timing of the ash-producing explosion signals seem to bear most directly on the nature of the plumbing system during the phreatic phase. The locations of the phreatic vents suggest that the 350-year-old Castle Peak dome (Young et al. 1996) initially provided a barrier to the formation of a single central conduit, such that it was easier to establish phreatic vents around its margins (Figs 4 and 10). Some barrier must also have existed between the magma and nearby soufrieres as evidenced by the lack

SEISMICITY AND GAS EMISSION DURING PHREATIC PHASE

of response of the soufrieres during this crisis (Hammouya et al. 1998; Boudon et al. 1998). The formation of the phreatic conduits may have isolated magma from the hydrothermal system by sealing the magmatic system from hydrothermal system by precipitation of vapour-transported silica (Boudon et al. 1998), thereby forcing volatiles preferentially through the conduits (Hammouya et al. 1998). We also favour a role for the conduits allowing magmatic gas to bypass the soufrieres, but are mindful that the phreatic vents did not open until three years after the seismic crisis began (Gardner et al. 1996; Young et al. 19986). Although the 18 and 28 July vents appeared to act independently of each other, COSPEC data from 12 August showed SO2 contributions from both vents (Fig. 7 and Table ) indicating that they were connected at some depth (Fig. 10). A vigorous steam plume emanated from the 28 July vent while it was active, whereas almost all the significant ash plumes (>1 km above the volcano) originated from the 18 July vent and crack area. We propose that the 28 July vent represented the path of least resistance for magmatic gas and was the source of the tremor, whereas the 18 July vent represented the path of least resistance for magma ascent. Both the late September dome and spine, and the dome that began forming in mid-November 1995 were extruded broadly in the area of the 18 July vent (see Watts et al. 2002). SO2 emission rates decreased slowly from late July to midAugust, and then dramatically to below detection level (<40 x 103 t d - 1 ; 0.5kgs - 1 ) after 19 August (Fig. 7). We favour the explanation that the slow decline in SO2 emission rates was the natural consequence of degassing a small volume of magma - as the magma degassed a carapace formed that thickened with time. Eventually the carapace thickened sufficiently to prevent gas beneath it from escaping to the surface, except during explosive events which would temporarily rupture the carapace. The decrease in SO2 emission rates after 19 August may reflect that the carapace had become a more effective seal or, alternatively, that explosive activity on 19 August may have changed the structure of the plumbing system by (a) blocking of the 28 July vent, or (b) opening fractures that allowed for the release of gas into the wall rock or water into the conduit. Either structual change would contribute to the sudden demise of the steam plume at the 28 July vent. An influx of groundwater could explain the reduced SO2 emissions by scrubbing, although behaviour of the 27 August vents is difficult to reconcile with such a model. After 19 August and for the remainder of the initial phreatic phase, SO2 emission rates were only above the detection level while the 27 August vents were active. That these vents were short-lived, suggests that the volume of gas beneath the carapace was small, or that the ruptured carapace quickly rehealed. We suggest that the data best support a model of a smallvolume intrusion of magma by mid-July, which led to the formation of the first phreatic vent on 18 July 1995. A second phreatic vent, which became the path of least resistance for rising gases opened on 28 July. The similarity in hypocentre depths of 2 km (Fig. 8) for edifice VT swarms during August and September 1995 indicates that the magma had stalled by early August. This depth may represent the location of brittle fracture at the top of the magma column or the depth of brittle fracture due to volatile overpressurization from magma at some depth below 2km. Volatile exsolution from the stalled magma resulted in formation of a degassed carapace that thickened with time, resulting in slowly declining SO2 emission rates. Declining SO2 emission rates due to degassed magma plugging the conduit have also been invoked at Redoubt Volcano (Casadevall et al. 1994) and at Mount Pinatubo (Daag et al. 1996). Volatile pressurization beneath the carapace probably drove ashproducing explosions in late August and early September 1995, and the opening of the 27 August vents on the north flank of Castle Peak dome. Continued exsolution of volatiles and (or) an influx of volatile-rich magma finally pushed the stalled magma towards the surface in late September 1995 resulting in the formation of the small dome and spine. The second phreatic phase, from 30 September to midNovember 1995, was similar in many respects to the mid-August to late September period of the initial phreatic phase. Seismicity

579

was dominated by small explosion signals and a daily episode (10-40 minutes) of tremor or discrete LP events. Two short-lived vents opened on 11 and 14 October (Watts et al. 2002) and an explosion on 4 November was accompanied by a significant ash plume. Unlike the initial phreatic period, however, significant shallow deformation occurred (Jackson et al. 1998). We suggest that data for the second phreatic phase also support a model of magma intrusion that stalled. This time, however, the magma stalled at shallower levels below the late September dome and spine. Slow influx of magma or accumulating gases resulted in deformation of the upper edifice. As during the initial phreatic phase, continued exsolution of volatiles or aseismic influx of magma from depth resulted in failure of the capping dome and spine. On 12 November, hybrid events reappeared, and intensified between 14 and 20 November 1995, coincident with the start of near-continuous dome extrusion (Young et al. 1998b). Summary (1)

(2) (3)

(4)

(5)

(6)

(7)

Heightened seismicity that began in 1992 was the fourth seismic crisis centred on Soufriere Hills Volcano in the past 100 years, and the only one during which (a) activity at nearby soufrieres remained unchanged, (b) new vents opened, and (c) a magmatic eruption ensued. SO2 emission rates of >200td - 1 (>2.3kgs - 1 ) measured in late July 1995 implicated magma intrusion as the cause of the seismic unrest. Hypocenters for all summit VT swarms clustered around 2 km, and similarities in waveforms during these swarms suggest that the earthquakes were being generated repeatedly from the same source. We propose that the magma stalled below 2 km from early August until late September 1995. No deformation was detected from early August to early September 1995, which suggests that any significant changes in volume at depths >3km below the volcano's summit, if they occurred at all, had already occurred by early August 1995 and that no permanent shallow deformation had occurred by early September 1995. The gradual decrease in SO2 emission rates from late July 1995 until 19 August 1995, at which time rates fell below detection thresholds (<40td - 1 or <<0.5kgs - 1 ), appears more consistent with degassing of a small volume of magma and development of a degassed carapace that thickened with time, than with SO2 being scrubbed by groundwater, i.e. removed by hydrolysis. We suggest the following model for the initial phreatic phase, (a) Decompressive degassing of the intruding magma led to shallow pressurization that forced opened phreatic vents on 18 and 28 July 1995. (b) Magma may have ascended to slightly shallower levels between 4 and 6 August when explosions and VT swarms were accompanied by SO2 emission rates of 810td - 1 (9.4kgs - 1 )andl200td - 1 (13.9 kgs -1 ).(c) After early August magma stalled in its ascent, and continued degassing of the stalled magma led to the formation of a carapace that thickened with time, (d) Around 19 August 1995, pressurized gases beneath the carapace drove explosions that sent ash plumes several kilometres above the volcano and effectively bled off available gas such that SO2 emissions dropped below the detection level. Continued exsolution of gas from magma beneath the carapace increased pressure that led to explosion signals accompanied by ash plumes to > 1 km above the volcano in late August and early September 1995 and formed short-lived vents on the flanks of Castle Peak dome, (e) Lastly, either continued exsolution of volatiles from magma beneath the carapace or, more likely, aseismic influx of volatile-rich magma from depth, finally pushed the stalled magma to the surface during vigorous hybrid activity in late September 1995. The second phreatic phase, between late September and midNovember 1995, was similar in most respects to the initial phreatic phase, except for the occurrence of significant shallow

C. A. GARDNER & R. A. WHITE

580

deformation (Jackson et al. 1998). We infer that magma ascended to shallow levels and stalled beneath the lateSeptember dome and spine. New magma slowly intruded into and deformed the upper edifice, eventually forcing the plug to fail in mid-November 1995, and initiating the start of nearcontinuous dome growth. Monitoring and managing the Soufriere Hills volcanic crisis during the initial phreatic phase was largely an effort by the Seismic Research Unit (SRU) from the University of West Indies, Trinidad, the US Geological Survey (USGS), individuals from several universities and from observatories in the Caribbean, and local residents. We thank all our colleagues during that time, especially W. Ambeh, R. Robertson and L. Lynch, all from the SRU, for the opportunity to work together during an often trying, but always interesting, time. We also thank T. Gerlach for numerous discussions regarding hydrothermal systems and magmatic degassing, and C. D. Miller for his Montserrat notebooks in which many important observations were recorded. T. Druitt, J. Ewert, C. Newhall, J. Stix and B. Voight provided constructive reviews that greatly helped to improve the manuscript.

References AMBEH, W. B., LYNCH, L. L., CHEN, J. & ROBERTSON, R. E. A. 1995. Contemporary seismicity of Montserrat, West Indies. In: ALI, W., PAUL, A. & YOUNGON, V. (eds) Transactions of the 3rd Geological Conference of the Geological Society of Trinidad and Tobago and the 14th Caribbean Geological Conference, vol. 1, 1-2. ASPINALL, W. P., MILLER, A. D., LYNCH, L. L., LATCHMAN, J. L., STEWART, R. C., WHITE, R. A. & POWER, J. A. 1998. Soufriere Hills eruption, Montserrat 1995-1997: volcanic earthquake locations and fault plane solutions. Geophysical Research Letters, 25, 3397-4000. BOUDON, G., VlLLEMANT, B., KOMOROWSKI, J-C,

ILDEFONSE, P. & SEMET,

M. P. 1998. The hydrothermal system at Soufriere Hills volcano, Montserrat (West Indies): Characterization and role in the on-going eruption. Geophysical Research Letters, 25, 3693-3696. CASADEVALL, T. J., JOHNSTON, D. A., HARRIS, D. M. ET AL. 1981. SO2 emission rates at Mount St. Helens from March 29 through December 1980. In: LIPMAN, P. W. & MULLINEAUX, D. R. (eds) The 1980 eruptions of Mount St. Helens, Washington. United States Geological Survey, Professional Papers, 1250, 193-200. CASADEVALL, T. J., DOUKAS, M. P., NEAL, C. A., MCGIMSEY, R. G. & GARDNER, C. A. 1994. Emission rates of sulfur dioxide and carbon dioxide from Redoubt Volcano, Alaska during the 1989-1990 eruptions. In: MILLER, T. P. & CHOUET, B. A. (eds) The 1989-1990 eruptions of Redoubt Volcano, Alaska. Journal of Volcanology and Geothermal Resources, 62, 519-530. CHOUET, B. 1988. Resonance of a fluid-driven crack; radiation properties and implications for the source of long-period events and harmonic tremor. Journal of Geophysical Research, 93, 4375-4400. CHOUET, B. A. 1994. Precursory swarms of long-period events at Redoubt Volcano (1989-1990), Alaska: Their origin and use as a forecasting tool. In: MILLER, T. P. & CHOUET, B. A (eds) The 1989-1990 Eruptions of Redoubt Volcano, Alaska. Journal of Volcanology and Geothermal Research, 62, 95-135. CHOUET, B. A. 1996. Long-period volcano seismicity; its source and use in eruption forecasting. Nature, 380, 309-316. DAAG, A. S., TUBIANOSA, B. S., NEWHALL, C. G. ET AL. 1996. Monitoring sulfur dioxide emission at Mount Pinatubo. In: NEWHALL, C. G. & PUNONGBAYAN, R. S. (eds) Fire and Mud: Eruptions and Lahars of Mount Pinatubo, Philippines. Philippine Institute of Volcanology and Seismology, Quezon City, and University of Washington Press. Seattle, 409-414. DEVINE, J. D., RUTHERFORD, M. J. & GARDNER, J. E. 1998. Petrologic determination of ascent rates for the 1995-1997 Soufriere Hills Volcano andesitic magma. Geophysical Research Letters, 25, 3673-3676. DOUKAS, M. P. & GERLACH, T. M. 1995. Sulfur dioxide scrubbing during the 1992 eruptions of Crater Peak, Mount Spurr volcano, Alaska. In: KEITH, T. E. C. (ed.) The 1992 Eruptions of Crater Peak Vent, Mount Spurr Volcano, Alaska. United States Geological Survey Bulletin, 2139, 47-57. ENDO, E. T. & MURRAY, T. L. 1991. Real-time seismic amplitude measurement (RSAM): a volcano monitoring and prediction tool. Bulletin of Volcanology, 53, 533-545.

GARDNER, C. A., CASADEVALL, T. J. & ROBERTSON. R. E. A. 1996. SO2 emissions during the phreatic phase of volcanic unrest at the Soufriere Hills Volcano, Montserrat, West Indies (abstract). The Second Caribbean Conference on Natural Hazards and Disasters. Publication No. 1. Unit for Disaster Studies, Jamaica. 33. GRAEDEL, T. E. 1977. The homogeneous chemistry of atmospheric sulfur. Review of Geophysics and Space Physics, 15. 421-428. HAMMOUYA, G., ALLARD, P., JEAN-BAPTISTE. P.. PARELLO. F.. SEMET. M. P. & YOUNG, S. R. 1998. Pre- and syn-eruptive geochemistry of volcanic gases from Soufriere Hills of Montserrat. West Indies. Geophysical Research Letters, 25, 3685-3688. HARLOW, D. H., POWER, J. A., LAGUERTA, E. P.. AMBUBUYOG. G.. WHITE, R. A. & HOBLITT, R. P. 1996. In: NEWHALL. C. G. & PUNONGBAYAN. R. S. (eds) Fire and Mud: Eruptions and Lahars of Mount Pinatubo, Philippines. Philippine Institute of Volcanology and Seismology. Quezon City, and University of Washington Press. Seattle. 385-305. HENLEY, R. W. & ELLIS, A. J. 1983. Geothermal systems ancient and modern: a geochemical review. Earth Science Reviews, 19. 1-50. JACKSON, P., SHEPHERD, J. B.. ROBERTSON. R. E. A. & SKERRITT. G. 1998. Ground deformation studies at Soufriere Hills Volcano. Montserrat I. Electronic distance meter studies. Geophysical Research Letters, 25. 3409-3412. LAHR, J. C., CHOUET. B. A., STEPHENS. C. D.. POWER. J. A. & PAGE. R. A. 1994. Earthquake classification, location, and error analysis in a volcanic environment: implications for the magmatic system of the 1989-1990 eruptions at Redoubt Volcano. Alaska. In: MILLER. T. P. & CHOUET, B. A. (eds) The 1989-1990 Eruptions of Redoubt Volcano, Alaska. Journal of Volcanology and Geothermal Research, 62, 137-151. LEE, W. H. K. & VALDES, C. M. 1985. HYPO71PC: a personal computer version of the HYPO71 earthquake location program. United States Geological Survey Open-File Report 85-749. MACGREGOR, A. G. 1949. Prediction in relation to seismo-volcanic phenomena in the Caribbean volcanic arc. Bulletin Volcanologique. 8. 69-86. McNuTT. S. R. 1996. Seismic monitoring and eruption forecasting of volcanoes: a review of the state-of-the-art and case histories. In: SCARPA. R. & TILLING, R. I. (eds) Monitoring and Mitigation of Volcano Hazards. Springer, Berlin, 99-146. MILLER, A. D. & MVO TEAM SEISMIC 1997. Seismicity of the Soufriere Hills Volcano, Montserrat, July 1995 to mid-1997. Montserrat Volcano Observatory Open-File Report . 97 24. MILLER, A. D., STEWART, R. C.. WHITE, R. A. ET AL. 1998. Seismicity associated with dome growth and collapse at the Soufriere Hills Volcano. Montserrat. Geophysical Research Letters, 25. 3401-3404. MURRAY, T. L., POWER, J. A., DAVIDSON, G. & MARSO. J. N. 1996. A PC-based real-time volcano data-aquisition and analysis system. In: NEWHALL, C. G. & PUNONGBAYAN, R. S. (eds) Fire and Mud: Eruptions and Lahars of Mount Pinatubo, Philippines. Philippine Institute of Volcanology and Seismology. Quezon City, and University of Washington Press, Seattle, 225-232. NEUBERG, J., BAPTIE. B.. LUCKETT, R. & STEWART. R. 1998. Results from the broadband seismic network on Montserrat. Geophysical Research Letters, 25, 3661-3664. NEWHALL, C. G. & PUNONGBAYAN. R. S. (eds) 1996. Fire and Mud: Eruptions and Lahars of Mount Pinatubo, Philippines. Philippine Institute of Volcanology and Seismology. Quezon City, and University of Washington Press, Seattle. PERRET, F. A. 1939. The volcano-seismic crisis at Montserrat 1933-1937. Publication of the Carnegie Institute. 512. POWER, J. A., WYSS. M. & LATCHMAN, J. L. 1998. Spatial variations in the frequency-magnitude distribution of earthquakes at Soufriere Hills Volcano, Montserrat. West Indies. Geophysical Research Letters, 25, 3653-3656. REIFFENSTEIN, R. J., HULBERT, W. C. & ROTH. S. H. 1992. Toxicology of hydrogen sulfide. Annual Review of Pharmacology and Toxicology, 109-134. SHEPHERD, J. B. & ASPINALL, W. P. 1983. Seismicity and earthquake hazard in Trinidad and Tobago, West Indies. International Journal of Earthquakes and Engineering Structual Dynamics, 11, 229-250. SHEPHERD, J. B., TOMLIN. J. F. & Woo. D. 1971. Volcano-seismic crisis in Montserrat, West Indies. Bulletin Volcanologique, 35. 143-163. SPARKS, R. S. J. & YOUNG. S. R. 2002. The eruption of Soufriere Hills Volcano, Montserrat (1995-1999): overview of scientific results. In:

SEISMICITY AND GAS EMISSION DURING PHREATIC PHASE DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London Memoirs, 21, 45-69. SPARKS, R. S. J., YOUNG, S. R., BARCLAY, J. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills Volcano, Montserrat, West Indies: November 1995 to December 1997. Geophysical Research Letters, 25, 3421-3424. STEPHENS, C. D., CHOUET, B. A., PAGE, R. A., LAHR, J. C. & POWER, J. A. 1994. Seismological aspects of the 1989-1990 eruptions at Redoubt Volcano, Alaska: the SSAM perspective. In: MILLER, T. P. & CHOUET, B. A. (eds) The 1989-1990 eruptions of Redoubt Volcano, Alaska. Journal of Volcanology and Geothermal Research, 62, 153-182. STERN, A. C., BOUBEL, R. W., TURNER, D. B. & Fox, D. L. 1984. Fundamentals of Air Pollution (second edition). Academic Press, New York. SYMONDS, R. B. & GERLACH, T. M. 1998. Modeling the interaction of magmatic gases with water at active volcanoes. In: AREHART, G. B. & HULSTON, J. R. (eds) Water-Rock Interaction: Proceedings of the 9th International Symposium of Water-Rock Interaction. Balkema, Rotterdam, 495-498. VOIGHT, B. 1990. The 1985 Nevado del Ruiz volcano catastrophe: anatomy and retrospection. Journal of Volcanology and Geothermal Research, 42, 151-188. VOIGHT, B., HOBLITT, R. P., CLARKE, A. B., LOCKHART, A. B., MILLER, A. D. & LYNCH, L. 1998. Remarkable cyclic ground deformation monitored in real-time on Montserrat, and its use in eruption forecasting. Geophysical Research Letters, 25, 3405-3408.

581

WATTS, R. B., HERD, R. A., SPARKS, R. S. J. & YOUNG, S. R. 2002. Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London Memoirs, 21, 115-152. WHITE, R. A., MILLER, A. D., LYNCH, L. & POWER, J. 1998. Observations of hybrid seismic events at Soufriere Hills Volcano, Montserrat: July 1995 to September 1996. Geophysical Research Letters, 25, 3657-3661. YOUNG, S. R., HOBLITT, R. P., SMITH, A. L., DEVINE, J. D., WADGE, G. & SHEPHERD, J. B. 1996. Dating of explosive volcanic events associated with dome growth at the Soufriere Hills volcano, Montserrat, West Indies. In: Second Caribbean Conference on Natural Hazards and Hazard Management, Kingston, Jamaica. MVO Open-File Report , 96/22. YOUNG, S. R., FRANCIS, P. W., BARCLAY, J. ET AL. 1998a. Monitoring SO2 emission at the Soufriere Hills volcano: Implications for changes in eruptive conditions. Geophysical Research Letters, 25, 3681-3684. YOUNG, S. R., SPARKS, R. S. J., ASPINALL, W. P., LYNCH, L. L., MILLER, A. D., ROBERTSON, R. E. A. & SHEPHERD, J. B. 1998b. Overview of the eruption of Soufriere Hills Volcano, Montserrat, July 18 1995 to December 1997. Geophysical Research Letters, 25, 3389-3392. ZAPATA, G., J. A., CALVACHE V, M. L., CORTES, J. ET AL. 1997. SO2 fluxes from Galeras Volcano, Columbia 1989-1995: Progressive degassing and conduit obstruction of a Decade Volcano. In: STIX, J., CALVACHE V, M. L. & WILLIAMS, S. N. (eds) Galeras Volcano, Columbia: Interdisciplinary Study of a Decade Volcano. Journal of Volcanology and Geothermal Research, 77, 195-208.

Spaceborne radar measurements of the eruption of Soufriere Hills Volcano, Montserrat G. WADGE1, B. SCHEUCHL1 2 & N. F. STEVENS 13 1

Environmental Systems Science Centre, Harry Pitt Building, University of Reading, Reading RG6 6AL, UK (e-mail: [email protected]) 2 Department of Electrical and Computer Engineering, University of British Columbia, Vancouver, British Columbia, V6T124, Canada 3 Institute of Geological and Nuclear Sciences Ltd, Gracefield Research Centre, Lower Hutt, New Zealand

Abstract: Radar measurements from space are used to help monitor the evolution of a Peleean eruption on Soufriere Hills Volcano, Montserrat, during 1996 to April 1999. Data from four radar systems are used: ERS-1, ERS-2, Radarsat and JERS-1. We demonstrate that ratio images of backscattered radar energy collected at different times provide useful qualitative summary maps of gross topographic change (e.g. infilling of valleys with deposits) and changes in backscattering properties with time. Radar phase data using the interferometry technique can also provide valuable change detection information. Phase coherence images for ERS-1 and ERS-2 pairs with only one day separation on 25 and 26 September 1997 and 8 and 9 April 1999 allow the areal extent of some pyroclastic flows deposits emplaced during those 24-hour periods to be mapped. Generally, the radar phase can only be retrieved from those parts of the volcano where the vegetation is destroyed by pyroclastic flow deposits and local slopes are not too steep. Quantitative information on the topography of the volcano can also be extracted from the phase data, though not routinely from the new lava dome that grew during 1995-1998. By comparing the radar-measured post-eruption topography with the pre-eruption topography the thickness of the pyroclastic flow deposits (up to 85 m in some valleys on the northeastern slopes) are mapped. Radar interferometry also supplies a means of measuring surface deformation between radar images.

During eruptive crises there is a need to observe surface volcanological phenomena. This may be made difficult or impossible because of factors such as inaccessibility, restricted daylight and cloud cover, and some high-latitude volcanic regions are particularly difficult in this regard (e.g. Alaska and the Aleutians; Lu & Freymueller 1998). Spaceborne remote sensing by radar offers a way of guaranteeing observations through clouds, essentially anywhere on the globe (Francis et al. 1996). In the case of the eruption of Soufriere Hills Volcano on the tropical island of Montserrat, the crisis during the 1995-1999 period was long-lived and the monitoring by the Montserrat Volcano Observatory (MVO) was hampered by persistent orographic cloud cover. Knowledge of the state of growth of the lava dome on the volcano was often missing for weeks at a time between the temporary liftings of the cloud. The changing topography of the volcano, a vital source of information on the eruption, can, potentially, be given by radar measurements. Synthetic aperture radar (SAR) is a technique that uses its own source of energy (microwaves) to illuminate the Earth's surface and measure the returning reflected energy (Henderson & Lewis 1998). There are three main sources of information that can be utilized from this returning signal: the phase, amplitude and polarization. Here we consider only the phase (angular measurement of the radar waveform) and amplitude (power of the returned signal). The image construction technique of SAR produces an inherent distortion of topography and a noisiness (speckle), which can make SAR images more difficult to understand than optical images. Images constructed from the amplitude of the returned signal give a proxy measurement of the topography through the orientation of the local surfaces scattering the energy back to the radar antenna and also the roughness of the surface. If the surface changes in a major way between images taken from the same place, then any changes in amplitude will reflect the changed topography or changed local scattering properties. Using a time series of SAR amplitude information in this way, we show here how some of the surface effects of the eruption can be mapped in a qualitative manner. Potentially, the phase of the radar signal gives the most valuable, quantitative information about the surface of a volcano, and this is exploited in SAR interferometry. The phase of a single SAR image carries no information that can be extracted usefully. However, two SAR images acquired from the same, or almost the same, viewpoint can be used to extract information on the surface topography or on the surface movements that have occurred between the acquisition of the two images (Bamler & Hartl 1998). This is because the two sets of phase interfere in proportion to the separation of the viewpoints (or baseline). The modulation of this interference by

the topography can be inverted using the knowledge of the imaging geometry. Any relative movement of the surface will also produce systematic changes in phase that can be extracted once the imaging geometry and topographic effects are removed. These capabilities can be used for a variety of geophysical applications (Massonnet & Feigl 1998). Volcanological applications to date have focused largely on basaltic volcanoes (e.g. Massonnet et al. 1995; Sigmundsson et al. 1999). This paper contains the initial results of a continuing study of the radar data at Soufriere Hills. The aim is to present an overview of the techniques and the results for the non-remote sensing specialist rather than a detailed evaluation. We describe here how publicly available Spaceborne radar data can be used to extract volcanological information from a volcano during an ongoing Peleean eruption at an andesitic stratovolcano. Analysis of the operational utility of the technique at Soufriere Hills Volcano is given in Wadge et al. (1999, 2000). A detailed study of the surface deformation results will be reported elsewhere.

The Soufriere Hills eruption from 1995 to 1999 Soufriere Hills Volcano is a medium-sized andesitic stratovolcano (Rea 1974), typical of hundreds situated above subduction zones. It was recognized from deposits of previous eruptions (the last occurring about 350 years ago), and from the proximity of the main town, Plymouth, that any new eruption of the volcano would produce considerable hazards from pyroclastic flows (Wadge & Isaacs 1988) (Fig.l). The eruption that began in July 1995 was in some respects typical of eruptions from many other volcanoes of Peleean type. From July to November 1995 the eruption was principally phreatic in nature. The main characteristic, however, was the growth of a lava dome over three years at the central, English's Crater. By March 1996 this dome became sufficiently unstable to generate the first pyroclastic flows that were mainly directed down radial valleys (Figs 2 and 3). The vigour of these flows generally increased through to the end of 1997, with a hiatus in late 1996 (Young et al. 1998). Occasionally, vertically directed explosions occurred (Fig. 2). One main characteristic was that the vigour of the volcanic activity, and specifically the rate of magma discharge, increased with time (Sparks et al. 1998). There was a huge collapse of the flank and dome in the SW sector at the end of 1997 (Sparks et al 2002). The succeeding three months saw the rapid regrowth of the dome, but few pyroclastic flows. After the cessation of magma supply in March

DRUJTT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 583-594. 0435-4052/02/S15 © The Geological Society of London 2002.

583

584

G. WADGE ET AL.

Fig. 1. Map of Montserrat as it was prior to the eruption period 1995-1999.

SPACEBORNE RADAR

MEASUREMENTS

585

Fig. 2. Timeline of the eruption of Soufriere Hills Volcano from 1996 to 1999. The lower part is a simplified graphic summary of the vigour of pyroclastic flow creation, magnitude expressed as four vertical intervals and frequency in the thickness of the line. The lighter shaded events are flows derived from sub-Plinian or Vulcanian explosions. The upper box shows the times of SAR acquistions used in this study: ERS is represented by thick lines (E), Radarsat by thin lines (R) and JERS-1 by dashed lines (J).

1998, the volcano was relatively quiet until July 1998 when frequent collapses of the dome resumed, diminishing in 1999. Magma extrusion recommenced in November 1999. Prior to the eruption, the upper slopes of the volcano were covered by forest and the lower slopes by a mixture of trees, agriculture and housing. By the end of 1997 much of the tree cover had been destroyed by pyroclastic flows, except on the southeastern and parts of the northwestern flanks of the volcano. The pyroclastic flow deposits contain large metric-sized blocks, but their surfaces are dominated by fine-grained ash from pyroclastic surges and ash fallout (Cole et al. 1998). Some river valleys are almost filled in places by deposits several tens of metres thick; elsewhere the flows have eroded the landscape.

SAR amplitude information Data and processing Four satellite-borne radars have provided data for this study: ERS-1 and ERS-2, Radarsat and JERS1 (Francis et al. 1996). Of these, the ERS SAR data are the most plentiful, contributed by two identical radars on satellites (ERS-1 and ERS-2) following one another on the same orbits with one-day separation. The main technical differences between these instruments are shown in Table 1. The wavelength of the radar affects sensitivity to the long-term stability of scattering processes on the ground; hence the longer wavelength radar of JERS-1 is much less noisy in areas of vegetation than the ERS radar

because it penetrates down to the more stable ground surface. However, the longer wavelength means that the phase is much less sensitive to surface deformation changes. The greater the incidence angle of the radar (measured from the vertical), the less the distorion of the image due to topography. Because SARs image to the side, the same region is viewed from different directions depending on whether the satellite is passing from south to north (ascending pass) or north to south (descending pass). ERS data are not recorded onboard the spacecraft. Because of this, and the distance of Montserrat from the station in Ecuador that receives the ERS data for Montserrat, only ascending pass (the radar views from the west) ERS data were available for Montserrat during the period July 1997 to April 1999 (no data were available for the first two years of the eruption) (Table 2). Ascending and descending data are potentially available from Radarsat SAR. There is only one suitable pair of descending JERS-1 images (Table 2). Figure 4 shows a Radarsat amplitude image of Montserrat taken from a descending pass in November 1996 (Wadge & Haynes 1998). The imaging geometry foreshortens the east-facing slopes and elongates those facing west. The growing delta created by successive pyroclastic flows entering the sea is evident off the east coast. Gross changes in the topography and the nature of the radar scatterers can be detected in these SAR amplitude images, but this requires comparison across two images, and a practised eye. Here we discuss these changes in a qualitative manner and illustrate some of them through images created by ratioing co-registered images with each other (Fig. 5). This has the advantage of suppressing the unchanged image information (e.g. most topography) and emphasizing those areas that have changed. The images in Figure 5 have

586

G. WADGE ET AL.

Fig. 3. Sketch-maps of the distribution of pyroclastic flow deposits at four periods during the 1995-1999 period (after Cole et al. 2002).

been filtered (low-pass, 5x5 kernel) and thresholds of 5 dB applied about a mean of OdB for no change. Thus the light areas have stronger backscatter in the later (numerator) scene than the earlier (denominator) scene and vice versa.

