Air Pollution Modeling and Its Application XIX
NATO Science for Peace and Security Series This Series presents the results of scientific meetings supported under the NATO Programme: Science for Peace and Security (SPS). The NATO SPS Programme supports meetings in the following Key Priority areas: (1) Defence Against Terrorism; (2) Countering other Threats to Security and (3) NATO, Partner and Mediterranean Dialogue Country Priorities. The types of meeting supported are generally "Advanced Study Institutes" and "Advanced Research Workshops". The NATO SPS Series collects together the results of these meetings. The meetings are coorganized by scientists from NATO countries and scientists from NATO's "Partner" or "Mediterranean Dialogue" countries. The observations and recommendations made at the meetings, as well as the contents of the volumes in the Series, reflect those of participants and contributors only; they should not necessarily be regarded as reflecting NATO views or policy. Advanced Study Institutes (ASI) are high-level tutorial courses intended to convey the latest developments in a subject to an advanced-level audience Advanced Research Workshops (ARW) are expert meetings where an intense but informal exchange of views at the frontiers of a subject aims at identifying directions for future action Following a transformation of the programme in 2006 the Series has been re-named and re-organised. Recent volumes on topics not related to security, which result from meetings supported under the programme earlier, may be found in the NATO Science Series. The Series is published by IOS Press, Amsterdam, and Springer, Dordrecht, in conjunction with the NATO Public Diplomacy Division. Sub-Series A. B. C. D. E.
Chemistry and Biology Physics and Biophysics Environmental Security Information and Communication Security Human and Societal Dynamics
http://www.nato.int/science http://www.springer.com http://www.iospress.nl
Series C: Environmental Security
Springer Springer Springer IOS Press IOS Press
Air Pollution Modeling and Its Application XIX
edited by
Carlos Borrego University of Aveiro, Portugal and
Ana Isabel Miranda University of Aveiro, Portugal
Published in cooperation with NATO Public Diplomacy Division
Proceedings of the 29th NATO/CCMS International Technical Meeting on Air Pollution Modelling and Its Application Aveiro, Portugal 24–28 September 2007
Library of Congress Control Number: 2008928516
ISBN 978-1-4020-8452-2 (PB) ISBN 978-1-4020-8451-5 (HB) ISBN 978-1-4020-8453-9 (e-book)
Published by Springer, P.O. Box 17, 3300 AA Dordrecht, The Netherlands. www.springer.com
Printed on acid-free paper
All Rights Reserved © 2008 Springer Science + Business Media B.V. No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, microfilming, recording or otherwise, without written permission from the Publisher, with the exception of any material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work.
Previous Volumes in this Mini-Series Volumes I-XII were included in the NATO Challenges of Modern Society Series. AIR POLLUTION MODELING AND ITS APPLICATION I Edited by C. De Wispelaere AIR POLLUTION MODELING AND ITS APPLICATION II Edited by C. De Wispelaere AIR POLLUTION MODELING AND ITS APPLICATION III Edited by C. De Wispelaere AIR POLLUTION MODELING AND ITS APPLICATION IV Edited by C. De Wispelaere AIR POLLUTION MODELING AND ITS APPLICATION V Edited by C. De Wispelaere, Francis A. Schiermeier, and Noor V. Gillani AIR POLLUTION MODELING AND ITS APPLICATION VI Edited by Han van Dop AIR POLLUTION MODELING AND ITS APPLICATION VII Edited by Han van Dop AIR POLLUTION MODELING AND ITS APPLICATION VIII Edited by Han van Dop and Douw G. Steyn AIR POLLUTION MODELING AND ITS APPLICATION IX Edited by Han van Dop and George Kallos AIR POLLUTION MODELING AND ITS APPLICATION X Edited by Sven-Erik Gryning and Millán M. Millán AIR POLLUTION MODELING AND ITS APPLICATION XI Edited by Sven-Erik Gryning and Francis A. Schiermeier AIR POLLUTION MODELING AND ITS APPLICATION XII Edited by Sven-Erik Gryning and Nadine Chaumerliac AIR POLLUTION MODELING AND ITS APPLICATION XIII Edited by Sven-Erik Gryning and Ekaterina Batchvarova AIR POLLUTION MODELING AND ITS APPLICATION XIV Edited by Sven-Erik Gryning and Francis A. Schiermeier AIR POLLUTION MODELING AND ITS APPLICATION XV Edited by Carlos Borrego and Guy Schayes AIR POLLUTION MODELING AND ITS APPLICATION XVI Edited by Carlos Borrego and Selahattin Incecik AIR POLLUTION MODELING AND ITS APPLICATION XVII Edited by Carlos Borrego and Ann-Lise Norman AIR POLLUTION MODELING AND ITS APPLICATION XVIII Edited by Carlos Borrego and Eberhard Renner
v
Preface In 1969, the North Atlantic Treaty Organization (NATO) established the Committee on Challenges of Modern Society (CCMS). The subject of air pollution was from the start one of the priority problems under study within the framework of various pilot studies undertaken by this committee. The organization of a periodic conference dealing with air pollution modelling and its application has become one of the main activities within the pilot study relating to air pollution. The first five international conferences were organized by the United States as the pilot country, the second five by the Federal Republic of Germany, the third five by Belgium, the fourth four by The Netherlands, the next five by Denmark and the last five by Portugal. This volume contains the abstracts of papers and posters presented at the 29th NATO/CCMS International Technical Meeting on Air Pollution Modelling and Its Application, held in Aveiro, Portugal, during September 24–28, 2007. This ITM was organized by the University of Aveiro, Portugal (Pilot Country and Host Organization). The key topics distinguished at this ITM included: Local and urban scale modelling; Regional and intercontinental modelling; Data assimilation and air quality forecasting; Model assessment and verification; Aerosols in the atmosphere; Interactions between climate change and air quality; Air quality and human health. The ITM was attended by 156 participants representing 36 countries from Asia, Australia, Europe as well as North and South America. Invited papers were presented by Alexander Baklanov, Denmark (On-line integrated meteorological and chemical transport modelling: advantages and prospectives), Ashok Gadgil, USA (Rapid Data Assimilation in the Indoor Environment: theory and examples from real-time interpretation of indoor plumes of airborne chemicals), Gregory Carmichael, USA (Predicting air quality: current status and future directions) and Michael Brauer, Canada (Models of exposure for use in epidemiological studies of air pollution health impacts). On behalf of the ITM Scientific Committee and as organizers and editors, we should like to thank all the participants who made the meeting so successful. Among the participants, we especially recognize the efforts of the chairpersons and rapporteurs. Finally, special thanks to the sponsoring Institution University of Aveiro, Portugal, and the sponsoring organizations NATO Committee on the Challenges of Modern Society and GRICES (Office for International Relations in Science and Higher Education, Portugal). A special grant was given by EURASAP (European Association for the Sciences of Air Pollution) to award a prize to young researchers for the best paper or poster. The next conference will be held in 2009 in the United States of America. Aveiro, Portugal Aveiro, Portugal
Ana Isabel Miranda (Local Conference Organizer) Carlos Borrego (Scientific Committee Chairperson) vii
The members of the Scientific Committee for the 29th NATO/SPS International Technical Meeting (ITM) on Air Pollution Modeling and Its Application
G. Schayes, Belgium D. Syrakov, Bulgaria D. Steyn, Canada S.-E. Gryning, Denmark N. Chaumerliac, France E. Renner, Germany W. Klug, Germany G. Kallos, Greece D. Anfossi, Italy T. Iversen, Norway
C. Borrego, Portugal (Chairman) A.I. Miranda, Portugal J.M. Baldasano, Spain P. Builtjes, The Netherlands H. Dop, The Netherlands S. Incecik, Turkey B. Fisher, United Kingdom S.T. Rao, United States F. Schiermeier, United States
ix
History of NATO/CCMS air pollution pilot studies
Pilot Study on Air Pollution: International Technical Meetings (ITM) on Air Pollution Modelling and Its Application Dates of Completed Pilot Studies: 1969 1975
- 1974 - 1979
1980
- 1984
Air Pollution Pilot Study (Pilot Country – United States) Air Pollution Assessment Methodology and Modelling (Pilot Country – Germany) Air Pollution Control Strategies and Impact Modelling (Pilot Country – Germany)
Dates and Locations of Pilot Study Follow-Up Meetings: Pilot Country United States (R.A. McCormick, L.E. Niemeyer) February 1971 Eindhoven, First Conference on Low Pollution The Netherlands Power Systems Development July 1971 Paris, Second Meeting of the Expert Panel on France Air Pollution Modelling
From 1972 to 2000 all of the meetings were entitled NATO/CCMS International Technical Meetings (ITM) on Air Pollution Modelling and Its Application October May June
1972 1973 1974
Paris, France Oberursel, Federal Republic of Germany Roskilde, Denmark
3rd ITM 4th ITM 5th ITM
Pilot Country Germany (Erich Weber) September 1975 Frankfurt, Federal Republic of Germany September 1976 Airlie House, Virginia, USA September 1977 Louvain-La-Neuve, Belgium August 1978 Toronto, Ontario, Canada October 1979 Rome, Italy
6th ITM 7th ITM 8th ITM 9th ITM 10th ITM
Pilot Country Belgium (C. De Wispelaere) November 1980 Amsterdam, The Netherlands August 1981 Menlo Park, California, USA September 1982 Ile des Embiez, France September 1983 Copenhagen, Denmark April 1985 St. Louis, Missouri, USA
11th ITM 12th ITM 13th ITM 14th ITM 15th ITM xi
xii
History of NATO/CCMS air Pollution Pilot Studies
Pilot Country The Netherlands (Han van Dop) April 1987 Lindau, Federal Republic of Germany September 1988 Cambridge, United Kingdom May 1990 Vancouver, British Columbia, Canada September 1991 lerapetra, Crete, Greece
16th ITM 17th ITM 18th ITM 19th ITM
Pilot Country Denmark (Sven-Erik Gryning) November 1993 Valencia, Spain November 1995 Baltimore, Maryland, USA June 1997 Clermont-Ferrand, France September 1998 Varna, Bulgaria May 2000 Boulder, Colorado, USA
20th ITM 21st ITM 22nd ITM 23rd ITM 24th ITM
All of the following meetings were entitled NATO/SPS International Technical Meetings (ITM) on Air Pollution Modelling and Its Application. Pilot Country – Portugal (Carlos Borrego) October 2001 Louvain-la-Neuve, Belgium May 2003 Istanbul, Turkey October 2004 Banff, Canada May 2006 Leipzig, Federal Republic of Germany September 2007 Aveiro, Portugal
25th ITM 26th ITM 27th ITM 28th ITM 29th ITM
List of Participants The 29th NATO/CCMS International Technical Meeting on Air Pollution Modeling and Its Application, Aveiro, Portugal, September 24–28, 2007.
Albania Mima, Marieta
Environmental Center for Administration and Technology Rr.A. Frasheri, Pall.16/ Shk.6/ Ap.53, Tirana
[email protected]
Australia Hurley, Peter
CSIRO Marine and Atmospheric Research 107-121 Station Street, Private Bag 1, Aspendale, 3195 Melbourne
[email protected]
Olaru, Doina
University of Western Australia Business School, 35 Stirling Highway 6009 Crawley
[email protected]
Austria Pechinger, Ulrike
ZAMG Environmental Meteorology Hohe Warte 38, A 1191 Vienna
[email protected]
Belgium Andy, Delcloo
Royal Meteorological Institute of Belgium (RMI) Observations Ringlaan 3, 1180 Ukkel
[email protected]
Clemens, Mensink
VITO NV, Integrated Environmental Studies Boeretang 200, 2400 MOL
[email protected]
Schayes, Guy
Université de Louvain, Department de Astronomy e Geophysics Chemin du Cyclotron 2, B-1348 Louvain-laNeuve
[email protected] xiii
xiv
Stijn, Janssen
List of Participants
VITO NV, Integrated Environmental Studies Boeretang 200, 2400 MOL
[email protected]
Brazil Escada, Marcos
Petroleo Brasileiro S.A., Environment Department Rodovia Presidente Dutra, km 143, 12223900 São José dos Campos
[email protected]
Bulgaria Batchvarova, Ekaterina
National Institute of Meteorology and Hydrology Blvd. Tzarigradsko Chaussee 66, 1784 Sofia
[email protected]
Prodanova, Maria
National Institute of Meteorology and Hydrology Blvd. Tzarigradsko Chaussee 66, 1784 Sofia
[email protected]
Syrakov, Dimiter
National Institute of Meteorology and Hydrology Blvd. Tzarigradsko Chaussee 66, 1784 Sofia
[email protected]
Canada Brauer, Michael
The University of British Columbia, School of Occupational and Environmental Hygiene 2206 East Mall, 3rd Floor, V6T1Z3 Vancouver
[email protected]
Flagg, David
York University, Department of Earth and Space Science and Engineering 230 Vaughan Road, M6C 2M6 Toronto
[email protected]
Gong, Wanmin
Environment Canada, Air Quality Research Division 4905 Dufferin Street, M3H 5T4 Toronto
[email protected]
List of Participants
xv
Sloan, James
University of Waterloo, Department of Earth Sciences 200 University Ave. W, N2L3G1 Waterloo
[email protected]
Steyn, Douw
The University of British Columbia, Department of Earth and Ocean Sciences 6339 Stores Road, V6T 1Z4 Vancouver
[email protected]
Talbot, Donald
Canadian Meteorological Center, Environment Canada 2121 Rte TransCanadienne, H9P 1J3 Dorval
[email protected]
Czech Republic Fuka, Vladimir
Charles University, Department of Meteorology and Environment Protection V Holesovickach 2, 18000 Prague
[email protected]
Halenka, Tomas
Charles University, Department of Meteorology and Environment Protection V Holesovickach 2, 18000 Prague
[email protected]
Zemankova, Katerina
Charles University, Department of Meteorology and Environment Protection V Holesovickach 2, 18000 Prague
[email protected]
Denmark Baklanov, Alexander
Danish Meteorological Institute (DMI), Meteorological Research Lyngbyvej 100, DK-2100 Copenhagen
[email protected]
Gryning, Sven-Erik
Risø National Laboratory, Wind Energy Department Frederiksborgueg 399, DK-4000 Roskilde
[email protected]
Hedegaard, Gitte
National Environmental Research Institute, Department of Atmospheric Environment Frederiksborgvej 399, 4000 Roskilde
[email protected]
xvi
List of Participants
Søren, Thykier-Nielsen
Risø National Laboratory, Wind Energy Department Frederiksborgueg 399, DK-4000 Roskilde
[email protected]
Sørensen, Jens Havskov
Danish Meteorological Institute (DMI), Research Department Lyngbyvej 100, DK-2100 Copenhagen
[email protected]
Estonia Kaasik, Marko
University of Tartu, Institute of Environmental Physics Ülikoooli 18, 50090 Tartu
[email protected]
Prank, Marje
University of Tartu, Institute of Environmental Physics Ülikoooli 18, 50090 Tartu
[email protected]
Teinemaa, Erik
Estonian Environmental Research Centre Air Quality Management Marja 4D, 10617 Tallinn
[email protected]
Finland Karppinen, Ari
Finnish Meteorological Institute, Research and Development P.O. Box 503, 00101 Helsinki
[email protected]
Siljamo, Pilvi
Finnish Meteorological Institute, Meteorological Research P.O. Box 503, 00101 Helsinki
[email protected]
Soares, Joana
Finnish Meteorological Institute Air Quality Research Erik Palmenin Aukio 1, 00560 Helsinki
[email protected]
Sofiev, Mikhail
Finnish Meteorological Institute, Air Quality Research Erik Palmenin Aukio 1, 00560 Helsinki
[email protected]
List of Participants
xvii
France Armand, Patrick
Commissariat à l’Energie Atomique (CEA), DIF/ DASE/ SRCE Bátiment G, 91680 Bruyères-le-Châtel
[email protected]
Blot, Romain
University of South – Toulon Var, LSEET-LEPI-ISITV Av. G. Pompidou, BP 62, 83162 La Valette du Var
[email protected]
Carvalho, Ana
Laboratoire de Météorologie Dynamique École Polytechnique, 91128 Palaiseau
[email protected]
Chaumerliac, Nadine
Université Blaise Pascal, LaMP CNRS 24 avenue des landais, 63170 Aubière
[email protected]
Chollet, Jean-Pierre
Université Joseph Fourier Grenoble BP 53, 38041 Grenoble Cedex
[email protected]
Didier, Damien
IRSN, DEI/SESUC/BNTA BP17, 92262 Fontenay aux Roses
[email protected]
Freve-Nollet, Valerie
Université des Sciences et technologies de Lille, PC2A Bat. C11, 59655 Villeneuve d’Ascq
[email protected]
Menut, Laurent
Laboratoire de Météorologie Dynamique, IPSL/CNRS École Polytechnique, 91128 Palaiseau
[email protected]
Quelo, Denis
IRSN, DEI/SESUC/BNTA BP17, 92262 Fontenay-aux-Roses
[email protected]
xviii
List of Participants
Queguiner, Solen
CEREA (EDF R&D/ENPC), MFEE 6 Quai Watier, F-78240 CHATOU
[email protected]
Terrenoire, Etienne
Université des Sciences et technologies de Lille, PC2A Bat. C11, 59655 Villeneuve d’Ascq
[email protected]
Vautard, Robert
LSCE/IPSL – Laboratoire CEA/CNRS/UVSQ Orme des Merisiers - Bât. 701, 91191 GIF SUR YVETTE
[email protected]
Georgia Tavamaishvili, Ketevan
The National Forensics Bureau, Department of Forensic Chemistry Ilia Chavchavadze ave. # 84, 0162 Tbilisi
[email protected]
Germany Aulinger, Armin
GKSS Research Center, Institute for Coastal Research Max-Planck-Str. 1, 21502 Geesthacht
[email protected]
Bewersdorff, Ines
GKSS Research Center, Institute for Coastal Research Max-Planck-Str. 1, 21502 Geesthacht
[email protected]
Graff, Arno
Umweltbundesamt, Air Quality Assessment Wörlitzer Platz 1, 06844 Dessau
[email protected]
Kerschbaumer, Andreas
Freie Universitaet Berlin, Institut fuer Meteorologie Carl-Heinrich-Becker-Weg 6-10, 12165 Berlin
[email protected]
Renner, Eberhard
Institute for Tropospheric Research, Modelling Permoserstraße 15, 04318 Leipzig
[email protected]
List of Participants
xix
Suppan, Peter
Institute for Meteorology and Climate Research, Atmospheric Environmental Research (IMK-IFU) Kreuzeckbahnstr.19, 82467 Garmisch-Partenkirchen
[email protected]
Wolke, Ralf
Institute for Tropospheric Research, Modelling Permoserstraße 15, 04318 Leipzig
[email protected]
Greece Astitha, Marina
University of Athens, School of Physics University Campus, Bldg PHYS-5, 15784 Athens
[email protected]
Bartzis, John
University of West Macedonia, Engineering and Management of Energy Resources Sialvera and Bakola Str., 50100 Kozani
[email protected]
Kallos, George
University of Athens, School of Physics University Campus, Bldg PHYS-5, 15784 Athens
[email protected]
Kushta, Joni
University of Athens, School of Physics University Campus, Bldg PHYS-5, 15784 Athens
[email protected]
Israel Kishcha, Pavel
Tel-Aviv University, Department of Geophysics Ramat Aviv, 69978 Tel-Aviv
[email protected]
Reisin, Tamir
Soreq Nuclear Research Center, Applied Physics 81800 Yavne
[email protected]
List of Participants
xx
Terliuc, Benjamin
Nuclear Research Center Negev, Environmental Researches P.O. Box 9001, 84190 Beer Sheva
[email protected]
Italy Anfossi, Domenico
CNR-ISAC, Corso Fiume 4, 10133 Torino
[email protected]
Belfiore, Giovanni
ISAC – CNR, Corso Fiume 4, 10010 Torino
[email protected]
Carnevale, Claudio
Università degli Studi di Brescia, Dipartimento di Elettronica per l’Automazione Via Branze 38, 25123 Brescia
[email protected]
Mircea, Mihaela
Istituto di Scienze dell Via Gobetti 101, Bologna
[email protected]
Pirovano, Guido
Cesi Ricera SPA Via Rubattino, 54, 20134 Milano
[email protected]
Trini Castelli, Silvia
CNR, National Research Council Corso Fiume 4, 10133 Torino
[email protected]
Trozzi, Carlo
Techne Consulting s.r.l. Via G. Ricci Curbastro 34, I00149 Roma
[email protected]
Japan Takigawa, Masayuki
Frontier Research Center for Global Change 3173-25 Showa-machi, Kanazawa-ku, 236-0001 Yokohama
[email protected]
Niwano, Masanori
FRCGC, JAMSTEC Atmospheric Composition Research Program 3173-25 Syowa-machi, Kanazawa-ku, 236-0001 Yokohama
[email protected]
List of Participants
xxi
Ohara, Toshimasa
National Institute for Environmental Studies, Asian Environment Research Group 16-2 Onogawa, 305-8506 Tsukuba
[email protected]
Kitada, Toshihiro
Toyohashi University of Technology, Department of Ecological Engineering 1-1 Hibarigaoka, Tempaku-cho, 441-8580 Toyohashi
[email protected]
Lithuania Ulevicius, Vidmantas
Institute of Physics, Environmental physics and chemistry Savanoriu 231, LT-02300 Vilnius
[email protected]
Norway Denby, Bruce
Norwegian Institute for Air Research (NILU), Urban Environment and Industry Department P.O. BOX 100, 2027 Kjeller
[email protected]
Guerreiro, Cristina
Norwegian Institute for Air Research (NILU), Urban Environment and Industry Department P.O. BOX 100, Instituttvn. 18, 2027 Kjeller
[email protected]
Saltbones, Jørgen Ingar
Norwegian Meteorological Institute, Meteorological Department P.O. Box 43, Blindern, NO-0313 Oslo
[email protected]
Portugal Amorim, Jorge Humberto
Universidade de Aveiro, Departamento de Ambiente e Ordenamento Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
Borrego, Carlos
Universidade de Aveiro, Departamento de Ambiente e Ordenamento Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
xxii
List of Participants
Carvalho, Anabela
Universidade de Aveiro, Departamento de Ambiente e Ordenamento Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
Caseiro, Alexandre
Universidade de Aveiro, Departamento de Ambiente e Ordenamento Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
Costa, Ana Margarida
Universidade de Aveiro, Departamento de Ambiente e Ordenamento Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
Coutinho, Miguel
Instituto de Ambiente e Desenvolvimento (IDAD) Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
Ferreira, Joana
Universidade de Aveiro, Departamento de Ambiente e Ordenamento Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
Lopes, Myriam
Universidade de Aveiro, Departamento de Ambiente e Ordenamento Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
Martins, Helena
Universidade de Aveiro, Departamento de Ambiente e Ordenamento Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
Martins, Vera
Universidade de Aveiro, Departamento de Ambiente e Ordenamento Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
List of Participants
xxiii
Miranda, Ana Isabel
Universidade de Aveiro, Departamento de Ambiente e Ordenamento Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
Monteiro, Alexandra
Universidade de Aveiro, Departamento de Ambiente e Ordenamento Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
Moreira, Cármen
Universidade de Aveiro, Departamento de Ambiente e Ordenamento Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
Tavares, Richard
Universidade de Aveiro, Departamento de Ambiente e Ordenamento Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
Tchepel, Oxana
Instituto Politécnico de Leiria, Escola Superior de Tecnologia e Gestão Morro do Lena, Alto do Vieiro, 2411-901 Leiria
[email protected]
Tomé, Mário
Instituto Politécnico de Viana do Castelo, Escola Superior de Tecnologia e Gestão Avenida do Atlântico, 4900-348 Viana do Castelo
[email protected]
Valente, Joana
Universidade de Aveiro, Departamento de Ambiente e Ordenamento Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
Santos, João
Universidade de Aveiro, Departamento de Ambiente e Ordenamento Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
xxiv
List of Participants
Santos, Pedro
Universidade de Aveiro, Departamento de Ambiente e Ordenamento Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
Silva, João Vasco
Universidade de Aveiro, Departamento de Ambiente e Ordenamento Campus Universitário de Santiago, 3810-193 Aveiro
[email protected]
Republic of Macedonia Alcinova Monevska, Suzana
Republic Hydrometeorological Institute, Department for Informatics and Telecommunications Skupi bb, 1000 Skopje
[email protected]
Russia Genikhovich, Eugene
Voeikov Main Geophysical Observatory, Department of Monitoring of Air Pollution Karbyshev Street, 7, 194021 St. Petersburg
[email protected]
Serbia Rajkovic, Borivoj
Faculty of Physics, Institute of Meteorology Dobracina 16, 11000 Belgrade
[email protected]
Vujadinovic, Mirjam
Faculty of Physics, Institute of Meteorology Dobracina 16, 11000 Belgrade
[email protected]
Slovenia Mlakar, Primoz
MEIS d.o.o. Mali Vrh pri Smarju 78, SI-1293 Smarje – Sap
[email protected]
Boznar, Marija Zlata
MEIS d.o.o. Mali Vrh pri Smarju 78, SI-1293 Smarje – Sap
[email protected]
List of Participants
xxv
South Korea Kim, Yong Pyo
Ewha Womans University, Environmental Science and Engineering 11-1 Daehyundong, Seoudaemungu, 120-750 Seoul
[email protected]
Roh, Woosub
Yonsei University, Atmospheric Science Department Seodaemongu sinchon-dong, 82 Seoul
[email protected]
Spain Alonso, Lucio
University of the Basque Country, Chemical & Environmental Engineering Alameda de Urquijo s/n, 48013 Bilbao
[email protected]
Arasa, Raul
University of Barcelona, Astronomy and Meteorology Avinguda Diagonal 647 7º Planta, 08028 Barcelona
[email protected]
Arnold, Delia
Technical University of Catalonia, Institute of Energy Technologies Av. Diagonal 647, 08028 Barcelona
[email protected]
Baldasano, Jose M.