Volcanological interpretation Wadge & Haynes (1998) described the effects of the eruption in the area of the dome and the Tar River valley that were evident in

Radarsat amplitude images from 1996. Figure 5 shows four images from the ERS and JERS-1 sensors, each created by ratioing two amplitude images from different periods of the eruption. In Figure 5a (2 September 1996 over 24 May 1997) derived from descending JERS-1 data, changes are evident (as dark and light bands) in the White River valley south of the dome (A), as well as weaker darktoned changes in the Tar River valley (B). In the case of the White River valley area, a channel 80m deep and 50m wide was cut into the cliffs above Galway's Soufriere during dome collapse on 30 and 31 March 1997 (Cole et al. 2002). Thus the dark areas in Figure 5(a) directly south of the dome correspond to slopes which steepened

Table 1. Some technical characteristics of the SARs used in this study Satellite

Period

ERS-1 ERS-2

Band A (cm)

Incident angle (°)

Repeat cycle (day)

Resolution (m)

Onboard tape recorder

Orbit stability

1991-2000 1995-

C/5.7

23

35(1)

28

No

Good

Radarsat

1995-

C/5.7

10-59

24

10-100

Yes

Poor

JERS-1

1992-1998

L/23.5

39

44

18

Yes

Moderate

SPACEBORNE RADAR MEASUREMENTS Table 2. SAR scenes used in this studv Satellite

Date

Pass

Track

Frame

Orbit/mode

ERS-2 ERS-2 ERS-1 ERS-2 ERS-1 ERS-2 ERS-2 ERS-2 ERS-2 ERS-2 ERS-2 ERS-2 ERS-2 ERS-2 ERS-2 ERS-2 ERS-1 ERS-2 ERS-2 ERS-2 ERS-2 ERS-1 ERS-2 ERS-1 ERS-2

18.7.97 7.9.97 25.9.97 26.9.97 11.10.97 12.10.97 31.10.97 5.12.97 21.12.97 9.1.98 25.1.98 1.3.98 5.4.98 24.4.98 29.5.98 14.6.98 2.7.98 7.8.98 23.8.98 11.9.98 1.11.98 4.3.99 5.3.99 8.4.99 9.4.99

Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend Ascend

347 75 347 347 75 75 347 347 75 347 75 75 75 347 347 75 347 347 75 347 75 347 347 347 347

315 333 315 315 333 333 315 315 333 315 333 333 333 315 315 333 315 315 333 315 333 315 315 315 315

11724 12454 32399 12726 32628 12955 13227 13728 13957 14229 14458 14959 15460 15732 16233 16462 36407 17235 17464 17736 18466 39914 20241 40415 20742

JERS-1 JERS-1

2.9.96 24.5.97

Descend Descend

425 425

273 273

-

Radarsat Radarsat Radarsat Radarsat Radarsat Radarsat Radarsat Radarsat Radarsat

4.3.96 15.5.96 23.11.96 17.12.96 31.8.97 1.11.97 4.11.97 9.2.98 29.3.98

Descend Descend Descend Descend Descend Descend Descend Ascend Ascend

-

_ -

F2 F2 F2 F2 S3 S6 S2 S5 S5

-

Date = day.month.year. Pass: ascend = satellite travels south to north, looks to east; descend = satellite travels north to south, looks to west. Track = number of the repeat swath (100km wide for ERS) imaged by satellite. Frame = number of the along-track subdivision (100km long) that images Montserrat. Orbit = number of times satellite has orbited Earth. Mode = variable image geometry classes for Radarsat: F2 = incidence angle range in degrees (39-42), nominal resolution = 10 m; S2 = incidence angle range in degrees (24-31), nominal resolution = 30m; S3 = incidence angle range in degrees (30-37), nominal resolution = 30m; S5 = incidence angle range in degrees (36-42), nominal resolution = 30 m; S6 = incidence angle range in degrees (41-46), nominal resolution = 30m.

587

primary volcanic deposits during this period by storms, particularly during the hurricane season (June to November). On the northeastern deposits (C), most of which had been emplaced earlier (June and September 1997), there are additional possible explanations: surface roughening (at the centimetre scale) due to preferential removal of fine ash from the surface (crust development was noted in some instances soon after emplacement) and gradual increase in the dielectric constant of the surface as it changed from hot and dry to cooler and wetter, as is consistent with the experimental findings of Adams et al. (1996). Major changes due to the collapse and debris avalanche down the White River valley on 26 December 1997, and accompanying pyroclastic surges on the southwestern sector of the volcano, are evident (D). The largely positive changes are due to removal of forest and partial infilling of the valley by thick debris avalanche and pyroclastic flow deposits. The first three months of 1998 saw the large collapse amphitheatre created by the 26 December 1997 event infilled by vigorous dome growth (Voight et al. 2002). The effects of this are clear in Figure 5c whose images span from immediately after the collapse (9 January 1998) to well after the cessation of dome growth (7 August 1998). The bright, curvilinear feature facing SW with a dark area to the NE (E) corresponds to the reflective surface of the collapse scar being effectively moved to the SW by new dome growth. We see exactly the opposite pattern of change in ratio images (not shown) that span the collapse itself (e.g. late 1997 to January 1998), as we would expect. To the east an ESE-trending linear feature (F) corresponds to the dome collapse scar produced on 3 July 1998 . Dark areas in and around the Tar River valley (G) represent new pyroclastic flow deposits from 3 and 27 July 1998. These new deposits must be 'smoother' to radar and hence darker in Figure 5c than the earlier, eroded deposits. However, unlike in Figure 5a a significant increase in the Tar River delta in July 1998 is not obvious. Figure 5d covers a period (14 June 1998 to 5 March 1999) after magma supply ceased in March 1998. Slight darkening in the Tar River valley (H) represents the pyroclastic flow deposits that were emplaced in the latter half of 1998, but by March 1999 they are much less evident than earlier (G in Fig. 5c). To interpret these amplitude data more closely would require a detailed quantitative study of the scattering properties of the eruption deposits (e.g. Adams et al. 1996), though it is worth emphasizing three general interpretational points that we make from this ratio image study: (1) major topographic change is characterized by parallel positive/negative ratio values (e.g. A, E, F in Fig. 5); (2) newly emplaced pyroclastic flow deposits lower the backscatter intensity relative to existing surfaces (e.g. G, H in Fig. 5); (3) pyroclastic flow deposits increase in backscatter intensity with time (e.g. C in Fig. 5). SAR interferometry Data and processing

in aspect away from the satellite between JERS-1 acquisitions, and the light areas represent slopes which steepen in aspect towards the sensor, resulting from the channel formed between these acquisitions. Changes in this area are also the result of infilling of the Galway's Soufriere with dome talus. The changes further downslope are due to the effects of the pyroclastic flow deposits which had begun to inundate increasingly further down the White River valley in April 1997, to forest removal and to valley infilling to depths of a few tens of metres. A strip of dark-toned change occurs on the Tar River delta, which is inferred to be due to a significant change in delta size after the large-scale dome collapse on 17 September 1996 (MVO 1996). Despite the image distortion there are similarities between Figure 5b (12 October 1997 over 1 November 1998) and the distribution of pyroclastic flow deposits at the start of 1998 (Fig. 3). From 12 October 1997 to 1 November 1998 the radar backscatter from the surface of the pyroclastic flow deposits evident in Figure 5b must have increased. One reason for this increase may be the erosion of

The phase information from SAR images is extracted by differencing two images that have slightly different viewpoints and/or are separated in time. Table 3 shows that we might have been able to create 111 phase difference images, or interferograms, from the ERS data of 1997 and 1998. Sixteen of these image pairs had viewpoints that were too widely separated (baselines > 100m), even for flat topography. The phase information content of the remaining interferograms depends on the constancy of the surface across the time interval of each pair. In general, rock/mineral and urban surfaces have constant scatterers, whilst vegetation and water-affected surfaces do not. Also noise tends to increase with time. For Montserrat, the combination of abundant, fast-growing vegetation, wet climate and the production of new volcanic deposits (dome, pyroclastic flows and ashfall) greatly reduces the number of coherent interferograms in this data set that have any useable information to 22 (Table 3). As part of the interferometric processing it is standard practice to generate an image map of the constancy of phase information, or coherence, that acts as a visual means of

588

G. WADGE ET AL.

period. However, the volcano was very active during this time (02:41 on 25 September to 02:41 on 26 September 1997), with Vulcanian explosions (Druitt et al. 2002; some 2cm sized clasts of pumice in the north of the island) coupled with fountain-collapse-generated pyroclastic flows in Gages Ghaut, Tuitt's Ghaut, Tar River valley and Tyre's Ghaut. Some of these flows are recorded in the tandem coherence image as ribbons of low coherence. The flow down Tuitt's Ghaut is particularly clear in Figure 5a and is seen to apparently bifurcate. Further to the SE, White's Ghaut also shows a lowcoherence ribbon suggesting undocumented new pyroclastic deposits. We can also detect the occurrence of the pyroclastic flow deposit created on 8 April 1999 (Norton et al. 2002) from the coherence image of the 8-9 April tandem pair (Fig. 6b). A ribbon of low coherence on the southern side of Tar River is seen to nearly reach the delta. These are the first reported examples of the emplacement of pyroclastic flow deposits being detected from ERS tandem coherence data.

Topographic information

Fig. 4. Radarsat amplitude image (F2 mode) of Montserrat taken on 23 November 1996. The image was acquired from a descending pass (nearly west-looking) and so the topography is foreshortened on east-facing slopes, c Canadian Space Agency.

gauging information quality. This image product can also be used as a measure of surface change detection in its own right. In general, on Montserrat the pyroclastic flow deposits produce good coherent surfaces for interferometry. However, this is reduced in time by the superposition of new flow deposits, new ashfall deposits or waterinduced erosion. Although each of the ERS satellites only returns to the same orbit track every 35 days, during our study the two satellites (ERS-1 and ERS-2) followed each other by only one day, and it is possible to form interferograms from images acquired in successive days by the two different satellites. Such 'tandem' interferograms tend to have higher coherent values than those created from images separated by 35 days or more. Figure 6a shows a coherence image for the tandem interferogram of 25/26 September 1997. The bright areas are of high phase coherence. The main factors reducing coherence in the Montserrat images are leaf movement (in forested areas), changes in surface water content and changes in surface cover (e.g. ashfall or flow deposits). The only areas that are coherent in Figure 6a are just to the north of Plymouth and, more importantly, the northeastern slopes of the volcano. This corresponds to the area covered by new pyroclastic flow deposits from the major collapse dome on 21 September 1997 which reached beyond the airport and the NE coast (Cole et al. 2002). The block-and-ash surface of this deposit preserved the same phase response across the 24-hour

There are two ERS tandem interferograms available from 1997 (25-26 September and 10-11 October) during periods of intense explosive (Vulcanian) activity (Druitt et al. 2002) and two from 1999 (4-5 March and 8-9 April) when there was no activity in the March case and only minor activity in the case of the small pyroclastic flow of 8 April (Norton et al. 2002). As we might expect from this contrast in activity, the 1999 pairs have better general coherence, partly because there are more surge deposits on the SW of the volcano as a result of the 26 December 1997 sector collapse, and partly because there were almost no new volcanic deposits or less airborne ash to decorrelate the scattering properties of the surfaces. The forested areas of Centre Hills and South Soufriere Hills still have no phase information. Figure 7 shows the interferogram of the ERS tandem pair from 4 and 5 March 1999. The phase fringes effectively represent the topography as contour intervals with a spacing of about 45m per fringe. This spacing is controlled by the baseline separation of the image pair (253 m). During the eruption no useful interferometric information was collected from the lava dome itself. The one-day period of ERS tandem acquisition on 25 and 26 September 1997. which might have yielded information, coincided with vigorous explosive activity (Druitt et al. 2002), which destroyed phase coherence in the summit area. Even in March 1999 the dome itself had a noisy phase signal over 24 hours (Fig. 7), though there is information right up to the northern side of the dome. This is partly due to slope effects, but may also be due to surface motion, such as block or talus instabilities. Lower down the volcano the surfaces of the pyroclastic flow deposits are coherent generally and the interferogram can be used to extract useful information. To convert the interferogram to a map of topographic heights, the phase fringes must have the 2rr cycle removed. We achieve this

Table 3. Numbers of ERS interferometry data products of Montserrat: July 1997 to November 1998 Frame 315

Frame 333

Maximum possible ERS-2 images Acquired ERS-2 images* Acquired ERS-1 images*

15 10 2

14 9 1

Maximum possible interferograms from those acquired ERS-1 and ERS-2

66

45

Non-baseline-limited interferogramsf

58

37

Coherent interferograms

17

5

* Reduction in data captured for operational technical reasons. The reduction in data due to the spacecraft orbits being too far apart. Subsequent data lost because of variable ground conditions.

Fig. 5. Images of the ratio of two radar amplitude (backscattered power) scenes of southern Montserrat for four separate intervals during the eruption. (a) A JERS-1 image ratio of 24 May 1997 divided by 2 September 1996. Note that, unlike the ascending pass acquisitions of the other three images, this is a descending pass set of acquisitions with an opposite sense of distortion of the topography and the position of the dome (©Japanese Space Agency MITI/NASDA). (b) An ERS-2 image ratio of 1 November 1998/12 October 1997. (c) An ERS-2 image ratio of 7 August 1998/9 January 1998. (d) An ERS-2 image ratio of 5 March 1999/14 June 1998. The dome is marked by an asterisk and the letters are discussed in the text. Note that image (c) is from a different track (track 347) from images (b) and (d) (track 075). This factor and the manual masking of the coast leads to possible inaccurate representation of the shape of the Tar River delta. © ESA 1997, 1998, 1999.

Fig. 6. Coherence images of the ERS tandem pairs of (a) 25 and 26 September 1997 and (b) 8 and 9 April 1999. The dark ribbons of no coherence to the NE of the volcano in (a) and the single area of no coherence in the southern part of the Tar River valley in (b) have been highlighted in white and correspond to pyroclastic flow deposits emplaced during these 24-hour periods.

590

G. WADGE ET AL.

Fig. 7. Interferogram of the ERS tandem acquisitions of 4 and 5 March 1999. For this interferogram each fringe cycle of colour represents about 45 m of topographic relief. Where the data are too noisy (coherence <0.2) the amplitude image is shown. Note that the radar imaging geometry distortion has not been corrected here.

phase unwrapping using a least-squares algorithm (e.g. Zebker & Lu 1998) and calibrate the resultant field with absolute values where the height is known (e.g. where the deposit grades to the coast at sea level). Careful inspection of Figure 7 reveals places where the fringe pattern is discontinuous and the continuity of the field breaks down. This happens particularly on the steep sides of the valleys where the quality of the phase data is low. Spurious values propagate from these discontinuities and so we have to edit the results manually. Also we might expect the coastline to exhibit the same fringe colour. However, the coastline of Montserrat has cliffs of different heights in many places and this affects the appearance of the fringes adjacent to coast in Figure 7. In Figure 8 we see the result of subtracting the topographic surface of the northern and eastern part of the volcano created with the interferogram of 4 and 5 March 1999 from the same topographic surface derived from the 1:25 000 scale contour map of the volcano prior to the eruption. This gives the thickness of the deposits created during the eruption. Because of difficulties in unwrapping the surface across noisy areas we split the area into two and unwrapped and calibrated each separately. The most striking feature of this is the sinuous infilling of the northeastern valley

system (Tyre's, Mosquito, Tuitfs, Paradise and Pea ghauts) by deposits up to 85 m thick. The apex of the delta has deposit depths of about 75m above sea level near the former shoreline. The area covered by the interferometric data shown here is not a map of the pyroclastic flow deposits and the negative values on the western side of the hills in the northernmost part of the region may be spurious. The uncertainty of the interferometry-derived topography is probably of the order of 10m root mean squared. This result can be compared with the equivalent change in topography measured by photogrammetry from aerial photographs taken on 8 February 1999 (Wadge, 2000) (Fig. 8). This is of similar accuracy to the interferometric digital elevation model (DEM). based on a comparison with the pre-emption DEM in areas of no known change, though there is a tendency for locally higher levels of inaccuracy across valleys because of a tendency by the photogrammetric algorithm to smooth the surface. Overall the results are generally similar, with the thickest deposits being found in Mosquito and Tuitf s Ghauts, the Tar River delta and the upper Tar River valley. However, there are some discrepancies. The thicknesses of the deposits derived interferometrically are generally greater by about 10-15m (e.g. the apex of the Tar River delta). This may be due

SPACEBORNE RADAR MEASUREMENTS

591

Fig. 8. Upper left: pyroclastic flow deposit thickness map for the NE slopes of the volcano created by differencing the unwrapped ERS tandem interferogram of 4 and 5 March 1999 from a preemption DEM created from the 1: 25 000 scale topographic mapping. Note that because of phase discontinuities north of the Tar River valley, the two regions had their phases unwrapped and heights calibrated separately. The extent of the Tar River delta in March 1999 shown by the dashed line is based on the amplitude and coherence images. The interferometric processing involves a filtering and masking operation that creates a buffer of null values at the thin edge of the delta. D, dome, TRV, Tar River valley; A, airport; MG, Mosquito Ghaut; TG, Tuitt's Ghaut; TyG, Tyre's Ghaut. Contours are at intervals of 100 m above sea level. Lower right: image of the change in topographic height (in metres) of Soufriere Hills Volcano from the start of the eruption to February 1999. Calculated from a pre-eruption digital elevation model and one created photogrammetrically from aerial photographs taken on 8 February 1999 (Wadge 2000).

to an error in height calibration. Also the deposits in the area of Tyre's Ghaut are apparently too thick over too large an area. This is probably due to local atmospheric noise.

Surface deformation Ground survey and global positioning system (GPS) measurements (Jackson et ai 1998; Shepherd et al 1998; Mattioli et al. 1998) at Soufriere Hills Volcano in 1996-1997 showed perhaps two scales of deformation. Within about 1 km of the dome and feeder conduit, non-elastic displacements of several tens of centimetres were recorded. During late 1997, tiltmeter records from Chance's Peak at the summit of the volcano showed cyclic, recoverable tilting every few hours, probably caused by very shallow (< 1 km) pressurization processes (Voight et al. 1998). The evidence for far-field deformation

due to deeper processes associated with a crustal magma chamber is less clear. GPS results are difficult to interpret because there are relatively few points, particularly to the south and east, within a limited distance from the volcano. They seem to show only minor displacements (a few centimetres) over periods of weeks to months, with little long-term consistent spatial pattern (Shepherd et al. 1998; Mattioli et al. 1998). Differential radar interferometry with ERS data is capable of measuring relative surface motion along the line-of-sight of the radar (about 23° from the vertical) at about the 1 cm level or better. To achieve this, the phase effect due to topography must be explicitly modelled and removed from the interferogram. Any residual fringes can be assumed to be due to relative surface motion between the image acquisitions. Figure 9 shows one such differential interferogram of the northeastern slopes of the volcano for a 36-day period: (2 July-7 August 1998). The topography derived from the 4 and 5 March 1999 ERS tandem pair (Fig. 8) has been used to

592

G. WADGE ET AL.

Fig. 9. Differential interferogram showing results from the northeastern slopes of the volcano for the period 2 July 7-August 1998. The topographic effect has been removed using the tandem ERS interferogram from March 1999. The colour cycle represents 28 mm, one half-wavelength of the ERS SAR, and the sense of relative movement is for increasing distance from the satellite to the volcano from red to green to blue. D, dome. Contours are at intervals of 100 m above sea level.

remove the topographic effect. There are no reported pyroclastic flows from this period and it is assumed that there have been negligible (< 10 m) changes of topography in this area of the volcano in the interlude. Figure 10 shows about one residual fringe remaining over most of the northeastern slopes, which means that much less than 28 mm (half the C-band wavelength) of motion relative to the satellite can have occurred. There is a residual anomaly to the north of volcano and also some evidence of a depth-related anomaly in the main river valleys. This latter feature could be due to

Fig. 10. Plot of the degree of coherence of the ERS SAR interferometric pairs from 1997 to 1999 over Montserrat. Coherence decreases, as predicted theoretically, with increasing perpendicular baseline separation. The decrease of coherence with period between each pair of images for this data set is distinctive and probably caused by the rate of erosion of the surfaces of the pyroclastic flow deposits.

inaccurate topographic correction or to relative subsidence of the deposits (cf. Stevens et al. 2001). Atmospheric noise may also play a part. This finding of less than 2-3 cm of vertical component of motion in 36 days at distances of 1-6 km from the dome is consistent with the GPS results of the far-field deformation. However, careful consideration of atmospheric, topographic and volcanological factors needs to be made before the value of this as a surface deformation monitoring technique at Soufriere Hills Volcano can be evaluated. Work is underway to this effect.

SPACEBORNE RADAR MEASUREMENTS

Discussion Although we do not have data from the earliest part of the eruption in 1995-1996, it is clear that it is the destruction of the vegetated valleys and lower slopes that has provided the ability to extract radar phase information at Soufriere Hills Volcano. This enabled the mapping of the thickness of the pyroclastic flow deposits. The measurement of any trans-eruption deformation field using radar interferometry, such as that achieved on largely vegetation-free basaltic volcanoes (e.g. Jonsson et al. 1999; Sigmundsson et al. 1999) is precluded by the lack of coherent, pre-eruption surfaces. The discovery that the surface of the new lava dome does not, generally, preserve phase over intervals of one day and greater is perhaps not surprising, but partial coherence could not be ruled out a priori. If radar interferometry is to capture the rapidly changing topography of an active lava dome routinely from space, then it must be done with a single-pass system with a frequent revisit capability (days to week) and high vertical resolution (c. 10m). Eventually, the dome surface will become sufficiently stable to preserve radar phase. When and where this first happens may supply valuable information on the thermomechanical behaviour of lava domes, in the way that is being explored for lava flows elsewhere (e.g. Stevens et al. 2001). We noted, in relation to the brightening of the amplitude images with time, the likely role of rain-induced erosion/resedimentation of the pyroclastic flow deposits. This is also a limiting factor for SAR interferometry. Figure 10 shows a plot of the coherence of interferometric pairs from 1997 to 1999. The tripartite classification of the interferograms is subjective, but clearly shows the loss of coherence with time and with perpendicular baseline separation. Generally, the pyroclastic flow deposits become incoherent after about 250 days and lose much of their coherence after 175 days. This is almost certainly due to erosion. For surface deformation studies where enough time must elapse to generate a measureable strain, this can be a serious limitation. Considerably more volcanological information can be extracted from this evolving data set. An important issue not addressed here is the operational utility of radar from the perspective of a volcano observatory. Can these data be received, processed and analysed fast enough to be of use in decision-making? Wadge et al. (1999, 2000) make it clear that the 'normal' data reception route for ERS SAR data via the regional ground receiving stations is too slow for timely input to the scientific decision-making process at MVO during periods of crisis. The future prospect for radar monitoring of volcanoes is likely to improve as more advanced radar systems are launched in the next few years, data become more accessible, and the processing easier to manage.

Conclusions (1)

Spaceborne radar data obtained from 1996 to April 1999 contain useful information on the changes associated with the contemporary eruption of Soufriere Hills Volcano. (2) Gross topographic changes during the eruption can be mapped qualitatively using ratio images of the radar amplitude data. (3) SAR interferometry with ERS data separated by one day produces images of phase coherence that can be used to map the deposits from volcanic events. We identified the outline of pyroclastic flow deposits emplaced during 25-26 September 1997 and 8-9 April 1999. (4) SAR interferometry allows quantitative measurements of flank topography to be made for areas denuded of thick vegetation. We show that the valleys on the northeastern flank of the volcano have up to 85 m of deposit, similar to results obtained with airborne photogrammetric data. (5) The topography of the new lava dome of Soufriere Hills Volcano cannot be mapped routinely with ERS data because the phase becomes incoherent over 24 hours. This is probably due to the mobility of surface blocks on the dome.

(6)

593

Surface deformation measurements can also be collected using SAR interferometry with images acquired weeks to years apart. However, the pyroclastic flow deposits, which are the main source of data, decorrelate after 100-200 days, probably mainly due to erosion. This reduces the potential amount of data for surface deformation monitoring considerably.

We thank our project colleagues A. Smith, M. Palmer, S. Lomas-Clarke, C. Riley, D. Rothery, P. Francis and S. Blake for their support. We are endebted to numerous colleagues associated with MVO for their longterm help and encouragement in applying the technique in this challenging environment. G.W. is grateful to ESA for supplying ERS data under ERS AO3-269 data grant. We are also grateful to P. Mouginis-Mark and NASA for supplying Radarsat data for 1997-1998 and to M. Haynes and NPA Group for the use of Radarsat data from 1996 under the CivInSAR programme. G.W. and N.F.S. acknowledge support under NERC grant GR3/ 11685, and G.W. and B.S. acknowledge support from the BNSC Earth Observation LINK programme R4/008. The manuscript benefited greatly from reviews by P. Mouginis-Mark, an anonymous reviewer and, particularly, T. Druitt.

References ADAMS, J., PERGER, W. F., ROSE, W. I. & KOSTINSKI, A. 1996. Measurements of the complex dielectric constant of volcanic ash from 4 to 19 GHz. Journal of Geophysical Research, 101, 8174-8185. BAMLER, R. & HARTL, J. 1998. Synthetic aperture radar interferometry. Inverse Problems, 14, R1-R54. COLE, P., CALDER, E. S., DRUITT, T. H., HOBLITT, R., ROBERSTON, R., SPARKS, R. S. J. & YOUNG, S. R. 1998. Pyroclastic flows generated by gravitational instability of the 1996-97 lava dome of Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 4325-3428. COLE, P. D., CALDER, E. S., SPARKS, R. S. J. ET AL. 2002. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 231-262. DRUITT, T. H., YOUNG, S. R., BAPTIE, B. ET AL. 2002. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306. FRANCIS, P. W., WADGE, G. & MOUGINIS-MARK, P. J. 1996. Satellite monitoring of volcanoes. In: SCARPA, R. & TILLING, R. I. (eds) Monitoring and Mitigation of Volcano Hazards. Springer, Berlin, 257-298. HENDERSON, F.M & LEWIS, A. J. (eds) 1998. Principles and Applications of Imaging Radars. Manual of Remote Sensing, Vol. 2 (third edition). Wiley, New York. JACKSON, P., SHEPHERD, J. B., ROBERTSON, R. E. A. & SKERRITT, G. 1998. Ground deformation studies at Soufriere Hills Volcano, Montserrat I: electronic distance meter studies. Geophysical Research Letters, 25, 3409-3412. JONSSON, S., ZEBKER, H., CERVELLI, P., SEGALL, P., GARBEIL, H., MOUGINISMARK, P. & ROWLAND, S. 1999. A shallow-dipping dike fed the 1995 flank eruption at Fernandina volcano, Galapagos, observed by satellite radar interferometry. Geophysical Research Letters, 26, 1077-1080. Lu, Z. & FREYMUELLER, J. T. 1998. Synthetic aperture radar interferometry coherence analysis over Katmai volcano group, Alaska. Journal of Geophysical Research, 103, 29 887-29 894. MASSONNET, D. & FEIGL, K. L. 1998. Radar interferometry and its application to changes in the Earth's surface. Review of Geophysics, 36, 441-500. MASSONNET, D., BRIOLE, P. & ARNAUD, A. 1995. Deflation of Mount Etna monitored by spaceborne radar interferometry. Nature, 375, 567-570. MATTIOLI, G. S., DIXON, T. H., FARINA, F., HOWELL, E. S., JANSMA, P. & SMITH, A. L.1998. GPS measurements of surface deformation around Soufriere Hills volcano, Montserrat from October 1995 to July 1996. Geophysical Research Letters, 25, 3417-3420. MVO (MONTSERRAT VOLCANO OBSERVATORY) Report of the explosive event of September 17-18 1996. Special Report 01. http://www.geo.mtu.edu/ volcanoes/west.indies/soufriere/govt/specrep/specrepO 1 .html NORTON, G. E., WATTS, R. B., VOIGHT, B. ET AL. 2002. Pyroclastic flow and explosive activity at Soufriere Hills Volcano, Montserrat, during a

594

G. WADGE ET AL.

period of virtually no magma extrusion (March 1998 to November 1999). In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London. Memoirs. 21. 467-481. REA, W. J. 1974. The volcanic geology and petrology of Montserrat. West Indies. Journal of the Geological Society, London, 130, 341-366. SHEPHERD. J. B., HERD. R. A.. JACKSON, P. & WATTS, R. 1998. Ground deformation measurements at Soufriere Hills Volcano, Montserrat II: rapid static GPS measurements June 1996 June 1997. Geophysical Research Letters, 25, 3413-3416. SIGMLNDSSON. F., DURAND. P. & MASSONNET, D. 1999. Opening of an eruptive fissure and seaward displacement at Piton de la Fournaise volcano measured by Radarsat satellite radar interferometry. Geophysical Research Letters, 26, 533-536. SPARKS, R. S. J., YOUNG, S. R., BARCLAY, J. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills Volcano, Montserrat, West Indies: November 1995 to December 1997. Geophysical Research Letters, 25, 3421-3424. SPARKS, R. S. J., BARCLAY, J., CALDER, E. S. ET AL. 2002. Generation of a debris avalanche and violent pyroclastic density currents on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano. Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London, Memoirs, 21, 409-434. STEVENS, N. F., WADGE, G., WILLIAMS, C. A., MORLEY, J. G., MULLER, J-P., MURRAY, J. B. & UPTON, M. 2001. Surface movements of emplaced lava flows measured by SAR interferometry. Journal of Geophysical Research, 106. 11 293-11 313. VOIGHT, B., HOBLITT, R. P., CLARKE, A. B.. LOCKHART, A. B., M l L L E R ,

A. D.. LYNCH, L. & McMAHON, J. 1998. Remarkable cyclic ground

deformation monitored in real-time on Montserrat. and its use in eruption forecasting. Geophysical Research Letters. 25, 3405-3408. VOIGHT, B.. KOMOROWSKI. J.-C.. NORTON. G. E. ET AL. 2002. The 26 December ( Boxing Day) 1997 sector collapse and debris avalanche at Soufriere Hills Volcano, Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano. Montserrat, from 1995 to 1999. Geological Society. London. Memoirs. 21. 363-407. WADGE. G. 2000. A DEM of the volcanic deposits of Soufriere Hills Volcano during 1999. MVO Open File Report OFR 00-07. WADGE. G. & HAYNES. M. 1998. Radar images growth of Soufriere Hills Volcano. Montserrat. International Journal of Remote Sensing. 19. 797-800. WADGE. G. & ISAACS. M. C. 1988. Mapping the volcanic hazards from Soufriere Hills Volcano. Montserrat. West Indies using an image processor. Journal of the Geological Society, London. 145. 541-551. WADGE. G.. SCHEUCHL. B.. CABEy. L.. PALMER. M. D.. RILEY. C.. SMITH. A. & STEVENS. N. F. 1999. Operational use of InSAR for volcano observatories: experience from Montserrat. Proceedings FRINGE'99. ESA Publication SP478. Noorwijk. Netherlands. WADGE. G.. CABEY. L.. LOMAS-CLARKE. C.. PALMER. M. D. & SMITH. A. 2000. Operational use of InSAR for volcano observatories: experience in local reception at Montserrat. Proceedings ERS-EMVISAT Symposium, 16-20 Oct. 2000. Gothenburg. Sweden. YOUNG. S. R.. SPARKS. R. S. J.. ASPINALL. W. P.. LYNCH. L. L.. MILLER. A. D., ROBERTSON. R. E. A. & SHEPHERD. J. B. 1998. Overview of the eruption of Soufriere Hills volcano. Montserrat. 18 July 1995 to December 1997. Geophysical Research Letters. 25. 3389-3392. ZEBKER. H. A. & Lu, Y. 1998. Phase unwrapping algorithms for radar interferometry: residue-cut, least-squares, and synthesis algorithms. Journal of the Optical Society of America. 15. 586-598.

The relationship between degassing and rockfall signals at Soufriere Hills Volcano, Montserrat R. LUCKETT 1 , B. BAPTIE1 & J. NEUBERG 2 1

British Geological Survey, Murchison House, West Mains Road, Edinburgh EH9 3LA, UK (e-mail: [email protected]) 2 School of Earth Sciences, University of Leeds, Leeds, LS2 9JT, UK

Abstract: One of the most common types of seismic event recorded during the eruption of Soufriere Hills Volcano from 1995 to 1999 is known as a rockfall signal because signals recorded when rockfalls were observed on the dome are of this type. Evidence is presented that two seismic sources contributed to these events. The action of falling debris on the dome generated seismicity between 2 Hz and 8 Hz, while many rockfall signals also have a marked spectral peak between 1 Hz and 2 Hz. Deployment of a pressure sensor near the volcano has shown that the 1-2 Hz energy was associated with degassing at the surface of the dome; however, the relative timing of gas escape and seismic signal showed that the first was not the direct source of the second. Resonance of the magma conduit linked to degassing at the surface is invoked as a probable source for the 1-2 Hz seismicity. The relative importance of the two seismic sources that contributed to rockfall signals is examined in the context of the behaviour of the volcano.

The eruption of Soufriere Hills Volcano on Montserrat in the West Indies started in July 1995 and a lava dome was extruded between November of that year and March 1998. The Montserrat Volcano Observatory has monitored volcanic seismicity since the start of the eruption using first one and then two networks of seismometers. The data in this paper were recorded by the second of these networks (Fig. 1), installed in October 1996 and known as the 'broadband array', because it includes several broadband seismometers (Neuberg et al. 1998). The broadband seismometers are Guralp CMG-40Ts with an effective bandwidth of 0.03 Hz to 30 Hz and the 1 Hz seismometers are Integra LA 100s with an effective bandwidth of 1-30 Hz. The data are telemetered in 24-bit digital format, giving a dynamic range of 145 dB. In December 1998 a Setra 276 pressure transducer was installed at seismic station MBLG; this can measure fluctuations in barometric pressure as small as 300 Pa. The seismic and pressure data are recorded by the same acquisition system and so the detection algorithm, which uses the seismic data to distinguish discrete events, includes the pressure trace in the resulting event files. In addition, all data are recorded continuously, allowing longer time windows to be examined when required. Various seismic signals were recorded during the 1995-1999 eruptive period (Miller et al. 1998) and these have been grouped into classes according to their shape, frequency content, duration, etc. One of the most common types of signal was classified as a 'rockfall signal' because such signals were recorded when rockfalls were observed on the dome. In this paper possible source mechanisms for these signals are considered, including interaction between the rockfall and the dome surface, degassing at the surface, and resonance within the dome. Comparison between seismic data and pressuresensor data is a vital tool in the separation of these possible sources. For the purposes of this paper the term long period' is defined as describing seismicity with frequencies between 1 and 2 Hz. This is to be consistent with existing Montserrat Volcano Observatory terminology and to allow the authors to refer to events classified on Montserrat as long-period rockfalls without confusion. This use of long period is specific to this paper and it is not equivalent to the term 'low frequency' used to describe Montserrat seismicity by Neuberg et al. (2000) or Baptie et al. (2002).

Description of rockfall signals Throughout the growth of the dome at Soufriere Hills Volcano, portions of dome material separated from the main edifice and avalanched down the volcano (Calder et al 2002; Cole et al. 2002). These rockfalls could be as small as a single block bouncing down the dome but more often consisted of many rocks obscured by a cloud of ash. At night individual rocks could be more clearly identified, as they were hotter and therefore brighter than the ash cloud.

In the most extreme cases pyroclastic flows resulted, with runout distances of several kilometres, but most rockfalls ended up as part of a growing talus slope at the base of the dome. Rockfalls continued during the period of no dome growth between March 1998 and November 1999 (Norton et al. 2002). Whenever a rockfall was observed, a signal was recorded on nearby seismometers, although this was not always of such a nature as to trigger the detection algorithm. Similar signals were classified as 'rockfall signals' even if no visual confirmation was available that a rockfall was taking place, either because there were no observers in place or because the rockfall was on the far side of the dome from whoever was watching. Rockfall signals (Fig. 2) are emergent; that is, the amplitude increases gradually at the onset of the signal, and have durations ranging between a few tens of seconds and many minutes. Rockfall signals can be divided into three subclasses based on waveform and frequency content, as explained in detail below. These subclasses are rockfalls with energy only above 2 Hz, rockfalls with an additional spectral peak between 1 and 2 Hz, and rockfalls where the 1-2 Hz energy starts before the rest of the signal, resulting in a precursory phase. The frequency content of rockfall signals is dominated by site effect, as can be seen by observing the difference in spectra at different stations (Fig. 3). However, in all cases most energy is concen trated between 2 and 8 Hz. Many events also have a peak in spectral amplitude between 1 and 2 Hz at all stations. In some cases this lowerfrequency part of a rockfall signal starts before the higher-frequency component, resulting in a precursor similar in frequency and waveform to the events called long-period earthquakes on Montserrat (Miller et al. 1998). Such events were classified as 'long-period rockfalls' by the staff at the observatory and, to be consistent with this established terminology, the frequency band between 1 and 2 Hz will from now on be described as 'long-period'. In other rockfalls, long-period energy is present in the spectra of a given event but there is no precursor in the time domain. There are thus three subclasses of rockfall signals: those with energy only between 2 and 8 Hz (Fig. 4a); those with an additional component between 1 and 2 Hz (Fig. 4b); and long-period rockfalls, where this long-period component starts before the rest of the signal (Fig. 4c). It was demonstrated using a video camera that there was no movement on the dome during the long-period precursors to longperiod rockfalls, but that rocks started moving at approximately the same time as the onset of the high-frequency signal. In some cases, however, a puff of ash or steam was observed to correspond to the long-period phase and to come from the same part of the dome as the rockfall. Later in the 1995-1999 period these venting episodes were recorded using a pressure sensor, as described in the next section. Similar signals to long-period rockfalls, although of a different scale, were associated with the Vulcanian explosions that occurred on Montserrat in August, September and October 1997 (Fig. 5).

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 595-602. 0435-4052/02/$ 15 The Geological Society of London 2002.

595

596

R. LUCKETT ET AL.

Fig. 1. Montserrat. including locations of the seismic stations and pressure sensor from which data are presented in this paper. The contour lines are at 100 m intervals.

Each explosion (Druitt et al. 2002) started with a vertical jet of material emerging from a crater near the top of the dome. Within a few seconds of this being observed, an approximately 1 Hz seismic signal started to be recorded. This signal continued for 10-20 seconds in isolation while the eruption column collapsed, then a higher-amplitude signal, from the resulting pyroclastic flows, was recorded in addition. When the pyroclastic flows came to rest the long-period signal could be seen to be continuing as tremor. This tremor was associated with ash-venting from the crater and lasted for up to several hours after the explosion. The seismic signals associated with the Vulcanian explosions are discussed further by Druitt et al. (2002). Analysis of rockfall signals If rockfall signals were caused purely by dome instability then, assuming a simple model where rockfalls occur as soon as the dome's slope becomes unstable, one would expect a correlation between the number of rockfall signals recorded and the speed of dome growth.

There is no such correlation, as can be seen in Figure 6, where the number of rockfall signals per day in 1997 is compared to the lava extrusion rate. This extrusion rate is based on estimates of dome volume using photographic and surveying techniques (Sparks et al. 1998; Sparks & Young 2002). These surveys could only be made at irregular and sometimes widely spaced intervals and so the extrusion rate is, at some times, only known very approximately. It is clear, however, that there is no correlation between extrusion rate and the number of rockfall signals. Some mechanism thus needs to be found in addition to purely gravitational forces for the triggering of rockfalls. Uhira et al. (1994) analysed seismic signals associated with pyroclastic flows on Unzen Volcano, Japan. They modelled the source of these signals as a sequence of forces resulting from removal of mass from the dome, the collision of this mass with the slope below, and the flow of the resulting fragments down the slope. At Montserrat their models explain those signals with no longperiod component and the higher-frequency component of those that do, but not the long-period energy that can precede any actual displacement of dome rock.

Fig. 2. A rockfall signal as recorded on the vertical components at the five seismic stations nearest the dome (Fig. 1). This signal was recorded at 13:21 local time on 3 May 1997.

Fig. 3. Normalized power spectra for the traces in Figure 2.

598

R. LUCKETT ET AL.

Fig. 4. Examples of the three subclasses of rockfall signals at station MBGH. In each case the upper trace is the raw ground velocity on the vertical component and the lower trace the same signal after low-pass filtering at 2 Hz to show the long-period component, (a) A rockfall signal with little energy below 2 Hz. (b) A rockfall that contains energy below 2 Hz. (c) A long-period rockfall with energy below 2 Hz present as a precursory wavelet, in addition to throughout the rest of the signal.

Fig. 5. Seismic signal associated with a Vulcanian explosion recorded at 20:57 local time on 5 August 1997. Ground velocity on the vertical component at station MBGE is shown.

ROCKFALLS AND DEGASSING ON MONTSERRAT

599

Fig. 6. The number of rockfall signals recorded each day throughout 1997 compared with a plot of lava extrusion rate (Sparks & Young 2002). The rockfalls plotted here are those triggered events defined as rockfalls by analysts at the observatory in their daily categorization of all triggered events.

Observations of the dome while long-period rockfall signals were being recorded have shown that the long-period, often almost monochromatic, precursor (Fig. 4c) is associated with the venting of gas. Once the rockfall has started, it is not possible to tell from visual observation whether gas-venting continues because too much ash is disturbed; however, the long-period energy continues throughout the event, so it is probable that degassing also continues. That the long-period energy is not contributed by the rockfall itself is indicated by the fact that other rockfalls on the same region of the dome correspond to seismic signals with very little such long-period energy, implying that the seismicity caused by the action of rocks at the surface does not include energy in this band. Many rockfall signals other than those of long-period rockfalls contain a large proportion of long-period energy, suggesting that many rockfalls are associated with gas-venting. The difference between these events and long-period rockfalls is only in the relative timing of the two phases. An association between rockfall signals containing long-period energy and gas-venting was demonstrated in December 1998 when a pressure sensor was installed as part of the seismic network. Figure 7 shows (a) a typical rockfall signal, (b) the same signal low-pass filtered with corner frequency of 2 Hz, and (c) the corresponding acoustic signal. As in this example, the delay between the start of the seismic signal and the pressure pulse is generally greater than 10 seconds. The dome and the sensor are 2 km apart, meaning that an acoustic wave would reach the sensor in less than 6 seconds. Therefore, although the two are obviously linked, the source of

the seismic signal is not gas-venting through a narrow opening at the surface. Neuberg et al. (2000) discuss long-period seismicity at Soufriere Hills Volcano and conclude that the source is resonance of interface waves at the walls of a fluid-filled conduit. The trigger for this resonance is unknown, but a possible explanation is that gas is injected into the conduit at depth or begins to escape from the conduit into cracks in the surrounding dome. Neuberg et al. (2000) also show how such a resonance would be focused by the shape of the dome to cause appreciable vibration at the surface. It is possible that, once such resonance started, it would increase the rate of degassing in the conduit (Hellweg 2000), introducing a feedback effect. In the case of long-period rockfalls the resonance occurs before the rockfall, as does gas-venting. However, because no pressure sensor was deployed when long-period rockfalls were being recorded, the relative timing of the long-period phase and gas release is unknown. The degassing and/or the seismic disturbance therefore start the rockfall. In contrast to long-period rockfalls, the long-period and the higher-frequency seismicity often start, as far as can be discerned from the seismograms, at the same time (Fig. 4b). If the same source is assumed as for long-period rockfalls, then the resonance must also start first in these events. If this resonance started slowly then the resulting seismic disturbance at the surface could cause a rockfall to start gradually. As both signals would be emergent it would not be possible to resolve which is recorded first. To accept

600

R. LUCKETT ET AL.

Fig. 7. A rockfall signal recorded at 14:34 local time on 19 December 1998 at station MBLG and the pressure-sensor recording at the same station, (a) The raw ground velocity on the vertical component, (b) The same signal low-pass filtered with a corner frequency of 2 Hz. (c) Atmospheric pressure, as recorded a few metres from the seismometer. Dashed lines have been added to show the time lag between the start of the seismic signal and the start of the pressure signal, in this case approximately 12 seconds. 1 hPa = 100 Pa.

this model it is necessary to explain why the long-period phase of long-period rockfalls does not cause a rockfall to occur until after a time interval of several seconds. One explanation is that longperiod rockfalls occur when the surface of the dome is temporarily more stable, at least locally. If two separate sources, one contributing energy mainly below 2 Hz in frequency and the other contributing mainly higher-frequency energy, combine to generate many rockfall signals, then it is interesting to monitor the spectral content of rockfall signals over time. Power spectra have been calculated for all the rockfall signals recorded between late 1996 and early 1999 and the energy content of each event divided into the proportion below 2 Hz and that above. Figure 8 shows the percentage of energy below 2 Hz averaged over each day for rockfalls in 1997, when dome growth was at its greatest. The rockfall data are compared with the lava extrusion rate, as in Figure 6. Times of highest dome extrusion rate correspond approximately to times when the greatest proportion of rockfall seismicity was below 2 Hz in frequency. The main exception to this is the peak in extrusion rate in January (days 0-30), when rockfalls generally comprised high-frequency energy. This can be explained by the type of growth at the volcano at this time, in the so-called 'Galway's Wall episode'. At the end of 1996 and the beginning of 1997 the dome was growing in the southwestern edge of English's Crater and rockfalls were occurring on the outside of the corresponding crater wall. No new material was involved in these rockfalls, which occurred on a crater wall being deformed as the growing dome pushed against it. The rockfalls were often triggered by near-surface earthquakes (Watts et al. 2002). Apart from this, the indication is that rockfall signals contain a greater element of resonance when the dome is growing fastest. This can be explained by the fact that when the magma in the conduit is fresher it is more gas-rich, leading to a

greater impedance contrast between the conduit walls and the magma. This in turn leads to the interface waves resonating in the conduit walls having greater amplitude. Thus the greater the rate of magma entering the conduit, the more energetic the resonance of the conduit. A trigger is still required to start each event, however, meaning that the number of events is controlled by something other than the rate of magma ascent and can be uncorrelated with it, as is observed. Conclusions (1)

The seismic signals known on Montserrat as rockfall signals have, in general, two separate sources. The first is resonance of a fluid-filled conduit, which contributes long periods (frequencies between 1 Hz and 2 Hz). The second is a combination of forces resulting from rocks moving on the surface of the dome, which contributes higher-frequency seismicity. (2) In some cases, the long-period seismicity precedes the higherfrequency component, and it is concluded that these rockfalls are started by seismic energy focused by the shape of the dome. This could also be the case for all rockfall signals which contain a relatively large proportion of long-period energy, but in most events the two components start too close together in time for a phase interval to be discerned. (3) The long-period resonance is associated with gas escaping into the atmosphere, but pressure sensor data show that there is a time delay between the seismicity and the venting. It is suggested that resonance of the conduit is started by gas escaping from the conduit into the dome, from where it vents into the atmosphere a few seconds later.

ROCKFALLS AND DEGASSING ON MONTSERRAT

601

Fig. 8. The percentage of rockfall energy below 2 Hz at MBWH, averaged for each day's events and compared with lava extrusion rate. As in Figure 6 rockfalls are those events identified as such by an analyst as part of the daily processing at the observatory. The frequency data are calculated from fast Fourier transforms, which are automatically carried out for all events over the time window delimited by the triggering algorithm (from 2 seconds before triggering until 10 seconds after detriggering has occurred).

(4)

Long-period energy is a more important component of rockfall signals when the dome is growing fastest, and thus the magma is most gas-rich.

The authors would like the thank the staff at the Montserrat Volcano Observatory, and R. White and another, anonymous, reviewer for their helpful comments. This paper is published by permission of the director of the British Geological Survey (NERC).

References BAPTIE, B., LUCKETT, R. & NEUBERG, J. 2002. Observations of lowfrequency earthquakes and volcanic tremor at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 611-620. CALDER, E. S., LUCKETT, R., SPARKS, R. S. J. & VOIGHT, B. 2002. Mechanisms of lava dome instability and generation of rockfalls and pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 173-190. COLE, P. D., CALDER, E. S., SPARKS, R. S. J. ET AL. 2002. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B.P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 231-262.

DRUITT, T. H., YOUNG, S. R., BAPTIE, B. ET AL. 2002. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat. In: DRUITT, T.H. & KOKELAAR, B.P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306. HELLWEG, M. 2000. Physical models for the source of Lascars harmonic tremor. Journal of Volcanology and Geothermal Research, 101, 183-198. MILLER, A. D., STEWART, R. C, WHITE, R. A. ET AL. 1998. Seismicity associated with dome growth and collapse at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3401-3404. NEUBERG, J., BAPTIE, B., LUCKETT, R. & STEWART, R., 1998. Results from the broadband seismic network on Montserrat. Geophysical Research Letters, 25, 3661-3664. NEUBERG, J., LUCKETT, R., BAPTIE, B. & OLSEN, K., 2000. Models of tremor and low-frequency earthquake swarms on Montserrat. Journal of Volcanology and Geothermal Research, 101, 83-104. NORTON, G. E., WATTS, R. B., VOIGHT, B. ET AL. 2002. Pyroclastic flow and explosive activity at Soufriere Hills Volcano, Montserrat, during a period of virtually no magma extrusion (March 1998 to November 1999). In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 467-481. SPARKS, R. S. J. & YOUNG, S. R., 2002. The eruption of Soufriere Hills Volcano, Montserrat (1995-1999): overview of scientific results. In: DRUITT, T. H. & KOKELAAR, B.P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 45-69.

602

R. LUCKETT ET AL.

SPARKS. R. S. J.. YOUNG, S. R.. BARCLAY. J. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills Volcano. Montserrat. West Indies: November 1995 to December 1997. Geophysical Research Letters, 25. 3421-3424. UHIRA, K.. YAMASOTO. H. & TAKEO. M.. 1994. Source mechanism of seismic waves excited by pyroclastic flows observed at Unzen volcano. Japan. Journal of Geophysical Research. 99, 17 757-17773.

WATTS. R. B.. HERD. R. A.. SPARKS. R. S. J. & YOUNG. S. R. 2002. Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano. Montserrat. from 1995 to 1999. Geological Society. London. Memoirs. 21. 115-152.

A model of the seismic wavefield in gas-charged magma: application to Soufriere Hills Volcano, Montserrat J. NEUBERG & C. O'GORMAN School of Earth Sciences, The University of Leeds, Leeds LS2 9JT, UK (e-mail: [email protected])

Abstract: A stationary two-phase magma model is used to derive the relationship between gas mass and volume fractions, gas and magma bulk densities, temperature and pressure as a function of depth. In turn, these parameters are used to obtain the vertical seismic velocity profile in the gas-charged magma. A two-dimensional finite-difference model of a magma-filled conduit embedded in an elastic medium is then employed to generate the seismic wavefield in and around the conduit. The high impedance contrast between gas-rich magma and surrounding rock results in the seismic energy being efficiently trapped in the conduit; this leads to the generation of a long-lived resonance of tens of seconds commonly observed as low-frequency earthquakes and harmonic tremor. During a single seismic event, a variety of different seismic radiation patterns along the conduit is observed, leading to the occurrence of several distinct seismic phases in the synthetic seismograms. Observations from several volcanoes show peaked amplitude spectra with integer harmonic overtones that exhibit a time-dependent gliding. These features are successfully modelled by varying the excess pressure and, consequently, the gas volume fraction and the seismic velocity, representing sudden degassing events, such as the reduction of pressure by ash-venting, a dome collapse or a Vulcanian explosion, or, in turn, the pressurization of the conduit prior to such events.

Recent observations of low-frequency seismic events at volcanoes such as Semeru in Indonesia (Schlindwein et al. 1995), Arenal in Costa Rica (Benoit & McNutt 1997) and Soufriere Hills on Montserrat (Miller et al. 1998; Neuberg et al. 1998), which occur in swarms and often precede volcanic eruptions, have provoked a wide investigation into the mechanism of these events. As in Baptie et al. (2000) and Luckett et al. (2002) we refer here to low-frequency seismic events and do not distinguish, within the family of lowfrequency events, between long-period events (LPs) and so-called hybrid events, since for Soufriere Hills on Montserrat there is a continuum linking LPs and hybrids (Neuberg et al. 2000). Lowfrequency events have a dominant frequency ranging from 0.2 Hz to 5 Hz and represent a resonating physical system comprising a fluid embedded in a solid medium (Ferrazzini & Aki 1987; Chouet 1988). The seismic wavefield is generated at the boundary between the fluid and the solid in the form of interface waves that allow the seismic energy trapped in the fluid to leak into, and propagate through, the elastic medium. Hence, the seismic wavefield of LPs, as observed at the surface, shows a complicated interference pattern of waves originating from various parts of the fluid-filled conduit or dyke, and interacting with the free surface and interfaces in the volcanic edifice. Simple resonator models (e.g. Hurst 1992; Seidl et al. 1981) based on point sources are, therefore, inadequate to describe such a wavefield. Models of resonating (fluid-filled) bodies with corresponding eigenfrequencies (so-called 'organ pipe modes'; Schlindwein et al. 1995) are equally misleading, as their resonant behaviour is merely based on the acoustic velocity of the body, rather than the interaction between fluid and elastic medium. The generation of the wavefield due to the leakage of seismic energy through the fluidelastic boundary is fundamentally different from a resonating body with an isotropic radiation pattern. The characteristics of the wavefield are controlled by the contrast of elastic parameters across the interface, as well as the width of the conduit compared to the seismic wavelength. Therefore, it is crucial for the interpretation of the seismic signals to understand the behaviour of the elastic parameters in various parts of the conduit, particularly in the presence of a free gas phase. Based on observations from Arenal volcano, Costa Rica, Benoit & McNutt (1997) propose a one-dimensional resonator model where changes in the gas bubble concentration control the acoustic velocity inside the conduit, and, therefore, the resonance behaviour or the seismic source. They propose that seismic energy is radiated from a displacement antinode of the standing seismic wave in the conduit. The essential difference from our model is that their tremor frequencies are controlled by the length of the conduit and the acoustic velocity, again 'organ pipe modes', while the seismic

wavefield in our model is generated by interface waves along the conduit wall. As shown by Ferrazzini & Aki (1987), the width of the conduit, as well as the elastic parameters, determine the dispersion properties of the wavefield and therefore the frequencies. A seismometer at the surface records only that part of the seismic energy which leaks out of the conduit as interface waves; the resonant part trapped in the conduit cannot be directly observed. In this aspect lies the fundamental difference between 'organ pipe' models, which interpret the seismic signal directly as the resonance inside the conduit, and models like the one we propose here where pressure perturbations inside the conduit go through a 'filter' when leaking out as interface waves. Low-frequency earthquakes on volcanoes exhibit particular features that make them indicative tools for the internal stage of volcanic activity (Lahr et al. 1994; Chouet et al. 1994; Chouet 1996; Voight et al. 1998; Neuberg et al. 1998). One striking feature is their close relationship to volcanic tremor, leading to two different classes of models: (i) volcanic tremor is generated by the repetitive triggering of LPs, yielding peaked frequency spectra with integer harmonics of degree 10 and higher; or (ii) tremor is a long-lived resonance that is triggered only once. The seismic energy is efficiently trapped in the conduit/dyke and released over a time of several tens of seconds. A closely related observation depicted in Figure 1 is the change of the amplitude spectrum with time. In this example from Montserrat up to eight gliding spectral lines can be identified between larger seismic events. In terms of the tremor models mentioned above, the time dependence indicates (i) a change in the trigger rate of the repetitive triggering model, or (ii) a change in the conduit parameters for the sustained resonance, respectively. A further investigation of these two tremor models (repetitive triggering or long-lived resonance) provides one of the motivations for this paper. Here we do not intend to focus on magma dynamics but merely on the impact of pressure changes on the seismic impedance contrast across the conduit walls, and, therefore, on the seismic wavefield. This study is motivated by observations from Montserrat, but results and conclusions are much more general in linking (changes of ) seismic signals directly to (changes of) magma properties. In the next section we focus on a stationary gas-charged magma model and determine pressure, gas content and density as a function of depth. These parameters are then used to determine the seismic velocity and impedance contrast across the conduit interface as a function of depth. Once the depth profiles of the relevant physical properties are established, we use in the following section a two-dimensional finite-difference method to simulate the seismic wavefield of such a conduit, filled with a gas-charged magma. To mimic a degassing event, we change the physical properties in

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 603-609. 0435-4052/02/$ 15 The Geological Society of London 2002.

603

J. NEUBERG & C. O'GORMAN

604

Fig. 1. Tremor example from Montserrat. seismic broadband station MBGA. 12 December 1997; vertical component; horizontal distance to dome is 2 km. Spectrogram: window length 27 seconds, overlap 6.8 seconds. Note the gliding of spectral lines in quiet tremor episodes. Up to eight integer harmonics can be identified in the spectrum.

the conduit accordingly, and study the impact on the seismic signals. In the final section we discuss the two tremor models in the light of the modelling results, and speculate about different trigger mechanisms leading to the random or repetitive excitation of seismic low-frequency events.

lava dome sitting on top of the magma column, a small collapse of which can then be represented by a decompression step - Pe. In a fixed volume at depth = the fractions of gas and liquid are given by: (2)

Physical parameters in gas-charged magma In order to obtain the depth-varying impedance contrast along the conduit walls, we need to determine the interaction between pressure, gas content and density in a gas-charged magma. No flow properties of the magma are taken into account because, for the short time the seismic waves travel in the conduit and the surrounding rock, steady-state conditions can be assumed. This assumption is only applicable to highly viscous magmas of rhyolitic and andesitic type with small discharge rates, i.e. a flow velocity much smaller than the seismic velocity. Additional dynamic pressure components due to internal magma processes or gas loss through the conduit wall are not taken into account. The depth-dependent pressure is given by: P(z) = Pe +

gp(z') di' Jo

(1)

where g is the acceleration due to gravity, and p(z) is the bulk density of the gas-liquid mixture. Pe is the excess pressure due to a

where p( is the (constant) density of liquid magma. pg the density of gas, and neg the exsolved gas mass fraction. The gas density is controlled by the gas law: (3)

with R being the gas constant, m the molecular mass of the relevant gas (SO2. CO2 or H 2 O), P the depth-dependent pressure, and T the magma temperature. If ntol is the total volatile mass fraction of exsolved and dissolved gas. and ndg the mass fraction of dissolved gas described by the solubility law (Shaw 1974):

(4) -6

-1/2

where the constant s = 4.\ x 10 Pa was calibrated experimentally by Shaw (1974) on a rhyolitic liquid with water as the only volatile in the system, then the amount of exsolved gas affecting the elastic parameters of the magma-gas mixture is given by: (5)

SEISMIC WAVEFIELD IN GAS-CHARGED MAGMA

605

The volume change: (10)

is absorbed by both liquid and gas phases: (11)

according to their bulk moduli: (12)

and with the gas volume Vg = xV and the liquid volume V( = (1 - x)v we obtain for the bulk modulus of the mixture:

(13) With Equation 7 we reintroduce the exsolved gas mass fraction of exsolved water neg and obtain:

(14) and by multiplying by p and taking the square root we obtain the inverse acoustic velocity I/a: Fig. 2. Physical parameters in a gas-charged magma in a conduit of 2000 m length. Note the high gas volume fraction in the upper 500 m of the conduit. The dashed line indicates the exsolution level.

(15) and therefore:

and therefore for small water fractions:

(16)

(16)

(6)

For the determination of the seismic properties in the gas-liquid mixture the relevant parameter is the volume fraction occupied by the gas, X, rather than its mass fraction. With M, Mg, V and Vg denoting mass and volume of bulk mixture and gas, respectively, Mg = Vgpg and therefore neg M = xVp g . The gas volume fraction x can be derived from the mass fraction neg weighted by the ratio of bulk and gas density: (7) Pg

Pressure, densities and the mass fraction of exsolved gas are dependent on each other. Hence, we use an iterative approach and start at the (linear) lithostatic pressure profile and work out gas content and densities, which are used in further iterations to determine the depth-dependent pressure in Equation 1. After not more than ten iteration steps the scheme converges and we obtain pressure, density, gas mass fraction and gas volume fraction as a function of depth. Figure 2 shows an example for arbitrary values of excess pressure of 5 MPa, a total water content of 2%, and a density of liquid magma of 2300 kg m-3.