Barcelona Supercomputing Center (BSC-CNS), Earth Sciences Jordi Girona 31, 08034 Barcelona
[email protected]
Casares-Long, Juan
Universidad de Santiago de Compostela, Department of Chemical Engineering Rua Lope Gomez de Marzoa, 15782 Santiago de Compostela
[email protected]
xxvi
List of Participants
De Blas Martín, Maite
University of the Basque Country, Chemical & Environmental Engineering Alameda de Urquijo s/n, 48013 Bilbao
[email protected]
Durana Jimeno, Nieves
University of the Basque Country, Chemical & Environmental Engineering Alameda de Urquijo s/n, 48013 Bilbao
[email protected]
Gangoiti Bengoa, Gotzon
University of the Basque Country, Chemical & Environmental Engineering Alameda de Urquijo s/n, 48013 Bilbao
[email protected]
García Fernández, José
University of the Basque Country, Chemical & Environmental Engineering Alameda de Urquijo s/n, 48013 Bilbao
[email protected]
Ilardia, Juan Luis
University of the Basque Country, Chemical & Environmental Engineering Alameda de Urquijo s/n, 48013 Bilbao
[email protected]
Navazo Muñoz, Marino
University of the Basque Country, Chemical & Environmental Engineering Alameda de Urquijo s/n, 48013 Bilbao
[email protected]
Saavedra, Santiago
Universidad de Santiago de Compostela, Ingeniería Química c/Lope Gómez de Marzoa, 15782 Santiago de Compostela
[email protected]
Saez de Cámara, Estíbaliz
University of the Basque Country, Chemical & Environmental Engineering Alameda de Urquijo s/n, 48013 Bilbao
[email protected]
San José, Roberto
Universidad Politécnica de Madrid, Environmental Software and Modelling Group Campus de Montegancedo, 28660 Boadilla del Monte, Madrid
[email protected]
List of Participants
xxvii
Soler, Maria Rosa
University of Barcelona, Astronomy and Meteorology Avinguda Diagonal 647 7º Planta, 080128 Barcelona
[email protected]
Souto, Jose A.
University of Santiago de Compostela, Chemical Engineering Lope Gómez de Marzoa – Campus Sur, 15782 Santiago de Compostela
[email protected]
Valdenebro Villar, Veronica
University of the Basque Country, Applied Mathematics Alameda de Urquijo s/n, 48013 Bilbao
[email protected]
Vargas, Arturo
Technical University of Catalonia, Institute of Energy Technologies Avda. Diagonal 647, 08028 Barcelona
[email protected]
Switzerland Andreani, Sebnem
Paul Scherrer Institute, Laboratory of Atmospheric Chemistry PSI, 5232 Villigen PSI
[email protected]
The Netherlands Barbu, Alina
Delft University of Technology, Delft Institute of Applied Mathematics Mekelweg 4, 2628 CD Delft
[email protected]
Builtjes, Peter
TNO, Air Quality and Climate P.O. Box 342, 7300 AH Apeldoorn
[email protected]
Manders, Astrid
RIVM, Laboratory for Environmental Monitoring P.O. Box 1, 3720 BA Bilthoven
[email protected]
Schaap, Martijn
TNO, B&O P.O. box 342, 7300 AH Apeldoorn
[email protected]
List of Participants
xxviii
Timmermans, Renske
TNO, Air Quality and Climate P.O. Box 342, 7300 AH Apeldoorn
[email protected]
Turkey Incecik, Selahattin
Istanbul Technical University øTÜ Ayaza÷a Kampüsü 34469 Maslak-Istanbul
[email protected]
Ukraine Bugaieva, Liudmyla
National Technical University of Ukraine, Chemical Technology Processes Cybernetics 37, Peremogy ave., 03056 Kiev
[email protected]
United Kingdom Dore, Anthony
Centre for Ecology ad Hydrology, Bush Estate Penicuik, EH26 OQB Midlothian
[email protected]
Fisher, Bernard
Environment Agency, Risk and Forecasting Science Kings Meadow Road, RG1 8DQ Reading
[email protected]
Hill, Richard
Westlakes Scientific Consulting Ltd, Environmental Science CA24 3LN Moor Row, Cumbria
[email protected]
USA Bullock, Russell
National Oceanic and Atmospheric Administration, Air Resources Laboratory, Atmospheric Sciences Modeling Division US EPA Mail Drop E243-03, 27711 Research Triangle Park, NC
[email protected]
Carmichael, Gregory
The University of Iowa 424 IATL, 52242 Iowa
[email protected]
List of Participants
xxix
Davidson, Paula
National Oceanic and Atmospheric Administration, NWS 1325 E-W Highway, 20910 Silver Spring MD
[email protected]
Duvall, Rachelle
Environmental Protection Agency, National Exposure Research Laboratory 109 TW Alexander Drive, 27711 Research Triangle Park
[email protected]
Gadgil, Ashok
Lawrence Berkeley National Laboratory, Environmental Energy Technologies Mailstop 90-3058, 1 Cyclotron Road, 94720 Berkley
[email protected]
Hanna, Steven
Harvard University, Exposure, Epidemiology and Risk Program 7 Crescent Avenue, 04046-7235 Kennebunkport, Maine
[email protected]
Hogrefe, Christian
University at Albany, Atmospheric Sciences Research Center 251 Fuller Road, 12203 NY, Albany
[email protected]
Isakov, Vlad
National Oceanic and Atmospheric Administration /EPA, Atmospheric Sciences Modeling Division 109 T.W. Alexander Drive, 27711 Research Triangle Park, NC
[email protected]
Luecken, Deborah
National Oceanic and Atmospheric Administration /EPA, Atmospheric Sciences Modeling Division 109 TW Alexander Drive MD E243-03, 27711 Research Triangle Park
[email protected]
Mathur, Rohit
National Oceanic and Atmospheric Administration /EPA, Atmospheric Sciences Modeling Division
xxx
List of Participants
109 T.W. Alexander Drive, 27711 Research Triangle Park, NC
[email protected] Mobley, David
National Oceanic and Atmospheric Administration /EPA, Atmospheric Sciences Modeling Division 109 TW Alexander Drive MD E243-03, 27711 Research Triangle Park
[email protected]
Napelenok, Sergey
National Oceanic and Atmospheric Administration /EPA, Atmospheric Sciences Modeling Division 109 T.W. Alexander Drive, 27711 Research Triangle Park, NC
[email protected]
Nolte, Chris
National Oceanic and Atmospheric Administration, Atmospheric Sciences Modeling Division 109 TW Alexander Drive, Mail Code E243-01, 27701 Research Triangle Park, NC
[email protected]
Pleim, Jonathan
National Oceanic and Atmospheric Administration /EPA, Atmospheric Sciences Modeling Division 109 T.W. Alexander Drive, 27711 Research Triangle Park, NC
[email protected]
Porter, P. Steven
University of Idaho, Civil Engineering 1776 Science Center, 83402 Idaho Falls
[email protected]
Rao, S. Trivikrama
National Oceanic and Atmospheric Administration /EPA, Atmospheric Sciences Modeling Division 109 TW Alexander Drive MD E243-02, 27711 Research Triangle Park
[email protected]
Sarwar, Golam
National Oceanic and Atmospheric Administration /EPA, Atmospheric Sciences Modeling Division
List of Participants
xxxi
109 TW Alexander Drive MD E243-03, 27711 Research Triangle Park
[email protected] Schiermeier, Frank
National Oceanic and Atmospheric Administration 303 Glasgow Road, 27511 Cary, NC
[email protected]
Uliasz, Marek
Colorado State University, Department of Atmospheric Science 1371 Campus Delivery, 80523-1371 Fort Collins, Colorado
[email protected]
Venkatram, Akula
University of California, Riverside Department of Mechanical Engineering A368 Bourns Hall, 92521 Riverside
[email protected]
Watkins, Timothy
National Oceanic and Atmospheric Administration /EPA, Atmospheric Sciences Modeling Division 109 TW Alexander Drive MD E243-03, 27711 Research Triangle Park
[email protected]
Venezuela Diaz, Luis
PDVSA INTEVEP, Department of Air Quality URB. Santa Rosa, 1201 Norte 4
[email protected]
Contents Preface......................................................................................................................vii List of Participants................................................................................................ xiii Chapter 1
Local and urban scale modeling..................................................... 1
1.1 On-line Integrated Meteorological and Chemical Transport Modelling: Advantages and Prospectives.......................................................... 3 Alexander Baklanov and Ulrik Korsholm 1.2 Modelling of the Urban Wind Profile........................................................ 18 Sven-Erik Gryning and Ekaterina Batchvarova 1.3 Development of a Lagrangian Particle Model for Dense Gas Dispersion in Urban Environment ................................................................... 28 G. Tinarelli, D. Anfossi, S. Trini Castelli, A. Albergel, F. Ganci, G. Belfiore and J. Moussafir 1.4 CFD and Mesoscale Air Quality Modelling Integration: Web Application for Las Palmas (Canary Islands, Spain) ....................................... 37 R. San José, J.L. Pérez, J.L. Morant and R.M. González 1.5 On the Suppression of the Urban Heat Island over Mountainous Terrain in Winter ............................................................................................. 46 Charles Chemel, Jean-Pierre Chollet and Eric Chaxel 1.6 Air Quality Management Strategies in Large Cities: Effects of Changing the Vehicle Fleet Composition in Barcelona and Madrid Greater Areas (Spain) by Introducing Natural Gas Vehicles........................... 54 María Gonçalves, Pedro Jiménez-Guerrero and José M. Baldasano 1.7 Evaluation of the Hazard Prediction and Assessment Capability (HPAC) Model with the Oklahoma City Joint Urban 2003 (JU2003) Tracer Observations......................................................................................... 63 Steven Hanna, Joseph Chang, John White and James Bowers 1.8 Origin and Influence of PM10 Concentrations in Urban and in Rural Environments ......................................................................................... 72 Andreas Kerschbaumer, Rainer Stern and Martin Lutz 1.9 Development and Application of MicroRMS Modelling System to Simulate the Flow, Turbulence and Dispersion in the Presence of Buildings.......................................................................................................... 81 S. Trini Castelli, T.G. Reisin and G. Tinarelli 1.10 Numerical Treatment of Urban and Regional Scale Interactions in Chemistry-Transport Modelling .................................................................. 90 R. Wolke, D. Hinneburg, W. Schröder and E. Renner xxxiii
xxxiv
Chapter 2
Contents
Regional and intercontinental modeling ..................................... 99
2.1 Contribution of Biogenic Emissions to Carbonaceous Aerosols in Summer and Winter in Switzerland: A Modelling Study .......................... 101 ù. Andreani-Aksoyo÷lu, J. Keller, M.R. Alfarra, A.S.H. Prévôt, J.J. Sloan and Z. He 2.2 A Regional Air Quality Model over the Kanto Region of Japan: The Effect of the Physics Parameterization on the Meteorological and Chemical Fields ...................................................................................... 109 Masanori Niwano, Masayuki Takigawa, Hajime Akimoto, Masaaki Takahashi and Mitsuhiro Teshiba 2.3 Regional Aerosol Optical Thickness Distribution Derived by CMAQ Model in the Siberian Forest Fire Emission Episode of May 2003................................................................................................... 118 Hee -Jin In, Yong Pyo Kim and Kwon-Ho Lee 2.4 Modelling the Deposition of Reduced Nitrogen at Different Scales in the United Kingdom ....................................................................... 127 Anthony J. Dore, Mark R. Theobald, Maciej Kryza, Massimo Vieno, Sim Y. Tang and Mark A. Sutton 2.5 Long-Term Simulations of Surface Ozone in East Asia During 1980–2020 with CMAQ and REAS Inventory .............................................. 136 Toshimasa Ohara, Kazuyo Yamaji, Itsushi Uno, Hiroshi Tanimoto, Seiji Sugata, Tatsuya Nagashima, Jun-ichi Kurokawa, Nobuhiro Horii and Hajime Akimoto 2.6 The Use of Meso-Scale Atmospheric Circulation Types as a Strategy for Modelling Long-Term Trends in Air Pollution. ................. 145 Douw Steyn, Bruce Ainslie, J.W. Kaminski, J.C. McConnell, Alberto Martilli and L. Neary 2.7 Development and Applications of Biogenic Emission Term as a Basis of Long-Range Transport of Allergenic Pollen ........................... 154 Pilvi Siljamo, Mikhail Sofiev, Tapio Linkosalo, Hanna Ranta and Jaakko Kukkonen 2.8 High Resolution Nested Runs of the AURAMS Model with Comparisons to PrAIRie2005 Field Study Data.................................... 163 Paul A. Makar, Craig Stroud, Brian Wiens, SunHee Cho, Junhua Zhang, Morad Sassi, John Liggio, Michael Moran, Wanmin Gong, Sunling Gong, Shao-Meng Li, Jeff Brook, Kevin Strawbridge, Kurt Anlauf, Chris Mihele and Desiree Toom-Sauntry 2.9 The Effect of Lateral Boundary Values on Atmospheric Mercury Simulations with the CMAQ Model .............................................................. 173 O. Russell Bullock, Jr.