Seismic velocity in bubbly liquids The seismic (acoustic) velocity in the gas-charged magma is given by:

(8) where B is the bulk modulus of the magma (liquid and gas) with bulk density p, representing its incompressibility or resistance to a pressure change dP:

(9)

In an approximation for a high gas volume fraction (x 1) Equation 13 reduces to the second term on the right-hand side. With the bulk modulus of gas that corresponds to the constant pressure at a given depth, Bg = P, and Equation 7 one obtains: (17)

in accordance with Wood (1932) and Leighton (1994). For very small quantities of exsolved gas (x 1) the density of the liquid dominates (p p ) and Equation 16 reduces to the acoustic velocity in the liquid:

(18) Figure 3 shows an example for the depth profiles for gas volume fraction x and seismic velocity a for a magma with total water content of 2%, a density of 2300 kg m-3 and a low excess pressure of 5 MPa, such that the exsolution level lies in the conduit. As soon as the first gas bubbles form at the exsolution level the seismic velocity decreases drastically due to the very small bulk modulus of gas. The almost linear increase of gas volume fraction in the upper part of the conduit is not matched by the seismic velocity, which remains low after the initial drop. Seismologically, this has two major implications: (i) as soon as gas is exsolved, the contrast of elastic parameters between bubbly magma and surrounding rock is huge, trapping seismic energy very efficiently in the magmatic foam; and (ii) the sharp velocity drop across the exsolution level provides an additional interface separating the conduit into two seismically different layers. The product of seismic velocity and density is usually referred to as impedance, because changes in the behaviour of seismic waves can be due to either velocity or density or a combination of both. However, it is difficult to invert seismic data unambiguously for

606

J. NEUBERG & C. O'GORMAN Table 1. Conduit parameters Density of country rock (kg m -3 ) Density of gas-free magma (kg m-3) P-wave velocity of country rock (m s -1 ) P-wave velocity of gas-free magma ( m s - 1 ) S-wave velocity of country rock ( m s - 1 ) S-wave velocity of gas-free magma ( m s - 1 ) Conduit temperature ( C) Excess pressure (MPa) Total gas mass fraction (%)

2600 2300 3000 1400 1725 0 1000 3-9 1-2

zone of exponential damping along a 35-grid-point region to limit reflections from model boundaries (Cerjan et al. 1985). The time increment used in the finite-difference code is fixed at 0.0002 s. There is no anelastic attenuation incorporated into the model. As shown by Neuberg el al. (2000). an open system permits seismic energy to escape too rapidly for resonance to occur; thus the model is closed in the sense that the fluid does not reach the surface. A plug of degassed, highly viscous magma or the existence of a lava dome correspond to such a model.

Conduit model parameters Fig. 3. (a) Schematic diagram, (b) gas volume fraction and (c) seismic velocity as a function of depth. The velocity drops drastically as soon as gas is exsolved.

velocity or density, but only for impedance. Therefore we refer to a contrast in density and/or elastic parameters across an interface as an impedance contrast. In order to understand the wavefield characteristics in and around the conduit, one has to take the direction of wave propagation into account. Due to the high horizontal impedance contrast between the conduit and surrounding rock, the seismic energy is trapped in the narrow conduit just like light in a fibreglass cable (Neuberg et al. 2000). This results in seismic wave propagation which is primarily vertical along the conduit. In contrast, seismic waves can penetrate the interface at the exsolution level much more easily due to their vertical incidence at this interface. The overall effect of the velocity drop is that seismic energy leaks into the surrounding elastic medium by very different amounts depending on the horizontal impedance contrast. This provides us with a complicated seismic source releasing different amounts of seismic energy from different parts of the conduit. These effects will be further investigated in the next section.

Application to conduit models This section describes the method used to generate synthetic seismograms for a conduit model, and the parameters chosen relevant for this process. Certain parameters are varied to examine their influence on the modelled system. The results are then compared with observational data.

The conduit with dimensions 20 m diameter by 2000 m is filled with a gas-charged rhyolitic magma surrounded by elastic rock. The gas in the magma is H2O. Locations of large, low-frequency events on Montserrat occur at about 2000 m below the base of the lava dome (Aspinall et al. 1998). but a shallow source depth for smaller events cannot be excluded. Therefore, we varied the source depth to identify corresponding changes in the seismic signature (Neuberg el al. 2002). Parameters which are appropriate for Montserrat and were used in the computation are listed in Table 1 (Aspinall et al. 1998. Sparks et al. 1998. Spera 2000).

Source type and position A Kupper wavelet with a centre-frequency of 1.0 Hz is used to represent the seismic source that triggers the system. The frequency range corresponds to the typical frequencies observed (Miller et al. 1998. Neuberg et al. 1998) although it has to be low enough to match numerical requirements imposed by the finite-difference method. If the seismic source is located in the region of the conduit where the magma is bubble-free, the energy is able to leak to the surrounding medium relatively quickly. This results in a strong first arrival, with consequent low-amplitude reverberations but a lack of resonance. Alternatively, if the seismic source is located in the foam region of the conduit, the majority of the energy is trapped due to the high impedance contrast. This results in strong repetitive resonance lasting for several tens of seconds. For this reason the position of the seismic source in the following models is fixed in the bubbly magma above the exsolution level. The source depth can vary considerably and does not have any impact on the characteristics of the resonance (Neuberg et al. 2000).

Variance of gas content and excess pressure Computation method The two-dimensional seismic response of an elastic medium to an impulsive point source is computed by a finite-difference code. The code solves the two-dimensional elastic equations of motion (Levander 1988) using a fourth-order staggered grid method. We chose a discrete grid spacing of 2 m corresponding to approximately 25 points per minimum wavelength of 50 m. We use absorbing boundary conditions at the sides and incorporate an attenuative

Once the fundamentally stable parameters for the system such as conduit size, shape and position have been established, we vary the two variables having the greatest impact on the system: the total gas content ntot and the excess pressure Pe. One should keep in mind that we focus here on the impact of pressure changes on the corresponding seismic velocity changes. Three different total gas contents of 1.5%, 1.75% and 2.0% are chosen, and their corresponding velocity profiles are calculated for a range of excess pressures from 3 MPa to 9 MPa (Figure 4).

SEISMIC WAVEFIELD IN GAS-CHARGED MAGMA

607

located in the conduit at a shallow depth of 350 m. The seismogram shows several distinct phases of different frequency contents. The snapshots in Figure 5 give a more detailed insight into how the energy of the seismic wavefield is distributed after the source has been triggered. After 1 second the wavefield is dominated by the direct P-wave and the first interaction with the free surface. The wavefronts are travelling downwards in the elastic medium faster than in the conduit. After about 8 seconds (second snapshot) most of the seismic energy propagating downwards in the conduit has reached the exsolution level and radiates waves from the conduit interface with relatively high frequency and small amplitude, controlled by the impedance contrast between bubble-rich magma and elastic rock. Most of the energy is trapped in the conduit. Once the seismic waves inside the conduit have reached the exsolution level where the impedance contrast has drastically decreased, much more energy can escape the conduit, leading to a high-amplitude wavefield with lower frequencies (third snapshot). While after 8 seconds the seismic signals have been generated by a resonating line source, after 12 seconds the part of the conduit below the exsolution level acts as a seismic point source. This leads to a distinct phase arrival in the seismogram which could easily be misinterpreted as a separate seismic event. This example demonstrates how different parts of the conduit can act as distinct secondary seismic sources where the amount and characteristic frequency of seismic energy leaking into the elastic medium is determined by the corresponding contrast of elastic parameters across the conduit wall. Particle motion analysis (not shown here) demonstrates that only the first weak P-arrival has a linear polarization, which is immediately followed by elliptically polarized waves often mistaken as surface waves. Distinct phase arrivals in the seismograms correspond to different 'source regions' along the conduitbut cannot be traced back like P-waves.

Spectral shift

Fig. 4. Seismic velocity profiles as a function of excess pressure and variable total gas content: 1.5%, 1.75% and 2.0%.

The depth of the exsolution interface increases as the total gas content of the system is raised. This increases the region of the conduit that exhibits a high impedance contrast with the country rock and enables more seismic energy to be trapped. A similar effect is observed if the excess pressure Pe is lowered, as the amount of gas exsolved is inversely proportional to the total pressure on the system. Figure 4 demonstrates that the system reacts more sensitively to pressure changes when the total gas content is higher. Due to the predominantly vertical propagation direction of the seismic waves up and down the conduit, the seismic energy is able to travel relatively easily through the vertical impedance contrast created by the exsolution interface, while the horizontal impedance contrast prevents an easy transfer of energy into the surrounding country rock. Because this horizontal contrast is higher in the foam, the bulk of the energy that leaks out originates from below the exsolution level. On Montserrat the exsolution is thought to commence in the magma chamber at 5 km depth, rather than in the conduit (Devine et al. 1998). In our numerical model such a situation can be simulated by velocity profiles that do not reach the pure liquid magma velocity of 1400 m s - 1 , as there is always an amount of exsolved gas present. This will be the case for a total gas content of 3% and above if an excess pressure of 5 MPa is assumed. The seismogram depicted in Figure 5 shows an example where several parts of the conduit act as secondary seismic sources with varying characteristics. A gas content of 2% and an excess pressure of 3 MPa was used, such that the exsolution level is reached at about 1850 m (see Fig. 4); the trigger source kicking of the resonance is

The phenomenon of gliding spectral lines, as seen in Figure 1, has been identified before as a volcanic source effect and has been qualitatively linked to degassing and bubble concentration (Benoit & McNutt 1997). In a more quantitative manner, we model this characteristic spectral shift of volcanic tremor by varying the elastic parameters of the fluid. Over the time duration of a few minutes it is impossible for the magma to degas internally to such an extent that the elastic parameters would be significantly affected. (L. Wilson, pers. comm. 1999). Therefore, the exsolved gas content is fixed at 5% and the excess pressure now drops from 5 MPa to 3 MPa. This simulates, for example, the release of pressure caused by an episode of ash-venting. As seen in the velocity profile in Figure 6, the drop in Pe has only a minimal effect on the velocity profile of the conduit, the difference in profiles being only 5 ms-1 in places. However, this change is enough to cause a slight shift in the spectral peaks towards lower frequencies and a decreasing spacing of spectral lines. In turn, gliding towards higher frequencies with increased spacing of spectral lines indicates an increase of conduit pressure. This is in good agreement with observations from Montserrat and elsewhere, and demonstrates that the excess pressure is a likely candidate amongst conduit parameters, a change of which can lead to gliding spectral lines.

Discussion and conclusions Bubble-rich magmas provide conditions where seismic energy is efficiently trapped in a conduit, leading to a long-lived resonance. Pressure perturbations propagate in the conduit with velocities much smaller than the acoustic velocity in the bubble-free magma. This leads to a resonance pattern, which exhibits a peaked spectrum with integer harmonics, but is fundamentally different from standing waves in a fluid-filled conduit. Seismic signals observed at the

608

J. NEUBERG & C. O'GORMAN

Fig. 5. Synthetic seismogram (top) and snapshots of seismic wavefield from a conduit model with an exsolution level at 1850 m. The seismic source at 350 m depth in the conduit is active for only 1 second. Dark and light wavefronts correspond to dilatational and compressional phases, respectively. Note the apparent seismic point source after 12 seconds just below the exsolution level, which radiates spherical waves into the elastic medium.

surface are a superposition of interface waves leaking from the conduit wall rather than a direct image of the wave pattern trapped in the conduit. Such signals are indicative of a highly periodic excitation mechanism which can be produced by either a long-lasting resonance, or a repetitive trigger mechanism (Neuberg 2000). In the case of seismic energy being trapped in bubble-rich magma, the seismic waves travel up and down the conduit walls and generate the periodic excitation as they are reflected at both ends of the conduit: at the upper end where the system is closed by the presence of a lava dome, and at the lower end of the conduit where the exsolution level with its sharp velocity contrast provides a reflector. In this way we synthesized resonant tremor sequences, based on realistic parameters, lasting for several tens of seconds. Of particular interest is the qualitative link between the observation of gliding spectral lines and the pressurization and decompression of the magmatic system. A shift of spectral lines towards lower frequencies corresponds to a decompression of the system. The pressure drop induces a decompression in the magma and the increased gas volume fraction results in a lower seismic velocity inside the conduit. Linked through a dispersion relation, the interface waves

along the conduit will slow down as well, resulting in a leaking wave pattern of lower frequency content. In turn, a pressurization of the system, whatever it may be caused by. will yield a smaller gas volume fraction, a higher seismic velocity in the conduit and. therefore, a higher velocity of interface waves. The wavefield observed at the surface will have a higher frequency content. The next generation of models will include magma viscosity and diffusion which control the time dependence of the changes in the magma after a decompression event. So far we assumed an immediate increase in gas volume fraction after decompression. Viscosity together with the distribution of bubble sizes will determine the frequency-dependent damping mechanism of the low-frequency events. With a resonance model we cannot explain the highly periodic occurrence of single low-frequency events with identical waveforms (White et al. 1998) over time periods of hours. Similarly, the merging of single low-frequency events into tremor requires a repetitive, non-destructive source mechanism. A feedback system is needed which provides the link between a conduit resonance and the repetitive triggering of low-frequency events. This will finally shed some further light on the nature of the actual trigger mechanism which kick-starts the conduit resonance.

SEISMIC WAVEFIELD IN GAS-CHARGED MAGMA

Fig. 6. A change in the excess pressure of P = 2 MPa produces a corresponding change in the seismic velocity in the conduit of 5 m s-l. The impact on the seismic signal is characterized by a gliding of spectral peaks.

We would like to thank L. Wilson for many interesting and enjoyable discussions on the physics of magma, and K. Olsen for supporting the FD calculations. Part of these simulations were carried out on the SGIOrigin 2000 at the University of California, Santa Barbara (NSF grant EAR 96-28682), with support from NSF grant EAR 96-28682. We thank the staff of the Montserrat Volcano Observatory (MVO) for providing the seismic data and their general assistance. The MVO is supported by the Department for International Development (UK) and the British Geological Survey. We are also grateful to S. McNutt and B. Chouet for their comments, which significantly improved the paper.

References ASPINALL, W. P., MILLER, A. D., LYNCH, L. L., LATCHMAN, J. L., STEWART, R. C., WHITE, R. A. & POWER, J. A. 1998. Soufriere Hills eruption, Montserrat 1995-1997: volcanic earthquake locations and fault plane solutions. Geophysical Research Letters, 25, 3397-3400. BAPTIE, B., LUCKETT, R. & NEUBERG, J. 2002. Observations of lowfrequency earthquakes and volcanic tremor at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 611-620. BENOIT, J. & McNuTT, S. 1997. New constraints on source processes of volcanic tremor at Arenal Volcano, Costa Rica, using broadband seismic data. Geophysical Research Letters, 24, 449-452. CERJAN, C., KOSLOFF, D., KOSLOFF, R. & RESHEF, M. 1985. A nonreflecting boundary condition for discrete acoustic and elastic wave equations. Geophysics, 50, 705-708.

609

CHOUET, B. A. 1988. Resonance of a fluid-driven crack: Radiation properties and implications for the source of long-period events and harmonic tremor. Journal of Geophysical Research, 93, 4373-4400. CHOUET, B. A. 1996. Long-period volcano seismicity: its source and use in eruption forcasting. Nature, 380, 309-316. CHOUET, B. A., PAGE, R. A., STEPHENS, C. D., LAHR, J. C. & POWER, J. A. 1994. Precursory swarms of long-period events at Redoubt Volcano (1989-1990), Alaska: Their origin and use as a forecasting tool. Journal of Volcanology and Geothermal Research, 62, 95-135. DEVINE, J. D., MURPHY, M. D., RUTHERFORD, M. J. ET AL. 1998. Petrologic evidence for the pre-eruptive pressure-temperature conditions, and recent reheating, of andesitic magma erupting at the Soufriere Hills Volcano, Montserrat, W.I. Geophysical Research Letters, 25, 3669-3672. FERRAZZINI, V. & AKI, K. 1987. Slow waves trapped in a fluid-filled infinite crack: implication for volcanic tremor. Journal of Geophysical Research, 92, 9215-9223. HURST, A. W. 1992. Stochastic simulation of volcanic tremor from Ruapehu. Journal of Volcanology and Geothermal Research, 51, 185-198. LAHR, J., CHOUET, C., STEPHENS, C., POWER, J. & PAGE, R. 1994. Earthquake classification, location and error analysis in a volcanic environment: implications for the magmatic system of the 1989-1990 eruptions at Redoubt Volcano, Alaska. Journal of Volcanology and Geothermal Research, 62, 137-151. LEIGHTON, T. G. 1994. The Acoustic Bubble. Academic Press, London. LEVANDER, A. R. 1988. Fourth-order finite-difference P-SV seismograms. Geophysics, 53, 1425-1436. LUCKETT, R., BAPTIE, B. & NEUBERG, J. 2002. The relationship between degassing and rockfall signals at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 595-602. MILLER, A., STEWART, R., WHITE, R. ET AL. 1998. Seismicity associated with dome growth and collapse at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, 3401-3404. NEUBERG, J. 2000. Characteristics and causes of shallow seismicity in volcanoes. Philosophical Transactions of the Royal Society, London, 358, 1533-1546. NEUBERG, J., BAPTIE, B., LUCKETT, R. & STEWART, R. 1998. Results from the broadband seismic network on Montserrat. Geophysical Research Letters, 25, 3661-3664. NEUBERG, J., LUCKETT, R., BAPTIE, B. & OLSEN, K. B. 2000. Models of tremor and low-frequency earthquake swarms on Montserrat. Journal of Volcanology and Geothermal Research, 101, 83-104. SCHLINDWEIN, V., WASSERMANN, J. & S C E R B A U M , F. 1995. Spectral analysis of harmonic tremor signals at Mt. Semeru volcano, Indonesia. Geophysical Research Letters, 22, 1685-1688. SEIDL, D., SCHICK, R. & RIUSCETTI, M. 1981.Volcanic tremors at Etna: a model of hydraulic origin. Bulletin of Volcanology, 44, 43-56. SHAW, H. R. 1974. Diffusion of H2O in granitic liquids: Part I, Experimental data; Part II, Mass tarnsfer in magma chambers. In: HOFFMAN, A. W. ET AL. (eds) Geochemical Transport and Kinetics. Carnegie Institute, Washington, 634, 139-170. SPARKS, R. S. J., YOUNG, S. R., BARCLAY, J. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills Volcano, Montserrat, West Indies: November 1995 to December 1997. Geophysical Research Letters, 25, 3421-3424. SPERA, F. J. 2000. Physical properties of magma. In: SIGURDSSON, H., HOUGHTON, B., McNuTT, S. R., RYMER, H. & STIX, J. (eds) Encyclopedia of Volcanoes. Academic Press, London, 171-190. VOIGHT, B., HOBLITT, R. P., CLARKE, A. B. LOCKHART, A. B., MlLLER, A. D.,

LYNCH, L & McMAHON, J. 1998. Remarkable cyclic ground deformation monitored in real-time on Montserrat, and its use in eruption forecasting. Geophysical Research Letters, 25, 3405-3408. WHITE, R., MILLER, A., LYNCH, L., POWER, J. &, MVO STAFF 1998. Observations of hybrid seismic events at Soufriere Hills volcano, Montserrat, West Indies: July 1995 to September 1996. Geophysical Research Letters, 25, 3657-3660. WOOD, A. B. 1932. A Textbook of Sound. Bell, London, 360-364.

Observations of low-frequency earthquakes and volcanic tremor at Soufriere Hills Volcano, Montserrat B. BAPTIE1, R. LUCKETT1 & J. NEUBERG2 1

Global Seismology Group, British Geological Survey, Murchison House, West Mains Road, Edinburgh, UK (e-mail: [email protected]) 2 School of Earth Sciences, University of Leeds, Leeds LS2 9JT, UK

Abstract: The 1995-1999 extrusive phase of the eruption of Soufriere Hills Volcano on Montserrat was dominated by lowfrequency volcanic earthquakes (also called long-period and hybrid earthquakes). These earthquakes have distinctive peaked spectra and commonly occur in swarms related to the pressurization of the upper part of the magma conduit. We use data from an array of broadband seismometers to examine spatial and temporal variation in the spectral properties of these earthquakes between January and August 1997. Although spectra are generally stable over long periods of time at a given reference point, we also find evidence for changes in the spectra with time and with event magnitude, which may be attributed to changes in the source. The relative amplitude of spectral peaks varies at different stations around the volcano, leading to the conclusion that the observed wavefield is a combination of both source and propagation effects. We also analyse a number of tremor episodes related both to swarms of low-frequency earthquakes and to volcanic explosions. During certain tremor episodes we observe harmonic spectra, with shifting spectral peaks whose frequencies are the same at all stations. This behaviour can be modelled by repetitive triggering of individual earthquakes, where the trigger frequency changes with time, as well as by time-dependent changes in the acoustic properties of the magma-filled conduit caused by pressurization or depressurization.

The eruption of Soufriere Hills Volcano on the island of Montserrat has resulted in a wealth of qualitative and quantitative data, including a unique seismological dataset, which can be used to study magmatic processes within the volcano. These data can be used to relate seismic events to the actual volcanic activity and better constrain models for volcanic earthquakes. The different phases of the eruption and the volcanological aspects are well documented (e.g. Young et al. 1998; Sparks & Young 2002; Kokelaar 2002). Observed seismicity at different stages of the eruption on Montserrat and the characteristics of the different types of seismic events are described in detail by Miller et al. (1998) and Aspinall et al. (1998). However, during the first extrusive phase of the eruption between 1995 and 1998, one type of earthquake, lowfrequency earthquakes, similar to 'B-type' volcanic earthquakes (Minakami 1974), or long-period earthquakes (Chouet et al. 1994), dominated the seismic activity. The frequency content of these signals generally ranges from 0.2 to 5.0 Hz, so the low-frequency' events discussed here should not be confused with truly broadband signals with energy in the 0.01 to 25 Hz range, such as those observed at Aso (Kaneshima et al. 1996) and Stromboli (Neuberg et al. 1994). In previous work (Miller et al. 1998; Neuberg et al. 1998; White et al. 1998), low-frequency events have been separated into long-period and hybrid earthquakes. Long-period earthquakes are emergent, with energy between 0.5 and 2 Hz. Hybrid earthquakes have an additional high-frequency phase (3-10 Hz) that precedes the long-period phase. However, Neuberg et al. (2000) show that a continuum exists between longperiod and hybrid earthquakes, and that the two types are actually end-members of a single event type. For this reason, we refer to all events of this type as low-frequency events. Aspinall et al. (1998) state that by late 1996 most seismic events were originating within a volume about 2 km in diameter that extended up to the surface from about 3 km below the peak of the volcano. These hypocentres were calculated routinely as part of the volcanic monitoring operation at the Montserrat Volcano Observatory (MVO) using the HYPO71 algorithm (Lee & Lahr 1978). Hypocentres for all located earthquakes between 1 October 1996 and 31 August 1997 are shown in Figure 1. Further work by Baptie (unpublished) suggests that hypocentres for swarms of lowfrequency earthquakes in 1997 are tightly clustered directly under the lava dome at depths as shallow as 1 km. As on other volcanoes, low-frequency earthquakes typically occur in swarms which can last for many hours and contain many hundreds of individual earthquakes. In some cases the time interval

between discrete events is observed to decrease until the events merge and form a continuous signal (Luckett et al. 1997). The similarity between the spectral characteristics of the tremor and the individual earthquakes is in keeping with observations from other volcanoes (Latter 1981; Fehler 1983), which suggests a close relationship between low-frequency earthquakes and volcanic tremor. Chouet (1985) suggests that harmonic tremor is the result of the superposition of the response of the volcanic conduit to a series of pressure impulses. Swarms of low-frequency earthquakes also correlate with inflation of the volcano measured by tiltmeters (Voight et al. 1999), indicative of pressurization. Figure 2 shows this correlation for the time period 20-30 June 1997. This suggests that the low-frequency earthquakes are closely connected to pressurization in the upper part of the magma conduit, probably caused as gas exsolution and crystal growth lead to the formation of a viscous plug at the top of the magma column and the resultant pressurization of magma and gas under this plug (Voight et al. 1998, 1999). Low-frequency earthquakes on Montserrat, like those observed on other volcanoes (Chouet et al. 1994; Tsuruga et al. 1997), have distinctive spectral peaks. Many authors have suggested that such spectral peaks are a source effect. This conclusion is generally supported by the coherence of the spectral peaks across an array of seismometers and, in some cases, by temporal variation in spectral properties (Aki et al. 1977; Benoit & McNutt 1997). Numerous models have been proposed to explain these features, generally involving the role of fluids and conduits of varying geometries. Chouet (1985, 1988) models the resonance of a magma- or gas-filled conduit, periodically excited by short-term pressure transients. Julian (1994) suggests that oscillations may be excited by non-linear fluid flow. Ferrazzini & Aki (1987) use an analytical approach to derive the dispersion relationship for what they referred to as 'slow waves' propagating along the fluid-solid interface of a fluid-filled crack. Neuberg et al. (2000) use a numerical approach to examine the effect of varying gas content and surface topography on the elastic wavefield generated by a point source in a fluid-filled conduit. The resultant wavefield is shown by the authors to be dominated by inhomogeneous waves propagating along the fluid-solid interface and leaking energy into the surrounding medium. These waves propagate through the medium and form a complex interference pattern due to interaction with the free surface. In keeping with the observations of spectral peaks at different stations on Montserrat, modelled spectra show different amplitudes at different distances from the conduit.

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 611-620. 0435-4052/02/$ 15 © The Geological Society of London 2002.

611

612

B. BAPTIE ET AL.

Fig. 1. Hypocentres for all located earthquakes between 1 October 1996 and 31 August 1997, determined using the HYPO71 algorithm. Contours in metres above sea level. Depth in kilometres above sea level.

However, interaction of the source wavefield with the propagation medium can also lead to similar peaked spectra. Gordeev el al. (1990), found that the spectral features changed over distances of less than 1 km and also that the dispersive nature of the observed wavefield was characteristic of surface waves. Goldstein & Chouet (1994) suggest that near-surface resonances are responsible for the peaked spectra observed on Puu Oo. Wegler & Seidl (1997) model the spectral peaks observed on Mount Etna by propagation in a low-velocity waveguide. In this paper we analyse both spatial and temporal variation in spectral characteristics of low-frequency earthquakes, recorded on an array of broadband seismometers between January and August 1997, and discuss the findings in the context of possible models. Volcanic activity at this time was dominated by rapid dome growth and incremental collapse of the lava dome. We also examine the relationship between low-frequency earthquakes and volcanic tremor, by examining a number of tremor episodes that occurred over the same time period. In particular, we examine the observation that volcanic tremor on Montserrat can be considered as the superposition of individual earthquakes. Finally, we present an example of tremor related to explosive activity in early 1999, after the end of the extrusive phase.

Data acquisition Data analysed in this paper were recorded on the seismometer array shown in Figure 3. Five Guralp CMG-40T, broadband seismometers are marked by the squares. Three Integra LA 100/F short-period seismometers, also included in the network, are marked by triangles. Data were transmitted to the observatory site as digital samples by UHF radio telemetry and recorded both continuously and in triggered mode. A sampling frequency of 75 Hz and and 24-bit digitization give an effective system bandwidth of 0.03-30 Hz and a dynamic range of up to 145 dB for the broadband stations, and 1-30 Hz and lOO dB for the short-period stations. The broadband stations each have three orthogonal sensors to monitor three components of ground motion, allowing reconstruction of the full vector wavefield. The remaining stations have vertical component sensors only. The network operated continuously from installation in October 1996 until September 1997 when the MVO moved location from south to north of the island (see Fig. 3) because of escalation in the volcanic activity. Stations MBGA and MBBE were destroyed in June 1997 by pyroclastic flows. Similarly, station MBGE was destroyed in December 1997. The network was reinstalled in April 1998 at close to the original capability.

OBSERVATIONS OF LOW-FREQUENCY EARTHQUAKES AND TREMOR

613

Fig. 2. Correlation of low-frequency earthquakes with tilt measurements recorded between 20 and 30 June 1997. The histogram shows the number of earthquakes per hour and the occurrence of the earthquake swarms. Tilt measurements were recorded at Chances Peak on the crater rim, approximately 1.5 km from the vent.

Spatial variability

Fig. 3. Acquisition geometry of the Montserrat broadband network. Three-component broadband stations are marked by squares. Short-period seismometers are marked by the triangles. The two positions of the MVO (North and South) are marked by stars. Soufriere Hills Volcano is indicated by the abbreviation SHV. Contours are in metres above sea level.

Following a convolutional model, the wavefield recorded at a particular station is a combination of source function, its interaction with the propagation medium and individual site effects. Figure 4 shows a typical example of the waveforms and spectra for the vertical component of ground velocity of a low-frequency earthquake recorded at 01:26 UTC (Co-ordinated Universal Time, 4 hours ahead of local time) on 4 April 1997. Channels have been sorted according to distance from the source, with the closest at the top. Higher frequencies are attenuated with increasing distance from the source. We observe slight differences in the relative amplitudes of the spectral peaks, both at different stations and on different components (not shown). This is in keeping with the wavefield generated by a point source in a closed, fluid-filled conduit (Ferrazzini & Aki 1987; Neuberg et al. 2000). As discussed earlier, energy leaks through the interface and propagates through the medium, forming a complex interference pattern through interaction with the free surface, which results in distinctive spectral peaks. However, it is difficult to model observed spectral peaks and speculate on possible source dimensions or geometry, because the observed peaks are a function not only of the conduit geometry, but also the magma properties, so a large number of models may fit any particular observation. To examine in detail the variations in spectral properties we calculated power spectra for all low-frequency earthquakes from January to August 1997 that were recorded as triggered events on the seismic acquisition system. Figure 5 shows peak frequency calculated for each event plotted against maximum amplitude at the five broadband stations. We define peak frequency as the frequency with the greatest amplitude given by the power spectrum of ground velocity for each earthquake. The maximum amplitude is calculated from the average amplitude in 2-second windows along the entire record. The peak frequency is not constant at any one station. Instead, the peak frequencies fall into a number of groups which are independent of earthquake magnitude. The frequency of the spectral peaks also changes with station location. Earthquakes with

614

B. BAPTIE ET AL.

Fig. 4. Velocity seismograms and normalized amplitude spectra for a single lowfrequency earthquake on 4 April 1997 01:26 UTC. The distance of each seismic station from the lava dome is given in brackets. Frequency is in Hertz.

a peak frequency of 0.2-0.5 Hz are dominated by ocean microseism. By comparison with other peaks, peak frequencies for these earthquakes are stable at all stations. The variability of the spectra decreases as earthquake magnitude increases. The peak frequencies of the largest earthquakes, at a particular station, are relatively stable over long periods of time, whereas those for the smaller earthquakes are more scattered. For a point source, such similar waveforms require a spatial variation in source position of less than one-quarter of the dominant wavelength, approximately 200m. This observation clearly requires a highly repeatable trigger mechanism. However, for a resonating conduit, the wavefield is independent of the position of the initial point source (Neuberg et al. 2000). In fact, the source may even be outside of the conduit. To explore the relationship between event magnitude and the spectral properties, we calculated the sum of the spectral amplitudes for the vertical component of ground motion in a number of discrete frequency bands. Gordeev et al. (1990) use a similar method to find evidence of increasing source dimension with increasing amplitude of volcanic tremor. Four frequency bands were analysed: 0.5-0.9 Hz; 0.9-1.4 Hz; 1.4-2.0 Hz; and 2.0-3.0 Hz. The contribution in each frequency band was normalized by the total spectral amplitude for each earthquake. We then plotted normalized amplitude in each frequency band against total amplitude for groups of low-frequency earthquakes at three separate stations (Fig. 6). Earthquakes analysed were from low-frequency swarms on 9-14 February 1997, 14 March 1997 and 4, 12 and 15 April 1997. Normalized amplitude is observed to increase or decrease in each

frequency band with increasing total amplitude. In particular, normalized amplitude in the 0.9-1.4 Hz band increases with total amplitude at all three stations, while decreasing in the other frequency bands. This shows that larger earthquakes have more energy focused at a single frequency for a particular station. Temporal variability Figure 7 shows the variation of the peak frequency with maximum amplitude measured at station MBGE for the first eight months of 1997. Calculations are made in the same way as those for Figure 5. We observe groups of earthquakes with the same peak frequency. Some of the peak frequencies remain stable with time, for example those at 1.21 Hz and 1.43 Hz. However, there are also peak frequencies which are not stable with time; for example, there are a number of events in January and February with a peak frequency of 0.85 Hz. This peak is not observed at any other time. Similarly, there are events in July and August with a peak frequency of 0.78 Hz, which is also unique. Peak amplitudes for these events are above the average peak amplitude of the microseism peak. Assuming no temporal change in the propagation medium, we suggest that this could be caused by a long-term change in the properties of the magma-filled conduit. Neuberg et al. (2000) have shown how changes in conduit properties can lead to changes in the observed spectral patterns at a single station. For example, increasing gas content will decrease the seismic velocity inside the conduit and change the wavelength of the resonating waves.

OBSERVATIONS OF LOW-FREQUENCY EARTHQUAKES AND TREMOR

615

Variability in tremor characteristics

Fig. 5. Peak frequency versus the maximum amplitude of ground velocity measured on the vertical component for all low-frequency earthquakes between January and July 1997. Calculated for the vertical component of ground motion at all broadband stations.

Fig. 6. The dependence of relative amplitude ( E/E) on total amplitude in four fixed-frequency bands: from 0.5-0.9 Hz; 0.9-1.4 Hz; 1.4-2.0 Hz; and 2.0-3.0 Hz. Measurements are shown for the vertical component of ground velocity at three stations, MBGE, MBGH and MBBE.

The first observation of individual low-frequency earthquakes getting closer together with time to form a continuous tremor signal was in July 1996 (Miller 1997). Further instances of this behaviour were recorded on the broadband network in December 1996 and throughout 1997. By looking at the time periods when swarms of low-frequency earthquakes occurred, we were able to determine all examples of individual events merging to form tremor between December 1996 and August 1997. To examine the tremor characteristics, we calculated spectrograms from continuous broadband data using overlapping amplitude spectra calculated using 2048 data samples and an overlap of 1024 samples. Figure 8a shows a velocity spectrogram calculated for station MBGE starting at 14:30 UTC (10:30 local time) on 25 June 1997. The tremor episode corresponds to the inflation of the volcano recorded by tiltmeters (Voight et al. 1999) and ends with the catastrophic collapse of the lava dome (Loughlin et al. 2002). Individual events on the seismogram gradually merge after about 42 minutes and form tremor. The peaks on the spectrogram for the tremor episode are the same as those for the individual events. After approximately 145 minutes the collapse occurs, as indicated by a sharp increase in amplitude on the seismogram. The characteristics of the spectrogram also change, with the strongly peaked spectra replaced by a much more broadband signal. The collapse lasts for about 20 minutes. So, in general, the spectral characteristics of these episodes are the same as those for individual events and the relative amplitudes of the spectral peaks vary strongly at different stations across the seismic network. However, during some episodes we observe spectral peaks composed of integer harmonics whose frequency changes with time. The frequencies of these peaks are the same at all stations. This behaviour has been modelled by Neuberg et al. (2000) as the superposition of identical low-frequency earthquakes, where the integer harmonics are controlled by the trigger frequency of the source. Changing frequency of the integer harmonics reflects a gradual change in the trigger frequency, rather than in the resonance parameters of the source. Figure 8b shows a second example of this type of behaviour observed on 6 February 1997. Up to six integer harmonics can be observed in this case. The fundamental frequency is observed to increase from 0.57 to 1.9 Hz over a period of about 12 minutes, then

616

B. BAPTIE ET AL.

The repetitive triggering mechanism may still be the cause of the harmonic behaviour, but the amplitudes of the individual events are too small to discriminate them from the background noise. We discussed earlier how the earthquake swarms occur at the same time as pressurization of the upper part of the conduit recorded by tiltmeters, and that this was followed by sudden depressurization linked to the collapse of material from the lava dome. Depressurization will result in an increase in the gas volume fraction and therefore a change in the seismic properties of the conduit by reducing the bulk modulus (Neuberg & O'Gorman 2002) of the magma. This means that a single trigger might generate a long-lived resonance whose spectral peaks change smoothly over time during the course of the depressurization event as the elastic parameters of the fluid change. A fourth type of tremor is shown in Figure 8d, from 13 August 1997. The spectrograms show up to nine integer harmonics with a spacing of approximately 1 Hz. The harmonic behaviour due to a repetitive triggering mechanism cannot be observed because the individual events are far enough apart in time to be imaged discretely. Since distance in time maps directly to the frequency domain, spectral lines are not observed. Rather, the individual events that constitute the tremor are themselves harmonic and are identical at all stations. Figure 9 shows an example of the waveforms and spectra of one of the harmonic events. Relative amplitudes of the spectral peaks are the same at all stations. The seismograms bear a strong resemblance to the tremor signals observed on Semeru Volcano by Schlindwein et al (1995), which were modelled by a resonating gas volume. The coherence of the observed harmonic spectra points to a source effect, in which individual low-frequency earthquakes are triggering harmonic oscillations of the conduit.

Tremor associated with explosive volcanism

Fig. 7. Temporal variation in peak frequency versus the maximum amplitude of ground velocity measured on the vertical component for low-frequency earthquakes between January and August 1997. Calculated for the vertical component of ground motion at station MBGE.

decrease again over a similar time period. This would correspond to an increase in the trigger frequency by a factor of 4. The relative amplitudes of the individual harmonics are observed to vary with both time and station position. Figure 8c shows a third example of tremor behaviour, this time from 12 February 1997. The first 35 minutes of this record consist of low-frequency earthquakes that have merged to form continuous tremor, marked by spectral peaks in the 1-2 Hz range and with additional lower-amplitude peaks at higher frequencies. As previously, the relative amplitude of the spectral peaks varies from station to station. The end of the swarm is marked by a sudden change in the amplitude of the seismic signal and also by a dramatic change in the spectral properties of the signal. We observe a number of harmonic spectral peaks, whose frequencies change with time. The gliding spectral lines initially show decreasing frequency for a few minutes, then a slight increase to a peak at 45 minutes from the end of the record. A distinctive pyroclastic flow signal coincides with this peak, marked by an increase in the amplitude of the seismic signal and by an increase in spectral energy at higher frequencies. These signals are generated by the collapse of material from the destabilized lava dome and can be recognized by their relatively broadband signals with peaks at approximately 3.5 Hz. Slight changes in the gliding spectral lines are also observed a few minutes later followed by another pyroclastic flow.

Figure lOa shows a seismic signal and spectrogram associated with a Vulcanian explosion on 5 August 1997 at 20:57 UTC (16:57 LT). The explosion was one in a sequence of 13 explosions during August 1997 (Druitt el al. 2002). Seismic signals for all the explosions are extremely similar. Before the explosion, random triggering of low-frequency earthquakes means that the frequency signature of the tremor episode leading up to the explosion is the same as that for individual low-frequency earthquakes, with peaks at 1.4 and 1.8 Hz. When the explosion occurs there is a clear change in the spectral properties observed on the spectrogram. The spectrogram shows that the explosion signal consists of an initial low-frequency phase which coincides with the ascent of a vertical eruption column. This is followed by higher frequency 'noise' as the column collapses under gravity and pyroclastic flows travel radially down the flanks of the volcano (Druitt et al. 2002). However, the low-frequency component continues as a tremor signal which lasts for over an hour after the initial explosion and gradually decays in amplitude. Also, the frequency of the tremor signal is identical to that of the initial explosion signal and varies at different stations around the volcano. In a closed conduit system, where the conduit is bounded above and below, energy from a single explosion can be trapped in the conduit and lead to resonance. However, in an open-conduit system, as must be the case after the vent has been cleared by the explosion, only short-lived resonance is possible as seismic energy rapidly leaks away at the free surface (Neuberg et al. 2000). Since the duration of the post-explosion tremor is too long to be caused by a single resonance of the conduit, kick-started by the initial explosion, we suggest that a feedback mechanism may exist whereby a succession of triggers and resonances leads to a longduration signal whose amplitude gradually decays with time. A different example of tremor associated with a volcanic explosion can be seen in Figure lOb. The seismogram shows a small explosion from 7 January 1999, by which time dome growth had temporarily ceased. The explosion signal, marked by the sharp

OBSERVATIONS OF LOW-FREQUENCY EARTHQUAKES AND TREMOR

617

Fig. 8. Velocity spectrograms (vertical component) for station MBGE. (a) Starting 14:30 UTC, 25 June 1997: individual earthquakes get closer together in time and merge to form a continuous tremor signal, (b) Starting 18:30 UTC, 6 February 1997, showing a tremor episode formed by coalescing low-frequency earthquakes. Integer harmonic spectral peaks appear after around 30 minutes. The frequency of these peaks increases with time, reaching a peak at 53 minutes. (c) Starting 11:30 UTC, 12 February 1997, showing tremor formed during a low-frequency earthquake swarm. Integer harmonic spectral peaks appear after around 35 minutes, after the end of the earthquake swarm. The frequency of these peaks increases slightly over the next 10 minutes, then decreases. (d) Starting 02:00 UTC, 13 August 1997, showing a low-frequency earthquake swarm. Harmonic spectral peaks result from spectra of individual earthquakes. Spectrograms were calculated using overlapping 2048 sample Fourier transforms.

618

B. BAPTIE ET AL.

Fig. 9. Velocity seismogram and spectrum from a single harmonic low-frequency earthquake on 03:16 UTC 13 August 1997.

increase in amplitude approximately 28 minutes after the start of the record, is again followed by a tremor signal which corresponds to continued gas- and ash-venting. However, the spectrogram shows that the time period leading up to the explosion is dominated by a sequence of gliding spectral lines whose frequency increases with time over a period of some 20 minutes until the explosion occurs. Again, there are two models to explain this behaviour: (1) an increase in the trigger frequency of individual earthquakes; or (2) a resonance with a change in the elastic properties of the magma. In the latter case, as pressure in the conduit builds up before the explosion, there is a decrease in the gas volume fraction and a corresponding increase in bulk modulus and seismic velocity that leads to a change in the observed seismic wavefield. This behaviour is discussed in detail by Neuberg & O'Gorman (2002).

Correcting for the instrument response to give ground displacement and restitution of the explosion signal to 100 seconds, reveals a relatively simple signal onset (Fig. 11), that is very similar at all stations, suggesting that this is due to source rather than propagation effects. Particle motion is in keeping with the expanding wavefront from an explosive source mechanism.

Discussion and conclusions We analysed low-frequency earthquakes and volcanic tremor from the November 1995 to March 1998 extrusive phase of the eruption of Soufriere Hills Volcano from January to August 1997. Volcanic

Fig. 10. Velocity spectrograms (vertical component) for station MBGE. (a) Starting 20:30 UTC, 5 August 1997, showing the signal generated by a Vulcanian explosion, (b) Starting 07:30 UTC, 7 January 1999, showing a small explosion from the post-extrusive phase of the eruption. Spectrograms were calculated using overlapping 2048 sample Fourier transforms.

OBSERVATIONS OF LOW-FREQUENCY EARTHQUAKES AND TREMOR

619

Fig. 11. Instrument corrected seismogram showing ground displacement from the explosion at 07:58 UTC on 7 January 1999.

activity over this period was characterized by unsteady and incremental dome growth and subsequent periodic collapse. In addition, we analysed a tremor episode from January 1999, related to ashventing, by which time the first phase of dome growth had ended. Similarities in the peak frequency at individual stations point to families of earthquakes with common propagation characteristics and a repeatable source mechanism. Peak frequencies are independent of earthquake magnitude. However, increasing magnitude results in a decrease in the variability of the spectral peaks. Also, the greater the magnitude, the more energy is focused in a particular frequency band. The observed temporal changes in the peak frequency could result from changes in conduit properties over time. We conclude that the spectral peaks of the low-frequency earthquakes are a result of the interaction of a point source with a magma-filled conduit. Changes in the spectral amplitudes at different azimuths and distances are a result of interference with the free surface. Our observations of volcanic tremor revealed a variety of characteristics. During the tremor episodes individual earthquakes were found to become more closely spaced until they merged together. If the triggering is random, then the tremor spectra display the same properties as those of the individual events.

Harmonic spectra and gradual changes of frequency with time can be modelled by repetitive triggering of identical low-frequency earthquakes where the triggering interval remains constant or changes smoothly with time. However, gradual changes in the frequency of the harmonic spectral peaks could also be caused by a single long-lived resonance and a time-dependent change in the elastic properties of the magma-filled conduit. For instance, episodes of pressurization or depressurization, either before an explosion or after a collapse of the lava dome, could lead to changes in the elastic properties of the conduit with time and therefore changes in the observed seismic wavefield. In a single case, we also observe harmonic tremor that is a result of superposition of individual earthquakes which are themselves harmonic. The scarcity of such harmonic earthquakes during the eruption suggests that the switch to harmonic behaviour is a consequence of a particular state in the conduit parameters which was rarely achieved during this phase of the eruption. We gratefully acknowledge the contribution of MVO staff. We thank also the two reviewers who provided helpful and improving suggestions. Published by permission of the Director, British Geological Survey (NERC).

620

B. BAPTIE ET AL.

References AKI, K., FEHLER, M. & DAS, S. 1977. Source mechanism of volcanic tremor: Fluid driven crack models and their application to the 1963 Kilauea eruption. Journal of Volcanology and Geothermal Research, 2, 259-287. ASPINALL, W., MILLER, A., LYNCH, L., LATCHMAN, J., STEWART, R., WHITE, R. & POWER, J. 1998. The Soufriere Hills eruption, Montserrat 1995-1997: volcanic earthquake locations and fault plane solutions. Geophysical Research Letters, 25(18), 3397-3400. BENOIT, J. & McNUTT, S. 1997. New constraints on source processes of volcanic tremor at Arenal volcano, Costa Rica, using broadband seismic data. Geophysical Research Letters, 24, 449-452. CHOUET, B. 1985. Excitation of a buried magmatic pipe: A seismic source model for volcanic tremor. Journal of Geophysical Research, 90, 1881-1893. CHOUET, B. 1988. Resonance of a fluid-driven crack: Radiation properties and implications for the source of long period events and harmonic tremor. Journal of Geophysical Research, 93, 4373-4400. CHOUET, B., PAGE, R., STEPHENS, C, LAHR, J. & POWER, J. 1994. Precursory swarms of long-period events at Redoubt volcano (1989-1990), Alaska: Their origin and use as a forcasting tool. Journal of Volcanology and Geothermal Research, 62, 95-135. DRUITT, T. H., YOUNG, S. R., BAPTIE, B. ET AL. 2002. Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Monserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 281-306. FEHLER, M. 1983. Observations of volcanic tremor at Mount St. Helens volcano. Journal of Geophysical Research, 88(B4), 3476-3484. FERRAZZINI, V. & AKI, K. 1987. Slow waves trapped in a fluid-filled infinite crack: implication for volcanic tremor. Journal of Geophysical Research, 92, 9215-9223. GOLDSTEIN, P. & CHOUET, B. 1994. Array measurements and modelling of sources of shallow volcanic tremor at Kilauea volcano, Hawaii. Journal of Geophysical Research, 99(B2), 2637-2652. GORDEEV, E., SALTYKOV, V., SINITSYN, V. & CHEBROV, V. 1990. Temporal and spatial characterisitics of volcanic tremor wavefields. Journal of Volcanology and Geothermal Research, 40, 89-101. JULIAN, B. 1994. Volcanic tremor: non-linear excitation by fluid flow. Journal of Geophysical Research, 99, 11859-11877. KANESHIMA, S., KAWAKATSU, H., MATSUBAYASHI, H. ET AL. 1996. Mechanism of phreatic eruption at Aso volcano inferred from near-field broadband seismic observations. Science, 273, 642-645. KOKELAAR, B. P. 2002. Setting, chronology and consequences of the eruption of Soufriere Hills Volcano, Monserrat (1995-1999). In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 1-43. LATTER, J. 1981. Volcanic earthquakes and their relationship to eruptions at Ruapehu and Ngauruhoe volcanoes. Journal of Volcanology and Geothermal Research, 9, 293-309. LEE, W. & LAHR, J. 1978. HYPO71 (revised): a computer program for determining hypocenter, magnitude and first motion pattern of local earthquakes. US Geological Survey Open File Report, 75-311. LOUGHLIN, S. C., CALDER, E. S., CLARKE, A. B. ET AL. 2002. Pyroclastic flows and surges generated by the 25 June 1997 dome collapse, Sou-

friere Hills Volcano, Monserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs. 21. 191-209. LUCKETT, R., BAPTIE, B., STEWART, R., NEUBERG, J.. MILLER, A. & LATCHMAN, J. 1997. Broadband seismicity on Soufriere Hills volcano, Montserrat. Eos (Transactions of the American Geophysical Union), 78, 441. MILLER, A. 1997. Seismicity of the Soufriere Hills Volcano, Monserrat, July 1995 to mid-1997. Montserrat Volcano Observatory Open File Report, 97/24. MILLER, A., STEWART, R., WHITE. R. ET AL. 1998. Seismicity associated with dome growth and collapse at the Soufriere Hills Volcano, Monserrat. Geophysical Research Letters. 25(18). 3401-3404. MINAKAMI, T. 1974. Seismology of volcanoes in Japan. In: CIUETT, A. L., LUONGO, G. & RAPOLLA, A. (eds) Physical Volcanology. Developments in Solid Earth Geophysics. Elsevier, Amsterdam, 1-27. NEUBERG, J. & O'GORMAN. C. 2002. A model of the seismic wavefield in a gascharged magma: application to Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 603-689. NEUBERG. J., LUCKETT, R., RIPEPE, M. & BRAUN. T. 1994. Highlights from a seismic broadband array on Stromboli Volcano. Geophysical Research Letters, 21(9). NEUBERG, J., BAPTIE, B., LUCKETT, R. & STEWART, R. 1998. Results from the broadband seismic network on Montserrat. Geophysical Research Letters, 25, 3661-3664. NEUBERG, J.. LUCKETT, R., BAPTIE. B. & OLSEN. K. 2000. Models of tremor and low-frequency earthquake swarms on Montserrat. Journal of Volcanology and Geothermal Research. 101. 83-104. SCHLINDWEIN, V., WASSERMANN. J. & ScHERBAUM. F. 1995. Spectral analysis of harmonic tremor signals at Mt. Semeru volcano. Indonesia. Geophysical Research Letters. 22. 1685-1688. SPARKS, R. S. J. & YOUNG, S. R. 2002. The eruption of Soufriere Hills Volcano, Monserrat (1995-1999): overview of scientific results. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs. 21, 45-69. TSURUGA, K., YOMOGIDA. K., HONDA. S.. ITO. H.. OHMINATO, T. & KAWAKATSU, H. 1997. Spatial and temporal variations of volcanic earthquakes at Sakurajima Volcano Japan. Journal of Volcanology and Geothermal Research, 75, 337-358.

VOIGHT. B., CLARKE, A., LOCKHART, A., MILLER. A., LYNCH, L. & McMAHON, J. 1998. Remarkable cyclic ground deformation monitored in real-time on Montserrat and it's use in eruption forecasting. Geophysical Research Letters, 25(18), 3405-3408. VOIGHT, B., SPARKS, R., MILLER, A. ET AL. 1999. Magma flow instability and cyclic activity at Soufriere Hills volcano. Montserrat. British West Indies. Science. 283, 1138-1142. WEGLER, U. & SEIDL, D. 1997. Kinematic parameters of the tremor wavefield at Mt. Etna. Geophysical Research Letters. 24. 759-762. WHITE, R., MILLER. A., LYNCH, L. & POWER. J. 1998. Observations of hybrid seismic events at the Soufriere Hills Volcano. Monserrat: July 1995 to September 1996. Geophysical Research Letters. 25(19). 3401-3404. YOUNG, S., SPARKS, R., ASPINALL, W., LYNCH, L., MILLER. A., ROBERTSON, R. & SHEPHERD, J. 1998. Overview of the eruption of Soufriere Hills Volcano, Monserrat, July 18 1995 to December 1997. Geophysical Research Letters. 25(18). 3389-3392.

Variation in HC1/SO2 gas ratios observed by Fourier transform spectroscopy at Soufriere Hills Volcano, Montserrat C. OPPENHEIMER1, M. EDMONDS2, P. FRANCIS3 & M. BURTON1, 4 1 Department of Geography, University of Cambridge, Downing Place, Cambridge CB2 3EN, UK (e-mail: [email protected]) 2 Department of Earth Sciences, University of Cambridge, Downing Street, Cambridge CB2 3EQ, UK 3 Department of Earth Sciences, The Open University, Milton Keynes, MK7 6AA, UK 4 Present address: Istituto Nazionale di Geofisica e Vulcanologia, Sezione de Catenia, U.F. Sistema Poseidon, Via Monti Rossi 12, 95030 Nicolosi, Catania, Italy

Abstract: We present here the results of open-path Fourier transform infrared (FTIR) spectroscopy of gases emitted from the lava dome of Soufriere Hills Volcano. Although measurement campaigns have been discontinuous, they do span a three-year period and provide strong evidence of secular change in the HC1/SO2 molar ratio from 5 in 1996 to <0.5 in 1999. The post1996 spectral data represent the only available measurements of gas ratios for the volcano's summit emissions, and complement SO2 emission rate data obtained by ultraviolet correlation spectroscopy (COSPEC), enhancing the interpretation of degassing at the volcano. The long-term decreasing HC1/SO2 ratio accompanied an increasing SO2 emission rate, and suggests a transition from degassing of andesitic to basaltic magma, or progressive tapping of a sulphur-rich vapour phase that was introduced by mafic magma, or that was already resident within the andesite magma reservoir. On timescales of minutes to hours, we observed variations in HC1/SO2 ratios associated with dome collapses. On 27 July 1998, for example, the HC1/SO2 ratio dropped from about 0.7 to 0.4 within minutes of a minor dome collapse. A much larger collapse event on 26 October 1998 was followed by a decrease in HC1/SO2 from 0.6 to 0.1, some 14 hours later. These changes are suggestive of transient degassing from a sulphurrich source region - larger collapses resulted in tapping of deeper sources, with exsolved gases taking longer to reach the surface. The sustained degassing and response to dome collapse events suggest a permeable upper conduit system established by syneruptive vesiculation, and efficient transfer of volatiles out of the magma chamber by degassing-driven convection in the lower part of the central conduit. Open-path FTIR spectroscopy represents a means for remote geochemical surveillance when access to vent regions is restricted for safety reasons, yielding valuable insights into degassing mechanisms.

Better knowledge of the character and role of volcanic gases is crucial to the interpretation and understanding of volcanic activity. In the case of lava-dome eruptions, such as that of Soufriere Hills Volcano, Montserrat, degassing behaviour (i.e., the style, timing and extent of volatile loss) may influence transitions between periods of repose, dome growth and explosion, essentially through modulation of gas separation and transfer out of the magma (Jaupart & Allegre 1991; Martel et al 1998; Nakada & Yoshinobu 1999). However, many factors can control these alternating states, including magma ascent rate, viscosity, permeability and degassing rate (Hammer et al. 1999), interaction with the hydrothermal system (Stix et al. 1993, 1997; Villemant & Boudon 1998, 1999), rheological stiffening due to gas exsolution and microlite growth (Nakada et al. 1995; Sparks 1997; Voight et al. 1999) and stick-slip behaviour along conduit walls (Denlinger & Hoblitt 1999). Gas data are essential to evaluate competing models and to support eruption prediction, volcano monitoring and identification of eruption cessation, but are difficult to obtain directly (Francis et al. 2000). In particular, the inherent dangers of collecting gas samples on growing lava domes explain why the existing database on syneruptive emissions is so sparse. Indeed, although the Soufriere Hills eruption has been the subject of sustained, intense geophysical surveillance, geochemical monitoring has been limited essentially to measurements of SO2 emission rates by ultraviolet correlation spectrometry (COSPEC; Young et al. 1998; Watson et al. 2000). However, interpretation of COSPEC data can be complicated by effects such as scavenging of SO2 by hydrothermal systems (Doukas & Gerlach 1995; Oppenheimer 1996) or by aqueous aerosol in plumes (Oppenheimer et al. 1998d). Experience on Montserrat has shown that it is often difficult operationally to discriminate between multiple explanations for observed changes in SO2 degassing, where the competing hypotheses may have very different implications for hazard assessment. For example, the elevated (>10 k g s - 1 ) SO2 fluxes observed following the cessation of dome growth in March 1998 (Sparks & Young 2002; Norton et al. 2002) were hard to interpret in the absence of other geochemical data. In order to reinforce understanding of degassing at Soufriere Hills Volcano, we have undertaken an experimental programme of Deceased

remote measurement of the tropospheric plume (Fig. 1) using a fieldportable Fourier transform infrared (FTIR) spectrometer. Volcanic and atmospheric gas species can be identified and quantified by their 'fingerprint' rotational-vibrational spectral lines. Our group has used this technique at several volcanoes, including Mount Etna and Vulcano, Italy (Francis et al. 1995, 1996, 1998) and Masaya, Nicaragua (Horrocks et al. 1999; Burton et al. 2000, 2001). Our first FTIR spectroscopic observations on Montserrat were made in 1996 (Oppenheimer et al. 1998a). These data were obtained by employing a Nernst glower (lamp) as an infrared source, over path lengths of 140-2000 m on the west flank of Soufriere Hills Volcano. HC1 was readily detected and quantified, but SO2 was not present above detection limits, providing a minimum estimate of the HC1/SO2 molar ratio of about 10 (but see below). Combining this result with the contemporaneous SO2 emission rate of around 2.7 k g s - 1 , and comparing it with the dome growth rate and chlorine content of dome lavas, suggested that the derived HC1 emission rate could be supported by degassing of lava entering the dome. Given uncertainties in calculations, this is consistent with the modest degassing excess (with respect to syneruptive volatile budgets) observed by Hammouya et al. (1998). Measurements obtained using the dome itself as the infrared source yielded similar results (Oppenheimer et al. 1998b). Following this initial campaign, we developed field and retrieval routines for solar occultation measurements of plume gases at Mount Etna and Masaya volcanoes (Oppenheimer et al. 1998c; Francis et al. 1998). This approach involves observation of the Sun (acting as infrared source) through the gas plume. We believed that such solar measurements at Soufriere Hills would yield better signal-to-noise ratios for target volcanic gas species (in terms of the number of molecules of plume gases relative to other atmospheric gas absorptions), and make it possible to detect and quantify SO2 as well as HC1. Our subsequent FTIR spectroscopy campaigns on Montserrat have all employed this technique, and we report the new results here. We examine the spectroscopic observations to identify evolutionary trends over two timescales: minutes to hours, and months to years. These are interpreted in the context of other geophysical and geochemical data collected by the Montserrat Volcano Observatory (MVO), in order to develop models for explaining observed degassing behaviour. As a representative of a wider class of intermediate to silicic volcanism, Soufriere Hills

DRUITT, T. H. & KOKELAAR, B. P. (eds) 2002. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 621-639. 0435-4052/02/$ 15 © The Geological Society of London 2002.

621

622

C. O P P E N H E I M E R ET AL.

Fig. 1. Airborne view of Soufriere Hills Volcano's gas and aerosol plume drifting over Plymouth on 1 February 1999.

offers a valuable opportunity to learn about syn- and post-effusive degassing trends associated with lava dome eruptions.

Synopsis of the eruption: July 1995-January 1999 As with many volcano surveillance histories, the geochemical monitoring at Montserrat has revolved primarily around the application of COSPEC and wind speed measurements to the

estimation of fluxes of SO2 in the plume. This section provides an overview of the eruption chronology for the period covered by the spectroscopic observations, and summarizes the SO2 emission rates measured by the MVO (Fig. 2). The eruption of Soufriere Hills Volcano began on 18 July 1995 with a series of phreatic explosions (Robertson el al. 2000). Over the month of August 1995, SO2 fluxes decreased from about 10. to less than 1 k g s - 1 . Sustained effusion of an andesitic lava dome began in mid-November 1995. By March 1996. collapses of portions of the dome generated pyroclastic flows that travelled eastwards down

Fig. 2. SO2 emission rates measured with COSPEC since the start of the eruption in July 1995. Values are plotted for each individual run. The application of COSPEC on Montserrat is discussed by Young el al. (1998). Watson el al. (2000) and Edmonds (2001). Some of the major events and episodes of the eruption are indicated. Asterisks indicate the periods of FTIR spectroscopic campaigns.

GAS SURVEILLANCE AT SOUFRIERE HILLS VOLCANO

623

Fig. 3. Map of Montserrat indicating place names mentioned in the text. FTIR spectra were collected from fixed positions beneath the plume, which generally travelled within the boundaries shown (large dashed lines). Most observations were made from the centre of Plymouth, and along the road between Cork Hill and Plymouth.