Contents
xxxv
2.10 Air Pollution Modelling with Perturbational Downscaling ................... 182 Eugene Genikhovich, Mikhail Sofiev, Guy Schayes and Irene Gracheva 2.11 Forest Fires Impact on Air Quality over Portugal.................................. 190 A.I. Miranda, A. Monteiro, V. Martins, A. Carvalho, M. Schaap, P. Builtjes and C. Borrego 2.12 The VetMet Veterinary Decision Support System for Airborne Animal Diseases ............................................................................................ 199 Jens Havskov Sørensen, Søren Alexandersen, Poul Astrup, Knud Erik Christensen, Torben Mikkelsen, Sten Mortensen, Torben Strunge Pedersen and Søren Thykier-Nielsen 2.13 Development and Verification of TAPM .............................................. 208 Peter Hurley 2.14 Development of Fire Emissions Inventory Using Satellite Data ........... 217 Biswadev A. Roy, George A. Pouliot, J. David Mobley, Thompson G. Pace, Thomas E. Pierce, Amber J. Soja, James J. Szykman and J. Al-Saadi 2.15 Toward a US National Air Quality Forecast Capability: Current and Planned Capabilities................................................................................ 226 Paula Davidson, Kenneth Schere, Roland Draxler, Shobha Kondragunta, Richard A. Wayland, James F. Meagher and Rohit Mathur 2.16 Two-Way Coupled Meteorology and Air Quality Modeling................. 235 Jonathan Pleim, Jeffrey Young, David Wong, Rob Gilliam, Tanya Otte and Rohit Mathur 2.17 Numerical Simulation of Air Pollution Transport Under Sea/Land Breeze Situation in Jakarta, Indonesia in Dry Season.................... 243 Toshihiro Kitada, Asep Sofyan and Gakuji Kurata 2.18 Synergetic or Non-Linear Effects in PM10 and PM2.5 Scenario Calculations for 2015 in Belgium .................................................................. 252 Clemens Mensink, Felix Deutsch, Jean Vankerkom and Liliane Janssen Chapter 3
Data assimilation and air quality forecasting ........................... 261
3.1 Rapid Data Assimilation in the Indoor Environment: Theory and Examples from Real-Time Interpretation of Indoor Plumes of Airborne Chemical .................................................................................... 263 Ashok Gadgil, Michael Sohn and Priya Sreedharan 3.2 Comparison of Data Assimilation Methods for Assessing PM10 Exceedances on the European Scale .............................................................. 278 Bruce Denby, Martijn Schaap, Arjo Segers, Peter Builtjes and Jan Horálek
xxxvi
Contents
3.3 An Observing System Simulation Experiment (OSSE) for Aerosols ................................................................................................... 287 Renske Timmermans, Martijn Schaap, Arjo Segers, Hendrik Elbern, Richard Siddans, Stephen Tjemkes, Robert Vautard and Peter Builtjes 3.4 Modelling of Benzo(a)pyrene Depositions over North Sea Coastal Areas: Impact of Emissions from Local and Remote Areas ............. 296 Ines Bewersdorff, Armin Aulinger, Volker Matthias and Markus Quante 3.5 Air Quality Forecasting During Summer 2006: Forest Fires as One of Major Pollution Sources in Europe ............................................... 305 Mikhail Sofiev, Pilvi Siljamo, Ari Karppinen and Jaakko Kukkonen 3.6 Comparison of Methods to Generate Meteorological Inputs for Modeling Dispersion in Coastal Urban Areas.......................................... 313 Akula Venkatram, Wenjun Qian, Tao Zhan and Marko Princevac 3.7 Developing a Method for Resolving NOx Emission Inventory Biases Using Discrete Kalman Filter Inversion, Direct Sensitivities, and Satellite-Based NO2 Columns ................................................................. 322 Sergey L Napelenok, Robert W. Pinder, Alice B. Gilliland and Randall V. Marin 3.8 A Suggested Correction to the EMEP Database, Regarding the Location of a Major Industrial Air Pollution Source in Kola Peninsula........................................................................................................ 331 Marko Kaasik, Marje Prank, Jaakko Kukkonen and Mikhail Sofiev 3.9 Fusing Observations and Model Results for Creation of Enhanced Ozone Spatial Fields: Comparison of Three Techniques .............. 339 Edith Gégo, P.S. Porter, V. Garcia, C. Hogrefe and S.T. Rao Chapter 4
Model assessment and verification............................................. 347
4.1 The Effect of Heterogeneous Reactions on Model Performance for Nitrous Acid............................................................................................. 349 Golam Sarwar, Robin L. Dennis and Bernhard Vogel 4.2 Saharan Dust over the Eastern Mediterranean: Model Sensitivity .......... 358 Pavel Kishcha, Slobodan Nickovic, Eliezer Ganor, Levana Kordova and Pinhas Alpert 4.3 Air Quality Ensemble Forecast Coupling ARPEGE and CHIMERE over Western Europe............................................................ 367 Ana C. Carvalho, Laurent Menut, Robert Vautard and Jean Nicolau 4.4 Uncertainty in Air Quality Decision Making........................................... 376 Bernard Fisher
Contents
xxxvii
4.5 Application of Advanced Particulate Matter Source Apportionment Techniques in the Northern Italy Basin ................................ 385 Marco Bedogni, Simone Casadei, Guido Pirovano, Giovanni Sghirlanzoni and Andrea Zanoni 4.6 Has the Performance of Regional-Scale Photochemical Modelling Systems Changed over the Past Decade? ..................................... 394 C. Hogrefe, J.-Y. Ku, G. Sistla, A. Gilliland, J.S. Irwin, P.S. Porter, E. Gégo, P. Kasibhatla and S.T. Rao 4.7 Application of a Regional Atmospheric Emission Inventory to Ozone and PM Modelling over the French North Region: The summer 2006 Heat Wave Case Study. ................................................... 404 E. Terrenoire and V. Fèvre-Nollet 4.8 Evaluating Regional-Scale Air Quality Models....................................... 412 Alice B. Gilliland, James M. Godowitch, Christian Hogrefe and S.T. Rao 4.9 Ozone Modeling over Italy: A Sensitivity Analysis to Precursors Using BOLCHEM Air Quality Model........................................................... 420 Alberto Maurizi, Mihaela Mircea, Massimo D’Isidoro, Lina Vitali, Fabio Monforti, Gabriele Zanini and Francesco Tampieri 4.10 Modelling Evaluation of PM10 Exposure in Northern Italy in the Framework of CityDeltaIII Project...................................................... 426 C. Carnevale, G. Finzi, E. Pisoni and M. Volta 4.11 Comprehensive Surface-Based Performance Evaluation of a Size- and Composition-Resolved Regional Particulate-Matter Model for a One-Year Simulation ................................................................. 434 M.D. Moran, Q. Zheng, M. Samaali, J. Narayan, R. Pavlovic, S. Cousineau, V.S. Bouchet, M. Sassi, P.A. Makar, W. Gong, S. Gong, C. Stroud and A. Duhamel 4.12 Comparison of Six Widely-Used Dense Gas Dispersion Models for Three Actual Railcar Accidents ............................................................... 443 Steven Hanna, Seshu Dharmavaram, John Zhang, Ian Sykes, Henk Witlox, Shah Khajehnajafi and Kay Koslan 4.13 A Statistical Approach for the Spatial Representativeness of Air Quality Monitoring Stations and the Relevance for Model Validation ...................................................................................................... 452 Stijn Janssen, Felix Deutsch, Gerwin Dumont, Frans Fierens and Clemens Mensink 4.14 Estimation of the Modelling Uncertainty Related with Stochastic Processes .............................................................................. 461 Oxana Tchepel, Alexandra Monteiro and Carlos Borrego
xxxviii
Contents
4.15 Development of a New Canadian Operational Air Quality Forecast Model .............................................................................................. 470 D. Talbot, M.D. Moran, V. Bouchet, L.-P. Crevier, S. Ménard, A. Kallaur and the GEM-MACH Team Chapter 5
Aerosols in the atmosphere......................................................... 479
5.1 Predicting Air Quality: Current Status and Future Directions ................. 481 Gregory R. Carmichael, Adrian Sandu, Tianfeng Chai, Dacian N. Daescu, Emil M. Constantinescu and Youhua Tang 5.2 Diagnostic Analysis of the Three-Dimensional Sulfur Distributions over the Eastern United States Using the CMAQ Model and Measurements from the ICARTT Field Experiment ................... 496 Rohit Mathur, Shawn Roselle, George Pouliot and Golam Sarwar 5.3 Heterogeneous Chemical Processes and Their Role on Particulate Matter Formation in the Mediterranean Region .......................... 505 Marina Astitha, George Kallos, Petros Katsafados and Elias Mavromatidis 5.4 Regional Coverage Modelling of Marine Aerosols Concentration in French Mediterranean Coastal Area .......................................................... 514 Romain Blot, Gilles Tedeshi and Jacques Piazzola 5.5 Formation of Secondary Inorganic Aerosols by High Ammonia Emissions Simulated by LM/MUSCAT ........................................................ 522 Eberhard Renner and Ralf Wolke 5.6 The Origins and Formation Mechanisms of Aerosol during a Measurement Campaign in Finnish Lapland, Evaluated Using the Regional Dispersion Model SILAM ........................................................ 530 Marje Prank, Mikhail Sofiev, Marko Kaasik, Taina Ruuskanen, Jaakko Kukkonen and Markku Kulmala 5.7 Modelling Regional Aerosols: Impact of Cloud Processing on Gases and Particles over Eastern North America and in Its Outflow During ICARTT 2004...................................................... 539 W. Gong, J. Zhang, M.D. Moran, P.A. Makar, S.L. Gong, C. Stroud, V.S. Bouchet, S. Cousineau, S. Ménard, M. Samaali, M. Sassi, B. Pabla, R. Leaitch, A.M. Macdonald, K. Anlauf, K. Hayden, D. Toom-Sauntry, A. Leithead and J.W. Strapp 5.8 On the Role of Ammonia in the Formation of PM2.5 ............................... 548 C. Mensink and F. Deutsch Chapter 6
Interactions between air quality and climate change............... 557
6.1 Linking Global and Regional Models to Simulate U.S. Air Quality in the Year 2050................................................................................ 559 Chris Nolte, Alice Gilliland and Christian Hogrefe
Contents
xxxix
6.2 Impacts of Climate Change on Air Pollution Levels in the Northern Hemisphere with Special Focus on Europe and the Arctic................................................................................................. 568 Gitte B. Hedegaard, Jørgen Brandt, Jesper H. Christensen, Lise M. Frohn, Camilla Geels, Kaj M. Hansen and Martin Stendel 6.3 Regional Climate Change Impacts on Air Quality in CECILIA EC 6FP Project .............................................................................................. 577 Tomas Halenka, Peter Huszar and Michal Belda Chapter 7
Air quality and human health .................................................... 587
7.1 Models of Exposure for Use in Epidemiological Studies of Air Pollution Health Impacts ............................................................................... 589 Michael Brauer, Bruce Ainslie, Michael Buzzelli, Sarah Henderson, Tim Larson, Julian Marshall, Elizabeth Nethery, Douw Steyn and Jason Su 7.2 Long-Term Regional Air Quality Modelling in Support of Health Impact Analyses............................................................................. 605 C. Hogrefe, B. Lynn, K. Knowlton, R. Goldberg, C. Rosenzweig and P.L. Kinney 7.3 A Modeling Methodology to Support Evaluation of Public Health Impacts on Air Pollution Reduction Programs................................... 614 Vlad Isakov and Halûk Özkaynak 7.4 Evaluating the Effects of Emission Reductions on Multiple Pollutants Simultaneously ............................................................................. 623 Deborah Luecken, Alan Cimorelli, Cynthia Stahl and Daniel Tong 7.5 Modelling of the Exposure of Urban Populations to PM2.5, NO2 and O3, and Applications in the Helsinki Metropolitan Area in 2002 and 2025 ........................................................................................................ 632 J. Kukkonen, P. Aarnio, A. Kousa, A. Karppinen, K. Riikonen, B. Alaviippola, M. Kauhaniemi, J. Soares, T. Elolähde and T. Koskentalo 7.6 The Importance of Exposure in Addressing Current and Emerging Air Quality Issues................................................................... 640 Tim Watkins, Ron Williams, Alan Vette, Janet Burke, B.J. George and Vlad Isakov Poster Session ....................................................................................................... 649 P1. Local and urban scale modelling ................................................................. 651 P1.1 Finite Volume Microscale Air-Flow Modelling Using the Immersed Boundary Method ................................................................... 651 V. Fuka and J. Brechler
xl
Contents
P1.2 Simplified Models for Integrated Air Quality Management in Urban Areas............................................................................................... 653 B. Sivertsen, A. Dudek and C. Guerreiro P1.3 Assessment of the Breathability in Urban Canyons Through CFD Simulations and Its Application to Sustainable Urban Design ............. 655 Mário Tomé, Ricardo J. Santos, António Martins and Mário Russo P1.4 Inter-Comparison of Gaussian Plume, Street Canyon and CFD Models for Predicting Ambient Concentrations of NOx and NO2 at Urban Road Junctions ................................................................................ 657 Richard Hill, Peter Jenkinson and Emma Lutman P2. Regional and intercontinental modelling .................................................... 659 P2.1 Local to Regional Dilution and Transformation Processes of the Emissions from Road Transport .......................................................... 659 Dimiter Syrakov, Kostadin Ganev, Reneta Dimitrova, Angelina Todorova, Maria Prodanova and Nikolai Miloshev P2.2 Application of Back Trajectories Using Flextra to Identify the Origin of 137Cs Measured in the City of Barcelona.................................. 661 Delia Arnold, Arturo Vargas, Petra Seibert and Xavier Ortega P2.3 The Role of Sea-Salt Emissions in Air Quality Models ........................ 663 Raúl Arasa, Maria R. Soler and Sara Ortega P2.4 SPECIATE – EPA’s Database of Speciated Emission Profiles............. 665 J. David Mobley, Lee L. Beck, Golam Sarwar, Adam Reff and Marc Houyoux P2.5 Regional Transport of Tropospheric Ozone: A Case Study in the Northwest Coast of Iberian Peninsula.................................................. 667 Santiago Saavedra, María R. Méndez, José A. Souto, José L. Bermúdez, Manuel Vellón and Miguel Costoya P2.6 Modelling of Atmospheric Transport of POPs at the European Scale with a 3D Dynamical Model Polair3D-POP ........................................ 669 Solen Quéguiner and Luc Musson-Genon P2.7 Evolution of the Ozone Episodes in Northern Iberia (Cantabric and Pyrenaic regions) Under West European Atlantic Blocking Anticyclones .................................................................................................. 671 V. Valdenebro, G. Gangoiti, A. Albizuri, L. Alonso, M. Navazo, J.A. García and M.M. Millán P2.8 High Temporal Resolution Measurements and Numerical Simulation of Ozone Precursors in a Rural Background ............................... 673 M. Navazo, N. Durana, L. Alonso, J. Iza, J.A, García, J.L. Ilardia, G. Gangoiti and M. De Blas
Contents
xli
P2.9 Nonlinearity in Source-Receptor Relationship for Sulfur and Nitrate in East Asia ................................................................................. 675 Woo-Sub Roh, Seung-Bum Kim and Tae-Young Lee P2.10 Modelling the Impact of Best Available Techniques for Industrial Emissions Control in Air Quality: Setting Up Inventories and Establishing Projections ....................................................... 677 R. Rodriguez, P. Maceira, J.A. Souto, J. Casares, A. Sáez and M. Costoya P2.11 Lake Breezes in Southern Ontario: Observations, Models and Impacts on Air Quality............................................................................ 679 David Flagg, Jeff Brook, David Sills, Paul Makar, Peter Taylor, Geoff Harris, Robert McLaren and Patrick King P2.12 High Time and Space Resolution Ozone Modelling in Regional Air Quality Management of a Complex Mountain Area Using Calgrid 2.44 ................................................................................ 681 Carlo Trozzi, Silvio Villa and Enzo Piscitello P2.13 Analysis of Atmospheric Transport of Radioactive Debris Related to Nuclear Bomb Tests Performed at Novaya Zemlya ..................... 683 Jørgen Saltbones, Jerzy Bartnicki, Tone Bergan, Brit Salbu, Bjørn Røsting and Hilde Haakenstad P2.14 Development and Application of a New Model for the Atmospheric Transport and Surface Exchange of Semi-Volatile Organics Using the CMAQ Model Framework. ................ 685 Fan Meng, Baoning Zhang, Fuquan Yang and James Sloan P2.15 Saharan Dust over Italy: Simulations with Regional Air Quality Model BOLCHEM ........................................................................... 687 Mihaela Mircea, Massimo D’Isidoro, Alberto Maurizi, Francesco Tampieri, Maria Cristina Facchini, Stefano Decesari and Sandro Fuzzi P3. Data assimilation and air quality forecasting............................................. 689 P3.1 Detection of a Possible Source of Air Pollution Using a Combination of Measurements and Inverse Modelling .............................. 689 Borivoj Rajkovic, Zoran Grsic and Mirjam Vujadinovic P3.2 Improving Emission Inventory in Lithuania.......................................... 691 Vidmantas Ulevicius, Vytautas Vebra, Kestutis Senuta and Svetlana Bycenkiene P4. Model assesment and verification................................................................ 693 P4.1 Tropospheric Ozone and Biogenic Emissions in the Czech Republic......................................................................................................... 693 K. Zemankova and J. Brechler
xlii
Contents
P4.2 Air Pollution Dispersion Modelling Arround Thermal Power Plant Trbovlje in Complex Terrain – Model Verification and Regulatory Planning................................................................................ 695 Marija Zlata Božnar, Primož Mlakar, Boštjan Grašiþ and Gianni Tinarelli P4.3 Development of a Quasi-Real-Time Forecasting System over Tokyo............................................................................................................. 697 Masayuki Takigawa, Masanori Niwano, Hajime Akimoto and Masaaki Takahashi P4.4 A Construction and Evaluation of Eulerian Dynamic Core for the Air Quality and Emergency Modelling System SILAM .................... 699 Mikhail Sofiev, Michael Galperin and Eugene Genikhovich P4.5 BOLCHEM Air Quality Model: Performance Evaluation over Italy........................................................................................................ 702 Alberto Maurizi, Mihaela Mircea, Massimo D’Isidoro, Lina Vitali, Fabio Monforti, Gabriele Zanini and Francesco Tampieri P4.6 Evaluation of an Operational Ensemble Prediction System for Ozone Concentrations over Belgium Using the CTM Chimere............... 705 Andy Delcloo and Olivier Brasseur P4.7 The Use of MM5-CMAQ-EMIMO Modelling System (OPANA V4) for Air Quality Impact Assessment: Applications for Combined Cycle Power Plants and Refineries (Spain) ............................ 707 R. San José, J.L. Pérez, J.L. Morant and R.M. González P4.8 Verification of Ship Plumes Modelling and Their Impacts on Air Quality and Climate Change in QUANTIFY EC 6FP Project ........... 709 Tomas Halenka, Peter Huszar and Michal Belda P5. Aerosols in the atmosphere........................................................................... 711 P5.1 Quantifying Source Contribution to Ambient Particulate Matter in Austria with Chemical Mass Balance Receptor Modeling ........................ 711 A. Caseiro, H. Bauer, I. Marr, C. Pio, H. Puxbaum and V. Simeonov P6. Interactions between air quality and climate change ................................ 713 P6.1 On the Effective Indices for Emissions from Road Transport............... 713 Kostadin Ganev, Dimiter Syrakov and Zahari Zlatev P7. Air quality and human health ...................................................................... 715 P7.1 A Multi-Objective Problem to Select Optimal PM10 Control Policies .......................................................................................................... 715 Claudio Carnevale, Enrico Pisoni and Marialuisa Volta
Contents
xliii
P7.2 What Activity-Based Analysis and Personal Sampling Can Do for Assessments of Exposure to Air Pollutants?............................................ 717 Doina Olaru and Jennifer Powell P7.3 Intake Fraction for Benzene Traffic Emissions in Helsinki................... 719 Joana Soares, Ari Karppinen, Leena Kangas, Matti Jantunen and Jaakko Kukkonen P7.4 Source Apportionment of Particulate Matter in the U.S. and Associations with In Vitro and In Vivo Lung Inflammatory Markers.......................................................................................................... 721 Rachelle M. Duvall, Gary A. Norris, Janet M. Burke, John K. McGee, M. Ian Gilmour and Robert B. Devlin P7.5 Air Pollution Assessment in an Alpine Valley ...................................... 723 Peter Suppan, Klaus Schäfer, Stefan Emeis, Renate Forkel, Markus Mast, Johannes Vergeiner and Esther Griesser P7.6 New Approaches on Prediction of Maximum Individual Exposure from Airborne Hazardous Releases ............................................... 725 John G. Bartzis, Athanasios Sfetsos and Spyros Andronopoulos P7.7 The Detroit Exposure and Aerosol Research Study .............................. 727 Ron Williams, Alan Vette, Janet Burke, Gary Norris, Karen Wesson, Madeleine Strum, Tyler Fox, Rachelle Duvall and Timothy Watkins Author Index......................................................................................................... 729 Subject Index…………… ……………………………..…………….............…..735
1.6 Air Quality Management Strategies in Large Cities: Effects of Changing the Vehicle Fleet Composition in Barcelona and Madrid Greater Areas (Spain) by Introducing Natural Gas Vehicles María Gonçalves, Pedro Jiménez-Guerrero and José M. Baldasano
Abstract Air quality modelling involves a strategy to manage air pollution in large cities, where air quality problems presently are mainly related to on-road traffic. Nowadays, one of the strategies to reduce emissions is based on the substitution of vehicles by introducing new technologies (e.g. cleaner fuels, hybrid vehicles, fuel cells, etc.). This work assesses the variation on air quality due to the substitution of specific vehicle fleets by natural gas vehicles in the two largest cities of Spain: Barcelona and Madrid. Six different scenarios are studied, focusing on the total or partial modification of public transportation vehicles (buses, taxis), freight vehicles and private vehicles. One scenario involving a combination of all of them is also studied. Under this perspective, the WRF/HERMES/CMAQ modelling system has been implemented and validated with a high resolution (1 km and 1 hour) in the area thanks to the calculation power of the MareNostrum super-computer of the Barcelona Supercomputing Center (94.21 TFlops peak). Daily average concentrations of NO2, SO2 and PM, both PM10 and PM2.5, and 8-hour average concentration for O3 and 1-hour maximum concentrations for these species are estimated both in Barcelona and Madrid Greater Areas. All the scenarios studied involve a reduction in NO2, SO2 and PM concentrations. Most important changes in air quality are registered when the combined scenario is implemented. Ozone concentrations remain approximately in the same levels as in the base case scenario, except for some VOC-limited areas where the reduction of NOx involves a slight O3 increase (under the 10%). A large reduction in PM concentration is observed for both cities when the 50% of commercial light vehicles is transformed. Results of the simulations for the combined scenario indicate that it is particularly effective in reducing PM10 (up to –43% in maximum hourly concentration at some points) and PM2.5 (up to –36%). Keywords Air quality modelling, alternative fuels, emissions, natural gas vehicles, urban pollution management
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
54
Air Quality Management Strategies in Large Cities: Effects of Changing the Vehicle Fleet
55
1. Introduction Air pollution is a major environmental problem in urban areas, where 80% dwellers in Europe live. The EU air quality standards (EU-AQS) are currently exceeded in some urban sites, particularly in terms of NO2 and PM10 (EEA, 2006). In spite of the unitary emissions reduction by vehicle achieved during last years, large contributions to atmospheric pollutants emissions still come from on-road transport (Costa and Baldasano, 1996; Oduyemi and Dadvison, 1998; Colvile et al., 2001; Crabbe et al., 1999; Ghose et al., 2004). Moreover some EU-AQS will be reduced in a near future (2010) and some of the emissions abatement strategies put into practice to reduce traffic emissions does not seem to be effective (Carslaw et al., 2007). Therefore public administrations focus their efforts on assessing the best alternatives to reduce urban on-road transport impacts. Different possibilities are being tested; among others the European Commission promotes the use of alternative fuels, specifically the natural gas as a fuel in the medium term (EC, 2001). The evaluation of the effects on air quality caused by the implementation of emission abatement strategies is essential in order to aid policy-makers and to establish the real efficacy of the different environmental plans. This work assesses the changes on air quality achieved with the introduction of natural gas vehicle fleets in two main cities of Spain: Barcelona and Madrid, using the WRF-ARW/HERMES/CMAQ simulation system. The evaluation is carried out in terms of ozone (O3), nitrogen dioxide (NO2), sulphur dioxide (SO2) and particulate matter (PM10 and PM2.5) concentrations. PM2.5 is assessed, in spite of not being still regulated by EU directives, due to its importance related with human health (Pope and Dockery, 2006).
2. Methods Air quality simulations are performed for an episode of photochemical pollution, selected considering air quality data monitored in Catalonia and Madrid, but also considering that the traffic situation must correspond to a normal day, avoiding weekends or holidays, in order to exclude distorting factors in traffic circulation. The chosen episode (June 17–18, 2004) fits in a typical summertime low-pressure gradient with very high levels of photochemical pollutants (especially O3 and PM10) over the Iberian Peninsula. The emissions scenarios were performed under a realistic approach, changing specific vehicle fleets: (1) transformation of the 100% of the urban buses to natural gas buses; (2) transformation of the 50% of taxis to natural gas vehicles (NGV); (3) transformation of 50% of intercity buses to natural gas buses; (4) transformation of a 50% of light commercial vehicles to NGV; (5) transformation of a 10% of private cars to natural gas cars; (6) transformation of 100% of heavy duty freight transport vehicles to NGV; and (7) combination of all the rest.
M. Gonçalves et al.
56
The Eulerian, three-dimensional model WRF/HERMES/CMAQ was applied with high spatial (1 km2) and temporal (1 hour) resolution. The use of fine scale was imposed by the necessity of assessing the subtle air quality variations in urban areas, as shown in the CityDelta project experience (Cuvelier et al., 2007), and in very complex terrains as Barcelona and Madrid Greater Areas (BGA and MGA) (Jiménez et al., 2005). The calculation was performed in a feasible time thanks to the MareNostrum supercomputer hold by the Barcelona Supercomputing Center (94.21 Tflops peak). The traffic module of HERMES emission model includes an intensive road net description, circulation and mobility data and the vehicle fleet composition for Spain for the year 2004, differentiating the specific composition of Barcelona and Madrid cities. The vehicle fleet is divided in 72 vehicle categories as a function of their age, the cubic capacity of their engine, the weight and the type of fuel they use. The EMEP-CORINAIR/EEA methodology was used in order to obtain the speed dependant emissions factors for each of them. The emission factors for NGV were obtained applying the emission reduction factors provided by the European NGV Association (Table 1) to the diesel Euro III emission factors. Table 1 Emission correction factors for the natural gas vehicles categories provided by the European Natural Gas Vehicles Association (ENGVA). NG category
Emission correction factor
Reference category CO
NMHC
NOx
PM
NG cars and LDVs
Euro III diesel cars and LDVs (<7.5 t)
0.53
0.44
0.18
0.05
NG HDVs
Euro III diesel HDVs (>7.5 t)
0.58
0.11
0.16
0.12
The 8-hour average O3 and 24-hour average NO2, SO2, PM10 and PM2.5 concentrations and 1-hour maximum O3, NO2, SO2, PM10 and PM2.5 concentrations were calculated for each scenario over the Barcelona and Madrid Greater Areas.
3. Results The emissions variation analysis indicates that the ozone precursors and the primary pollutant emissions decrease in all scenarios, especially in the combined case (scenario 7), when the changes in the vehicle fleet are more pronounced (up to 26.1% of vehicles changed in BGA and up to 23.1% in MGA). The main reductions affect PM and NOx emissions. The simulation results for the base case were validated with air quality data from several stations in the city areas and accomplished the EU directives (1999/30/CE, 2002/3/CE) and US-EPA guidelines related to O3 concentrations developed during the years 1991 and 2005. In the BGA, the station of Barcelona-Example shows a
Air Quality Management Strategies in Large Cities: Effects of Changing the Vehicle Fleet
57
negative MNBE for the primary pollutants (–7.78% for NO2; –23.25% for SO2 and –14.21% for PM10) due to the important street-scale influence of on-road traffic. The O3 is closely overestimated by the model (MNBE 8.94% for tropospheric O3). At the same time, the 1-hour peaks are slightly underestimated (UPA –12.82% for O3; –2.84% for NO2; –18.18% for SO2 and –8.47% for PM10). The model also tends to underestimate the concentration of the different pollutants for the stations of the MGA, as Getafe (MNBE –3.96% for O3; –14.58% for NO2; –20.02% for SO2 and –10.33% for PM10). The results of the evaluation confirm the need for working with fine grids in areas where the influence of on-road traffic is important; it becomes essential for addressing air quality processes in urban and industrial areas. With respect to the results of the base case scenario simulations, Figure 1 indicates that main problems in both Greater Areas are related to NO2 and PM10 concentrations; moreover their maximum concentrations are located over the road net which reflects the influence of traffic emissions on urban air quality. The difference between the 8-hour O3 and 24-hour NO2, SO2, PM10 and PM2.5 average concentrations and 1-hour maximum concentration for these pollutants, in the base case scenario and the different scenarios tested, over BGA and MGA, were estimated. All emissions scenarios exhibit a slight increase in O3 (the maximum hourly concentration rises up to 7.8% at some points of BGA), due to the NOx reduction in VOC-controlled areas; while the NO2, SO2 and PM concentrations decrease. The combined scenario involves the largest reductions in all cases (Figure 2). For instance the NO2 24-hour average concentration decreases up to –23.2%, and the maximum hourly concentration of SO2 up to –20.7%, of PM10 up to – 42.8% and of PM2.5 up to –35.6% at some points of the study areas. PM reductions are especially remarkable, indicating the on-road traffic origin of the largest part of the particulate matter concentrations that are registered in the urban areas. These reductions are larger over MGA, where the traffic has a larger weight as pollutant emissions source than in BGA: on average PM2.5 decreases – 2.2% over the MGA and a –0.2% over the BGA, while the PM10 reductions achieve the –14.9% in MGA and –6.6% in BGA Concerning SO2 reductions in MGA are also larger than in BGA, up to –8.7% average reductions in the combined scenario versus –1.5%. Again, the emission sources composition is an essential factor that affects the emissions abatement strategies effects. On one hand, MGA is characterized by a commercial and tertiary activity, meanwhile BGA presents a more important industrial component, which has a direct effect on the weight of traffic contribution to SO2 emissions and indirectly involves that a traffic-emissions abatement strategy could be more effective in reducing SO2 concentrations in Madrid than in Barcelona. Nevertheless, currently SO2 levels do not represent a problem in urban air quality.