624

C. OPPENHEIMER ET AL.

the Tar River valley (Fig. 3). These first reached the coast on 12 May 1996. In July-August 1996, the rate of dome growth peaked at 10 m3 s-1 and averaged 4 m3 s-1, SO2 emission rates increased from about 1-2 kg s-1, to around 10 kg s-1, and dome collapses increased in magnitude (Calder et al. 2002). Collapse of c. 11 x 106 m3 of the dome on 17 September 1996 initiated an explosive eruption that propelled an ash plume to an estimated altitude of 15 km. Lava effusion resumed on 1 October 1996 at a rate of around 5 m3 s-1, piling up the dome sufficiently by March 1997 for it to feed the first pyroclastic flows down the White River Valley on the SW flank of the volcano (Fig. 3). As the dome continued to grow and overwhelm the pre-existing English's Crater topography, further collapses issued pyroclastic flows down drainages on the north and NE flanks in June 1997. Sulphur dioxide fluxes peaked at more than 50 kg s-1, one day before the devastating pyroclastic flows of 25 June 1997. This was the last COSPEC measurement to be made for over a year. The only estimates of sulphur dioxide emission rate (a few kilograms per second) in the intervening period derive from measurements of SO2 concentrations in Plymouth, obtained with diffusion tubes (a reasonably tight correlation was obtained between SO2 concentrations at a particular site in Plymouth and the contemporary SO2 flux measurement). Pyroclastic flows reached Plymouth in August 1997, and Bramble Airport, about 6 km NE of the summit (Fig. 3), on 21 September 1997. By 25 December 1997, the dome volume stood at 113 x 106 m3, but this was reduced by 55 x 106 m3 on 26 December following a sector collapse of the SW flank of English's Crater (Sparks et al. 2002; Voight et al. 2002). The deficit had been regained by the second

week of March 1998, at which time dome growth ceased. Most of the gas geochemical data reported here pertain to the following period of non-effusive activity prior to the growth of the second dome, which began in November 1999. A collapse on 3 July 1998 removed around 20-25 x 106 m3 of the dome (Calder et al. 2002) releasing a minimum of 1.1 x 107 kg of SO2, according to radiative transfer modelling of ultraviolet observations by the spaceborne Total Ozone Mapping Spectrometer (Mayberry et al. 2002). COSPEC surveillance resumed on 8 July 1998, and revealed SO2 fluxes above 20 kg s-1 for several weeks. This strong degassing was fed by vigorous gas vents located within the 3 July 1998 collapse scar (Fig. 4b). Further sporadic collapses reduced the dome volume to around 77 x 106 m3 by the end of January 1999, which marks the end of the geochemical observation period discussed here. At this time, SO2 fluxes were elevated again, at around lO kg s-1, and sustained by vents on or near the floor of the 3 July 1998 collapse scar (Fig. 4a). Methodology The principal characteristic of our field methodology described here is use of the solar disc as the source of infrared radiation, i.e. the Sun is viewed through the volcanic plume (Fig. 5a; Francis et al. 1998; Oppenheimer et al. 1998c). This provides a strong signal across the infrared region of interest, particularly at higher frequencies where HC1 and HF absorb. The solar occultation technique is important for two reasons: first, it is a comparatively safe

Fig. 4. Views of gas vents within the scar formed by the 3 July 1998 dome collapse in (a) January 1999 and (b) July 1998.

GAS SURVEILLANCE AT SOUFRIERE HILLS VOLCANO

625

Fig. 4. (continued)

and straightforward way to observe plume compositions through time; second, it opens up the possibility of investigating plume chemistry, since measurements can be made at different distances from source and hence for variable plume ages (Oppenheimer et al. 1998c). We conducted several brief campaigns on Montserrat spanning a few days or weeks at a time in 1996, 1998 and 1999 (Figs 1,2). Only the first observations, in 1996, employed the infrared lamp (Fig. 5b). All subsequent campaigns involved solar measurements. The observation periods coincided with dome extrusion in 1996, and with the hiatus in dome growth after March 1998 (Fig. 2). (Measurement campaigns have been sustained since November 1999: for details see Edmonds 2001; Edmonds et al. 2001.) Observation points were determined according to wind direction, accessibility and safety (Fig. 3). We used the same instrument (manufactured by MIDAC, California) over this period, with the exception of the addition of an indium antimonide (InSb) detector in 1998, though we continued to collect spectral data with our original mercury-cadmium-telluride (MCT) detector (Fig. 6). The InSb detector delivers improved performance over the MCT detector at mid-infrared frequencies, and we selectively analyse here only the solar occultation spectra obtained with the InSb detector. Details of the instrument, field data collection and analytical methods are summarized in Francis et al. (1995, 1998) and Oppenheimer et al. (1998c), but our basic approach was to measure gases in absorption (as they absorb photons from the source, in contrast to observations of emission spectra, e.g. Love et al. 1998). Retrievals were performed by using a forward model (AOPP 2001) to simulate transmittance spectra for the atmospheric path, using the HITRAN96 database of spectral line positions and strengths. We found the best fit to the field data using an optimal estimation, non-

linear, least-squares algorithm (Rodgers 1976). The retrieval output is reported here as column concentrations of target gases in parts per million (by volume) multiplied by metres (i.e. ppm m, the product of gas concentration and atmospheric path length; 1 ppm m would be equivalent to a uniform concentration of 1 ppm of gas along a 1 m path at a given atmospheric pressure and temperature). These are equivalent (and readily converted to) to the number of molecules of target gas species per unit area along the observation path. Although we standardized sampling and retrieval techniques as far as possible, several issues are important for analysis and intercomparison of datasets. (i)

Different approaches to retrievals are required for the active (lamp) and solar occultation spectra. This arises because absorptions in the solar spectra occur through the whole atmosphere, whereas the active spectra constitute absorptions across a short atmospheric path, typically 100-2000 m in length. In addition, absorptions within the solar atmosphere need to be fitted for solar occultation spectra. (ii) As the solar zenith angle changes, so does the atmospheric path length, whereas once the lamp and spectrometer are positioned, the path length through the atmosphere remains constant. (iii) The changing path for solar measurements and variations in wind speed also implies that the plume being sampled is of variable age. (iv) There is a choice of absorption features for retrieval of SO2 amounts - a very strong band centred at 1360 cm - 1 , a strong one at 1150 cm - 1 , and a weak one at 2500 cm - 1 . The signal in solar spectra is predominantly shortwave and hence the 2500 cm-1 feature is preferred. All features may be analysed in

626

C. O P P E N H E I M E R ET AL. active spectra, the preference depending on whether sufficient SO2 is present to use the weaker 2500 cm-1 feature, and on the levels of water contamination in the 1150cm -1 and 1360 cm-1 regions (which depend largely on the path length). In particular, the 1360 cm-1 feature can only be exploited at very short path lengths (< 100 m) because of strong contamination of the signal by the presence of water vapour. These differences in contaminating spectral lines, in addition to the different spectral response functions of the MCT and InSb detectors and non-linear response of the MCT detector, result in small variations in retrievals obtained at the different wavelengths.

Our FTIR spectroscopic work at Soufriere Hills has presented us with the greatest challenges of all our field campaigns to date. On the practical side, we have been unable to obtain safe access to monitoring sites that would allow observation paths rich in plume gases. Additionally, cloud frequently developed in the lee of the summit, hampering efforts to obtain solar occultation data. Working at, or near, sea level in the humid tropics ensured that both active and solar spectra were particularly contaminated by water absorptions, further complicating retrievals. In the case of active data collection (Fig. 5b), inspection of spectra indicated that at distances exceeding about 150 m, and at frequencies lower than around 1500 cm - 1 , radiation from the artificial lamp was effectively swamped by infrared emission from the atmosphere itself. In such circumstances, we believe it is best to operate across very short paths (< 100 m) to minimize water contamination, and to utilize the 1360 cm-1 SO2 absorption feature. Alternatively, the infrared source

could be modulated by the interferometer before being directed across the atmospheric path to a detector. For the solar spectra, we had to contend with measuring SO 2 at abundances close to the minimum detection limit, such that fitting contaminating lines accurately became critical. Adequate results for SO 2 were obtained by incorporating a spectral fit for solar atmospheric lines, and by simultaneously fitting N 2 O, CH4. O3 and H2O in the 2460-2540 cm-1 spectral window (Fig. 7a). HC1 was retrieved along with N2O. CH4. and H2O in the 2740-2860 cm-1 range (Fig. 7b). Uncertainties in the spectroscopically derived column concentrations (and gas ratios) are difficult to quantify. We can consider the error of an individual retrieval for a particular gas. or the error of a best-fit regression through many data points. Errors arise in measurements, retrieval procedures and initializing parameters. Paton Walsh el al. (1997) conducted a detailed uncertainty budget based on measurements by several FTIR spectrometers operating side by side, and found uncertainties for several molecules of up to 10%. We have implemented retrievals with spectra containing known gas concentrations (obtained using gas calibration cells for SO 2 and CO) and have found the algorithm to perform to accuracies of 5-10% for individual measurements (Horrocks el al. 2001). We also operated the FTIR spectrometer and COSPEC side by side in July 1998 in order to compare abundances of SO 2 retrieved by the two instruments. These indicated close agreement, though this approach to intercalibration is unsatisfactory because of the difficulty in identifying baseline, and baseline drift, for the COSPEC data, and the differing fields of view of the two instruments. We continue to work on traceable calibration of our FTIR spectroscopic measurements.

Fig. 5. Photographs showing application of (a) solar occultation technique on 7 September 1998 at a site north of Plymouth, and (b) active technique in July 1996. along a road 3 km west of the summit (the lamp is located approximately 500m from the spectrometer).

GAS SURVEILLANCE AT SOUFRIERE HILLS VOLCANO

627

Fig. 5. (continued)

but suggest conservative error estimates for the gas ratios reported here of approximately 10%.

Trends in HC1/SO2 ratios We have re-analysed the 1996 spectral data (Oppenheimer et al. 1998a) and confirmed that SO2 is not present above detection limits (in any of the 1150, 1360, and 2500 cm-1 absorption bands). However, we now believe that only the very shortest path-length data are suitable for SO2 retrievals because of strong emission from the warm atmosphere at low frequencies competing with emission from the infrared lamp. Based on more conservative estimates of the corresponding SO2 detection limits, this implies an HC1/SO2 molar ratio of 5 but still consistent with other measurements obtained in the gas phase that year, including dome fumarole samples (Hammouya et al. 1998) and filter pack data (Allen et al. 2001). Our next data were collected in July 1998, four months after dome growth ceased. For the solar spectra (1998-1999), we can retrieve both HC1 and SO2 column concentrations. No other volcanic gases (e.g. HF) have been identified above detection limits in any of the spectra. Figure 8 and Table 1 combine all the results obtained by solar occultation with the InSb detector. Column concentrations of HC1 were typically in the range of 20-200 ppm m, with a peak of around 500 ppm m observed on 29 July 1998. Corresponding figures for SO2 are a range of 50-400 ppm m and peaks of over 800 ppm m observed on 29 July and 5 September 1998. These are comparatively weak signals compared with our solar occultation measurements elsewhere, for example at Mount Etna, where we have measured HC1 and SO2 column concentra-

tions of up to 1400 and 6000 ppm m, respectively (Francis et al. 1998). As in previous studies (e.g. Francis et al. 1998), the best-fit regression through the origin was determined in order to yield the molar ratio of HC1/SO2 for the time period of interest. Alternatively, of course, gas ratios can be computed for individual spectra. We carried out retrievals in near-real time, and found that the HC1/ SO2 molar ratio had decreased significantly since 1996, to about 0.6. Subsequent measurements obtained in September-November and in February 1999 confirmed this falling trend in the HC1/SO2 ratio. We discuss this pattern in the next section before examining shorter timescale trends.

Long-term trends (months to years) The time series of mean daily HC1/SO2 ratios obtained by solar occultation data is shown in Figure 9. Following the order of magnitude decrease in the HC1/SO2 ratio between 1996 and 1998, the trend continued downwards from an average of 0.63 in July 1998, to 0.58 in September 1998, to 0.39 in October-November 1998, and 0.32 in January 1999 (though note that these mean values for the periods hide some rapid variation discussed in the next section). The 1996 and 1998-1999 periods contrast considerably in terms of volcanic activity. In July 1996, the dome was growing at a rate of up to 10 m3 s-1, whereas our subsequent measurements were obtained after the March 1998 cessation of the first phase of dome growth. Interpreting this trend is particularly important for assessing the implications of the sustained, high SO2 output that was first noted when COSPEC observations resumed in July 1998 (after a 12-month hiatus in observations; Fig. 2). Following a major

628

C. O P P E N H E I M E R ET AL.

Fig. 7. Gas retrievals for spectral windows spanning absorption features for (a) SO2. and (b) HC1. The observed signal and model fits are shown at the top of each graph. Dimensionless transmittance spectra for each retrieved gas are shown beneath, with values offset for clarity. Note that observed and modelled spectra are almost indistinguishable.

Fig. 6. Details of measurement technique: (a) interior view of the spectrometer, which measures approximately 30 x 15 x 15 cm. The HeNe laser tracks the interferometer mirror and provides a wavelength calibration, (b) Principle of the Michelson interferometer. Incoming broadband light is split into a through ray and a reflected ray by the beamsplitter. The reflected ray hits the fixed mirror and is directed to the detector. The through ray hits the moving mirror (travelling at a fixed speed. V) introducing a path difference when it is recombined with the through ray. This results in interference according to mirror position and wavelength, (c) Principle of infrared absorption spectroscopy. The strength, shape and distribution of rotational-vibrational absorption lines for target gas molecules depends on their concentration along the observation path, and their pressure and temperature.

dome collapse on 3 July 1998, SO2 emission rates exceeded 10 kg s-1 for much of the period up to January 1999 (Fig. 2). These high values were enigmatic since they confirmed a general increase in SO2 emission throughout the course of the eruption (for comparison, typical emission rates in mid-1996 were below 3 kg s-1), and beyond the end of active dome growth, contrary to expectations for degassing of a single batch of magma. Observations at other intermediate and silicic volcanoes have indicated decreasing post-effusive SO2 emission rates, e.g. Mount St Augustine, Alaska (Symonds et al. 1990), Redoubt, Alaska (Casadevall et al. 1994) and Galeras, Colombia (Zapata et al 1997).

Conversion of molar ratios to mass ratios, and multiplication by contemporaneous SO 2 fluxes obtained by COSPEC (Fig. 2). indicates that HC1 emission rates decreased from around 9 kg s -1 in mid-1996, to between 0.5 and 4 kg s-1 in 1998-1999 (Fig. 10). In other words, between 1996 and 1998-1999. the HC1 emission rate decreased while that for SO 2 increased. Unfortunately, interpretation of these patterns is not straightforward, in part because understanding of Cl and S degassing from magmas is far from complete. The solubilities of HC1 and SO2 in magma are controlled by several intensive parameters, and are influenced by the crystallization and degassing history of the melt. In particular. SO 2 fluid-melt partitioning is strongly controlled by oxygen fugacity (e.g. Kress 1997: Scaillet et al. 1998). The detailed behaviour of Cl in silicate melts depends also on magma composition, but especially on the content of any vapour phase and the magma ascent rate. It is well known (e.g. Metrich & Rutherford 1992) that Cl has a strong affinity for a water vapour phase, if present, such that deep exsolution of water vapour could extract Cl from the melt. On the other hand. Symonds et al. (1996) explained an increasing Cl S trend observed in gas emissions from Showa Shinzan volcano. Japan, several years after emplacement of a lava dome, as the result of second boiling in which Cl was preferentially exsolved from a residual melt already depleted in S. This latter result is in marked contrast to the declining HC1 SO 2 ratio in gas emissions that we have observed at Soufriere Hills. This observation alone indicates a deeper source of degassing than the lava dome itself. Only a decade after dome emplacement at Showa Shinzan did Cl S decrease, reflecting increased hydrothermal contribution (Symonds et al. 1996). Furthermore, at Soufriere Hills. SO 2 emission remained at high levels in the latter half of 1998 and early 1999 (Fig. 2). even with a large fraction of the first dome removed through gravitational collapse, and with degassing visibly concentrated at vents located beneath the excavated base of the

GAS SURVEILLANCE AT SOUFRIERE HILLS VOLCANO

629

Fig. 8. HC1 and SO2 column concentrations in the plume for all solar InSb spectra obtained in 1998 and 1999 except where SO2 errors exceeded 50%. Typical errors (retrieved simultaneously with gas column concentrations) are 10 ppm m for HC1 and 50 ppm m for SO2. The units reported here (ppm m) represent the product of a nominal plume concentration and the plume path length, for the estimated plume temperature and pressure. This quantity, termed column concentration, is equivalent to, and can be readily converted to, the number of molecules of target gas per unit area in the observation path beneath the plume. The prominent data trends with HC1/S02 ratios of around 0.6-0.7 and 0.2-0.3 represent measurements collected on days with unusually high column concentrations of gas in the plume.

1995-1998 dome (Fig. 4). Below, we outline several possible explanations for the observed long-term decrease in HC1/SO2 ratio, and then consider the relevance of each one for Soufriere Hills. The processes are not necessarily mutually exclusive. (i) (ii)

(iii)

(iv) (v)

(vi)

Fractional degassing of a single magma batch through time (e.g. chlorine removed early, sulphur late). This could be promoted by substantial water degassing early in the eruption. Changing interaction between magmatic and hydrothermal fluids through time. For example, a possible explanation for initially high, then decreasing, chlorine is early assimilation and volatilization of a shallow seawater source of chlorine that was subsequently progressively depleted or isolated from the magmatic system. Changes in speciation of sulphur (e.g. from H2S to SO2). The detection limit for H2S with open-path FTIR spectroscopy is high because the absorption lines are relatively weak and strongly overprinted by water absorptions. It is feasible that, if sulphur were predominantly emitted as H2S in 1996 emissions, our HC1/SO2 ratio is not a good guide to Cl/ (total S) values. Changes in the composition and/or volatile content of degassing magma, e.g. chlorine-rich andesite to sulphur-rich basalt. Change in the depth of the degassing source, i.e. from shallow to deep, progressively tapping a sulphur-rich vapour phase. This could result from progressive depressurization of the magmatic system by shedding of dome lava through time, and/or changes in permeability of the conduit and dome magma. Differences in diffusivity of chlorine and sulphur in the melt, i.e. HC1 has a higher diflfusivity than SO2.

Process (i) has been identified at other lava-dome eruptions, e.g. Mount St Helens, USA, where C and S contents of magmatic gases decreased through time (Gerlach & Casadevall 1986), in parallel with decreasing SO2 emission rate (Casadevall et al. 1983). It does not adequately explain observations at Soufriere Hills because of the low sulphur content of the andesitic magma (Edmonds et al. 2001). If the andesitic magma contributed to the sulphur flux at all, it was probably by degassing of a pre-existing S-rich volatile phase during the syneruptive period (Young et al. 1998). Studies of the 1976 and 1986 eruptions of Mount St Augustine Volcano, Alaska, indicated very high Cl emissions, comparable to those estimated for Soufriere Hills in 1996 (Oppenheimer et al. 1998a). Symonds et al. (1990) and Kodosky et al. (1991) suspected that part of the Cl emission observed at Mount St Augustine derived from assimilation of a shallow seawater source. However,

this conclusion was reached indirectly, based on isotopic evidence for a seawater component to the H2O emission from the volcano. In the case of Montserrat, it is unlikely that substantial mixing of magmatic gases and seawater was possible, because the plumbing system feeding the dome was apparently sealed ofT from the hydrothermal system by precipitation of vapour-phase silica, which limited degassing through country rocks (Boudon et al. 1998). Hammouya et al. (1998) also concluded that the hydrothermal system was largely isolated from the central intrusive conduits under Soufriere Hills because of the very limited disturbance of geochemical signatures of the soufrieres (fumarole fields) on the

Table 1. HCl/SO2 molar ratios retrieved from solar occultation spectra Date

HC1/S02 (SO2 error <50%)

HC1/S02 (SO2 error < 100%)

25 July 1998 26 July 1998 27 July 1998 28 July 1998 29 July 1998 30 July 1998 31 July 1998 3 September 1998 5 September 1998 6 September 1998 7 September 1998 8 September 1998 24 October 1998 25 October 1998 26 October 1998 27 October 1998 31 October 1998 3 November 1998 4 November 1998 5 November 1998 9 November 1998 10 November 1998 29 January 1999 31 January 1999

0.75 0.69 0.50 0.60 0.58 0.65 0.55 0.34 0.24 0.61 0.67 0.48 0.47 0.61 0.21 0.49 0.31 0.20 0.49 0.46 0.46 -

0.75 0.63 0.50 0.60 0.58 0.65 0.54 0.37 0.24 0.61 0.68 0.48 0.47 0.62 0.24 0.50 0.31 0.12 0.21 0.52 0.46 0.46 0.33 0.11

The results indicate best-fit linear regressions through the origin for HC1 versus SO? plots. HC1-SO2 pairs were excluded from the fits for SO2 errors exceeding 50% and 100%. Note that computed ratios for the two error classes are generally very close, indicating that even when poor quality data are included in the fit, the results are realistic.

630

C. OPPENHEIMER ET AL.

Fig. 9. Daily mean HC1/SO2 molar ratios for all InSb spectra obtained during the non-dome-building period in 1998 and 1999 except where SO2 errors exceeded 50%. The late January 1999 data are of poor quality, since only the plume edge was sampled (low column concentrations of plume gases), but there remains an indication of decreasing HC1 SO2 over this six-month period. The downward trend was confirmed by FTIR spectroscopic measurements on 30 October and 3 November 1999 when the HC1 SO2 ratio reached around 0.2.

flanks of the volcano. There were four well developed soufrieres on Montserrat prior to their engulfment by tephra and lahars in the recent eruption (Fig. 3). These soufrieres emitted negligible chlorine, indicating that magmatic chlorine was being condensed within the subsurface hydrothermal system. High 3 He/ 4 He ratios did, however, indicate a magmatic component in the discharges (Chiodini et al. 1996). We cannot exclude the possibility that the NaCl brine circulating in the hydrothermal system (Chiodini et al. 1996) could have provided some of the Cl emitted in the earlier stages of the recent eruption, but we believe that any contribution would have been strongly limited by the aforementioned isolation of the magmatic plumbing system from the hydrothermal system. In any case, it is not necessary to appeal to an external source of chlorine: pre-eruption volatile contents for Cl of 0.23-0.42 wt% were estimated for the andesitic magma from glass inclusion data (Devine et al. 1998; Young et al. 1998), which are sufficient to account for the observed HC1 emission. This is consistent with Cl solubility experiments carried out using recently erupted Soufriere Hills andesite which indicate that any pre-eruptive vapour phase contained negligible Cl (Signorelli & Carroll 2001). We therefore discount the significance of process (ii) for Soufriere Hills.

Fig. 10. Daily mean SO2 emission rates (squares) and corresponding HC1 emission rates (triangles) obtained from measured HC1 SO2 ratios by FTIR spectroscopy (Fig. 2). COSPEC data were collected infrequently and values have been interpolated on days for which FTIR spectra, but not COSPEC data, were available.

Argument (iii) can be discounted because of the oxidized nature of the Soufriere Hills andesite (Devine et al. 1998), and the absence of H 2 S in gas samples collected directly from the active lava dome in 1996 (Hammouya et al. 1998). Even if H2S were the predominant species of magmatic sulphur gases, it would have substantially oxidized in the atmosphere by the time the plume encountered the observation path. We believe that process (iv) provides the best overall explanation for the available data. Petrologic evidence provides strong indications that the eruption was triggered by intrusion of fresh mafic magma into a reservoir of cooler, highly crystallized andesitic magma (Murphy et al. 1998). which reheated (by around 50 C; Devine et al. 1998) and remobilized the more evolved host magma. The intruding mafic magma may also have played a role in triggering the eruption by supplying fresh volatiles. The initially strong HC1 gas emission could have been supported by the andesitic magma with its high pre-eruptive chlorine contents of about 3000 ppm (Edmonds et al. 2001). Falling HC1 SO2 ratio, and rising SO2 emission rate, can then be interpreted as the results of depletion of the available chlorine reservoir as the andesite erupted and degassed, concomitant with enhanced sulphur degassing from an existing vapour phase in the mafic magma. Neither basaltic nor

GAS SURVEILLANCE AT SOUFRIERE HILLS VOLCANO andesitic magma could have yielded the observed sulphur by contemporary degassing of melt, because glass inclusions contain less than SO ppm of sulphur (Edmonds et al 2001). Sparks et al. (1998) suggested that the increasing lava effusion rate in the first two years of the eruption required sustained intrusion of fresh basaltic magma into the reservoir to maintain high driving pressure in an open system. Because the andesitic and basaltic magmas do not appear to have mixed greatly - only about 1-2% of the erupted magma was basalt (Murphy et al. 1998) sulphur contained in a vapour phase in the basalt accumulating at the base of the reservoir may have only gradually migrated to shallow levels, resulting in the decreasing HC1/SO2 ratio. In this way, the HC1/SO2 trend could also be seen to reflect an increase in the depth of degassing, i.e. process (v) above. Post-eruptive degassing from the magma reservoir (located > 5 km depth; Devine et al. 1998) has probably been facilitated by the creation of a gaspermeable structure within the conduit by syneruptive vesiculation (e.g. Hammer et al. 1999). Experimental evidence has shown that sulphur diffuses an order of magnitude more slowly than Cl in silicic melt with water contents of 4-5 wt% (Bai & Van Groos 1994; Watson 1991; Baker & Rutherford 1996). If the source of all Soufriere Hills volatiles apart from H2O has been basalt, it is feasible that the degassing history has been partly controlled by diffusion. This would provide a decreasing Cl/S ratio through time and an increasing SO2 emission rate. If process (vi) has any relevance to Soufriere Hills, we believe it is subsidiary to process (iv) because, by itself, it fails to account for the decreasing HC1 emission rate observed (Fig. 10).

Short-term trends ( minutes to hours) In this section, we consider rapid variations in observed gas ratios. These could reflect: (i) (ii)

real, at-source variations in gas ratios; spatial heterogeneity in the plume arising from geochemically distinct individual vents on, and around, the dome and collapse scar, sampled as wind direction changes such that the optical path moves between plume centreline and margins; (iii) variable rates of HC1 and/or SO2 depletion due to atmospheric effects (gas diffusion into an aqueous phase) or plume composition (ash concentration and deposition, etc); (iv) rapid airborne scavenging of HC1 and/or SO2 and sampling of an ageing plume as measurements are taken towards the end of the day (when solar elevation drops and the plume is effectively being measured further away from its source). Conclusively discriminating between these different processes is difficult. We are duly cautious in attributing cause and effect because we have observed only a limited number of volcanic events with far from ideal temporal coverage of data. However, process (iv) does not appear to be significant over the distances we observed, since there is no consistent relationship between air mass factor and retrieved ratios, even at low solar elevations. Processes (ii) and (iii) are more difficult to quantify, but the plume was generally well mixed by the time it intersected solar observation paths. We also recorded observations of shifting wind direction to assess the impact of any source variability, as well as the visual appearance and ash content of the plume. As will be seen in the following subsections, which examine specific time periods of FTIR spectroscopic observation, we do see changes in gas ratios coincident with variations in volcanic activity that are strongly suggestive of genuine at-source fluctuations in gas chemistry.

HCI/SO 2 ratio changes associated with a minor lava dome collapse, 27 July 1998. On the morning of 27 July 1998, COSPEC measurements indicated an SO2 emission rate of about 13.4 kg s-1, 2.5 times higher than the preceding measurement on 25 July. In the afternoon of 27 July, excellent conditions (favourable wind direction and little

631

cloud) prevailed for FTIR spectroscopic observations, which were obtained from a location in Old Towne (7 km NW of the summit; Fig. 3). Data collection began at 13:45 local time. At 14:09, a dome collapse occurred, initiating a pyroclastic flow that travelled eastwards as far as the delta of pyroclastic deposits accumulated offshore from the Tar River valley (Fig. 3). The accompanying ash cloud reached our observation point at about 14:20, obscuring the Sun. Up to this time, the HC1/SO2 ratio had been around 0.68 (Fig. l la,b). Measurements resumed at 14:50 once the ash cloud had passed. The post-collapse HC1/SO2 ratio was significantly lower (about 0.37), but also more variable on timescales of a few minutes. Data collection ended around 15:10 due to cloud development. A dome collapse might alter the plume composition in several ways: (i) (ii)

(iii)

release of volatiles from the pyroclastic flow and deposit; exposure and depressurization of dome lava immediately beneath the failure surface, initiating degassing through the fresh surface; removal of material from the dome disturbs pressure gradients in the conduit, triggering exsolution, facilitating gas separation from melt, tapping gas trapped in 'pockets', and/or increasing conduit permeability.

Mechanisms (i) and (ii) enhance volatile exsolution from already strongly degassed lavas. Since there is very little sulphur in the andesite, only the HC1 emission should increase appreciably. Mechanism (iii), on the other hand, might be expected to favour SO2 emission by depressurizing a deeper source region. The observed decrease in HC1/SO2 ratio soon after the collapse favours process (iii). Further indication that the change in ratio resulted from enhanced SO2 degassing is seen in Figure 1 1 c, which shows the column concentrations of SO2 and HC1 immediately prior to, and after, the collapse. HC1 column concentrations do not shift appreciably from lOO ppm m whereas SO2 values approximately double from 150 ppm m before the collapse, to 300-400 ppm m about 40 minutes after. This also suggests that the reason for the shift in HC1/SO2 ratio is not the result of altered plume chemistry, since one would expect HC1 to be preferentially scavenged on to liquid-coated ash particles over SO2. Instead, HC1 column concentrations remain stable after the collapse event (Fig. l 1 c). This is consistent with the observation that SO2 emission rates (measured by COSPEC) often increased following dome collapses (Edmonds 2001; Edmonds et al. 2001). The next set of FTIR spectroscopic measurements was obtained in the early afternoon of 28 July, by which time the HC1/SO2 molar ratio had recovered to 0.6-0.7. This would imply that, if the change in ratio was initiated by depressurization of the conduit system, then resealing and pressurization occurred within 24 hours of the event. The rapid drop in HC1/SO2 ratio, within minutes of the dome collapse, suggests a shallow source of enhanced sulphur degassing, possibly indicating tapping of a gas-rich region within the conduit.

HCl/SO2 ratio decrease during an inactive period, 5 September 1998. Activity of the volcano was subdued in early September 1998. There was little seismicity, with only a very few volcanotectonic, longperiod or rockfall signals being recorded each day, and SO2 fluxes measured by COSPEC dropped to around l-2 kg s-1, the lowest levels then observed since the major dome collapse on 3 July 1998 (Fig. 2; Norton et al. 2002). Spectral data were collected daily from 3 to 8 September (inclusive), and each batch of measurements yielded HC1/SO2 molar ratios around 0.5, with one exception. On 5 September, solar occultation data were collected from Plymouth beginning at 15:12, and yielded HC1/SO2 ratios of 0.4-0.5 (Fig. 12a, b). There was then a 75 minute gap in observations, which resumed at 16:42 with an HC1/SO2 ratio of 0.21. The ratio climbed consistently for several minutes, then dropped over a 50 minute period to 0.17. The increasing ratio was associated with an eightfold increase in SO2 column concentrations, in tandem with a more modest rise in HC1 column concentrations (Fig. 12c), but there was no obivous relationship between activity of the volcano and

632

C. O P P E N H E I M E R A T A L .

Fig. 11. HC1 and SO 2 r e t r i e v a l s for 27 July 1998: (a) plots of plume c o l u m n c o n c e n t r a t i o n s of HC1 versus SO2: before (circles) and after ( squares ) a m i n o r dome collapse at 14:09; (b) i n d i v i d u a l HC1 SO 2 r a t i o s versus time: (c) column concentrations of S()2 ( t r i a n g l e s ) and HCl (crosses) in the p l u m e versus t i m e . Representative 1 error bars are shown in ( a ) .

GAS SURVEILLANCE AT SOUFRIERE HILLS VOLCANO

Fig. 12. HC1 and SO2 retrievals for 5 and 6 September 1998: (a) individual HC1/SO2 ratios versus time on 5 September; (b) HC1 and SO2 column concentrations and best-fit regressions for mid-afternoon (triangles) and late afternoon (circles) on 5 September, and 6 September (crosses); (c) column concentrations of SO2 (triangles) and HC1 (crosses) versus time. Some error bars in (b) are omitted for clarity.

633

634

C. O P P E N H E I M E R ET AL.

these variations. The plume was unusually optically dense during this observation period. COSPEC measurements obtained on 4 and 7 September 1998 yielded comparatively low SO2 emission rates of 2.4 and 1.5 kg s-1, respectively. The increases in HC1 and SO2 column concentrations throughout the afternoon of 5 September could reflect the increasing path length through the plume as the Sun set, and/or intersection of the observation path progressively closer to the plume centreline. Over the observation period, the solar elevation dropped from 58 to 23 . Other factors being equal, this would represent a doubling of the path length through the plume, and hence column concentrations of plume gases. This cannot explain, therefore, the eight-fold increase in SO2 column concentrations observed (from 120 to 1000 ppm m; Fig. 12c). In fact, the last values of SO2 obtained on the afternoon of 5 September 1998 are the highest in our entire dataset. This suggests that the overall decrease in HC1/SO2 did not result simply from preferential scavenging of HC1 by aerosol with time (distance from the vent), but that a real change in gas chemistry occurred at source. No geophysical indicators have been matched to this switch in degassing behaviour. The minor excursion in HCl/ SO2 ratio around 16:40-16:50 is also enigmatic.

HCl/ SO2 ratio changes following a large dome collapse, 26 October 1998. Activity increased in October 1998 with five significant collapses that initiated pyroclastic flows down the volcano's main drainages. Dome collapses were typically followed by periods of volcanic tremor and ash-venting and. occasionally, by swarms of volcanotectonic earthquakes. At 00:51 on 26 October 1998. the fourth collapse of the month occurred. The seismic signal of this event lasted for 12 minutes, and was followed by an extended period of tremor. Pyroclastic flows travelled down the Tar River Valley, reaching the sea, and down the White River valley (Fig. 3). We collected spectral data continuously from 13:45 to 17:00 on 26 October 1998. A consistent set of observations on October 25 yields an average HCl SO2 ratio of 0.66. This is similar to the ratio retrieved at the beginning of the 26 October measurements, but the ratio decreased monotonically through the afternoon to approximately 0.1 (Fig. 13a). While HCl column concentrations decreased from about 60 to 30 ppm m SO2 column concentrations initially showed considerable scatter around 120 ppm m but climbed steadily from about 15:45 to more than 200 ppm m (Fig. 13b). Again, this does not appear to be an artefact of the lengthening atmospheric path (as the Sun set) because a similar climbing SO 2 trend

Fig. 13. HCl and SO2 retrievals following a large dome collapse at 00:51 on 26 October 1998: (a) individual HCl SO 2 ratios versus time: (b) column concentrations of SO2 (triangles) and HCl (crosses) in the plume versus time.

GAS SURVEILLANCE AT SOUFRIERE HILLS VOLCANO

after mid-afternoon was observed on other days, e.g. 24 October (Fig. 14a) when HC1 column concentrations kept pace, maintaining a more or less stable HC1/SO2 ratio (Fig. 14b). We speculate that the observed drop in HC1/SO2 ratios signals a real at-source trend, but the reason for such a sustained decline nearly 14 hours after the dome collapse is unclear. One explanation is that this larger shedding of material induced degassing from a sulphur-rich magma at such a depth that the tapped gas did not immediately reach the surface. The HC1/SO2 ratio had recovered to 0.52 when the next measurements were obtained around 13.00 on 27 October 1998. HCl/SO2 ratio changes following a large dome collapse, 31 October 1998. A further significant dome collapse occurred at 04:18 on 31 October 1998, sending a pyroclastic flow down the White River valley, which reached Galway's Soufriere, and another in the Gages valley. The second half of October witnessed enhanced seismicity; on 31 October itself, 24 rockfall events and 16 volcanotectonic earthquakes were recorded. Spectral data were acquired from 08:40 to 11:40 and reveal a low HC1/SO2 ratio of about 0.3 (Fig. 15a). The time-series plot of HC1/SO2 ratio indicates, however, an oscil-

Fig. 14. HC1 and SO2 retrievals for 24 October 1998: (a) column concentrations of SO2 (triangles) and HC1 (crosses) versus time; (b) individual HC1/SO2 ratios versus time. In contrast to the second half of the afternoon on 26 October (Fig. 13), the increase in SO2 column concentrations was tracked by increasing HC1, maintaining a stable HC1/SO2 ratio.

635

lation for individual measurements between 0.2 and 0.4, with decreasing period (Fig. 15b). The corresponding column concentrations of SO2 and HC1 show some correlation with the HC1/SO2 ratio, for example, the small increase in the HC1/SO2 ratio at about 10:05 (labelled 'B' in Fig. 15b) coincided with increases in column concentrations of both species. However, at other times, the ratio behaved differently from the individual column concentrations. For example, at time 'A' indicated on Figure 15b, the HC1/SO2 ratio stabilized around 0.35 after column concentrations of HC1 and SO2 had begun decreasing. At time 'C, HC1/SO2 had decreased sharply to about 0.25, coincident with an increase in SO2. The fluctuations in column concentrations of gas can be explained in at least two ways. (i)

The plume centreline intersected the observation path twice during the 3 hour observation period due to variable wind direction. In this case, the variation in HC1/SO2 ratio could reflect sampling of different sources on the dome with distinct geochemical signatures. (ii) Variations in degassing rates occurred at source. This could have arisen from discrete venting episodes tapping gas pockets within the shallow magmatic system.

636

C. OPPENHEIMER ET AL.

Fig. 15. HC1 and SO2 retrievals following a large dome collapse at 04:18 on 31 October 1998: (a) HC1 and SO2 column concentrations in the plume and best-fit regression for all the data on this day: (b) column concentrations of SO2 and HCl, and HC1 SO2 ratio versus time. The oscillations and periods labelled 'A'. 'B' and 'C' are discussed in the text.

Discussion We have observed an order of magnitude decrease in HC1/SO2 ratios in the gas plume of Soufriere Hills Volcano between 1996 and 1998-1999. The HC1/SO2 ratio trend and changes in emission rates of HC1 and SO2 through time are inconsistent with degassing of a single batch of magma. Instead, they suggest that degassing was initially supported by the water- and chlorine-rich andesite, and subsequently by sulphur-rich basaltic magma, which progressively invaded the feeder reservoir up to at least late 1997, or early 1998. The strong Cl emission early in the eruption was predominantly sustained by syneruptive, low-pressure degassing of andesite. If and when the resupply of basaltic magma ceases, the emission rates of SO2 should drop exponentially, though it is likely that permeability of the volcanomagmatic system also determines SO2 fluxes into the atmosphere. COSPEC operation was very intermittent in 1999, but emission rate measurements prior to the appearance of the second dome in November 1999 did not reach the elevated levels observed in the second half of 1998 (Edmonds 2001). Furthermore, the lack of any extrusive activity between March 1998 and November 1999, even with substantial loss of dome material through collapses,

suggests cessation of fresh intrusions of mafic magma, possibly starting in 1997. If the mafic melt now in the reservoir contains appreciable chlorine, then the lower emission rates of HC1 in 19981999 imply limited exsolution to date, due to the depth of the source, but eventually second boiling may cause an increase in HC1/SO2 ratio, as observed at Showa Shinzan (Symonds et al 1996). In fact, it appears that Cl can exert a strong influence on second boiling because of the strong dependence of Cl solubility on Si, Ca, Al and Mg content of the melt (Webster et al. 1999). Should falling SO2 emission rates be accompanied by inflation, however, this could indicate closing of the magmatic system to gas loss, rather than reduced magma resupply, with quite different implications for hazard assessment. Shorter-timescale fluctuations in HC1 SO2 ratios associated with dome collapses appear to signify transient tapping of sulphur-rich gas from the conduit or reservoir. Larger collapses resulted in more significant decreases in HC1 SO2 ratios, but with longer delays between event and geochemical signature. This can be explained in terms of the greater length scales for the exsolved gases to traverse. Boudon et al. (1998) concluded that gas loss from magma across the conduit walls was probably limited by hydrothermal sealing at

GAS SURVEILLANCE AT SOUFRIERE HILLS VOLCANO

shallow depths. If this remains the case then gas loss must occur primarily along the conduit walls, or through permeable conduit magma. The observation that the HC1/SO2 ratio recovered within less than 24 hours implies that either degassing was initiated by transient exsolution, resulting from the pressure change on dome shedding, or that gas loss, once initiated, was rapidly inhibited through rheological changes, possibly induced by the exsolution event itself. Alternatively, the depressurization events enabled small pockets of pre-existing sulphur-rich vapour to escape. A critical issue at Soufriere Hills is the source of degassing when the dome is not growing. Between the 3 July 1998 dome collapse (which removed approximately 20% of the 1995-1998 dome) and the end of the non-effusive period in November 1999, of the order of 2 x 108 kg of SO2 were emitted in the summit plume. This amount could not have been supported by the dome or conduit magma, which, in itself, immediately points to a reservoir source for the SO2. SO2 emission rates remained above 2 kg s-1 right at the end of the non-effusive period, requiring a permeable network in, or along the walls of, the conduit. Such permeability may have developed through syneruptive vesiculation. We consider also that convection may occur in the lower part of the conduit, providing a means for extracting gas from the reservoir (e.g. Kazahaya et al. 1994). Stevenson & Blake (1998) argued that convection driven by degassing (and consequent density increase of the gas-poor magma) could account for the estimated H2O and CO2 flux at Mount St Helens in the months following its 18 May 1980 eruption. This is a plausible means to extract gas from the magma reservoir beneath Soufriere Hills Volcano, since the intensive parameters are similar to those calculated for Mount St Helens. Additionally, the reheating of the andesitic magma as a result of heat transfer from the basalt will have significantly reduced its viscosity, favouring convection, while the parallel cooling and crystallization of the basalt will have provided the volatiles. The mafic magma could support high sulphur degassing until either the system closes or the sulphur reservoir is depleted. Conclusions Remote measurements of HC1/SO2 ratios in the gas plume emitted by Soufriere Hills Volcano have revealed both long-term and shortterm fluctuations that can be interpreted, respectively, in terms of the deep (magma reservoir) and shallow (conduit and dome) degassing regimes. Improved interpretation of subsurface processes is possible when multiple gas species are quantified. A combination of FTIR spectroscopy with COSPEC has yielded information on gas fluxes as well as ratios. The order of magnitude drop in HC1/SO2 molar ratio from >5 in 1996 to c. 0.6 in the second half of 1998 accompanied an increase in SO2 from c. 3 kg s-1 in mid 1996 to c. lO kg s-1 in mid-1998, and a decrease in the HC1 flux from 9 to less than 3 kg s-1 over the same period. We interpret this in terms of an increasing contribution of SO2 from a reservoir-level basaltic source, and decreasing contribution of HC1 after the end of growth of the first dome in March 1998. HC1 degassing was supported primarily by syneffusive degassing of andesitic magma in the dome and upper conduit. On shorter timescales (from minutes to hours), we observed decreases in HC1/SO2 ratios in the plume following dome collapses. These can be explained by transient degassing of sulphur-rich regions triggered by depressurization and exsolution and/or changes in the permeability of the dome and conduit magma. There is some evidence in the delay between collapse event and plume signature that larger collapses tapped deeper sources of sulphur than smaller collapses. Although our spectroscopic campaign was carried out primarily on an experimental basis, our findings were made available in nearreal time and contributed to operational efforts of the MVO. This helped with assessment of volcanic hazards on Montserrat, notably during the hiatus in lava effusion between March 1998 and November 1999, when it was not known if dome growth would resume. At this time, geochemical signals represented the only

637

obviously anomalous readings from the volcano. These data and their interpretation represent the first real monitoring role played by FTIR spectroscopy at a volcano. Although waning gas emissions will make it more difficult to measure plume gases from Soufriere Hills, increasing safety will permit closer access to vent regions enhancing the gas signal. We strongly suggest that our results on Montserrat warrant the application of Fourier transform spectroscopy for geochemical surveillance on a much more routine and frequent basis elsewhere at degassing volcanoes. The FTIR spectroscopic campaign has been sustained through the second phase of dome growth beginning in November 1999 and is reported elsewhere (Edmonds 2001). Because of the dynamic relationships and feedbacks between volcano seismicity, deformation and degassing (e.g. Gardner & White 2002; Luckett et al. 2002; Neuberg & O'Gorman 2002), the most careful interpretations of volcanic activity require fully integrated geophysical, geodetic and geochemical surveillance programmes. This work was supported by NERC grant GR9/03608, and was greatly enhanced by the generous cooperation of the MVO and BGS. We warmly thank all our colleagues on Montserrat, especially the Chief Scientists in post during our visits, and those who helped with data collection in the field (including D. Pyle, J. Devine, T. Syers, L. Pollard and T. Christopher). A number of colleagues have contributed with beneficial and constructive reviews (B. Rose, S. Williams and T. Druitt) and insightful discussion (including J. Barclay, S. Sparks, L. Horrocks and P. Allard). We are also grateful to Adrian Hayes and Owen Tucker in the Department of Geography at Cambridge for technical and cartographic support, respectively.

References ALLEN, A. G., BAXTER, P. J. & OTTLEY, C. J. 2000. Gas and particle emissions from Soufriere Hills Volcano, Montserrat, West Indies: Characterization and health hazard assessment. Bulletin of Volanology, 62, 8-19. AOPP 2001. Reference Forward Model. Atmospheric, Oceanic and Planetary Physics, Department of Physics, University of Oxford, World Wide Web address: http://www.atm.ox.ac.uk/RFM/ BAI, T. B. & VAN GROOS, A. F. K. 1994. Diffusion of chlorine in granitic melts. Geochimia et Cosmochimia Acta, 58, 113-123. BAKER, L. L. & RUTHERFORD, M. J. 1996. Sulphur diffusion in rhyolite melts. Contributions to Mineralogy and Petrology, 123, 335-344. BOUDON, G., VlLLEMANT, B., KOMOROWSKI, J.-C., ILDEFONSE, P. & SEMET,

M. P. 1998. The hydrothermal system at Soufriere Hills volcano, Montserrat (West Indies): characterisation and role in the on-going eruption. Geophysical Research Letters, 25, 3693-3696. BURTON, M. R., OPPENHEIMER, C., HORROCKS, L. A. & FRANCIS, P. W. 2000. Remote sensing of CO2 and H2O emission rates from Masaya Volcano. Nicaragua. Geology, 28, 915-918. BURTON, M. R., OPPENHEIMER, C., HORROCKS, L. A. & FRANCIS, P. W. 2001. Diurnal changes in volcanic plume chemistry observed by lunar and solar occultation spectroscopy. Geophysical Research Letters, 28, 843-846. CALDER, E. S., LUCKETT, R., SPARKS, R. S. J. & VOIGHT, B. 2002. Mechanisms of lava dome instability and generation of rockfalls and pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 173-190. CASADEVALL, T. J., ROSE, W. I., GERLACH, T. J., GREENLAND, L. P., EWERT, J., WUNDERMAN, R. & SvMONDS, R. 1983. Gas emissions and the eruptions of Mt. St. Helens through 1982. Science, 221, 1383-1385. CASADEVALL, T. J., DOUKAS, M. P., NEAL, C. A., MCGIMSEY, R. G. & GARDNER, C. A. 1994. Emission rates of sulphur dioxide and carbon dioxide from Redoubt volcano, Alaska during the 1989-1990 eruption. Journal of Volcanology and Geothermal Research, 62, 519-530. CHIODINI, G., CIONI, R., FRULLANI, A., GUIDI, M., MARINI, L., PRATI, F. & RACO, B. 1996. Fluid geochemistry of Montserrat island, West Indies. Bulletin of Volcanology, 58, 380-392. DENLINGER, R. P. & HOBLITT, R. P. 1999. Cyclic eruptive behaviour of silicic volcanoes. Geology, 27, 459-462.

638

C. OPPENHEIMER ET AL.

DEVINE, J. D., MURPHY, M. D., RUTHERFORD, M. J. ET AL. 1998. Petrologic evidence for pre-emptive pressure-temperature conditions and recent reheating, of andesite magma erupting at Soufriere Hills volcano, MONTSERRAT, W. I. Geophysical Research Letters, 25, 3669-3672. DOUKAS, M. P. & GERLACH, T. M. 1995. Sulphur dioxide scrubbing during the 1992 eruptions of Crater Peak, Mount Spurr Volcano, Alaska. In: KEITH, T. E. (ed.) The 1992 Eruptions of Crater Peak Vent, Mount Spurr Volcano, Alaska. United States Geological Survey Bulletin, 2139, 47-57. EDMONDS, M. 2001. Sulphur and chlorine degassing at the Soufriere Hills Volcano, Montserrat. PhD thesis, University of Cambridge. EDMONDS, M., PYLE, D. & OPPENHEIMER, C. 2001. A model for degassing at the Soufriere Hills Volcano, Montserrat, West Indies, based on geochemical data. Earth and Planetary Science Letters, 186, 159-173. FRANCIS, P., MACIEJEWSKI, A., OPPENHEIMER, C., CHAFFIN, C., & CALTABIANO, T. 1995. SO2:HC1 ratios in the plumes from Mt. Etna and Vulcano determined by Fourier transform spectroscopy. Geophysical Research Letters, 22, 1717-1720. FRANCIS, P., CHAFFIN, C., MACIEJEWSKI, A. & OPPENHEIMER, C. 1996. Remote determination of SiF4 in volcanic plumes: a new tool for volcano monitoring. Geophysical Research Letters, 23, 249-252. FRANCIS, P., BURTON, M. & OPPENHEIMER, C. 1998. Remote measurements of volcanic gas compositions by solar FTIR spectroscopy. Nature, 396, 567-570. FRANCIS, P., HORROCKS, L. & OPPENHEIMER, C. 2000. Monitoring gases from andesite volcanoes. Philosophical Transactions of the Royal Society of London, 358, 1567-1584. GARDNER, C. A. & WHITE, R. A. 2002. Seismicity, gas emission and deformation from 18 July to 25 September 1995 during the initial phreatic phase of the eruption of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 567-581. GERLACH, T. M. & CASADEVALL, T. J. 1986. Fumarole emissions at Mount St. Helens volcano, June 1980 to October 1981; degassing of a magmahydrothermal system. Journal of Volcanology and Geothermal Research, 28, 141-160. HAMMER, J. E., CASHMAN, K. V., HOBLITT, R. P. & NEWMAN, S. 1999. Degassing and microlite crystallization during pre-climactic events of the 1991 eruption of Mt. Pinatubo, Philippines. Bulletin of Volcanology, 60, 355-380. HAMMOUYA, G., ALLARD, P., JEAN-BAPTISTE, P., PARELLO, F., SEMET, M. P. & YOUNG, S. R. 1998. Pre- and syn-eruptive geochemistry of volcanic gases from Soufriere Hills of Montserrat, West Indies. Geophysical Research Letters, 25, 3685-3688. HORROCKS, L., BURTON, M., FRANCIS, P. & OPPENHEIMER, C. 1999. Stable gas plume composition measured by OP-FTIR spectroscopy at Masaya volcano, Nicaragua, 1998-1999. Geophysical Research Letters, 26, 3497-3500. HORROCKS, L. A., OPPENHEIMER, C., BURTON, M. R., DUFFELL, H. R., DAVIES, N. M., MARTIN, N. A. & BELL, W. 2001. Open-path Fourier transform infrared spectrometry of SO2: An empirical error budget analysis, with implications for volcano monitoring. Journal of Geophysical Research, 106, 27647-27659. JAUPART, C. & ALLEGRE, C. 1991. Gas content, eruption rate and instabilities of eruption regime in silicic volcanoes. Earth and PlanetaryScience Letters, 102, 413-429. KAZAHAYA, K., SHINOHARA, H. & SAITO, G. 1994. Excessive degassing of Izu-Oshima Volcano: magma convection in a conduit. Bulletin of Volcanology, 56, 207-216. KODOSKY, L. G., MOTYKA, R. J. & SYMONDS, R. B. 1991. Fumarolic emissions from Mount St. Augustine, Alaska: 1979-1984 degassing trends, volatile sources and their possible role in eruptive style. Bulletin of Volcanology, 53, 381-394. KRESS, V. 1997. Magma mixing as a source for Pinatubo sulphur. Nature, 389, 591-593. LOVE, S. P., GOFF, F., COUNCE, D., SIEBE, C. & DELGADO, H. 1998. Passive infrared spectroscopy of the eruption plume at Popocatepetl Volcano, Mexico. Nature, 396, 563-567. LUCKETT, R., BAPTIE, B. & NEUBERG, J. 2002. The relationship between degassing and rockfall signals at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 595-602.

MARTEL, C., PICHAVANT, M., BOURDIER, J.-L., TRAINEAU. H.. HOLTZ, F. & SCAILLET, B. 1998. Magma storage conditions and control of eruption regime in silicic volcanoes: experimental evidence from Mt. Pelee. Earth and Planetary Science Letters. 156, 89-99. MAYBERRY, G. C.. ROSE, W. I. & BLUTH. G. J. S. 2002. Dynamics of volcanic and meteorological clouds produced on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT. T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London, Memoirs, 21, 539-555. METRICH, N. & RUTHERFORD. M. J. 1992. Experimental study of chlorine behavior in hydrous silicate melts. Geochimia et Cosmochimia Acta. 56, 607-616. MURPHY, M. D., SPARKS, R. S. J.. BARCLAY. J. ET AL. 1998. The role of magma mixing in triggering the current eruption at the Soufriere Hills volcano, Montserrat. West Indies. Geophysical Research Letters. 25. 3433-3436. NAKADA, S. & YOSHINOBU, M. 1999. Petrology of the 1991-1995 eruption at Unzen; effusion pulsation and groundmass crystallization. Journal of Volcanology and Geothermal Research. 89. 173-196. NAKADA, S., MOTOMURA, Y. & SHIMIZU. H. 1995. Manner of magma ascent at Unzen volcano (Japan). Geophysical Research Letters. 22. 567-570. NEUBERG, J. & O'GORMAN, C. 2002. A model of the seismic wavefield in gas-charged magma: application to Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21. 603-689. NORTON, G. E.. WATTS. R. B., VOIGHT. B. ET AL. 2002. Pyroclastic flow and explosive activity at Soufriere Hills Volcano, Montserrat. during a period of virtually no magma extrusion (March 1998 to November 1999). In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs. 21. 467-481. OPPENHEIMER, C. 1996. On the role of hydrothermal systems in the transfer of volcanic sulphur to the atmosphere. Geophysical Research Letters. 23, 2057-2060. OPPENHEIMER. C.. FRANCIS, P. & MACIEJEWSKI. A. 1998a. Spectroscopic observation of HC1 degassing from Soufriere Hills volcano. Montserrat. Geophysical Research Letters. 25. 3689-3692. OPPENHEIMER. C., FRANCIS, P. & MACIEJEWSKI. A. 1998b. Volcanic gas measurements by helicopter-borne fourier transform spectroscopy. International Journal of Remote Sensing. 19. 373-379. OPPENHEIMER. C., FRANCIS, P.. BURTON, M.. MACIEJEWSKI. A. & BOARDMAN, L. 1998c. Remote measurement of volcanic gases by Fourier transform infrared spectroscopy. Applied Physics B. 67. 505-515. OPPENHEIMER, C., FRANCIS. P. & STIX. J. 1998d. Depletion rates of SO2 in tropospheric volcanic plumes. Geophysical Research Letters. 25. 2671-2674. PATON WALSH. C., BELL, W., GARDINER, T. ET AL. 1997. An uncertainty budget for ground-based Fourier transform infrared column measurements of HC1, HF. N2O. and HNO 3 deduced from results of side-byside instrument intercomparisons. Journal of Geophysical Research. 102, 8867-8873. ROBERTSON, R. E. A., ASPINALL, W. P.. HERD. R. A.. NORTON. G. E.. SPARKS, R. S. J. & YOUNG. S. R. 2000. The 1995-1998 eruption of the Soufriere Hills Volcano. MONTSERRAT. W. I. Philosophical Transactions of the Royal Society of London. 358. 1619-1637. RODGERS, C. D. 1976. Retrieval of atmospheric temperature and composition from remote measurements of thermal radiation. Reviews of Geophysics and Space Physics. 14. 609-624. SCAILLET. B., CLEMENTE, B.. EVANS, B. W. & PICHAVANT. M. 1998. Redox control of sulphur degassing in silicic magmas. Journal of Geophysical Research. 103, 23 937-23 949. SIGNORELLI, S. & CARROLL, M. R. 2001. Experimental constraints on the origin of chlorine emissions at the Soufriere Hills volcano. Montserrat. Bulletin of Volcanology. 62. 431-440. SPARKS, R. S. J. 1997. Causes and consequences of pressurisation in lava dome eruptions. Earth and Planetary Science Letters. 150, 177-189. SPARKS, R. S. J. & YOUNG. S. R. 2002. The eruption of Soufriere Hills Volcano, Montserrat (1995-1999): overview of scientific results. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society. London. Memoirs, 21, 45-69.

GAS SURVEILLANCE AT SOUFRIERE HILLS VOLCANO SPARKS, R. S. J., YOUNG, S. R., BARCLAY, J. ET AL. 1998. Magma production and growth of the lava dome of the Soufriere Hills Volcano, Montserrat, West Indies: November 1995 to December 1997. Geophysical Research Letters, 25, 3421-3424. SPARKS, R. S. J., BARCLAY, J., CALDER, E. S. ET AL. 2002. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boxing Day) 1997 at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 409-434. STEVENSON, D. S. & BLAKE, S. 1998. Modelling the dynamics and thermodynamics of volcanic degassing. Bulletin of Volcanology, 60, 307-317. STIX, J., ZAPATA, J. A., CALVACHE, M. ET AL. 1993. A model of degassing at Galeras Volcano, Colombia 1988-1993. Geology, 21, 963-967. STIX, J., TORRES, R. A., NARVAEZ, L., CORTES, G. P., RAIGOSA, A. J., GOMEZ, D. M. & CASTONGUAY, R. 1997. A model of vulcanian eruptions at Galeras Volcano, Colombia. Journal of Volcanology and Geothermal Research, 77, 285-303. SYMONDS, R. B., ROSE, W. I., GERLACH, T. M., BRIGGS, P. H. & HARMON, R. S. 1990. Evaluation of gases, condensates, and SO2 emissions from Augustine volcano, Alaska: the degassing of a Cl-rich volcanic system. Bulletin of Volcanology, 52, 355-374. SYMONDS, R. B., MIZUTANI, Y. & BRIGGS, P. H. 1996. Long-term geochemical surveillance of fumaroles at Showa-Shinzan dome, Usu volcano, Japan. Journal of Volcanology and Geothermal Research, 73, 177-211. VILLEMANT, B. & BOUDON, G. 1998. Transition from dome-forming to plinian eruptive styles controlled by H2O and Cl degassing. Nature, 392, 65-69.

639

VILLEMANT, B. & BOUDON, G. 1999. H2O and halogen (F, Cl, Br) behaviour during shallow magma degassing processes. Earth and Planetary Science Letters, 168, 271-286. VOIGHT, B., SPARKS, R. S. J., MILLER, A. D. ET AL. 1999. Magma flow instability and cyclic activity at Soufriere Hills volcano, Montserrat, British West Indies. Science, 283, 1138-1142. VOIGHT, B., KOMOROWSKI, J.-C., NORTON, G. E. ET AL. 2002. The 26 December (Boxing Day) 1997 sector collapse and debris avalanche at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to 1999. Geological Society, London, Memoirs, 21, 363-407. WATSON, E. B. 1991. Diffusion of dissolved CO2 and Cl in hydrous silicic to intermediate magmas. Geochimia et Cosmochimia Acta, 55, 1897-1902 WATSON, I. M., OPPENHEIMER, C., VOIGHT, B. ET AL. 2000. The relationship between degassing and deformation at Soufriere Hills volcano, Montserrat. Journal of Volcanology and Geothermal Research, 98, 117-126. WEBSTER, J. D., KINZLER, R. J. & MATHEZ, E. A. 1999. Chloride solubility in basalt and andesite melts and implications for magmatic degassing. Geochimia et Cosmochimia Acta, 63, 729-738. YOUNG, S., FRANCIS, P. W., BARCLAY, J. ET AL. 1998. Monitoring SO2 emission at the Soufriere Hills volcano: implications for changes in eruptive conditions. Geophysical Research Letters, 25, 3681-3684. ZAPATA, J. A., CALVACHE, M. L., CORTES, G. P. ET AL. 1997. SO2 fluxes from Galeras Volcano, Colombia 1989-1995; progressive degassing and conduit obstruction of a Decade Volcano. Journal of Volcanology and Geothermal Research, 77, 195-208.