58
M. Gonçalves et al.
Fig. 1 8-hour O3 average concentration and 24-hour NO2, SO2 and PM10 average concentration in the base case scenario over the BGA (up) and the MGA (down)
Air Quality Management Strategies in Large Cities: Effects of Changing the Vehicle Fleet
59
Fig. 2 Difference of 8-hour O3 average concentration and 24-hour NO2, SO2 and PM10 average concentration between the combined and the base case scenario over the BGA (up) and the M GA (down)
60
M. Gonçalves et al.
Concerning individual scenarios tested, introducing a 50% of natural gas commercial light vehicles is the most effective in reducing NO2 concentrations in BGA (up to –2.5% on average); nevertheless in MGA better results are achieved when changing the 10% of the private cars fleet (up to –8.2% on average). These differences reflect the necessity of assessing the best strategies to be applied locally; taking into account the specific characteristics of the study area, in this case the different vehicle fleet composition between both cities (larger percentage of cars in Madrid than in Barcelona) entails this behaviour. In order to reduce SO2 and PM concentrations, the most effective measure involves the substitution of the oldest diesel and petrol light commercial vehicles (Figure 2), which are important contributors to these species emissions, by natural gas light duty vehicles (up to 7.5 t). In this case the PM10 hourly maximum concentration is reduced up to –20.8% at some points of BGA and up to 14.6% at some points of MGA, and the maximum hourly concentration of PM2.5 up to –11.3% at BGA and up to –12.2% at MGA.
4. Summary and Conclusion The NO2, SO2 and PM concentrations decrease in all scenarios, both for BGA and MGA domains. The largest reductions are achieved in the combined scenario (– 23% in NO2, –21% in SO2, –43% in PM10 and –36% in PM2.5 at some points of the study areas). The individual scenarios that prove to be more effective in reducing NO2, SO2 and PM concentrations are the transformation of 50% of commercial light vehicles and the 10% of private cars. Changing the 100% of the urban buses, the 50% of the intercity buses and the 100% of the heavy duty freight transport vehicles do not involve a vehicle fleet change larger than 1.5%, so that emissions (less than 5%) and air quality variation is not remarkable (less than 3%). Large cities are typically VOC-controlled areas; therefore the reduction in NOx concentration causes the increase of O3 maximum hourly and 8-hour average concentrations (less than 10% in all study cases). The most effective individual scenarios in reducing NO2 concentration are the 50% of commercial light vehicles change in Barcelona (up to –18%) and the 10% of private cars change in Madrid (up to –10%), because the vehicle fleet of Madrid is mainly composed of diesel and petrol private cars and taxis (82% versus a 66% in Barcelona). PM and SO2 concentrations reduction is especially noticeable when changing the commercial light vehicles into NGV. The specific local characteristics of the affected areas must be taken into account when assessing the effects of environmental management strategies. This work confirms the importance of having detailed emissions inventories and designing the environmental strategies under a realistic approach to obtain representative results. In the present case the traffic emissions reduction strategies are more effective in improving air quality in MGA than in BGA. The simulation system WRFARW/HERMES/CMAQ can be applied to assess the efficacy of abatement emissions
Air Quality Management Strategies in Large Cities: Effects of Changing the Vehicle Fleet
61
strategies in terms of air quality variation and due to its high resolution (1h-1km2) detects subtle changes in urban areas. Acknowledgments The authors gratefully acknowledge E. López for the implementation of HERMES. Also the authors thank to Natural Gas Corporation and the European Natural Gas Vehicles Association (ENGVA) for the collaboration in the assessment of the emission factors for NGV. This work was funded by the projects CICYT CGL2006-0803 and CICYT CGL2006-11879 of the Spanish Ministry of Education and Science and CALIOPE project 441/2006/3-12.1 of the Spanish Ministry of the Environment.
References Carslaw DC, Beevers SD, Bell MC (2007) Risks of exceeding the hourly EU limit value for nitrogen dioxide resulting from increased road transport emissions of primary nitrogen dioxide, Atmospheric Environment 41, 2073–2082. Colvile RN, Hutchinson EJ, Mindell JS, Warren RF (2001) The transport sector as a source of air pollution, Atmospheric Environment 35, 1537–1565. Costa M, Baldasano JM (1996) Development of a source emission model for atmospheric pollutants in the Barcelona area, Atmospheric Environment 30A, 2, 309–318. Crabbe H, Beaumont R, Norton D (1999) Local air quality management: a practical approach to air quality assessment and emissions audit, The Science of the Total Environment 235, 383–385. Cuvelier C et al. (2007) CityDelta: A model intercomparison study to explore the impact of emission reductions in European cities in 2010, Atmospheric Environment 41, 189–207. EC (2001) Communication of the European Commission of 07/11/2001 on an action plan and two proposals for directives to foster the use of Alternative Fuels for Transport, starting with the regulatory & fiscal promotion of biofuels. Brussels 7.11.2001 COM (2001) 547, 47 pp. EEA (2006) Transport and environment: facing a dilemma. TERM2005: indicators tracking transport and environment in the European Union. EEA Report nº 3/2006, 56 pp. Ghose MK, Paul R, Banerjee SK (2004) Assessment of the impacts of vehicular emissions on urban air quality and its management in Indian context: the case of Kolkata (Calcutta), Environmental Science and Policy 7, 345–351. Jiménez P, Jorba O, Parra R, Baldasano JM (2005) Influence of high-model grid resolution on photochemical modelling in very complex terrains, International Journal of Environment and Pollution 24, 180–200. Oduyemi K, Dadvison B (1998) The impacts of road traffic management on urban air quality, The Science of the Total Environment 218, 59–66. Pope CA, Dockery DW (2006) Health effects of fine particulate air pollution: lines that connect, Journal of Air and Waste Management Association 56, 709–742.
62
M. Gonçalves et al.
Discussion S.T. Rao: Have you also looked at the concentrations of toxic air contaminants (e.g., benzene, acetaldehyde, formaldehyde), not just criteria pollutants (e.g., ozone, CO, NO2) as changed the mobile source emissions from biofuels, electric, etc.? J. Baldasano: We have focused on O 3, NOx and PM, because they represent the most important problems related to air quality presently in the cities of Madrid and Barcelona (Spain).
1.4 CFD and Mesoscale Air Quality Modelling Integration: Web Application for Las Palmas (Canary Islands, Spain) R. San José, J.L. Pérez, J.L. Morant and R.M. González
Abstract The integration of sophisticated mesoscale air quality modelling systems, such as MM5-CMAQ and new generation of Computational Fluid Dynamics (CFD) modelling tools has been developed in this contribution. We have used an advanced and adapted version of the MIMO model (U. Karlsruhe, Germany) which is a sophisticated CFD model, to simulate the air concentrations at urban level with 10 m spatial resolution over the city of Las Palmas (Canary Islands, Spain). The CFD code receives the traffic emission data every second produced by a cellular automata model (CAMO). The integrated CFD model is called MICROSYS. This model is an Eulerian 3D tool which is running in diagnostic mode once every minute. The boundary conditions are obtained from the well-known MM5-CMAQ running over the city in prognostic mode. The MM5-CMAQ (OPANA V4) model is run with 1 km spatial resolution covering a domain of 16 × 16 km over the city. This system is operating in forecasting mode since 2004 and is operated over the Internet. The forecasting information for meteorology and air quality concentrations for the following 72 hours is used by MICROSYS to simulate the expected air concentrations at street level for the next three days. The system operates under daily basis and produces the detailed forecasting information at 6:00 GMT everyday. The Internet service includes a sophisticated VRML (Virtual Reality Modelling Language) tool to visualize in a 3D mode the air concentrations at street level by an Internet client. The VRML tool runs on the client server. We present also some comparative results related to the use of shared 64 bits memory machines and single 32 bits one-processor machines for CFD runs. Keywords Air pollution, CFD, modelling, numerical simulations
1. Introduction The use of urban air quality models based on Computational Fluid Dynamics tools (CFDs) requires receiving frequent metrological and air quality information from the numerical boundaries. The CFDs models are in fact embedded into the mesoscale air quality applications from which they receive the proper boundary C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
37
38
R. San José et al.
conditions. The advances on the capability of Computational Fluid Dynamics models and Air Quality Modelling Systems during the last decade have been quite substantial. The increase on computer capabilities and on the knowledge of turbulence parameterization and numerical schemes has also been very important during the last ten years. On the other hand, there is a considerable public interest on information related to the “real” pollution they are exposure on when they are walking in the street going to work or even during the period they are driving a car from/to work or other daily activities. At street level the differences in the concentration values at both sides of a street can be important, particularly, for instance, on relation to photochemical production during summer time in Mediterranean regions. In this contribution we have used the CFD model MIMO (U. of Karlsruhe (Germany)) and the mesoscale air quality modelling system MM5-CMAQ-EMIMO (NCEP/EPA /Technical University of Las Palmas de Gran Canaria (Canary Islands, Spain)) to simulate the impact of different emission reduction scenarios in the downtown area of Las Palmas de Gran Canaria (Canary Islands, Spain) City. These complex systems could evaluate the impact of several urban strategic emission reduction measures such as reduction of private traffic, increase of public transportation, impact on introduction of new fuel cell vehicles, etc. Also, they could be used for analysis of pollution concentrations at different heights (buildings) and on different areas of urban neighbourhoods. Air dispersion in urban areas is affected by atmospheric flow changes produced by building-street geometry and aerodynamic effects. The traffic flow, emissions and meteorology are playing also an important role. Microscale air pollution simulations are a complex task since the time scales are compared to the spatial scales (micro) for such a type of simulations. Boundary and initial conditions for such simulations are also critical and essential quantities to influence fundamentally the air dispersion results. Microscale Computational Fluid Dynamical Models (CFDM) are playing an increasing role on air quality impact studies for local applications such as new road and building constructions, emergency toxic dispersion gases at urban and local scale, etc. Microscale air dispersion simulations are applied to predict air-flow and pollution dispersion in urban areas. Pullen et al. (2005). Different combinations and applications appear in the literature such as Pospisil et al. (2004) by integrating a Lagrangian model and a traffic dynamical model into a commercial CFD code, Star-CD to simulate the traffic-induced flow field and turbulence. In this contribution we have applied the microscale dispersion model MIMO (Ehrhard et al., 2000) to create an operational air quality forecasting system based on then web in Las Palmas de Gran Canaria (Canary Islands, Spain). The MIMO CFD code has been adapted and incorporated into a mesoscale air quality modelling system (MM5-CMAQ-EMIMO) to fit into the one-way nesting structure. MM5 is a meteorological mesoscale model developed by Pennsylvania State University (USA) and NCAR (National Centre for Atmospheric Research, USA) (Grell et al., 1994). The CMAQ model is the Community Multiscale Air Quality Modelling System developed by EPA (USA) (Byun et al., 1998) and EMIMO is the Emission Model developed by San José R. et al. (2003). MM5 is a well recognized non-hydrostatic mesoscale meteorological models which uses global
CFD and Mesoscale Air Quality Modelling Integration: Web Application for Las Palmas
39
meteorological data produced by global models such as GFS model (NCEP, USA) to produce high resolution detailed three dimensional fields of wind, temperature and humidity which are used in our case as input for the photochemical dispersion model CMAQ (San José et al., 1997). In addition of MM5 output data, EMIMO model produces for the specific required spatial resolution, hourly emission data for different inorganic pollutants such as particulate matter, sulphur dioxide, nitrogen oxides, carbon monoxide and total volatile organic compounds VOCs. The VOCs are splitted according to SMOKE (Sparse Matrix Operator Kernel Emissions) Williams et al. (2001) and Coats (1995). The CFD and mesoscale models solve the Navier-Stokes equations by using different numerical techniques to obtain fluxes and concentrations at different scales. Mesoscale air quality models cover a wide range of spatial scales from several thousands of kilometres to 1 km or so. In this contribution we have applied the MM5-CMAQ-EMIMO model over Las Palmas de Gran Canaria (Canary Islands, Spain) domain to obtain detailed and accurate results of the pollutant concentrations at this spatial resolution in forecasting mode and the MIMO CFD model over a 1 × 1 km domain with several spatial resolutions ( 2–10 m ) and different vertical resolutions. MM5-CMAQ-EMIMO data serves as initial and boundary conditions for MIMO modelling run. In Figure 1 we observe the spatial architecture for the application of the MM5CMAQ-EMIMO mesoscale air quality modelling system. In Figure 2 we show a detailed diagram of the EMIMO modelling system. EMIMO is currently operating with the so called Version 2 which includes the CLCL2000 with 44 different landuse types with 100 m spatial resolution. EMIMO 2.0 also uses the CIESIN 30’’ (CIESIN, 2004), population database and the Digital Chart of the World 1 km land use database to produce adequate emission data per 1 km grid cell per hour and per pollutant. In this case EMIMO has used the detailed GIS database provide by the City of Las Palmas de Gran Canaria (Canary Islands, Spain). In order to apply the EMIMO CFD model, we need detailed information related to the building structure in the 1 km grid cell. This information is shown in Figure 3 for the total of the Las Palmas de Gran Canaria (Canary Islands, Spain) Community (Spain). The height of the buildings is not included in this file and it has been estimated directly for this experiment. A cellular automata traffic model (CAMO) has been developed. CAMO – which has been included into the EMIMO modelling system – is based on transitional functions defined in a discrete interval t as follows:
s (t 1) p ( s (t ), a (t )) u (t ) v( s (t ))
(1)
where s(t + 1), s(t) is a defined sate, a(t) is an input symbol and u(t) is an output symbol. We have used the Moore neighbourhood with eight different surrounding cells where each cell – representative of a vehicle – can move on. The whole system focusing on the 1 × 1 km urban area in Las Palmas de Gran Canaria (Canary Islands, Spain) downtown is called MICROSYS system. We have selected a
R. San José et al.
40
subdomain of 300 × 300 m with 5 m spatial resolution and 15 vertical layers for this particular experiment (Figure 4). The first 10 layers are equally spaced with 5 m spatial resolution up to 50 m in height and the last five layers are located at 55, 61.55, 68.20, 75.52 and 83.58 m in height.
Fig. 1 MM5-CMAQ-EMIMO architecture for this application
Fig. 2 EMIMO model basic architecture
CFD and Mesoscale Air Quality Modelling Integration: Web Application for Las Palmas
41
Fig. 3 E00 vector file for Las Palmas de Gran Canaria (Canary Islands, Spain) Community
Fig. 4 Subdomain view for this experiment in Las Palmas de Gran Canaria (Canary Islands, Spain) City downtown area
42
R. San José et al.
2. Results The system is operating under daily basis as follows: (a) the MM5-CMAQEMIMO mesoscale air quality forecasting system is operating under daily basis, starting at 01:00 GMT and running to complete six simulation days; (b) the MICROSYS CFD system is operating 1 minute every hour during the last 72 hours of simulation (future time). This 72 steady state simulations produce pollution concentrations in a 3D domain. Since the system is mounted for seven 1 × 1 km domains, covering the entire city, we have to multiply these 72 simulations times seven which makes 504 steady state simulations. Figure 5 shows a picture of the web site when showing a 1 × 1 km area in the n north of the city of Las Palmas de Gran Canaria (Canary Islands, Spain). On the right side, we observe a full user control panel to select several functions and capabilities to visualize the results. In Figure 6 we observe a detailed view for NO2 over an area of the District Ciudad del Mar (Las Palmas, Canary Islands, Spain) and a time series in a cross of streets in such a district. Finally in Figure 7, we see an example.
Fig. 5 OPANA model including the MICROSYS CFD for Las Palmas (Canary Islands, Spain) as seen in the web site for a 1 × 1 km area in the north of the Municipality
CFD and Mesoscale Air Quality Modelling Integration: Web Application for Las Palmas
43
Fig. 6 Detailed view of the District: Ciudad del Mar in Las Palmas (Canary Islands, Spain) for NO2 concentrations on June 14, 2007 at 08:00 GMT. Also we see the time series in a cross of streets in such a district
Fig. 7 VRML view on Internet for an area in the North of the city observing the NO2 concentrations in a 3D context. The user can navigate through the streets with different views
44
R. San José et al.
3. Conclusions The MM5-CMAQ-EMIMO modelling system has been used to provide detailed initial and boundary conditions to a system called MICROSYS which is composed by the MIMO CFD microscale dispersion model and CAMO which is a cellular automata traffic model. The results show that the air quality modelling system offers realistic results although no comparison with eddy-correlation measurement system has been performed in the area. The tool can be used for many air quality impact studies but in particular for traffic emission reduction strategies. Acknowledgments We would like to thank Professor N. Moussiopoulos (AUTH, Greece) for providing the MIMO model. Also to EPA/PSU/NCAR for providing the MM5-CMAQ modeling system code.
References Byun DW, Young J, Gipson G, Godowitch J, Binkowsky F, Roselle S, Benjey B, Pleim J, Ching JKS, Novak J, Coats C, Odman T, Hanna A, Alapaty K, Mathur R, McHenry J, Shankar U, Fine S, Xiu A, Lang C (1998) Description of the Models-3 Community Multiscale Air Quality (CMAQ) model. Proceedings of the American Meteorological Society 78th Annual Meeting Phoenix, AZ, January 11–16, 1998, pp. 264–268. CIESIN (2004) Center for International Earth Science Information Network (CIESIN). Global Rural-Urban Mapping Project (GRUMP): Urban/Rural population grids. CIESIN, Columbia University, Palisades, NY. http://sedac.ciesin.columbia.edu/gpw/ Coats CJ, Jr (1995) High Performance Algorithms in the Sparse Matrix Operator Kernel Emissions (SMOKE) Modelling System, Microelectronics Center of North Carolina, Environmental Systems Division, Research Triangle Park, NC, 6 pp. Ehrhard J, Khatib IA, Winkler C, Kunz R, Moussiopoulos N, Ernst G (2000) The microscale modelo MIMO: Development and assessment. Journal of Wind Engineering and Industrial Aerodynamics, 85, 163–176. Grell G, Dudhia J, Stauffer D (1994) A Description of the Fifty Generation Penn State/NCAR Mesoscale Model (MM5). NCAR Tech. Note, TN-398 + STR, 117 pp. Pospisil J, Katolicky J, Jicha M (2004) A comparison of measurements and CFD model predictions for pollutant dispersion in cities. Science of the Total Environment. 334–335; 185–195. Pullen J, Boris JP, Young T, Patnaik G, Iselin J (2005) A comparison of contaminant plume statisticsfrom a Gaussian puff and urban CFD model for two large cities. Atmospheric Environment, 39, 1049–1068.
CFD and Mesoscale Air Quality Modelling Integration: Web Application for Las Palmas
45
San José R, Prieto JF, Castellanos N, Arranz JM (1997) Sensitivity study of dry deposition fluxes in ANA air quality model over Las Palmas de Gran Canaria (Canary Islands, Spain) mesoscale area, Measurements and Modelling in Environmental Pollution, Ed. San José and Brebbia, pp. 119–130. San José R, Peña JI, Pérez JL, González RM (2003) EMIMO: an emission model, 292–298, Springer-Verlag. ISBN: 3-540-00840-3. Williams A, Caughey M, Huang H-C, Liang X-Z, Kunkel K, Tao Z, Larson S, Wuebbles D (2001) Comparison of emissions processing by EM-S95 and SMOKE over the Midwestern U.S. Preprint of International Emission Inventory Conference: One Atmosphere, One Inventory, Many Challenges. Denver, CO, May 1–3, pp. 1–13.
Discussion A. Baklanov: What was the resolution of the finest grid of MM5 used for boundary conditions for your CFD model runs? For correct downscaling with CFD it should be a city scale. However you use the MM5 version without any urban parameterization. So, your boundary conditions for CFD obstacle-resolved model within a city are not correct, because they don’t consider urban features. R. San José: You are right that we do not use the urbanized version of MM5 (although we have been working with it). The reason for that is the important increase on computer time which is a very sensitive issue for our operational forecasting system. However, the CFD model is running with 5 m resolution for the CFD domain (1 × 1 km) (we have 7 CFD 1 × 1 km domains for covering the whole city). We believe that this is a good approach although the computer limitations do not allow us to run MM5 urban instead of classic MM5. M. Mircea: What emission inventory is used from EU to urban scale? How is done the nesting? R. San José: We use EMEP 2000 50 km spatial resolution. We go down to 1 km by using EMIMO model approach which uses population and km of roads per squared km (mixed top-down and bottom-up approaches). The nesting for this application is done as one-way nesting. Using CAMx, we have used the two-way nesting approach.
1.9 Development and Application of MicroRMS Modelling System to Simulate the Flow, Turbulence and Dispersion in the Presence of Buildings S. Trini Castelli, T.G. Reisin and G. Tinarelli
Abstract A modelling system for the simulation of the flow and dispersion from the mesoscale down to the urban microscale is under development. This modelling system is a microscale version of the regional off-line system RMS (RAMSMIRS-SPRAY) – MicroRMS. A modified version of RAMS6.0 is used, in which a Cartesian grid and the ADaptive Aperture method are implemented for defining the presence of buildings in arbitrarily steep topography and where alternative versions of the k-İ turbulence closure model were incorporated. After RAMS, the Lagrangian stochastic dispersion model MicroSPRAY is applied, specially devoted to simulate accidental gaseous releases at microscales, including the presence of obstacles and buildings. At present, the efforts are focused on the development of a micro-version of the interface code MIRS, calculating the surface and boundary layers’ parameters and the Lagrangian variables. The first step in the project was to harmonize the treatment of buildings between RAM6.0 and MicroSPRAY approaches. Here we present the first tests of MicroRMS prototype, applied to the MUST exercise of Cost732 Action, a flow and dispersion field test carried out in the Great Basin Desert (USA) in 2001, where 120 standard containers were set up in a regular array of obstacles. Keywords Buildings, COST732 Action, flow and dispersion modelling, microscale, MUST experiment
1. Introduction Modelling atmospheric flows and pollutant dispersion in urban areas is a problem of peculiar characteristics, due to the complexity and heterogeneity of the urban site configuration. In particular, small-scale fluid dynamics superposes to the atmospheric larger scale flow and turbulence and the dispersive processes strongly depend on the specific structure of the urban fabric. Advanced computational fluid dynamics (CFD) models are generally applied to simulate the flow structure and the C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
81
82
S. Trini Castelli et al.
pollutant diffusion around buildings, obstacles or in urban canyons. In recent years, we have been proposing an alternative approach, where the flow was simulated in different cases by using an atmospheric model – RAMS6.0 (Reisin et al., 2007). In this approach, a complete description of the atmospheric processes is ensured, since boundary layer, radiation and moist processes, together with the interaction between the surface and the soil, are included. Moreover, for further developments, it is possible to take advantage of the several capabilities offered by atmospheric models, like data assimilation and nudging. We use the latest version of RAMS6.0, where a Cartesian grid is implemented and the so called ADaptive Aperture method (Walko and Tremback, 2002) is used for defining the presence of buildings and dealing with arbitrarily steep topography, enabling simulation at very high resolution, in the order of metres. To harmonize with the CFD approach and accounting for CFD results, we implemented in RAMS6.0 not only a standard version of the k-İ turbulence closure model (Trini Castelli et al., 2001, 2005) but also its renormalization group (RNG k-İ) version (Reisin et al., 2007). Test-simulations of the flow and turbulence were performed using both closure schemes and some results are discussed here. To simulate the dispersion process, we are developing a modelling system which will be the microscale version of the regional off-line system RMS (RAMS-MIRS-SPRAY, see for instance Trini Castelli et al., 2003), hereafter called MicroRMS, interfacing RAMS6.0 with MicroSPRAY Lagrangian particle dispersion model (Tinarelli et al., 2007). The advantage of the off-line coupling lays in the independency of the meteorological and dispersion models. This enables the flexibility of considering different dispersion scenarios with the same meteorology. At present, our efforts are focused on the development of a micro-version of the interface code MIRS (Trini Castelli and Anfossi, 1997), calculating the surface and boundary layer’s parameters and the Lagrangian variables (variances of the velocity fluctuation, local velocity decorrelation time scales, etc.) and harmonizing the treatment of buildings between RAM6.0 and MicroSPRAY. Having such a preprocessor offers a number of options in calculating the boundary-layer and turbulence parameters (Physick and Trini Castelli, 2005). MicroSPRAY is especially devoted to simulate accidental gaseous releases at microscale. It considers different emission geometries and conditions, different micrometeorological situations and the possible presence of obstacles, such as buildings. New modules treating the physics of non-neutral gases are also under implementation in the model. MicroRMS is conceived so to provide the simulation of the flow and dispersion encompassing all relevant scales, synoptic, mesoscale and down to the urban microscale. In this work, we present the preliminary test of MicroRMS prototype, applied in the frame of the MUST exercise of COST732 Action (http://www.mi.unihamburg.de/cost732).
2. MUST Case in COST732 Action The following description of the MUST case is an extract from a document by courtesy of Dr. Bernd Leitl, who prepared it for the COST732 community. The
Development and Application of MicroRMS Modelling System to Simulate the Flow
83
Mock Urban Setting Test – MUST data set provides flow and dispersion data measured within an idealized urban roughness. The experimental setup is based on an extensive field test carried out on a test site of the US Army in the Great Basin Desert in 2001 (Yee and Biltoft, 2004). A total of 120 standard size shipping containers were set up in a nearly regular array of 10 by 12 obstacles, covering an area of around 200 m by 200 m. The containers were 12.2 m long, 2.54 m high and 2.42 m wide and formed an idealized roughness. The exact location and orientation of each of the individual obstacles were documented with sufficient accuracy. At the centre of the container array, a so-called VIP van was placed, serving as collection point for sampled wind and concentration data. The size of the VIP van differed significantly from the size of the surrounding containers. The terrain of the field site is characterized as ‘flat open terrain’, an ideal horizontally homogenous roughness formed by bushes and grass land with a height of approximately 0.5–1 m. Other orographical structures, like dunes, were assumed to have no significant effect on the approach flow conditions at the test site. The nearest significant mountains are located 12 and 24 km far from the experimental field. The terrain slope is documented to be 0.5 m per km, rising to the south. Wind tunnel (WT) measurements within a scaled model of the MUST configuration were carried out for instance by Bezpalcova and Harms (2005). The laboratory data represent the reference dataset used in the COST732 exercise.
3. Configuration of the Modelling System and Simulations Since we are primarily interested in testing the applicability of our modified version of MicroRMS in real terrain, in this work we kept the geometry and the initial conditions as for the field test case. The preliminary runs were performed for the 0° and –45° (that in the real field was –41°) inflow cases, referring to the available WT dataset sketched in Figure 1, since at present the real field MUST data are under processing and not yet available.
Fig. 1 Sketch of the geometry and of the different flow direction in the MUST WT experiment (Courtesy of Dr. B. Leitl)
84
S. Trini Castelli et al.
The measured field mean wind profile for the inflow is available at three levels up to 16 m and the data are processed on the basis of measurements at the south tower, during the days of September 25 and 26, 2001. The observed values are given as normalized over the wind at a reference height of 8 m. Since RAMS needs an input profile higher than the top of the domain, we extrapolated logarithmically a profile from the observed data available. In Figure 2 we plot the profile used as input in RAMS for both 0° and –45° cases, the field observed data and the inflow data used in the wind tunnel experiment, to highlight the differences. The wind tunnel values are normalized with respect to the wind at a reference height of 7.29 m. The same initial speed profile was used for the 0° and for the –45° simulations. In the comparison with the WT data the U-component of the measured wind velocities is always aligned with the mean approach flow wind direction. Differently from the approach in CFD simulations (see, for instance, Milliez and Carissimo, 2007), no initial turbulent kinetic energy (TKE) profile was input and RAMS develops its own TKE field starting from a minimum initial threshold. RAMS simulation domain extends 420 m in the longitudinal dimension with a grid size of 0.6 m and 320 m in the latitudinal direction with a grid size of 1 m. In the vertical there are 35 levels with a resolution of 0.2 m up to a height of 3.3 m, then stretched up to a total height of 25 m. The 120 containers’ locations and sizes were set according to the data provided for the MUST case.
Fig. 2 Input wind profile in RAMS6.0 simulation (solid line), against observed field MUST data (diamonds) and approach flow profile in wind tunnel experiment (solid + dot line)
We recall that RAMS is not built to produce steady-state conditions. However, we set the initial boundary condition so to approach as close as possible a steady state solution and we verified that a quasi-steady flow was reached after 4 minutes of simulation. The time step during the whole simulation was 4 •10-3 s. In these simulations we run RAMS without using the modified boundary conditions at the buildings that we implemented, for which in a so called ‘influence region’ around the buildings a logarithmic interpolation between the values at the building face and the values of the variable in the ambient atmosphere near the buildings is imposed (Reisin et al., 2007).
Development and Application of MicroRMS Modelling System to Simulate the Flow
85
4. Results and Discussion The main goal of these preliminary runs with MicroRMS prototype is to verify the possibility of using this modelling system in such high resolutions and in a configuration typical to CFD studies. The present results need thus to be considered as indicative and qualitative, while a more quantitative analysis will be conducted running final simulations against the real field data. In Figure 3 the streamlines of the wind field produced by RAMS in the –45° case are plotted. The structure of the simulated flow appears to be plausible and its consistency can be verified looking at the variables’ profiles in some points of the domain. Fig. 3 Detail of the streamlines of the horizontal mean flow field at 1m level by RAMS6.0
In Figure 4 a comparison among two RAMS simulations, using k-İ or RNG-k-İ closure models, and the WT measured data in a point in the central part of the building arrays, located between two buildings, is presented. The simulated pattern of the wind profile well captures the values observed in the wind tunnel for both closures, while the TKE shows a maximum at a higher position than the observed one. The k-İ closure produces higher values for the TKE, while the RNG-k-İ tends to underestimate the observed range. The difference between the two closures, k-İ producing higher TKE values than RNG-k-İ, is consistent with many results from literature. In this specific case, with such highly complex configuration of buildings, apparently k-İ closure is better reproducing the turbulent field.
86
S. Trini Castelli et al.
Fig. 4 Case 0°. Wind profile in RAMS6.0 simulation at (x,y) = (–45.201 m, 3.577 m) (solid lines) using k-İ and RNG k-İ closures, against measured profiles in wind tunnel experiment at (x,y) = (–44.925 m, 4.05 m) (dots)
Fig. 5 Case –45°. Wind profile in RAMS6.0 simulation at (x,y) = (–26.100 m, –5.50 m) (solid lines) using RNG k-İ closure, against measured profiles in wind tunnel experiment at (x,y) = (–25.875 m, –5.25 m) (dots)
In Figure 5 an analogous comparison is proposed for the –45° case run with the RNG-k-İ model, in a different point than for the 0° case, but always located in the central part of the domain. The simulated wind profile shows higher values than the observed WT data, in particular in the middle heights. The difference may be related to the initial wind profile input in the model, which seems to be better representative for the 0° case than for the –45° case. The TKE comparison confirms the results found for the 0° case. The meteorological fields produced by RAMS were then processed through the prototype micro-version of MIRS so to make the grid structures and the assimilation of the building data compatible between RAMS and MicroSPRAY. In this preliminary test, since only the TKE was made available from RAMS, we made the assumption of isotropy for turbulence and the standard deviation Vi of the wind
Development and Application of MicroRMS Modelling System to Simulate the Flow
87
velocity fluctuations were set as V u V v V w 2k 3 , where k is the TKE. The Lagrangian time scales were then calculated using also the TKE dissipation İ, as T L 2V 2 C 0 H , where C0 is the Lagrangian structure function constant that here was assigned as C0 =4. For this first run, aimed at testing the functionality of the models’ chain, we referred to the MUST experiment named 2681829 (–41°), during which the atmosphere was characterized by a neutral stratification (Monin-Obukhov length L = 28,000 m), the source was located in (x,y) = (–77.46 m, 67.47 m) at 1.8 m height and the emission was continuous, with a release rate of 225 l/minute, lasting 15 minutes. As an example, in Figure 6 a picture of the 3D particle plume is reported, showing the interaction of the plume with the building array. The direction of the plume and its spread appear to be consistent with the expectations. To present the result on the concentration fields, we are repeating the simulation changing the inflow as for the real field case, which is –41°. Fig. 6 Case –45°. MicroSPRAY simulated plume dispersion, case 2681829
5. Conclusions First results using the microscale version of the regional off-line modelling system RMS (RAMS-MIRS-SPRAY) – MicroRMS, were presented. These encouraging results proved the capabilities of MicroRMS to simulate pollutant dispersion in complex urban configurations as in the MUST experiment. At present a new version of the interface parameterisation code, MIRS4.0, is under development, to harmonize the treating of buildings and obstacles between RAMS and MicroSPRAY models and to include parameterisations for the boundary layer and turbulence suited to the urban scale. Further investigations, related to the effect of the buildings’ boundary conditions on the flow, the turbulence and the dispersion and to the possible deflection of the plume induced by the buildings, are planned within the frame of the MUST real field experiment and of the COST732 Action activity. Acknowledgments The authors like to thank Professor M. Schatzmann, Dr. B. Leitl, Dr. J. Franke and all the COST732 community for the precious cooperation
88
S. Trini Castelli et al.
and for making available the MUST WT data and the information about MUST field experiment.
References Bezpalcova K, Harms F (2005) EWTL Data Report/Part I. Summarized Test Description Mock Urban Setting Test. Environmental Wind Tunnel Laboratory, Center for Marine and Atmospheric Research, University of Hamburg Milliez M, Carissimo B (2007) numerical simulations of pollutant dispersion in an idalized urban area, for different meteorological conditions, Bound.-Layer Meteorol. 122, 321–342 Physick W, Trini Castelli S (2005) Lagrangian Particle Models. Link with meteorological models, Section 11.2.5. in ‘Air Quality Modelling. Theories, Computational Methods and Available Databases and Software’, vol II – Advanced Topics, Zannetti P Ed., pp. 116–118, EnviroComp Institute and Air & Waste Management Association, Pittsburgh, USA. Reisin T, Altaratz Stollar O, Trini Castelli S (2007) Numerical simulations of microscale urban flow using the RAMS model. Developments in Environmental Science, Vol. 6, 32–44, Borrego C and Renner E Eds., Elsevier, Amsterdam, NL. Tinarelli G, Brusasca G, Oldrini O, Anfossi D, Trini Castelli S, Moussafir J (2007) Micro Swift-Spray (MSS), a new modelling system for the simulation of dispersion at microscale. Air Pollution Modelling and its Applications XVII, 449–458, Borrego C and Norman A Eds., Springer, New York, USA. Trini Castelli S, Anfossi D (1997) Intercomparison of 3D turbulence parameterisations as input to 3D dispersion Lagrangian particle models in complex terrain. Il Nuovo Cimento, 20C(3), 287–313. Trini Castelli S, Ferrero E, Anfossi D (2001) Turbulence closures in neutral boundary layers over complex terrain. Bound.-Layer Meteorol., 100, 405–419. Trini Castelli S, Anfossi D, Ferrero E (2003) Evaluation of the environmental impact of two different heating scenarios in urban area. Int. J. Environ. Pollut., 20, 207–217. Trini Castelli S, Ferrero E, Anfossi D, Ohba R (2005) Turbulence closure models and their application in RAMS. Environ. Fluid Mech., 5, 169–192. Walko R, Tremback C (2002) The Adaptive Aperture (ADAP) Coordinate. 5th RAMS Workshop and Related Applications, Santorini, Greece. Yee E, Biltoft CA (2004) Concentration fluctuation measurements in a plume dispersing through a regular array of obstacles. Bound.-Layer Meteorol. 111, 363–415.
Discussion A. Venkatram: The model underestimates the wind speed and overestimates the TKE. You concluded that you were “happy” with the results? Why? What is the significance of the 30s time scale in the Lagrangian
Development and Application of MicroRMS Modelling System to Simulate the Flow
89
particle model? The predictions are independent of the averaging time? Trini Castelli: These preliminary simulations were mainly aimed at verifying the possibility of using RAMS model for simulations of complex configuration with buildings, at very high resolution, and to check the micro version of our RMS modelling system. Therefore, we were “happy”, because this preliminary test was successful and results were reliable: we are not aware of other works performed with RAMS at this kind of resolution in such complex condition. We tested different turbulence closures at the 18 available measuring points and while in some cases we had underestimation of velocity and overestimation of TKE, in some other cases the results were good, alternately for the different closures. Surely we need to deeper investigate the reason why, for instance, k-İ closure works well in the upflow part of the domain while RNG-k-İ gives much better results in the downflow part of it. Since this is a steady-state case, the predictions of concentration are actually independent on the averaging time. The 30 s time scale for averaging was used to collect a sufficient number of particles to obtain stable concentration fields. Since the steadiness was reached after about 60 s, after this time interval the time scale does not affect concentration fields anymore, and it is used just to smooth the statistical fluctuations due to discretization in particles. Furthermore, this time scale was used for graphical reasons, to produce the plume-dynamics and concentration field animation since the initial time. S. Hanna: Your dispersion results for the MUST case with the mean wind at 45º angle to the obstacle orientation suggested that the plume direction was the same for the particles at heights below and above the obstacle height. The observation showed a significant shear. It will be interesting to see the results of your comparison of the model simulations to the observations. Trini Castelli: We recall that the height of the emission source in the case we considered was 1.8 m and the height of the buildings is 2.54 m. The 3D animation shown does not allow appreciating the shear of the plume because particles at any vertical level are plotted together, but when ground level concentration contours are plotted a deviation from the initial direction of the plume is clearly registered, indicating the shear of the plume below the obstacles’ height. However, further investigation is needed and a comparison between measured and predicted concentrations is under process.
1.3 Development of a Lagrangian Particle Model for Dense Gas Dispersion in Urban Environment G. Tinarelli, D. Anfossi, S. Trini Castelli, A. Albergel, F. Ganci, G. Belfiore and J. Moussafir
Abstract A new version of the Lagrangian particle model MicroSpray is proposed. It simulates the dense gas dispersion in situations characterized by the presence of buildings, other obstacles, complex terrain, and possible occurrence of low wind speed conditions. Its performances are compared to an atmospheric CFD model output and to a field experiment (Thorney Island). Keywords Dense gas dispersion, lagrangian particle model, tracer experiment 1. Introduction Accidental release and dispersion of hazardous material may cause severe environmental problems. Thus, correctly simulating the distribution of these hazardous substances is important. This is mostly accomplished by empirical models or, in some specific cases, by computational fluid dynamics (CFD) models. Another way, here proposed, is offered by Lagrangian particle dispersion (LPD) models. The LPD approach is a compromise between the complexity and CPU time demanding of CFD models and the simpler integral models. Thus, in this work, a new version of the LPD MicroSpray model, especially oriented to deal with dense gas dispersion in urban environment, is described. Then the comparison between the MicroSpray simulations and those of the CFD Mercure in two academic flat terrain cases is shown. Finally, some preliminary comparisons of MicroSpray predictions against the tracer Thorney Island field experiment No 8 are presented.
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
28
Development of a Lagrangian Particle Model for Dense Gas Dispersion in Urban
29
2. Brief Outline of MicroSpray and Mercure Models
2.1. MicroSpray MicroSpray is part of the model system MSS that includes MicroSWIFT and MicroSpray. MicroSWIFT is an analytically modified mass consistent interpolator over complex terrain. Given topography, meteorological data and buildings, a mass consistent 3D wind field is generated (Moussafir et al., 2004; Brusasca et al., 2005). It may also prescribe diagnostic turbulence parameters to be considered by MicroSpray inside the flow zones modified by obstacles. MicroSPRAY is a LPD model directly derived from SPRAY code, which is based on a 3D form of the Langevin equation for the random velocity (Tinarelli et al., 1994, 2000), and is able to take into account the presence of obstacles. Micro-Swift takes obstacles or buildings into account by setting as impermeable some of the cells of the terrain following grid where meteorological fields are defined. MicroSpray has been extended to deal with dense gas dispersion. We recall that an emitted cloud of hazardous material initially denser than the ambient air, begins to disperse under the action of its own buoyancy and momentum (horizontal, vertical or oblique in any direction). Then, its excess of density reduces as ambient air is entrained. Finally, at some distance downwind, transition to passive dispersion takes place. An important issue is the spread at the ground due to gravity. These effects are simulated into MicroSpray by implementing new algorithms. Concerning the initial phase, the following five conservation equations are integrated for each particle at each time step, based on Glendening et al. (1984): º d ªUp (1) Mass us b2 » E us « dt ¬ U a ¼ Energy
d us b 2 B dt
>
@
Up 2 N us wp b 2 Ua
(2)
Vertical momentum
d ª Up 2 º « u s wp b » dt ¬ Ua ¼
x horizontal momentum
d dt
ªUp º us b2 u p » « U ¬ a ¼
E us ua
(4)
y horizontal momentum
d dt
ªUp º us b2 v p » « U ¬ a ¼
E u s va
(5)
B b 2 us
(3)
2 2 2 where: u s u p 2 v p 2 w p 2 , B g U a U p U a , Ua u a v a wa , u e >D 1 u s D 2 Ua @ and E 2 b u e ; a , p refer to air and plume, respectively, B is the buoyancy, E represents the entrainment rate, b the plume radius and ue is
G. Tinarelli et al.
30
the entrainment velocity. The first three equations were derived by Hurley and Manins (1995) and inserted into the TAPM model (Hurley, 2005). Regarding the gravity spreading, we remind that when a dense plume reaches the ground a horizontal momentum is generated by its weight, which thus tends to spread the plume. We simulate this last process by an empirical method. The module of the horizontal velocity, U g , is computed as (Eidsvik, 1980): (6) U g D1 B H where D 1 2 and H is the mean height of the column above each particle. The two horizontal velocity components U gs , Vgs , due to the gravity spread are: U gs U g cos J and V gs U g sin J .
2.2. Mercure The Mercure model (Carissimo et al., 1997) is the atmospheric adaptation of the CFD code ESTET developed by Electricité. de France (EDF), commonly used for industrial CFD applications at EDF R&D. MERCURE code has undergone extensive validation (see: Riou, 1987; Duijm and Carissimo, 2001; Moon and Albergel, 1997). Relevant aspects of the code include: 3D flow simulation, influence of terrain and obstacles, multiple fluids and full non-hydrostatic formulation. MERCURE solves the classic Navier-Stokes equations system with adaptations for multiple fluids and for passive scalar tracer variables. A conservation relation for thermodynamic energy (enthalpy or virtual potential temperature) is optionally solved. Solving the thermal energy equation implies that thermal buoyancy (or dense) effects are included in the solution. Turbulence closure is by means of supplementary equations for the conservation of turbulent kinetic energy and dissipation using the k-H model. The conservation equations are discretized using a combination of finite difference and finite-volume methods solving separately each type of operator. As default, the inlet boundaries are Dirichlet for all parameters and the outlet boundaries are zero gradient for all parameters. Important aspects of the MERCURE setup for this study include: x Ideal gas equation of state x Boussinesq approximation is used, implying that density variations only affect the flow through buoyancy (or dense) terms x Gravity is the only retained volume force (Coriolis effects are ignored) x Thermal forcing due to radiative flux divergence is negligible
Development of a Lagrangian Particle Model for Dense Gas Dispersion in U rban
31
3. MicroSpray Validation
3.1. Comparison with MERCURE Simulations with MERCURE have been performed in two different flow conditions. The first one refers to a mean wind velocity at 10 m (1.5 m s-1) and the second one to a higher wind at 10 m (5 m s-1). In both simulations, a neutral turbulence has been considered considering a logarithmic wind profile, horizontally homogeneous. In order to verify that the two models behave similarly in a base case, both MERCURE and MicroSpray have firstly considered a continuous emission located 10 m above the ground without dense gas effects. Then the two models simulated an emission two times denser than air. Figure 1 shows ground level concentrations (glc) obtained in the base case by the two models. Maximum values are almost identical and also the overall pattern of the impact at ground is very similar. V=5.0m/s U Air
max = 1.59E-2 kg/kg
U Air
max = 1.64E-2 kg/kg
MERCURE
Spray
V=1.5m/s U Air
U Air
max = 4.32E-3 kg/kg
max = 4.66E-3 kg/kg
Fig. 1 glc obtained by MERCURE (above) and MicroSpray (below) in lower (left) and higher (right) wind speed in the base case (no dense plume)
Figure 2 shows instead the comparison of the dense emission case. In this condition, an emission from a stack having a diameter of 2.17 m and a vertical exit velocity of 1.14 m s-1 has been considered. The source is located at (0, 0, 10 m) and the plume is followed until about 400 m downwind the source.