Index Page number in italics refer to Figures; page numbers in bold refer to Tables accretionary lapilli 63, 437, 441, 444, 450, 497, 523, 540 accumulation rate, tephra 504-506 advanced very high resolution radiometer (AVHRR) 541, 542, 549 aerosol index (AI) 545-546 aggregation of tephra 506-508, 514 accretionary lapilli 437, 441, 444, 450, 497, 523, 540 effect on tephra fallout modelling 525-526, 535 processes 523-524 air quality monitoring programme analytical data 558-565 testing schedule 559 airborne ash see ash plumes also tephra fallout airport 215, 222-223 inundation 34 management 19-20 alert system 12, 13, 80, 81-82, 85 amphibole 186 see also hornblende andesite 51, 64-65 composition 45, 99, 156, 186, 496-497 mineralogy 55 modelling parameters 157, 160 rheology 139, 156-157 40 Ar/ 39 Ar geochronology 65, 100-109 ash masses in clouds 540 ash plumes 175 behaviour 483 character 287-289 defined 191, 211, 483 deposits 253-254 dynamics of 430 formation 286, 287 height 292, 293, 303, 472, 474, 476, 519, 526-527, 561 June 25, 1997 event 205, 220 modelling 324-342 structure 492 see also tephra fallout ash-cloud surges see pyroclastic surges ash-venting 12, 17, 39, 477-478, 491 airborne particle concentration 560, 561 autobrecciation 186 avalanche see debris avalanche Aymer's Ghaut 449 ballistic blocks 281, 286, 300-301 basalt magma 64-66 bedforms in pyroclastic density current deposits 446-448 Belham River valley 15, 40, 62, 176, 797, 198, 266, 268-269, 276, 277 Bethel 196, 797, 198, 277, 221 block-and-ash flows see dome-collapse pyroclastic flows block-poor pyroclastic density current deposits 415, 438 block-rich pyroclastic density current deposits 415, 439 Boxing Day (1997) event 8, 16, 17, 53, 133 see debris avalanche also Galway's Wall collapse also pyroclastic density currents also volcanogenic meteorological (VM) clouds Bramble village 196, 221, 225 brightness temperature difference (BTD) technique 542-543, 550 building structure damage 52, 797, 198, 204, 216, 217, 219, 221, 222, 453-454 13 C signature of fumaroles 5 Caribbean Disaster and Emergency Response Agency 19

Castle Peak Dome 5, 6, 24, 25, 45, 94, 97, 111, 117, 116, 121, 122, 723, 126, 727, 729, 132, 135, 136,409, 570 cataclastite 63, 245 Centre Hills 5, 93, 94, 95 40 Ar/ 39 Ar geochronology 101, 109 Chances Peak 5, 45, 54, 94, 97, 116, 121, 130, 353, 475, 476 40 Ar/ 39 Ar geochronology 105, 106 petrology 99 tiltmeter readings 180, 192, 193, 213, 355, 356 clinopyroxene 51, 99, 156, 496 co-pyroclastic flow plumes defined 483, 517 coastal fans 254-256, 258 cohesion avalanche deposits 393 tuffs 394 collapse structure 349-350 conduit studies application of seismic wavefield model 606-607 diameter 160 flow modelling during explosion 310-316 flow modelling during extrusion 159-164 geometry 155-156 initial conditions modelling 320-345 pressure in explosions 303-304 surrounding seismic wavefield model 604-608 convective overturning 65 cooling and dome morphology 146, 151 Cork Hill 14, 267, 268-269 correlation spectrometer (COSPEC) 80, 472, 473, 622-624, 626 crack measurement 355-356 crater-wall landslides 177 cristobalite health hazard 64, 558, 565 cross-stratification 444 in pyroclastic surge deposits 249, 252, 260 crystals content in magma 310 current direction indicators 451-454 growth rate 160 nucleation 158-159 dacite 99, 106 deaths 224, 263 debris avalanche 415-416, 419, 435 deposits block facies 373 Chartreuse Subunit 375-376, 382, 383, 386, 389 distribution 367-373 Fumarolic Unit 375-376, 384, 393-394 geotechnical properties 392-394 granulometry 391-392, 393 Halo Unit 376-377 hummocks 380, 416 matrix facies 373, 374 megablock facies 373, 382, 384 non-cohesive facies 378 relief and morphology 380-387 Soufriere Subunit 375, 382, 384, 389 Talus Unit 377-380, 382, 383, 386 texture 373-380 volume 416, 420 dynamics of 430 emplacement 391, 404 flow behaviour 404 modelling emplacement 401-402 precursory activity 363-365, 409-411 relation to pyroclastic current 387-391 velocity 391, 430, 435 deformation 59, 79, 165-167, 283, 294, 471-472 Galway's Wall 351-354 information from radar 590-591 phreatic phase 569, 576 see also tilting and tiltmeter data degassing of magma 59, 148 exsolution of gas 138-139 and induced crystallization 146-149, 151

monitoring by MVO 80 and rockfalls 599, 600 see also gas dense rock equivalent (DRE) 176, 342 density clasts in block-and-ash flow deposits 246, 247 crystals 160 dome bulk density 467 magma 310 tuff 394 deuterium enrichment 65 diffusion coefficient, in tephra fallout numerical model 522, 525 dixytaxitic texture 55 dome see lava dome dome-collapse pyroclastic flows 9, 26, 27, 60, 191, 235, 241, 477 block-and-ash flows 14, 50, 51, 52, 61, 62, 175, 191, 194, 195, 797, 198, 199, 200, 216, 235-238, 259, 283 damage caused 225 defined ix, 211 deposit distribution 272, 214, 233, 234, 236 deposit lithology 242-245 deposit particle size 203, 242, 244 eyewitness accounts 216 Farrell's Plain 235 front velocities 195, 236 geographic distribution 238 impact 238 morphology 238-242, 256-258 Spanish Point 243 Tar River valley 231, 234, 244, 250 temperature 245 Tuitt's Ghaut 231, 235, 250, 296 volumes 264 White River valley 231, 235, 242 causes 187-188 cyclicity of activity 180 generation 186-188 in 1996 14, 26, 232 in 1997 15-17, 29, 61, 192, 232 compared with other flows 207 hazard notes 207-208 particle size analysis 203 pulses 195-198, 199, 200 volume 205-206 in 1998 17, 232, 473, 474 in 1999 232 magnitude-frequency relationships 183-186 rockfalls 192, 231, 235-236, 247, 355, 470, 471, 472, 475, 476 classification 174-175, 775 duration 183, 185 magnitude 184-185 pyroclastic flow trigger 187-188 seismic signals 59, 177-178, 595-600 runout 175, 245, 522 seismicity and flow generation 177-180 and tephra 485-488, 493-494 tephra health hazards 517 terminology defined ix velocity 177 volume 174, 182, 523 Dry Ghaut pyroclastic density current 438 surge-derived pyroclastic flow 269, 273, 274, 277 Dyer's bridge 265-267, 268 Dyer's River valley 792 dynamic pressure 200, 224-225, 252 earthquakes 9, 12, 177, 567-568 explosion signals 59, 471, 475, 476 phreatic phase 571, 573, 575 hybrid earthquake swarms 59,176, 177, 179-180, 283, 354, 470-471, 472, 603, 611 dome-collapse events 191, 192, 794, 195, 213, 264, 269 phreatic phase 571, 573, 574

INDEX

642

earthquakes (continued) long-period earthquakes 59, 177-179, 354-355, 470-471, 472, 603, 611 phreatic phase 571, 573, 575 rockfall signals 595 rockfall seismic signals 59, 177-178, 192, 231, 235-236. 241, 355, 470. 471, 472, 475. 476 analysis 596-600 characteristics 595-596 tremor 177, 572, 573. 575, 603, 611, 615-618 volcanotectonic earthquakes 59, 177, 354, 470, 471, 472, 474, 475, 476 phreatic phase 571, 572, 573. 574 electronic distance measurement (EDM) 78. 471, 472 emergency management 1, 7-8, 12-13, 18-19, 227-230 English's Crater 5, 6, 45, 52, 94, 95, 97, 111, 116, 119, 121, 123, 351, 231, 235, 264, 269, 350, 409, 570 erosion pyroclastic density currents 451-454, 454-455 pyroclastic flows 254, 257, 260 eruption chronology 5-12, 13-18, 48-49, 167-168 eruption stages 117, 121-136, 155, 167-168 evacuation procedure 1, 78, 14, 15, 19, 21 exit pressure 310, 311, 312, 313, 314 exit velocity 314, 156 explosive activity 9-11, 14, 167 cyclicity 301-302 dome fragmentation 460-462 intensity 56, 148 interval 281, 293 mechanisms 302-303 seismic signals 59, 177, 294-296, 471, 475, 476 phreatic phase 571, 573, 575 sub-Plinian 489, 494 see also Vulcanian explosions 489-491, 494-496 extrusion features 141-150 rates 180-182, 187, 193 eyewitness accounts of pyroclastic flows 215-223, 265, 412-413 airport 215, 222-223 Bethel 2/7, 221 Bramble village 221, 225 Dyer's 219, 223, 225 Farrell's Yard 215, 217-218. 225 Harris 219-220 St George's Hill 215 Smoky Hill 220 Streatham 215, 219, 227, 223-224, 225, 265 Trant's Yard 221-222 Tuitt's 215, 221 Windy Hill 218-219, 223, 225 failure forecast method (FFM) 358-359 fallout pumice 297 fan formation 254-256, 258 Farm River valley 196, 201 Farrell's Plain ash plume 254 block-and-ash flows 235 June 25, 1997 surge 265, 266 Farrell's Yard 15, 215, 217-218, 225 Fergus Mountain 96, 421, 422, 423, 429, 446 financial impacts 1 flank instability see lava-dome collapse fluidization effects 225, 454 Fort Ghaut 15 1997 flows 176, 283 rockfalls 235 tephra fallout modelling 532, 533, 534 fountain-collapse pyroclastic flows 9-11, 35, 235, 290, 291, 320, 477 defined ix 1997 events 15, 33, 60 runout 522 volume 523 White River valley 247 fountain-collapse pyroclastic surges 286, 287, 290, 295

Fourier transform infrared spectrometry (FTIR) 472, 621, 624-637 fragmentation wave 460-462 Francis, Peter William xiii-xv friction angle avalanche deposits 393 tuffs 394 FTIR spectroscopy see Fourier transform infrared spectroscopy fumaroles 5, 26, 286 Gages Dome 40 Ar/ 39 Ar geochronology 105 petrology 99 Gages Ghaut 213 Gages Mountain 5, 45, 94, 97, 116. 475 Gages Soufriere 569-570 Gages valley, 1997 flows 282, 283 Gages Wall 351 tuff geotechnical properties 394. 395. 396 Galway's Mountain 5, 45, 54, 94, 97, 116 40 Ar/ 3 9 Ar geochronology 105 petrology 99 Galway's Soufriere 351, 353, 363. 366, 569-570 Galway's Wall collapse 130, 365 cause 403-404 chronology 366, 411-413 chute 254. 257 description 351 early deformation 351-354 early stress signs 349 instability monitoring 354-356

people's perceptions 226-227 sector collapse prediction 357-359 surge-derived pyroclastic flows 207-208 zonation and microzonation 10. 11. 12. 13. 14. 81, 82. 83 hazardous deposit threshold 529 HAZMAP model 518-519 output 523 parameters 519-523 results 529-536 HC1 detection 226. 621 ratio with SO 2 627-635 3 He 4 He signature of fumaroles 5 health hazards and monitoring 1. 517. 557. 558-565 hornblende 51. 55, 99. 156. 243. 283. 496 Hot River fan 254 hot springs, precursor activity 5. 26 Hurricane Hugo 5. 7. 19 Hurricane Luis 575 hybrid earthquake swarms 59. 150. 283. 354. 470-471. 472. 603. 611 dome-collapse events 191. 192. 194. 195. 213. 264, 269 phreatic phase 571. 573. 574 hypersthene 99 impedance contrast 605-606 inclusions, mafic 55. 64 injuries 224. 225-226 isentropic wind trajectory 548 isopach maps, tephra fallout 453. 490. 494. 495

map 371, 372 mitigation 403 modelling 458-463 sector collapse buildup 356-357 sector collapse event 357, 417, 418 sector collapse hazard prediction 357-359 stability assessment post-collapse 366, 372, 396-398. 41-415 precollapse 352. 363-365, 371. 395-396. 409-411 tuff geotechnical properties 394, 395, 396 volume 366, 415 Galway's Wall 1996 crisis 15, 24. 124 Garibaldi Hill 5. 93, 98, 116 40 A r 3 9 A r geochronology 106. 109 gas bubble behaviour modelling 308-315 charged magma wavefield modelling 603-608 emission monitoring 226, 474 correlation spectrometer (COSPEC) 80, 472. 473, 622-624, 626 Fourier transform infrared spectroscopy 472. 621,624-637 see also HC1; H2S; SO2 expansion 186-187 exsolution effects in explosion modelling 334 role in modelling dome pressurization 458-460 venting and rockfalls 599. 600 see also degassing of magma geochemistry 99 geochronology using 40 Ar/ 39 Ar 65. 100-112 geostationary observational environmental satellite (GOES) 539-542, 549 geotechnical property testing 405 avalanche deposits 392-394 tuffs 394-395 Germans Ghaut 276, 431, 438, 443 Gingoes Ghaut 276, 423, 438, 449 glass in pumice and rhyolite 138. 157, 186 grading in pyroclastic density current deposits 445-446 grain aggregation in tephra see aggregation granulometry and grain-size analysis see particle size ground deformation see deformation also tilting and tiltmeter data H2S 226, 569 hazard assessment and management 83-84 dome-collapse pyroclastic flow 207-208 map 2/2

kaolinite 375 lahars 17. 40. 41. 177 lapilli 64. 437 lava dome 52. 54. 61. 193 chronology of activity 512. 115-136. 472-476. 478-480 conduit size 155-156 explosive fragmentation 460-462 extrusion rate 181-182. 187. 193. 310 extrusive morphology 141-150 feature map 469 height variation with time 154. 155 instability 57 lobes 115. 118. 124. 126. 128. 129. 131. 137 Easter Lobe 126. 136. 138 pancake lobes 148. 149 Santa lobe 125-126. 131. 134 shear lobes 122, 124. 141-142. 148. 181. 186 Venus lobe 124 megaspines 122. 124. 725. 126. 729. 132. 133. 141. 148. 149 pressurization 156.458-460 rheology and petrology 136-141 role of degassing 59-60. 146-149. 151 shear faults 143 spines 38. 54. 60. 119. 121. 122. 123. 132. 148. 149. 472 volume variation with time 49. 80. 777. 154. 755, 191. 193. 472 whaleback structures 727. 122. 148. 149 lava-dome collapse 727. 174-177. 263-264. 349-350 airborne particle concentration 560, 561. 565 chronology 9-12. 264-273. 282. 285. 435. 472-476 effect on gas emission 628-629. 631. 634-635. 636

extrusive morphology 141-143 lava-dome growth 9. 25. 38, 56. 116. 120. 154-155. 191 chronology 14. 15. 17. 52. 54. 284 cycles 143-145 maps 119. 120. 125. 127. 128. 131. 137. 140. 141. 144. 146 models 145-149 relation to pyroclastic flow 180-183. 246 relation to rockfalls 235-238. 599. 601 Lesser Antilles arc 2 levees 256-258

INDEX lobes see under lava dome long-period seismic events 59, 354-355, 470-471, 472, 603, 611 phreatic phase 571, 573, 575 rockfall signals 595 mafic inclusions 55, 64 magma crystal content 160, 310 degassing 59-60 density 310 discharge rate 155, 181-182, 187, 193, 310, 320 eruption modelling see under numerical modelling extrusion pulsation pattern 56-58 extrusion rates 180-182, 187, 193 movement in phreatic phase 576-577 permeability 157-158, 159-161 porosity 60, 157 pressure 59, 60, 156, 160, 162, 164, 224-225, 310 temperature 60, 156, 160, 310, 320 viscosity 60, 156, 160, 161-162, 186 volume variation with time 49, 58, 320 water content 60, 156, 160, 186, 310 magma chamber depth 160, 310 pressure 156, 160, 310 megaspines 122, 124, 725, 126, 129, 132, 133, 141, 148, 149 microzonation 12, 82 modelling see numerical modelling Montserrat administration 71-73 airport 19-20 climate 5 island emergency schedule 7-8, 12-13 map 3, 46, 468, 584, 623 physiography 5, 21, 22, 45, 93 risk management maps 10, 11, 82, 83, 84 setting 2 volcanic chronology 5-12, 6-7, 95 Montserrat Volcanic Observatory (MVO) administration 71-73 crisis management 80-84 evolution 73-77 locations 72 monitoring work 77-80 public information work 84-88 morphology of extrusive features 141-150 Morris' village 429, 448, 451 Mosquito Ghaut 15, 213, 237 1997 pyroclastic flows 62, 176, 192, 194, 195, 196, 198, 200, 202, 203, 283, 293 block-and-ash flows 792, 231, 237 co-pyroclastic flow plumes 486 pyroclastic surge 237, 249 rockfalls 235 mud rain 523 nucleation 158-159, 160 numerical modelling conduit flow 308-316 equations 159, 168-169 parameters defined 154-158 results 159-164 conduit initial conditions 320-334, 344-345 video record and model compared 334-342 current velocities 431-432 debris avalanche emplacement 401-402 deformation 398-401 dome pressurization and fragmentation 458-462 eruption behaviour 159-164 parameters employed 160, 310, 325, 519, 604-605 particle transport 462-463 seismic waves 603-609 stratified density current 463-464 tephra fallout 518-536 obituary xiii-xv Old Road Bay 428, 430 olivine 99

orthopyroxene 51, 55, 99, 156, 186, 283, 496 overpressure of magma 59, 60, 156, 160, 162, 164, 224-225

Pan-Caribbean Disaster Prevention and Preparedness Program 18, 19 Paradise Ghaut 257 Paradise River valley 265 particle aggregation see aggregation particle size ash plumes 205, 327-342 avalanche deposits 391-392, 393 flow deposits block-and-ash 203, 242, 244 pumice-and-ash 247 pyroclastic density current 441, 444, 445, 448-450 pyroclastic surge 203-205, 250, 255 surge-derived pyroclastic flow 205, 206, 273, 275 tephra 484, 500-504, 519, 526 particle transport modelling 462-465 Pea Ghaut 196, 198, 200 Perches Mountain 5, 45, 94, 97, 116, 476 40 Ar/ 39 Ar geochronology 105, 106 petrology 99 permeability of magma 157-161, 160 phenocrysts 51, 55, 64, 99, 156, 186, 496 effect of concentration 139-140 size effect on eruption 162 phreatic phase 567 explosions 9, 13-14, 24, 485 monitoring 567-579 plagioclase 51, 55, 99, 156, 186, 243, 283, 496, 508, 570, 572 plate tectonic setting 2 plumes see ash plumes also tephra fallout Plymouth 41, 52 fan 256 pumice-and-ash flows 244, 247 ponding in pyroclastic density current deposits 439, 442 porosity of eruption products 60, 157, 167 pressures associated with magma chamber 156 in conduit in explosions 303-304 pressurization of dome, numerical modelling 458-460 pseudotachylite 63, 64, 245 pumice-and-ash flows 247, 287, 295 deposits distribution map 238, 240 lithology 247-248 morphology 246-247 particle size distribution 247 volumes 298 impacts 246 temperature 245 terminology ix velocity 248 pyroclastic density currents 11-12, 17, 37, 60, 269, 387, 389, 390-391, 416-428, 435-455 deposits bedforms 446-448 block-poor 415, 438 block-rich 415, 439 grading 445-446 granulometry 448-450 lithofacies 436-439 particle size 441, 444, 445, 448-450 stratification 444-445, 445 dynamics 430-432 erosional features 451-454, 454-455 regional variations 439-444 tar formation 423, 387 terminology ix-x velocity 275, 276, 430, 431-432, 457 pyroclastic flows 53 calendar of events 232 causes 173, 187-188 components 175 cyclicity of 180 damage 225

643 defined 9, 211 deposits distribution maps 16, 213, 236, 586 volume 49, 174, 175, 176 development over 1995-98 period 7-8, 231-235 erosional features 254, 257, 260 eyewitness accounts 215-223, 265 fluidization 225, 454 generation 177-180, 186-187 magma extrusion rate effect 180-183 magnitudefrequency relations 183-186 precursory activity 213 runout 176, 182, 183-185, 522 velocity 173, 177 Tar River valley events 173, 174, 175, 176, 179, 185-186, 234, 247 terminology ix, 231 velocity 215, 216, 225 with Vulcanian explosion 246-249 see also dome-collapse pyroclastic flows also fountain-collapse pyroclastic flows also surge-derived pyroclastic flows pyroclastic surges 52, 62, 175, 254, 420, 474, 476 behaviour 207 causes 187 damage caused 52, 225, 252 defined 211 deposits 260, 265 particle size 203-205, 206, 273, 275 volumes 265, 269 development 259-260 Dry Ghaut event 269-273 effects of 249 events September 17 1996 250-251 June 25 1997 191, 195, 198-205, 265 July 3 1998 251-252 eyewitness accounts 216 features 249-250, 254 Tar River valley 234 temperature 245 terminology ix topographic effects 195, 198-203, 225, 267, 269-270, 277, 272 velocities 225, 249, 274 pyroclastics dispersion modelling 324-342 see also under Vulcanian explosions pyroxene see clinopyroxene also orthopyroxene quartz 51, 99, 283, 496, 558, 565 radar see synthetic aperture radar rainfall 5, 563 real-time seismic amplitude measurement (RSAM) 178, 179, 180, 355, 356, 358-359 reverse grading 445 rheology and relation to petrology 136-141 Richmond Hill 14, 98 ridge and furrow morphology 256-258 risk assessment 83-84 risk management maps 10, 11, 82, 83, 84, 212 see also zones and microzones in hazard management Roche's Bluff 5, 98, 101, 116 40 Ar/ 39 Ar geochronology 105, 109 Roches Mt 251 rockfalls 192, 231, 235-236, 247, 355, 470, 471, 472, 475, 476 classification 174-175, 775 duration 183, 185 magnitude 184-185 pyroclastic flow trigger 187-188 seismic signals 59, 177-178 analysis 596-600 characteristics 595-596 St George's Hill 5, 93, 98-99, 776, 215 40 Ar/ 39 Ar geochronology 109 St Patrick's village 423, 425, 426, 430, 451 satellite imaging of fallout tephra 484-485, 487, 488, 490

INDEX

644

scanning electron microscopy (SEM) of fallout tephra 484, 506-511, 512, 513 scoria-fall deposit 96 search and rescue procedures 224 sector collapse 11-12 see Galway's Wall collapse also English's Crater sedimentology pyroclastic density current deposits 435-455 seismic signals and earthquakes 9, 12, 58, 177, 567-568 explosion signals 59, 471, 475, 476 phreatic phase 571, 573, 575 hybrid earthquake swarms 59, 167, 176, 177, 179-180, 283, 354, 470-471, 472, 603, 611 dome-collapse events 191, 192, 194, 195, 213, 264, 269 explosive phase events 294-296, 298, 299 phreatic phase 571, 573, 574 long-period earthquakes 59, 167, 177-179, 354-355, 470-471, 472, 603, 611 phreatic phase 571, 573., 575 rockfall signals 595 low frequency signals 167, 611-619 monitoring networks of MVO 77 precursor seismicity analysis 358-359 real-time seismic amplitude measurement (RSAM) 178, 179, 180, 355, 356, 358-359 relation to emplacement features 148 relation to Galway's Wall collapse 397-398, 411-412, 414 rockfall seismic signals 59, 177-178, 192, 231, 235-236, 241, 355, 470, 471, 472, 475, 476, 595-600 sub-Plinian phase 308 tremor 177, 572, 573, 575, 603, 611, 615-618 volcanotectonic earthquakes 59, 177, 354, 470, 471, 472, 474, 475, 476 phreatic phase 571, 572, 573, 574 wavefield modelling in gas-charged magma 603-608 settling velocity in tephra fallout numerical model 520-521 shear strength testing 405 avalanche deposits 392-394 tuffs 394-395 shock waves 343 Silver Hills 5, 93, 94, 95 40 Ar/ 39 Ar geochronology 100 slickensides 63-64, 245 slope stability, testing 392-398 smectite 375 SO2 emission 65-66, 226, 470, 472, 473, 474 HC1/SO2 ratios 621, 622-624, 627-635 phreatic phase 568, 569-570, 573, 575-576, 577-578 TOMS analysis 543-548, 552, 553, 554 solar occultation technique 624-625 Soufriere Hills Volcano 93, 94, 95, 374, 375 activity summary 89, 263-264, 467, 517-518, 557-558, 583-585, 622-624 andesite composition 45, 99 chronology of events 13-18, 48-49, 118, 153-154, 173, 349, 472-476, 478-480 domes 116, 373 eruption style 5, 9-12 flow deposit map 4 flow character and stratigraphy 45, 96, 98, 99 historical activity 6, 349 topography 5 soufrieres precursor activity 5, 26 temperatures 569-570 South Soufriere Hills 5, 93, 94, 95, 96, 421 40 Ar/ 39 Ar geochronology 101-106, 109 Spanish Point block-and-ash flow 99, 799, 243 fan 256 spines 38, 54, 60, 119, 121, 122, 123, 132, 148, 149, 472

Spring Estate 442, 446 stratification in pyroclastic density current deposits 444-445, 455

Streatham village eyewitness location 215, 219, 227, 223-224, 225, 265 pyroclastic surge deposits 204, 252, 265 sub-Plinian eruption (1996) 10, 14, 307, 308, 315 surge-derived pyroclastic flows 60, 62-63, 195, 390-391, 420, 428 deposits particle size 205, 206, 273, 275 temperature 245 volume 205, 265, 269 events compared 276-278 June 25 1997 265-268 Dec 26 1997 269-273, 274 hazard analysis 207-208, 263 physical parameters 276 terminology ix topographic effects 265, 269, 270 velocity 276 synthetic aperture radar (SAR) 583 amplitude information 585-587 interferometry 587-594 Tar River Soufriere 569-570 Tar River valley 97, 241 block-and-ash flows 50, 234, 244 erosion features 254 fan 256, 258 flow chronology 231, 235, 293 flow deposits map 234 pyroclastic flows 173, 174, 175, 176, 179, 185-186 pyroclastic surges 234, 249, 252 rockfalls 241 tephra fallout modelling 532, 533, 534 temperature deposits block-and-ash flow 245 pumice-and-ash flow 245, 248 pyroclastic density current 428, 430 pyroclastic surge 245, 249, 252 surge-derived pyroclastic flow 245 magma 60, 156, 310 soufrieres 569-570 tensile strength, tuff 394 tephra fallout 64 deposits accumulation rate 504-506 aggregation of grains 506-508, 514 bedforms 446-448 composition 558 distribution 484, 490, 493-496 lithofacies 435-444, 497-500 morphology of grains 508-511 particle morphology 508-511, 514-515 particle size distribution 448-451, 500-504, 519, 526 petrology 496-497 stratigraphy 444-446 exposure guidelines 558 generation process 485-491 health hazard 517, 557, 558-564 numerical modelling 518-536 plume structures 492 see also ash plumes tilting and tiltmeter data 297, 471 Chances Peak 180, 192, 193, 213, 355-356, 356 English's Crater 283 installations 78, 79-80 phreatic phase 569, 576 relation to low frequency earthquakes 611 topography effect on airborne particle concentration 563 pyroclastic flows 225 pyroclastic surges 195, 198-203. 225, 267, 269-270, 277, 272 surge-derived pyroclastic flows 265, 269, 270 information from radar 589-590 total ozone mapping spectrometer (TOMS) 543-548, 552, 554 Trant's village 201, 218

Trant's Yard 176. 221-222 tree damage 453 pyroclastic density currents 427, 428 pyroclastic flows 247 pyroclastic surges 198, 216, 249. 251-252, 270, 277, 272 surge-derived pyroclastic flows 265, 267, 268, 269, 270, 277, 272. 276 tremor seismic signals 177. 572, 573. 575, 603, 611, 615-618 Trials village 422. 423. 424 tridymite and health hazard 558, 565 tsunamis 15, 17, 357-358. 428-430 Tuitt's 215, 221, 252 Tuitt's Ghaut 213, 237 1997 flows 176, 192, 203, 283, 293 block-and-ash flows 231, 235, 237, 250 fountain-collapse pyroclastic flows 247 pyroclastic surge 237, 249 surge-derived pyroclastic flow 276 tephra fallout modelling 532, 533, 534 Tyre's Ghaut 203, 532. 533, 534 UK Government involvement 1, 15, 18, 19, 20-21, 66. 71-73. 74. 75. 76. 81. 86 vegetation cover 5 velocity of flows block-and-ash flows 236. 245 debris avalanche 391. 430 explosion products 281. 300, 303 pumice-and-ash flows 248 pyroclastic density currents 275, 276, 430, 457 modelling 431-432 pyroclastic surges 225, 249, 274, 276, 277 surge-derived pyroclastic flows 276 vent parameter modelling 332 vesicularity of explosion clasts 294 volcanic clouds data analysis and results AVHRR 541, 542 BTD 542-543 GOES 540-542 TOMS 543-548 results discussed 548-554 volcanic hazard see hazard volcanic plume see ash plumes volcanic tremor 177, 572. 573, 575, 603, 611 relation to low frequency seismics 615-618 volcanoes worldwide compared with Montserrat Bezymianny (Russia) 402-403 Egmont (New Zealand) 403 Galeras (Colombia) 56, 64 Lascar (Chile) 56, 64 Mont Pelee (Martinique) 45, 56, 149-150, 263, 457 Mount Dutton (Alaska) 64 Mount Lamington (Papua New Guinea) 56. 149, 457 Mount Merapi (Java) 150. 173, 174, 263 Mount Pinatubo (Philippines) 56 Mount St Augustine (Alaska) 403, 629 Mount St Helens (USA) 56, 149. 179. 186, 402-403, 457. 558. 565 Mount Sakurajima (Japan) 558, 565 Mount Spurr (Alaska) 558 Mount Unzen (Japan) 56, 64, 149, 173, 174, 179, 186, 263 Popocatapetl (Mexico) 56. 64 Redoubt (Alaska) 64 Santiaguito (Guatemala) 56. 64. 173 Shiveluch (Russia) 403 Socompa (Chile) 403 Soufriere (Guadeloupe) 402-403 volcanogenic meteorological (VM) clouds 549-552 volcanotectonic seismic events 59, 354, 470. 471, 472. 474. 475. 476 phreatic phase 571. 572. 573. 574 volumes block-and-ash flow deposits 264 debris avalanche debris 416, 420

645

INDEX dome collapses 176, 184, 187, 205-206, 523, 527-528 explosion products 58, 296-299, 523, 527-528 pyroclastic flow deposits 174, 175, 179 pyroclastic surge deposits 264, 269 surge-derived pyroclastic flow deposits 205, 265, 269 tephra 496 Vulcanian explosions 33, 36, 53, 56, 64, 74, 235, 264, 307-308, 478, 489-491, 494-496 airborne particle concentration 560, 561, 565 characteristics 284, 285-287, 294, 319-320, 321, 322, 323 conduit pressure 303-304 cyclicity 301-302 episodes first (Aug 1997) 128-129, 153, 283-284, 289-293 second (SeptOct 1997) 130-131, 154, 285 exit velocities 300-301 initiation 283, 304 mechanisms 302-303

modelling the conduit flow 308-316 initial conditions 320-345 plume character 287-289 products 293-294 pyroclastic flows associated 246-249 seismicity 294-296, 297, 298, 299 relation to volcanic tremor 616-618 shock waves 343 tephra analysis 501-503, 572, 513 health hazards 517 volume 296-299, 523 water in magma 60, 156, 186, 160, 310 in volcanogenic meteorological (VM) clouds 552 White River valley 364, 368, 378, 423, 428, 443, 446-447, 449 deposits block-and-ash flow 231, 235, 242, 357, 369, 370, 376, 377

fountain-collapse pyroclastic flow 247 pyroclastic density current 386, 417, 418, 420, 421 pyroclastic flow 17, 176, 236, 241, 283, 379 pyroclastic surge 269 fan 256 tephra fallout modelling 532, 533, 534 White's Ghaut 47, 176 wind direction effects 483, 487, 489 effect on airborne particles 560, 561 resuspension effects 562 in tephra fallout numerical model 521, 530, 535 trajectory model 548 Windy Hill 15, 218-219, 223, 225, 249, 265 witness observations 215-223, 412-413 zones and microzones in hazard management 10, 11, 12, 13, 14, 81, 82, 83 zoning of phenocrysts 55, 64