G. Tinarelli et al.
32 V=5.0m/s U = 2 * UAir
U = 2 * U Air
max = 0.21 kg/kg
max = 0.225 kg/kg
MERCURE
V=1.5m/s U 2 * U Air
max = 0.005 kg/kg
Spray
U = 2 * U Air
max = 0.0078 kg/kg
Fig. 2 glc obtained by MERCURE (above) and MicroSpray (below) in lower (left) and higher (right) wind speed in the dense gas case
In the lower wind speed case, MERCURE shows an evident splitting of the plume at ground, and a large horizontal spread due to the gravity effects. MicroSpray shows qualitatively a similar result, even if less pronounced. Both splitting and plume spread are present and the maximum glc is correctly captured. In the higher wind speed case, both models show a less pronounced spread effect at ground, due to the higher ventilation causing a more efficient entrainment. Maxima glc are still comparable and impact area seems closer to the source in MERCURE simulation.
3.2. Comparison with Thorney Island Exp. 8 The main information on this experiment, needed for this work, were found in two Data Set Reports: Rediphem (Nielsen and Ott, 1996) and MDA (Hanna et al., 1991). A mixture of Freon-12 and Nitrogen (3,958 kg) was instantaneously emitted from a cylinder (diameter = 14 m, height = 13 m), without any initial momentum. 42 samplers, located at different heights (0.4, 2.4, 4.4 and 6.4 m) in the range 70–500 m downwind the source, collected the emitted tracer for 660 s. The initial tracer concentration was 1 mol/mol. Wind speed was 2.4 m s-1 at 10 m and the wind heading was about 18 degrees to the left of the array centerline. Other important data were: friction velocity u* = 0.126 ms-1 and roughness length z0 = 0.012 m. The stability was Pasquill category D. No turbulence data were given. The relative emission density, Ue/Ua, where Ue and Ua, are the emission and ambient densities, was equal to 1.63. A computation domain of 200 × 800 × 200 m was considered. MicroSwift had horizontal grid spacing of 2 m and a stretched grid in the vertical. No obstacles
Development of a Lagrangian Particle Model for Dense Gas Dispersion in U rban
33
have been included in the MicroSwift simulation, since no exact information of their presence and dimensions were available. A logarithmic wind profile, horizontally homogeneous, was reconstructed on the basis of the above reported values of u* and z0, that were also used to reconstruct the turbulence fields. Twenty thousand particles were released at t = 0 uniformly distributed within the source cylinder centered at × = 200 m and y = 0 m and then their trajectories were calculated.
Fig. 3 Plant view of the smoke emitted after 1s (left) and 10s (right) obtained during the experiment (upper part) and glc obtained by MicroSpray (lower part)
Finally, concentration at sampler locations was computed. Figure 3 shows a qualitative comparison between simulation and experiment. In the upper part, two photographs show a plant view of the behaviour of the emitted puff 1s and 10s after the emission. The smoke is suddenly moved towards the ground, while the horizontal spread takes place emptying the internal part of the column and giving to the puff a
G. Tinarelli et al.
34
circular shape. This circular puff is then transported along the mean flow direction. MicroSpray gives a similar behaviour. Ground level concentration field patterns show a circular form and the horizontal spread is similar to the experimental one. Also the movement of the puff along the mean direction is correctly captured. Table 1 Statistical indexes – Thorney Island Exp. 8 preliminary simulation. CC
NMSE
FB
MG
VG
FA2
FA5
0.69
2.38
0.65
1.40
1.02
0.55
0.86
Table 1 shows some statistical indexes: correlation coefficient (CC), fractional bias (FB), geometric mean bias (MG) and variance (VG), factor of 2 (FA2) and 5 (FA5). We show both FB, NMSE and MG,VG even if, since observed concentrations vary over three orders of magnitude, the logarithmic indexes are more appropriate (Hanna et al., 1991). With reference to this, it is known that a “perfect” simulation would have MG = VG = 1. Table 1 shows that VG is rather good whereas MG (and, obviously, FB as well) is less satisfactory, indicating an overall underestimation. The general behaviour of the tracer experiment is correctly captured (VG = 1.02 and the values of CC and FA2 are satisfactory too). It is worthwhile to recall that no information on the turbulence characteristics of the experiment and of the wind profile were available (only knowing the value of the upwind mean wind at 10 m).
4. Conclusion In this paper, we have presented a new version of the LPD model MicroSpray devoted to simulate the dense gas dispersion. We have compared its prediction both with the CFD model Mercure and with a tracer experiment (Thorney Island Exp.8). Some preliminary results obtained above presented in flat terrain suggest that MicroSpray is able to perform correct simulations of dense gas dispersion both in academic cases and real field situations.
References Brusasca G, Tinarelli G, Oldrini O, Anfossi D, Trini S Castelli, Moussafir J (2005) Micro-Swift-Spray (MSS) a new modelling system for the simulation of dispersion at microscale. General description and validation. In: Air Pollution Modelling and its Applications XVII, C Borrego and D Steyn (eds.), Kluwer/Plenum, New York, in press Carissimo B, Dupont E, Musson-Genon L, Marchand O (1997) Note de Principe du Code MERCURE. Version 3.1, Electricité de France, EDF HE-33/97/001, EDF publications, France
Development of a Lagrangian Particle Model for Dense Gas Dispersion in U rban
35
Duijm NJ, Carissimo B (2001) Evaluation methodologies for dense gas dispersion models. In: The handbook of hazardous materials spills technology. M Fingas (ed.), McGraw-Hill, New York Eidsvik KJ (1980) A model for heavy gas dispersion in the atmosphere. Atmos. Environ., 14, 767–777 Glendening, JW, Businger, JA, Farber, RJ (1984) Improving plume rise prediction accuracy for stable atmospheres with complex vertical structure. J. Air Pollut. Control Assoc., 34, 1128–1133 Hanna SR, Strimaitis DG, Chang JC (1991) Hazard response modeling uncertainty (a quantitative method). Vol. 2, Evaluation of commonly used hazardous gas dispersion models. Sigma Research Corporation for AFESC, Tyndall AFB, FL, and API, Report Nos. 4545, 4546, and 4547, 338 pp Hurley PJ, Manins PC (1995) Plume rise and enhanced dispersion in LADM. ECRU Technical Note No. 4, CSIRO Division of Atmospheric Research, Australia Hurley PJ (2005) The Air Pollution Model (TAPM) Version 3. Part 1: Technical Description. CSIRO Atmospheric Research Technical Paper No. 71 Moon D, Albergel A (1997) The use of the MERCURE CFD code to deal with an air pollution problem due to building wake effects Journal of Wind Engineering and Industrial Aerodynamics, Volumes 67–68, April–June 1997, pp 781–791, Computational Wind Engineering Moussafir J, Oldrini O, Tinarelli G, Sontowski J, Dougherty C (2004) A new operational approach to deal with dispersion around obstacles: the MSS (MicroSwift-Spray) software suite, 9th International Conference on Harmonisation within Atmospheric Dispersion Modelling for Regulatory Purposes Garmisch 1–4 June 2004 Nielsen M, Ott S (1996) A collection of data from dense gas experiments, Riso Report 845(EN) Riou Y (1987) Comparison between the MERCURE-GL code calculations, wind tunnel measurements and. Thorney lsland field trials, J. Hazardous Mater. 16, 247–265 Tinarelli G, Anfossi D, Brusasca G, Ferrero E, Giostra U, Morselli MG, Moussafir J, Tampieri F, Trombetti F (1994) Lagrangian particle simulation of tracer dispersion in the lee of a schematic two-dimensional hill. J. Appl. Meteorol., 33 (6), 744–756 Tinarelli G, Anfossi D, Bider M, Ferrero E, Trini Castelli S (2000) A new high performance version of the Lagrangian particle dispersion model SPRAY, some case studies. In: Air Pollution Modelling and its Applications XIII, SE Gryning and E Batchvarova (eds.), Kluwer/Plenum, New York, pp 499–507
36
G. Tinarelli et al.
Discussion P. Builtjes: Is my impression correct that MicroSpray is better than CFDMercure, and if the models give similar results, why do you use MicroSpray and not Mercure? D. Anfossi: I would not say that MicroSpray is better than CFD-Mercure, but I certainly say that that the two models give similar results. The advantage of using MicroSpray is related to the computing time: a few minutes against many hours. B. Fisher: The treatment of the boundary condition at the surface is difficult in a Lagrangian particle model. How is this boundary condition treated within the CFD code Mercure? D. Anfossi: In MERCURE the boundary conditions are treated as follows. For the wind: wall law at ground (neutral case here); for the other variables: Neumann condition (null gradient)
1.7 Evaluation of the Hazard Prediction and Assessment Capability (HPAC) Model with the Oklahoma City Joint Urban 2003 (JU2003) Tracer Observations Steven Hanna, Joseph Chang, John White and James Bowers
Abstract Results are presented of an evaluation of the Hazard Prediction and Assessment Capability (HPAC) suite of models in an urban environment using data from the Joint Urban 2003 (JU2003) Field Experiment in Oklahoma City (OKC). JU2003 included 29 separate SF6 tracer continuous releases (of 30-minute duration) on ten days from a point source near ground level in or immediately upwind of the built-up downtown area. The ten Intensive Operating Periods (IOPs) consisted of six daytime periods and four nighttime periods. Tracer was sampled at over 100 locations at distances ranging from 0.1 to 4 km from the source. The current study tests two alternate urban configurations of HPAC and four optional meteorological inputs. The two HPAC configurations were the Urban Dispersion Model (UDM) and the Urban Canopy (UC) options. The four meteorological data options were basic default National Weather Service (NWS) data, a single averaged wind, a single upwind anemometer and radiosonde, and detailed three-dimensional winds from a meteorological model, MM5. The evaluations of the maximum 30-minute averaged concentrations on six downwind distance arcs are summarized in this paper. In most cases, the MM5 meteorological inputs yielded the best HPAC results. Also, in general, the UC urban option produced higher concentrations, by about a factor of two, than the UDM urban option. The UDM urban option performed better during the night IOPs and the UC urban option performed better during the day IOPs. There is an obvious day-night difference in the model biases, with most options overpredicting during the night and most options underpredicting during the day, suggesting that they are overstating the relatively-small observed day-night difference in near-ground urban stability and in tracer concentrations. Keywords Air quality model evaluation, HPAC model, JU2003 field experiment, urban dispersion models
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
63
64
S. Hanna et al.
1. Introduction and Approach This paper presents the methods and some of the results of an evaluation of the Hazard Assessment and Prediction Capability (HPAC) transport and dispersion modeling system (DTRA, 2004; Sykes et al., 2007) using the Joint Urban 2003 (JU2003) continuous release tracer data. Allwine et al. (2004) summarize the JU2003 field experiment, which took place in Oklahoma City during July 2003. Clawson et al. (2005) describe the JU2003 SF6 observations. The ten IOPs had three different source release locations: Botanical Garden (upwind of the downtown area) for IOPs 03 through 07; Westin Hotel (in the built-up downtown area) for IOPs 01, 02, and 08; and Park Avenue (in a street canyon in the downtown area) for IOPs 09 and 10. IOPs 01–06 took place during the day and IOPs 07–10 took place during the night. During each of the ten JU2003 IOPs, three continuous releases of SF6 of 30minute duration were made at 2-hour intervals, except that only two releases were made during IOP01. Hanna et al. (2007) present the results of a similarity analysis of the JU2003 wind, turbulence, and continuous-release concentration data. Instantaneous (puff) releases of SF6 also took place during each IOP, but these releases are not discussed or analyzed here (see Zhou and Hanna, 2007, for the results of an analysis of the along-wind diffusion of the puffs). Samplers were set out on a rectilinear grid in the built-up downtown area at distances less than 1 km from the source. Samplers were also set out on three concentric arcs, covering an angular range of about 120° at distances 1, 2, and 4 km to the north of the downtown area. The specific sampler locations changed from one IOP to the next, depending on the release location and the wind direction. Figure 1 presents a map of the sampler locations. The averaging time for the samplers was adjustable and was generally set to 5, 15, or 30 minutes. The analysis in this paper uses 30-minute averaged concentrations, C, normalized by the source emission rate, Q. In the downtown area, where the samplers were on a rectangular grid, the authors subjectively assigned each sampler to one of three effective “arc” distances: 0.30, 0.62, and 0.85 km. The data from the sampling arcs at 1, 2, and 4 km were also used. We consider only the street-level samplers. We have found that the first two trials of the daytime IOP05 are more representative of nighttime stable conditions, though the releases took place during the early morning. The IOP05 trials with relatively high observed C/Q were caused by relatively low mixing depths (less than 200 m). Consequently, IOP05 trials 1 and 2 are removed from some parts of the analysis of the daytime IOPs or are included in the nighttime category. Also, in the evaluations reported here, data from IOPs 01 and 02 are not included because of problems with setting up the input data. Previous analyses (see Hanna et al., 2007) of the JU2003 observed 30-minute averaged maximum concentrations on several downwind distance arcs showed that values of C/Q were generally about three times higher during the night IOPs than during the day IOPs. We hypothesized that this relatively small difference is due to the relatively small differences in near-ground stability, ranging from slightly unstable during the day to slightly stable during the night.
Evaluation of the Hazard Prediction and Assessment Capability (HPAC) Model
65
Fig. 1 Map of SF6 sampler locations during JU2003. The samplers that were used in the current paper are marked by solid triangles, and are the Field Research Division - Air Research Laboratory (FRD-ARL) continuous samplers described by Clawson et al. (2005)
This paper considers two urban HPAC options (Urban Dispersion Model (UDM) and Urban Canopy (UC)). The paper also considers four meteorological input options: BDF - Basic National Weather Service (NWS) default MED - Mesoscale Meteorological Model-Version 5 (MM5) MEDOC outputs AVG - Average wind speed and direction from all anem (Hanna et al., 2007) UPWND - Wind speed and direction from DPG Portable Weather Instrumentation Data System (PWIDS) #15 on the Post Office, located just upwind of the downtown area, with observed mixing heights based on the upwind Pacific Northwest National Laboratory (PNNL) radiosonde data.
2. HPAC Options and Inputs HPAC was used to calculate SF6 tracer concentrations at the locations of the bag samplers within the central business district (CBD) and the three outer arcs (at 1, 2, and 4 km). As discussed above, the model runs used two HPAC urban model
S. Hanna et al.
66
configurations and four different sets of meteorological data. Model source parameters included the initial Gaussian spread ıy and ız (set to 0.232 m), the tracer release duration (30 minutes), the release height (1.9 m), and the mass dissemination rate, Q (obtained from the JU2003 database). Cloud cover was set based on the standard hourly NWS observations. Surface moisture was set to “normal” and gridded terrain elevation was used except for the HPAC model runs that used the MM5 gridded data (the MED meteorological option), which already contained the terrain information. With two exceptions, boundary layer calculations were set to default. The gridded MM5 MEDOC output files include the mesoscale model’s boundary surface heat flux and boundary layer depth estimates. In the case of the UPWND meteorological option, the boundary layer depth was estimated by the authors from PNNL radiosonde soundings for each continuous release. The PNNL radiosonde data set was selected because of the close proximity of the balloon release site to PWIDS #15 and its location upwind of the CBD. Default values were used for other HPAC inputs such as the Bowen ratio (2) albedo (0.16), canopy height (30 m, the value used for the Urban 2000 study), and canopy flow index (2). All HPAC runs except those using the MED meteorological option used HPAC’s SWIFT diagnostic wind field model to derive mass-consistent wind fields. The SWIFT default parameters were used in these runs with the exception of the wind field update interval, which was set to 1 hour for the BDF and AVG meteorological options, and to 10 minutes for the UPWND meteorological option. Other HPAC input parameters were: x x x
Conditional averaging time set to 30 minutes No large scale variability Sampling height set to 1.5 m (the height of the bag samplers)
The BOOT statistical model evaluation software (Chang and Hanna, 2004) was used to compare predicted and observed arc maximum 30-minute averaged C/Q paired and unpaired in space and time. The limited results presented here focus on the maximum C/Q on a given downwind arc. The following performance measures were used, where we let X = C/Q: Fractional Bias FB = 2<Xo–Xp>/(<Xo> + <Xp>) Normalized Mean Square Error NMSE = <(Xo–Xp)2>/(<Xo> <Xp>) Fraction of Xp within a factor of two of Xo (FAC2) Geometric Mean MG = exp(
–) Geometric Variance VG = exp (<(lnXo–lnXp)2> Subscripts p and o refer to predicted and observed, and the symbol < > represents an average. Residual plots were also used in the evaluation, where the ratio of Xp/Xo was plotted versus downwind distance, x. The five lines on the box plot represent the 98th, 84th, 50th, 16nd, and 2nd percentiles, respectively, for the group of data considered.
Evaluation of the Hazard Prediction and Assessment Capability (HPAC) Model
67
3. Results The BOOT software was used to calculate the performance measures and generate the residual plots as defined above for the eight model configurations (two urban model options and four alternate meteorological inputs). Table 1 presents the performance measures, calculated separately for the day and night IOPs. Figure 2 shows the residual plots for urban options UDM and UC coupled with the MED meteorological option. It was necessary to split the analysis into day and night because the first comparisons showed that most of the eight model configurations had a tendency to overpredict at night and underpredict during the day. Warner et al. (2007) found similar biases in their evaluations of HPAC with the JU2003 data for a wider variety of urban model options and meteorological inputs. The MM5 MEDOC inputs were better able to account for the low mixing depths than the other meteorological options. Some basic conclusions from the performance measures and residual plots are: Table 1 Performance measures for evaluations of HPAC with the JU2003 data. See text for definitions of performance measures, meteorological options, and model options. Note that when FB = –2/3, there is a mean factor of two overprediction, and when FB = +2/3, there is a mean factor of two underprediction. FB = 0 and MG = 1 indicate an unbiased model. IOP03, IOP04, & IOP06 (daytime only, excluding IOP05) Met Options Model Options FB NMSE FAC2 BDF UDM 0.88 3.4 19% BDF UC -0.29 1.6 59% MED UDM 0.91 3.9 50% MED UC -0.38 1.9 70% AVG UDM 1.02 4.9 46% AVG UC 0.52 1.4 63% UPWND UDM 0.44 1.2 74% UPWND UC -0.89 3.9 44%
MG 2.28 1.41 1.81 0.89 2.15 1.59 1.04 0.47
IOP07-IOP10 (nighttime only) Met Options Model Options BDF UDM BDF UC MED UDM MED UC AVG UDM AVG UC UPWND UDM UPWND UC
MG 1.58 1.45 1.02 0.36 0.32 0.11 0.38 0.17
x
x
FB -1.15 -1.56 -0.47 -1.19 -1.35 -1.74 -1.30 -1.62
NMSE 20.1 34.0 7.2 13.0 31.0 85.2 22.8 38.0
FAC2 32% 5% 49% 25% 28% 3% 29% 5%
VG 3 2 2.1 1.6 2.4 1.6 1.5 2.3
VG 1.3E+07 1.3E+10
3.4 22.6 9.8 385.0 6.9 116.0
For most model and meteorological input combinations, there is a tendency to overpredict by an approximate factor of 3 or 4 at night and underpredict by an approximate factor of 2 during the day. For the daytime IOPs (03, 04, and 06), MED-UC and UPWND-UDM tend to have the least bias, lowest scatter and highest FAC2; and little trend with x. MED-UDM has an underprediction tendency of about a factor of 3 or 4 at small x. UDM and UC simulations of C/Q tend to
68
S. Hanna et al.
agree for all meteorological options at x > 1 km (outside of the built-up area). For all meteorological options, the ratio of the concentrations predicted by UC to UDM is about 3 or 4 at small x (inside of the builtup area). UDM simulations are always lower than the observed values (by factor of 2 to 5) at small x for all meteorological options. For the nighttime IOPs (07, 08, 09, and 10), MED-UDM has the least bias, lowest scatter, and highest FAC2; and has little trend of the residuals with x. Unlike the daytime runs, UDM and UC simulations do not agree as well at large x. The same bias occurs at all x. For all meteorological options, the ratio of the concentrations predicted by UC to UDM is about 2 at all x. The AVG and UPWND meteorological options lead to large mean overpredictions of a factor of 3–10. Our HPAC evaluations to date and the evaluations reported by other groups (e.g., Warner et al., 2007) have confirmed that urban HPAC overpredicts during the night and underpredicts (by a smaller amount) during the day. This type of behavior suggests that the model may be using too broad of a diurnal range in stabilities. The analyses by Hanna et al. (2007) of sonic anemometer data (including calculations of surface heat fluxes and Monin Obukhov lengths L) suggest that the stability in the built-up downtown area of OKC is near neutral, and usually slightly unstable, throughout the diurnal cycle. This result can be attributed to the strong mechanical mixing due to the buildings and anthropogenic heat input. At night, the slightlyunstable near-surface urban boundary layer is capped at a height of about 200 m by a more stable layer representative of the upwind boundary layer. It is hypothesized that urban HPAC would do better if it used a more nearly-neutral stability parameterization throughout the diurnal cycle, which would ameliorate the nighttime overpredictions and daytime underpredictions. Currently, the HPAC meteorological preprocessor assumes that the sensible heat flux in the upwind area is also valid in the urban area. (When used with MM5 inputs, HPAC uses the heat fluxes computed by the mesoscale model.) Use of the upwind sensible heat flux in the urban area is probably reasonable during the day, but is not valid at night, which may explain the larger biases at night. Note that the HPAC meteorological preprocessor does modify the friction velocity u* based on the increased roughness in the urban area. The observed arc-maximum 30-minute average normalized concentration data from JU2003 indicate that, on average in the CBD (x < 1 km), the daytime C/Q’s are about a factor of 3 smaller than the nighttime C/Q’s. At 1 km < x < 4 km, the day-night difference in C/Q increases to about a factor of 8, due to the increasing nighttime stability in the suburbs as the plume travels out of the CBD. The factor of 3 difference in CBD C/Q is consistent with an assumption of slightly unstable conditions during the day and very slightly-stable conditions during the night in the CBD. If a dispersion model is going to do well with these data, it must be able to simulate the observed factor of 3 day-night difference in C/Q. But because most HPAC model options showed large overpredictions during the night and large underpredictions during the day, it is concluded that those options are overstating the day-night difference in near-ground urban stability.
Evaluation of the Hazard Prediction and Assessment Capability (HPAC) Model
69
Finally, we hesitate to overinterpret these results. Just because a model combination seems to do better than another, it could be because that model tends to over or underpredict with respect to the other models and may have a compensating error. Until additional analyses are carried out, such as investigation of the effective wind speeds being used by each meteorological input option, it is difficult to untangle the possible effects of different meteorological inputs and decide which is more realistic. For example, it could be that the model that is currently performing better is using an effective wind speed that is too low or too high (i.e., there are compensating errors).
Fig. 2 Residual plots (Xp/Xo vs x) for HPAC urban options UDM and UC with MM5 MEDOC meteorological inputs, for day (top) and night (bottom) IOPs. The significant lines on the box plots indicate, from bottom to top, the 2nd, 16th, 50th (i.e., the median), 84th, and 98th percentiles of the n Xp/Xo points at that arc distance
Acknowledgments This study was supported by the U.S. Defense Threat Reduction Agency, with Rick Fry as project manager.
70
S. Hanna et al.
References Allwine KJ, Leach M, Stockham L, Shinn J, Hosker R, Bowers J, Pace J (2004) Overview of Joint Urban 2003 – An atmospheric dispersion study in Oklahoma City. Preprints, Symposium on Planning, Nowcasting and Forecasting in the Urban Zone. American Meteorological Society, January 11–15, 2004, Seattle, Washington. Chang JC, Hanna SR (2004) Air quality model performance evaluation, Meteorol. Atmos. Phys. 87, 167–196. Clawson K, Carter R, Lacroix D, Biltoft C, Hukari N, Johnson R, Rich J, Beard S, Strong T (2005) Joint Urban 2003 (JU2003) SF6 Atmospheric Tracer Field Tests, NOAA Tech Memo OAR ARL-254, Air Resources Lab., Silver Spring, MD, 162 pp. + Appendices. DTRA (2004) HPAC Version 4.04.04 (DVD Containing Model and Accompanying Data Files), DTRA, 8725 John J. Kingman Road, MSC 6201, Ft. Belvoir, VA 22060-6201. Hanna S, White J, Zhou Y (2007) Observed winds, turbulence, and dispersion in built-up downtown areas of Oklahoma City and Manhattan. Bound.-Lay. Meteorol. 125, 441–468. Sykes RI, Parker S, Henn D, Chowdhury B (2007) SCIPUFF Version 2.3 Technical Documentation. L-3 Titan Corp, POB 2229, Princeton, NJ 08543-2229, 336 pp. Warner S, Platt N, Urban J, Heagy J (2007) Comparisons of transport and dispersion model predictions of the JU2003 field experiment. J. Appl. Meteorol. Climatol., in press. Zhou Y, Hanna S (2007) Along-wind dispersion of puffs released in a built-up urban area. Bound.-Lay. Meteorol. 125, 469–486.
Discussion B. Denby: Did you come to any conclusion in regard to “suitability for application” from this study? i.e., Does HPAC do what it is designed to do? Did you also run HPAC in operational mode for this campaign? S. Hanna: The HPAC model’s suitability for application has been shown for many types of scenarios prior to this study. Most of these evaluations with field data were carried out by the model developers (Sykes et al., 2007) and have been published in peerreviewed journals. HPAC has also been shown to be suitable for stable conditions in urban areas in prior evaluations using observations from Salt Lake City. The current evaluation involves tracer releases during the day and night in Oklahoma City, using various input meteorology options. The model is seen to do best
Evaluation of the Hazard Prediction and Assessment Capability (HPAC) Model
71
using its preferred meteorological inputs – mesoscale model simulations, in this case the MM5 model. Thus HPAC does appear to do what it is designed to do. During the field experiment, HPAC was run in operational mode, using MM5 inputs, in order to aid in deciding how much tracer gas to release, where to place the samplers, and other aspects of the experiment. Those runs were made by NCAR and not by the authors of the current paper. A. Venkatram: You described several ways of choosing the meteorological inputs for the model. Is there a basis for selecting the inputs that would be most representative of the “real” situation? S. Hanna: In a “real” situation, HPAC would be run using the meteorological model inputs. The HPAC system is set up so that real-time meteorological simulations by five or six models are continually made and stored on the Meteorological Data Server (MDS) at NCAR. Whenever HPAC must be run, the user can access these model simulations via the internet from anywhere in the world. S.T. Rao: Have you looked at measurements at the roof top level? Does the model capture the concentrations at roof top levels better than those as the street levels? S. Hanna: Yes, we have analyzed the roof top concentration measurements, which were taken on about five tall downtown buildings. However, the current paper uses only the surface concentration measurements. In general though, the rooftop concentrations are significant fractions of the surface measurements (e.g., 0.01–0.5), with the larger ratios found for buildings that are the farthest distance from the source and/or have elevations at the low end of the range.
1.2 Modelling of the Urban Wind Profile Sven-Erik Gryning and Ekaterina Batchvarova
Abstract Analysis of meteorological measurements from tall masts in rural and urban areas show that the height of the boundary layer influences the wind profile even in the lowest hundreds of meters. A parameterization of the wind profile for the entire boundary layer is formulated with emphasis on the lowest 200–300 m and presented here. Results are shown from applying the parameterization of the wind profile on independent measurements from an urban experimental campaign that was carried out in Sofia, Bulgaria in 2003. Keywords Boundary layer height, Sofia experiment, wind profile, urban boundary layer
1. Introduction Analysis of profiles of meteorological measurements from a 160 m high mast at the National Test Site for wind turbines at Høvsøre (rural, Denmark) and at a 250 m high TV tower at Hamburg (urban, Germany) shows that the wind profile based on surface-layer theory and Monin-Obukhov scaling is valid up to a height of 50 to 80 m. At higher levels deviations from the measurements progressively occur, being most pronounced at atmospheric neutral conditions as illustrated in Figure 1. 300 200
Height (m)
100
Hamburg-urban 50 30
-200500 m (neutral) 200
20
10 5
10 15 20 25 Normalised wind speed u/u (12m) *
30
Fig. 1 Measurements of normalized wind speed from urban Hamburg compared to predictions by surface layer theory with Monin-Obukhov scaling for the stability according to Dyer (1974)
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
18
Modelling of the Urban Wind Profile
19
A parameterization of the wind profile is formulated that accounts for the influence of the height of the boundary layer. Application of the wind profile parameterisation requires – in addition to the height of the boundary layer – only the usual surface layer scaling parameters such as friction velocity, roughness length and the Obukhov stability scale. Figure 2 illustrates the performance of the parameterization of the wind profile in an urban area. 300 200
Height (m)
100
Hamburg-urban 50 30
-200500 m (neutral) 200
20
10 5
10 15 20 25 Normalised wind speed u/u (12m) *
30
Fig. 2 Measurements from urban Hamburg, as in the figure above, compared with the theory presented in this manuscript
2. Theory A detailed derivation of the wind profile is given in Gryning et al. (2007). The starting point for the analysis is the general expression for the wind profile for the homogeneous, stationary boundary layer:
u * Nl
du dz
(1)
where u and z are wind velocity and height above ground, u is the local friction * velocity, N the von Karman constant and l is the local length scale. In the surface layer the local friction velocity can be considered constant u
*
(2)
u*0 ,
where u*0 is the friction velocity near the ground. Above the surface layer the friction velocity diminishes and becomes small at the top of the boundary layer where it is in this study approximated as
u* z u *0 1 z z i
,
(3)
S.-E. Gryning, E. Batchvarova
20
where z i is the boundary-layer height. The wind profile length scale, l , is composed of three terms. In the surface layer the first length scale (I) is taken to increase linearly with height with a stability correction following Monin-Obukhov similarity. Above the surface layer the second (II) length scale L MBL becomes independent of height but not of stability, and at the top of the boundary layer the third length scale (III) is assumed to be negligible. A simple model for the combined length scale that controls the wind profile and its stability dependence is formulated by inverse summation: 1 l
1 L
SL
1 L
MBL
1
(4)
L UBL
I II III where LSL represents the length scale in the surface layer, LMBL in the middle of the boundary layer and LUBL the upper part of the boundary layer. Figure 3 illustrates the behaviour of the length scales for a 1,000 m deep neutral boundary layer. It can be seen that surface layer scaling, LSL N z (term I) is applicable up to about 50 m, where the influence of LMBL (term II) becomes noticeable. The height of the boundary layer (term III) can already be seen to influence the length scale at about 150 m height. I+II+III
1000
I+II
Height (m)
800
600
400
I
200
0 0
20
40 Length scale (m)
Fig. 3 Profiles of the length scale for neutral conditions with z i
60
80
1, 000 m and z0
0.05 m . The
dashed-dotted line corresponds to the surface layer scaling only. The dashed line includes the effect of LMBL 150m and the full line all three terms in the formulation of the length scale. The value of LMBL of is from Gryning et al. (2007)
Modelling of the Urban Wind Profile
21
3. Wind Profile Inserting Eq. (3) into Eq. (1) using Eq. (4) for the wind profile length scale l , and then integrating along z between the roughness length z 0 and height z yields for the neutral wind profile: u*0 §¨ § z · z z z §¨ ¸ ln¨ N ¨ ¨© z 0 ¸¹ L MBL, N z i ¨ 2 L MBL, N © ©
u z
·· ¸¸ ¸¸ ¹¹
(5)
The function LMBL and its parameterization are discussed in Section 4. For atmospheric stable conditions the functional form of the length scale with stability correction for the surface layer reads: (6)
§ · ¨1 bz / L ¸ z © ¹
L SL
where b ~ 5 is an empirical constant and L is the Obukhov length scale: u*30
L
(7)
N g T w' T ' 0
where T is temperature, g / T is the buoyancy parameter and w'T ' 0 is the kinematic heat flux at the surface. The corresponding wind profile for stable conditions reads: u z u*0
· 1 §¨ §¨ z ·¸ b z § ¨1 z ¸ z z ln ¨ ¸ ¨ ¨ z L © 2 z i ¸¹ L MBL z i N © © 0¹
§ z ¨ ¨ 2L © MBL
·· ¸¸ . ¸¸ ¹¹
(8)
For atmospheric unstable conditions the wind profile can be expressed as: 1 §¨
u( z)
N¨
u*0
©
ln(
z z0
§ z ¨L ©
) \ ¨
· ¸ z z ¸ L MBL z i ¹
§ z ¨ ¨ 2L © MBL
·· ¸¸ ¸¸ ¹¹
(9)
where the stability correction for the surface boundary layer is §z· ©L¹
\¨ ¸
§1 x x2 2 ¨© 3 3
ln¨
x
· ¸ 3 arctan§¨ 1 2 x ·¸ S ¨ ¸ ¸ 3 © 3 ¹ ¹
1 12 z L 1/ 3 .
where
(10)
S.-E. Gryning, E. Batchvarova
22
4. Parameterization of LMBL At the top of the boundary layer the wind profile conforms to the geostrophic wind. A parametrization of LMBL was achieved Gryning et al. (2007) by use of the geostrophic drag law (Rossby similarity theory) that relates the wind speed at the top of the boundary layer to the friction velocity near the ground. In barotropic, stationary conditions, the geostrophic drag law can be written
G
1
u*0
N
ª §u «ln¨ * 0 « ¨ fz ¬ © 0
2 º · ¸ B P » A 2 P ¸ » ¹ ¼
(11)
where G is the geostrophic wind speed, and A and B are the resistance law functions depending on the dimensionless stability parameter P u* 0 f L where f is the Coriolis parameter. For the neutral atmosphere the values A | 4.9 and B | 1.9 are rather well established but their stability dependence is still a matter of discussion. Applying the wind profiles Eqs. (5), (8) and (9) at the top of the boundary layer, z i , and eliminating G u*0 by use of Eq. (11) a parameterisation of LMBL can be obtained. Owing to ambiguity in the formulation of the A and B functions of the geostrophic drag law, an empirical fit of the dependence between u*0 f LMBL , u*0 f z 0 and u*0 f L will be devised. The roughness and atomspheric stability dependence of u*0 f LMBL can be approximated as (Gryning et al., 2007): u*0 f LMBL
§ § ¨ 2 ln¨ u*0 ¨ fz ¨ © 0 ©
§ u f L 2 · · ¸ 55 ¸ exp¨ *0 ¸ ¸ ¨ 400 ¹ ¹ ©
· ¸ . ¸ ¹
(12)
Figure 4 shows a comparison between the empirical fit in Eq. (12) and measurements. It is noted in Gryning et al. (2007) that the data fit to Eq. (12) is better for stable than for unstable conditions. Knowing u*0 , z 0 and L as well as f allows LMBL to be determined from Eq. (12) and the wind profile can then be estimated from Eqs. (5), (8) and (9). The height of the boundary layer can be taken from measurements, from an estimate from a validated meteorological preprocessor, e.g. Batchvarova and Gryning (1991) or it can be approximated by z i | 0.1u*0 f (Gryning et al., 2007).
Modelling of the Urban Wind Profile
23
Parametrization of (u*0/fLMBL)
50
40
30
20
10
0 0
10
20
30
40
Measurements of (u*0/f LMBL)
50
Fig. 4 Measurements of the normalized length scale u*0 f LMBL plotted against the parametrization of u*0 f LMBL given with Eq. (12). Circles represent unstable, triangles neutral and squares stable conditions
5. Sofia 2003 Boundary Layer Experiment The parameterizations were applied on data from an urban area boundary layer experiment in Sofia September/October 2003 (Batchvarova et al., 2004). The measurements took place at the National Institute of Meteorology and Hydrology (NIMH) located in the south eastern part of the city. The urban characteristics for the site are spread over 3 km to south and east, about 10 km to north and about 20 km to the west. A view of the area surrounding the measurements site is show in Figure 5. The experiment comprised high resolution boundary layer radiosoundings to determine the boundary layer growth and measurements with sonic anemometers at two heights at NIHM. The two sonic anemometers were mounted on the research tower of NIMH at 20 and 40 m height above ground level (Batchvarova et al., 2007). High resolution (3–4 ms-1 ascent rate) radiosoundings were performed with 1–2 hours interval under convective conditions. Five days with such conditions were identified during the period 18 September–8 October 2003, two of the days (29 September and 3 October) were suited for analysis in the context of this analysis. Figure 6 shows wind speed profiles from 29 September and 3 October 2003 for conditions where the wind passes over the urban Sofia before it reaches the measuring site. The meteorological conditions were unstable for both of the days, 29 September represent relative high and 3 October low wind speed conditions; some characteristic parameters are given in Table 1.
S.-E. Gryning, E. Batchvarova
24
Fig. 5 Impression of the Sofia urban area. The meteorological tower at NIMH is seen on the left Sofia 29 September 2003
1000
Sofia 29 September 2003
800
Height (m)
17 LT
LT
100 15
Height (m)
600
400
200
0
10 2
4 6 8 10 Normalized wind speed u/u*
2
12
Sofia 3 October 2004
1000
4 6 8 10 Normalized wind speed u/u
12 *
Sofia 3 October 2004
800
Height (m)
Height (m)
600
100 18 LT 17
400
200
LT
10
0 4
5 6 7 8 Normalized wind speed u/u*
9
4
5 6 7 8 Normalized wind speed u/u*
9
Fig. 6 Profiles of wind speed on 29 September (upper panels) and 3 October (lower panels). The wind speed profiles are for convenience shown with both a logarithmic (left) and linear (right) height scale. Dots and triangles represent measurements from two radiosoundings each day. The corresponding parameterized wind profile is given by a solid line for the dots and a dashed line for the triangles
Modelling of the Urban Wind Profile
25
It can be seen that in all cases the wind speed increases near the ground as function of height. For the high wind case (29 September) the increase in wind sped continues up to a height of 100–300 m above which the wind speed starts decreasing. In the low wind case on 3 October the decrease starts at a lower height. The predicted wind profile near the ground is better predicted for the windy experiment on 29 September as compared to the low wind speed conditions on 3 October. Further comparisons of the new parameterisation with wind profile data are to be made in order to assess the influence of surface non-homogeneities on the vertical wind profile. Table 1 Parameters from the experiments on 29 September and 3 October 2003. LT means local time, LT = GMT + 3. From measured meteorological parameters
29 September 3 October
15 LT 18 LT 17 LT 18 LT
Wind speed at 20 m height (m/s)
Friction velocity u* (m/s)
Obukhov length (m)
3.01 2.70 2.18 1.46
0.75 0.66 0.48 0.50
–140 –400 –100 –150
Boundar y-layer depth (m) 1,000 1,200 735 870
Estimated from wind profile Roughness length z 0 (m) 1.8 0.8 1.3 2.1
Acknowledgments The work is related to activities of the authors in COST728, COST732 and COST735. The Sofia boundary layer experiment was supported by a Swiss-Bulgarian Institutional partnership 7IP065650.
References Batchvarova E, Gryning S-E (1991) Applied model for the growth of the daytime mixed layer. Bound-Layer Meteorol 56:261–274 Batchvarova E, Gryning S-E, Rotach MW, Christen A (2004) Modelled aggregated heat fluxes compared to turbulence measurements at different heights. Proc. 9th Int. Conference on Harmonisation within atmospheric dispersion modeling for regulatory purposes, 1–4 June 2004, Garmisch Partenkirchen, Germany, vol 2, pp 7–12 Batchvarova E, Gryning S-E (2006) Progress in urban dispersion studies. Theor Appl Climatol 84:57–67 Dyer AJ (1974) A review of flux-profile relationships. Bound.-Layer Meteorol 7:363–372 Batchvarova E, Gryning S-E, Rotach M, Christen A (2007) Comparison of aggregated and measured turbulent fluxes in an urban area. In: Air pollution modeling and its application 17. 27. NATO/CCMS international technical meeting, Banff (CA), 24–29 Oct 2004. Borrego, C.; Norman, A.-L. (eds.), Springer, New York, pp 363–370 Gryning S-E, Batchvarova E, Brümmer B, Jørgensen HE, Larsen SE (2007) On the extension of the wind profile over homogeneous terrain beyond the surface boundary layer. Bound.-Layer Meteorol 124:251–268
26
S.-E. Gryning, E. Batchvarova
Discussion R. Hill: How sensitive is the calculation of wind speed profile to the boundary layer height? Could simple schemes for estimating boundary layer height be used? S.-E. Gryning: Limiting ourselves to the wind profile in the lowest few hundred metres the effect of the height of the boundary layer is important, especially during stable conditions. The wind profile formulation generally overpredicts the measurements during stable conditions if the effect of the boundary layer height is neglected. A simple scheme for the boundary layer height is definitely to prefer if the alternative is to neglecting it. S. Hanna: Is your wind profile formula limited to cities that do not have tall buildings? If there is an extensive area of tall buildings, such as in Manhattan or São Paulo, it seems as if the buildings will extend up above the so-called “surface boundary layer”. S.-E. Gryning: We derive the wind profile for the entire boundarylayer assuming stationary and homogeneous conditions. In a more relaxed formulation it can be considered in the so-called neighbourhood approach for large neighbourhoods. The neighbourhood concept is illustrated below (Batchvarova and Gryning 2006).
Fig. Schematics of the boundary layer structure over an urban area. The vertical and horizontal patterns represent the underlying surface of the neighbourhoods of tall and low buildings, respectively. Broad spaced patterns represent the urban internal boundary layers where advection processes are important. Fine spaced patterns show the inertial sublayers that are in equilibrium with the underlying surface and where Monin-Obukhov scaling applies. The forward slash pattern represents the roughness sublayer that is highly inhomogeneous both in its vertical and horizontal structure. The dotted pattern represents
Modelling of the Urban Wind Profile
27
adjustment zones between neighbourhoods with large accelerations and shear in the flow near the top of the canopy. Above the height where the internal boundary layers are intermixed the effects of the individual neighbourhoods cannot be distinguished any more – the so-called blended layer
If the area occupied by tall buildings is large enough to be considered as a neighbourhood, there will be large pressure adjustment zones downwind from the buildings but in principle the neighbourhood concept is applicable. For single tall buildings the neighbourhood concept is not applicable close to them. B. Fisher: Have you considered parameterizations describing the transition from one idealised urban boundary layer to another, say, the rural profile? S.-E. Gryning: The case of changes from one surface condition to another (i.e. rural-urban) was discussed at the ITM conference in 2006 in Leipzig (Gryning, S.-E.; Batchvarova, E., Turbulence, atmospheric dispersion and mixing height in the urban area, recent experimental findings. In: Air pollution modeling and its application XVIII. 28th NATO/CCMS International Technical Meeting, Leipzig (DE), 15– 19 May 2006. Borrego, C.; Renner, E. (eds.) (Elsevier, Amsterdam, 2007) (Developments in Environmental Science, 6) pp. 12–20). When the internal boundary layer reaches its equilibrium height the wind profile proposed here is applicable. R. San Jose: I’m not sure that with your approach the entrainment effects at the top of the boundary layer produced by the heat flux entrainment coming from the layer above the boundary layer. These effects can be very important and cannot be “felt” from surface. S.-E. Gryning: The entrainment from the top influences the boundary layer height and thus is included in the wind profile. A. Venkatram: Why is the Coriolis parameter (f) relevant to your formulation? The boundary layer is rarely “neutral” in the sense that the heat flux is zero. The scale (u*/f) does not scale with measured boundary layer height under most conditions. S.-E. Gryning: The u*/f scale is unavoidable because it relates to the geostrophic drag law. Trying to formulate the geostrophic drag law with the measured boundary layer height zi is not realistic (just try). Furthermore we tried to scale LMBL with both u*/f and zi, and despite a large scatter the u*/f scaling turned out to be the most convincing. We believe that proper scaling of the boundary layer height with u*/f should include the Brunt-Vaisala frequency in the free air above the boundary layer.
1.10 Numerical Treatment of Urban and Regional Scale Interactions in ChemistryTransport Modelling R. Wolke, D. Hinneburg, W. Schröder and E. Renner
Abstract The physical and chemical processes that determine the distribution of air pollutants occur on a wide range of temporal and spatial scales. Multiscale models can provide finer resolution in certain key regions, e.g. around large sources. The paper focuses on some numerical aspects of modelling urban and regional scale interactions as well as on requirements on the used parameterisations in this context. Multiblock grid techniques (“two-way nesting”) and implicit-explicit time integration schemes are suitable for an efficient numerical treatment of such scale interactions. In the online coupled model system LM-MUSCAT, both approaches are implemented for the chemistry-transport code. Gas phase processes, especially the formation of photooxidants, as well as the transport and the transformation of particulate matter, can be investigated. The advantages of the multiblock technique to establish the interactions between different scales in a natural way are demonstrated for one selected scenario in the Saxony area. The influence of grid resolutions on the simulation results is discussed.
Keywords Air quality modelling, cooling tower emissions, formation of secondary inorganic aerosol, model coupling, parallel computing, PM10 loading
1. Introduction The physical and chemical processes in the atmosphere are very complex. They occur simultaneously, coupled and in a wide range of scales. These facts have to be taken into account in the numerical methods for the solution of the model equations. The numerical techniques should allow the use of different resolutions in space and also in time. The lack of adequate resolution limits the ability for accurate modelling of individual processes and their interactions. For example, when plumes are injected into coarse grid cells in regional models with a uniform grid, the emitted material is diluted immediately within the cell and the details of the near field chemistry are lost. Multiscale models can provide finer resolution in certain key regions, e.g. around large sources or urban areas.
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
90
Numerical Treatment of Urban and Regional Scale Interactions in Chemistry-Transport
91
Multiblock grid techniques and implicit-explicit (IMEX) time integration schemes are suitable for an efficient treatment of scale interactions (Wolke and Knoth, 2000). The multiscale chemistry-transport code MUSCAT (MUltiScale Chemistry Aerosol Transport) is presented which is based on these techniques. The meteorological fields are generated simultaneously by the non-hydrostatic meteorological model LM (Local Model), which is the operational regional forecast model of the German Weather Service and offers the possibility to use different meteorological parameterisations in dependence on the spatial model resolution. Exemplarily, the different issues are discussed for one selected application. The study enables a detailed quantification of the contributions of cooling tower emissions to particle concentration levels in specific Saxonian urban areas. These investigations require a detailed process description as well as a high spatial resolution especially in the region around the cooling towers. The nearly constant daily emission rates of the power plants in Boxberg and Lippendorf were separately averaged for the individual cooling towers and for the simulated summer and winter periods. Supported by measurements, 12% of the emitted SO2 mass is considered as directly emitted sulphate. The plume rise was determined by using the model of Schatzmann and Policastro (1984).
2. The Chemistry Transport Code LM-MUSCAT The modelling department of the IfT has developed the state-of-the-art multiscale model system LM-MUSCAT (Wolke et al., 2004a). It is qualified for process studies as well as the operational forecast of pollutants in local and regional areas (Heinold et al., 2007; Wolke et al., 2004b). The model system consists of two online coupled codes. LM. The forecast model LM is a non-hydrostatic and compressible meteorological model and solves the governing equations on the basis of a terrain-following grid (Doms and Schättler, 1999; Steppeler et al., 2003). The model includes the dynamic kernel for the atmosphere as well as the necessary parameterisation schemes for various meteorological processes, boundary conditions and surface exchange relations. It describes the atmospheric flow and phenomena between the meso- and micro-scale (i.e. grid resolutions from 50 km to 50 m), in particular near-surface properties, convection, clouds, precipitation, orographical and thermal wind systems. The model is capable of self-nesting and offers the possibility to use different meteorological parameterisations in dependence on the spatial model resolution. MUSCAT. Driven by the meteorological model, the chemistry transport model MUSCAT treats the atmospheric transport as well as chemical transformations for several gas phase species and particle populations. The transport processes include advection, turbulent diffusion, sedimentation, dry and wet deposition. Due to the online coupling between LM and MUSCAT, the calculations exploit the actual properties of the atmosphere. The implicit-explicit time integration scheme of MUSCAT operates independently from the meteorological model, thus allowing
92
R. Wolke et al.
for autonomous time steps and different horizontal grid resolutions in selected regions of the model domain. For this purpose, the required LM fields (e.g., wind, temperature, humidity, exchange coefficients) are interpolated temporarily and spatially. The chemical part of MUSCAT contains the gas phase mechanism of RACM considering 76 reactive gas species with 239 reactions (Stockwell et al., 1997). Secondary particle formation and appropriate interactions with the gas phase are also included. For the description of the particle size distribution and aerosol dynamical processes the modal aerosol model M7 (Vignati et al., 2004) is used. In this approach, the total particle population is aggregated from seven log-normal modes with different compositions. Alternatively, a more simplified mass based approach (similar to EMEP) can be applied, especially for screening studies to evaluate PM. The radiation activity and biogenic emissions are calculated autonomously, whereas information on the cloud cover, temperature, and other meteorological parameters is taken from LM. The anthropogenic emissions in Saxony are accounted for by several spatial registers of point, line, and area sources (including plume rise). Annual averages of emissions are disaggregated in time (monthly, weekly, daily). EMEP emission data were used in the outer region. Coupling. Both models work widely independent and have their own separate time step control. Coupling between meteorology and chemistry transport takes place at each horizontal advection time step only. Inter-processor communication is realised by means of MPI. The parallelisation is performed by domain decomposition techniques, but LM and MUSCAT use different horizontal grid structures. In MUSCAT, finer and coarser resolutions can be used for individual sub-domains in the multiblock approach (Knoth and Wolke, 1998). This structure originates from dividing the meteorological grid into rectangular blocks of different sizes. The code is parallelised by distributing these blocks on the available processors. In the past, the “concurrent” coupling scheme has been used in LM-MUSCAT, where both models operate concurrently on distinct sets of processors (Wolke et al., 2004a). Since an adaptive time step control is applied in MUSCAT, the overall workload fluctuates during runtime, especially at scenarios with highly dynamical behaviour of the simulated chemical processes. These fluctuations led to processor idle time at the synchronisation points of the two models. To achieve a higher efficiency, an alternative coupling scheme has been implemented, which is based on the “sequential” approach (Lieber and Wolke, 2008). Benefits are an increased performance and a simplified model start-up, since no processor partitioning (determination of processors for LM and MUSCAT) has to be defined a priori.
3. Formation and Dispersion of Secondary Aerosols by Cooling Towers Motivation Fine and ultra-fine particles are suspected to cause damages in human health and natural environment. Several methods to reduce emissions by traffic, industry, agriculture and other sources have been successfully realised. Nowadays,
Numerical Treatment of Urban and Regional Scale Interactions in Chemistry-Transport
93
the focus of environmental sciences and politics is directed towards understanding the physico-chemical formation and growth processes of secondary particles rather than the primary production. Modifying factors in this context are the meteorological conditions, and the variable effective heights of the dominant emission sources. Therefore, long-term real weather simulations were performed in a mesoscale industrial region of Saxony. To investigate, to which extent the plume exhaust of the natural draft cooling towers in Boxberg and Lippendorf contributes to the formation of secondary aerosols. In spite of the relatively low emissions of primary particulate matter (PPM), interferences of the air quality are expected from the formation of sulphate and nitrate aerosols by the SO2 and NOx gas emissions and reactions with ammonia.
Fig. 1 Model region of Central Europe and nesting of the Saxony area year
Model setup The study mainly aims at problems of the secondary formation and growth of inorganic particles with sizes below 10 ȝm (PM10). The dominant contribution to mass accretion is provided by the heterogeneous condensation of gaseous compounds on pre-existing aerosols, involving pollutants such as ammonia and sulphuric or nitric acid (generated by complex pathways from the precursor species SO2 and NO2). The model system LM-MUSCAT was applied in a nested hierarchy with the superior control by the global reanalysis data of GME (Figure 1). The innermost region of interest covers an area of 240 × 156 km (Saxony) with variable resolution between 2.8 and 0.7 km, where the finest grid was arranged
94
R. Wolke et al.
around the dominant emission sites Boxberg and Lippendorf (Figure 2). The outer region extends over about 1,200 × 1,000 km (Central Europe) with a constant resolution of 8 km in the main part. The model simulation in this domain generates the initial and boundary concentrations for the inner region.
Fig. 2 Structure of grid refinement in the multiblock MUSCAT grid
Two typical summer/winter periods in 2002 each of 36 days were chosen for the simulations. The plume rise of the cooling towers in dependence on the meteorological conditions was accounted for by utilizing the adequate model of Schatzmann and Policastro (1984). One virtual neutral tracer per cooling tower was additionally emitted for identification purposes. Considering the meteorological and emission/immission conditions within the reference periods, two eight-day intervals were selected, for which the simulations on Saxony were repeated with the cooling tower emissions switched off. Thus, the difference between the switchon/off simulations allows extracting the direct and indirect influence of the corresponding emissions on the immission situation (Blanchard and Hidy, 2005). The outstanding SO2 production rates by the sites Boxberg (1.2 t/h) and Lippendorf (2.1 t/h) as compared to the extremely small dust emissions (0.04, 0.06 t/h, respectively) actually require the examination of the secondary formation of particulate matter. In this respect, the assumed partitioning of the cooling tower SO2 emission data (88% SO2, 12% SO4 ions) is of great significance for the results.
Numerical Treatment of Urban and Regional Scale Interactions in Chemistry-Transport
95
Fig. 3 PM10 concentration in Saxony caused by the emissions in Boxberg and Lippendorf at August 20th 2002, 9:00 LT (left side) and 12:00 LT (right side)
Simulation results The results are summarised as follows (an example is shown in Figure 3):
x
x
x
x
Significant direct or indirect influences of the cooling tower gas emissions on the particle immission situation can be noticed only within limited zones of very narrow (highly concentrated) plumes and mainly for several hours about noon in summer. The intrinsic particle formation caused by the considered emission sites frequently reaches peak values of about 10 ȝg/m3 PM10 (maximal 20 ȝg/m3), consisting predominantly of ammonium sulphate. Primary particles and ammonium-nitrate formation contribute to only 10%. The small probability for maximum plume intensities at a fixed location on the surface decreases the temporal averaged particle exposition down to 0.1 ȝg/m3 PM10. Advanced investigations concerning the formation of particulate sulphate from the gas phase within wet (saturated) plumes have not offered significant particle growth because of the limited length of stay in the steam plume (see also Engelke et al., 2007).
4. Conclusion Grid refinement can be used to describe the distribution of pollutant concentrations more accurately at locations where this is required, e.g. near concentrated emission sources or urban areas. On the other hand, uniform refinement leads to unacceptable high computational costs. Therefore local refinement is favoured. The time-integration scheme has also taken into account the temporal heterogeneity of atmospheric processes. The multiblock technique combined with IMEX time integration schemes are well suited for the numerical treatment of strong gradients and scale interactions in an efficient way.
96
R. Wolke et al.
Acknowledgments The work was supported by the LfUG of Saxony, the ZIH Dresden and the NIC Jülich. Furthermore, we thank the DWD Offenbach for good cooperation.
References Doms G, Schättler U (1999) The Nonhydrostatic Limited-Area Model LM (LokalModell) of DWD: Part I: Scientific Documentation (Version LM-F90 1.35), Deutscher Wetterdienst, Offenbach, 1999. Blanchard CL, Hidy GM (2005) Effects of SO2 and NOx emission reductions on PM2.5 mass concentrations in the southeastern United States, J. Air Waste Manage. Assoc. 55, 265–272. Engelke T, Hugo A, Renner E, Schmidt F, Wolke R, Zoboki J (2007) Mixing of plumes with ambient background air: effects of particle size variations close to the source. In: C Borrego and E Renner, Eds., Air Pollution Modeling and Its Application XVIII, Leipzig, Germany, 15–19 May 2006, Elsevier, Amsterdam, The Netherlands, pp. 621–630. Heinold B, Helmert J, Hellmuth O, Wolke R, Ansmann A, Marticorena B, Laurent B, Tegen I (2007) Regional Modeling of Saharan Dust Events using LMMUSCAT: Model Description and Case Studies. J. Geophys. Res. 112, D11204, doi:10.1029/2006JD007443. Knoth O, Wolke R (1998) An explicit-implicit numerical approach for atmospheric chemistry-transport modelling, Atmos. Environ. 32, 1785–1797. Lieber M, Wolke R (2008) Optimizing the coupling in parallel air quality model systems, Environ. Mod. Software 23, 235–243. Schatzmann M, Policastro AJ (1984) An advanced integral model for cooling tower plume dispersion, Atmos. Environ. 18, 663–674. Steppeler J, Doms G, Schättler U, Bitzer HW, Gassmann A, Damrath U, Gregoric G (2003) Meso-gamma scale forecasts using the nonhydrostatic model LM. Meteorol. Atmos. Phys. 82, 75–96. Stockwell RW, Kirchner F, Kuhn M, Seefeld S (1997) A new mechanism for regional atmospheric chemistry modeling, J. Geophys. Res. 102, 25847–25879. Vignati E, Wilson J, Stier P (2004) M7: An efficient size-resolved aerosol microphysics module for large-scale aerosol transport models, J. Geophys. Res. 109, D22202, doi:10.1029/2003JD004485. Wolke R, Knoth O (2000) Implicit-explicit Runge-Kutta methods applied to atmospheric chemistry-transport modelling, Environ. Mod. Software 15, 711– 719. Wolke R, Knoth O, Hellmuth O, Schröder W, Renner E (2004a) The parallel model system LM-MUSCAT for chemistry-transport simulations: coupling scheme, parallelisation and application. In: GR Joubert, WE Nagel, FJ Peters, and WV Walter, Eds., Parallel Computing: Software Technology, Algorithms, Architectures, and Applications, Elsevier, Amsterdam, The Netherlands, pp. 363–370.
Numerical Treatment of Urban and Regional Scale Interactions in Chemistry-Transport
97
Wolke R, Hellmuth O, Knoth O, Schröder W, Heinrich B, Renner E (2004b) The chemistry-transport modeling system LM-MUSCAT: description and CITYDELTA applications. In: C Borrego and S Incecik, Eds., Air Pollution Modeling and its Application XVI, Kluwer/Plenum, New York, pp. 427–439.
Discussion B. Fisher: You should make clear that for the power stations you consider the exhaust gases are emitted into the cooling tower plume to enhance buoyancy. This is an uncommon arrangement. Usually the stack and cooling power plumes do not interact? R. Wolke: In Germany, a special technique is used in power stations which are equipped with natural-draft cooling towers. The flue gases are piped directly into the centre of cooling towers. Then the exhausted gas-steam mixture contains gas phase species (e.g., CO, NOx and SO2), the directly emitted primary particles and, especially, an excess of ‘free’ sulphate ions in water solution, which after the desulphurisation steps remain non-neutralised by cations. All gaseous and particulate pollutants are lifted higher up into large varying effective heights by the hot steam. In this way, the additional building of high stacks can be avoided. Of course, strong emission limits have to be maintained also.
1.5 On the Suppression of the Urban Heat Island over Mountainous Terrain in Winter Charles Chemel, Jean-Pierre Chollet and Eric Chaxel
Abstract Most of the urban areas generate an Urban Heat Island (UHI) because of urban-rural differences such as in albedo or heat capacity. The UHI effect is often more noticeable either at night or in winter since it may balance the effects of stable stratification. Over complex topography, terrain-induced exchange processes interfere with the urban-scale circulations. Also, in a valley environment, temperature changes are larger than over flat terrain, so that the UHI may be either enhanced or suppressed depending upon the stratification of the atmosphere. Numerical simulations were conducted in the area of the Grenoble valley (France) for a selected PM10 pollution episode in February 2005 using a set of numerical codes (e.g. WRF and METPHOMOD). The UHI index was usually found to be in the order of 5 K at night and 3 K during the day. Ground surface measurements of radon were used to investigate the role of the UHI in the dispersion of locally emitted primary pollutants. Keywords Numerical simulation, urban heat island, valley
1. Introduction Urban climate differs from their countryside counterparts in many ways. Basically urbanization results in enhanced mixing, modified wind field, and increased cloudiness (e.g. Bornstein and Lin, 2000). Furthermore, most of the urban areas are warmer than their rural surroundings. This thermal effect is commonly referred to as Urban Heat Island (UHI) effect. Urban-rural differences in albedo, moisture, roughness, and heat capacity are the main contributors to the UHI (e.g. Runnalls and Oke, 2000). In addition waste heat from urban activities has also been identified as a key factor of the formation of the UHI (see for instance Sailor et al., 2003). The UHI effect is often more noticeable either at night or in winter since it may balance the effects of stable stratification. Over complex topography, the circulations induced by the terrain may interfere strongly with the urban-scale circulations, which are driven by the land-use heterogeneities. In our study, we focus on the Grenoble urban area, which population exceeds 300,000 inhabitants. The metropolitan area is located at the merging of three deep and steep-side valleys in the western part of the French Alps. Temperature inversions are frequent in winter C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
46
On the Suppression of the Urban Heat Island over Mountainous Terrain in Winter
47
and can last for many days. In that case, a cold air pool forms over the city leading to stagnant conditions. Such stable layers capping the whole Grenoble valley area are totally different from the deep mixed layers observed in summer (Chaxel and Chollet, 2007a). Nonetheless, both thermal storage by the urban canopy layer and the topography heating may contribute to mix the lowest layers of the atmosphere and induce horizontal circulations, which then can enable pollutants to be transported a few kilometres away. The characteristics of the urban environment can significantly impact levels of air pollution under stable conditions. The general aim of our study is to characterize and quantify the balance between the effects of the UHI and the stable stratification. To our knowledge, neither observational nor numerical studies have been conducted so far in order to quantify this balancing effect in a valley environment. A few numerical studies were performed to investigate both atmospheric dynamics and air quality in the Grenoble valley (e.g. Couach et al., 2003, 2004; Chaxel, 2006). The present study relies on numerical simulations, which were conducted for a selected PM10 pollution episode in February 2005. The modelling system, which consists in meteorological and chemistry-transport models, is described in Section 2. In Section 3, results from the modelling system are discussed by focusing on the balance between the UHI and stable stratification effects. Measurements of ground surface concentration of radon are used in Section 4 to investigate the role of the UHI in the dispersion of primary pollutants during the episode. Finally concluding remarks are given in Section 5.
2. The Modelling System The modelling framework consisted in state-of-science meteorological models (MetMs) and chemistry-transport models (CTMs). These models were implemented in a nested structure using three nest levels with grid cell horizontal resolutions of 18, 6 and 2 km. The outermost domain covered most of Europe. So it was large enough to handle synoptic scale forcing that is suitable for analysis of long lasting air pollution episodes. Several numerical codes were considered, and included MM5 (Grell et al., 1995), WRF (Skamarock, 2007), ARPS (Xue et al., 2000) for the MetMs and CHIMERE (Schmidt et al., 2001) and METPHOMOD (Perego, 1993) for the CTMs. The MetMs codes have strong similarities (e.g. terrain following coordinate system, non hydrostatic option) but differ in many ways (e.g. the degree of complexity in the representation of the urban canopy). MM5 treat the urban area by modifying the surface properties (e.g. roughness), whereas WRF contains an urban canopy parameterization scheme in the NOAH surface energy balance model. For comparison, the BEP urban canopy model (Martilli et al., 2002) was implemented in the ARPS code. As for the CTMs, CHIMERE was used for regional-scale calculations and provided initial and boundary conditions for METPHOMOD, which code was used at the local scale. Note that METPHOMOD use a Cartesian grid. A baseline of model performance was established for several pollution episodes in both summer and winter (see for
48
C. Chemel et al.
instance Couach et al., 2003; Chaxel, 2006; Chaxel and Chollet, 2007b). Previous studies were mainly focused on gas chemistry, with a peculiar attention to benzene. Particulate matters were treated as passive species, for which an emission inventory was derived from that of oxides of nitrogen (NOx) for traffic and domestic and district heating. Note that particle sedimentation was not considered. Fig. 1 Temperature time series from 6 to 10 February 2005 at CEA: measurements from 2 to 100 m above ground level (solid lines) and results from METPHOMOD (dots)
3. The Winter Episode The modelling system was setup to simulate a PM10 pollution episode during the first two weeks of February 2005. Despite sunny weather prevailed during that period, a long-lasting temperature inversion was maintained over the valley. The urban area is located at the convergence of three valleys, so that the Grenoble valley is often referred to as a Y-shaped valley. Hence, atmospheric dynamics is completely different from that in a closed basin. Indeed, the valley wind system is controlled by the interactions of the synoptic flow with the flows from the Voreppe Cluse (NW), the Gresivaudan valley (NE) and the uneven slopes in the Southern part of the valley.
On the Suppression of the U rban Heat Island over Mountainous Terrain in Winter
49
Fig. 2 Horizontal wind vectors at 10 m above ground level (shading denotes the temperature in °C in the range –5 to 5°C only and solid lines indicates the topography with 250-m interval contours) from WRF at 1900 UTC on 7 February 2005. (a) Without the urban canopy model and (b) with the urban canopy model
50
C. Chemel et al.
From 5 to 7 February 2005, the synoptic wind below 750 hPa at Lyon/Saint Exupéry airport (100 km North-West of Grenoble) was from the South while from 8 to 15 February 2005, it was from the North. The modelling system underwent a validation step using available observations. In Figure 1, the 2-m temperature from METPHOMOD from 6 to 10 February is compared with measurements from a 100m mast. The agreement is found very good. The characteristics of the urban boundary layer were related to those of the boundary layer in a rural area (Charavines). The 2-m temperature from WRF, using either an urban canopy model or not, is displayed in Figure 2 for the 7th of February 2005. The effect of the urban canopy model is found rather weak on that day. The UHI at 1900 UTC on that day is about 3 K. During the two-week episode, the UHI index was usually found to be in the order of 5 K at night and 3 K during the day. The vertical structure of the atmosphere strongly depends on the interaction of the valley wind system with the synoptic flow (see for instance Chemel et al., 2007). When the synoptic wind comes from the South, this interaction determines the distribution of pollutants within the valley.
4. Radon Radon is a heavy radioactive gas whose concentration is usually low in the atmosphere. However this gas is of special interest because of its toxicity and elevating cancer risk. Contrary to biogenic or anthropogenic airborne pollutants, radon is emitted from the Earth surface at a rather constant rate and is a long lifespan species. Hence it may be reasonably considered as a passive tracer with respect to atmospheric dynamics. The air quality agency ‘APPA Dauphine Savoie’ measures continuously the concentration of radon in Grenoble since 1991. Perrino et al. (2001) derived an index, which quantifies atmosphere stability, from the time derivative of radon concentration. Peak values of radon concentration are very likely to be attributed to a lack of vertical mixing. During the episode, there is a clear correlation between benzene and radon peak values. This demonstrates, at least for the period under consideration, that peak values of regulated pollutants are due to meteorological conditions and not to an increase in emissions. Alpha total and beta total radiations are displayed in Figure 3 in two locations (3 km away one from the other). A very good correlation between the two stations is found for the peak values when an inversion layer lies above the city (namely from 4 to 11 February and from 17 to 20 February). Concentration of radon is compared to the concentration of a passive scalar uniformly emitted at the ground surface in Figure 4. The variation in the scalar concentration follows the variation of radon measurement from 4 to 11 February. Note that it does not make sense to compare the amplitude of the time series because the emission rate of the passive tracer was arbitrary. Slight oscillations at night are observed in the measurement time series whilst being not captured by the model. This discrepancy at night could by attributed to unresolved small-scale mixing due to residual turbulence or gravity waves.
On the Suppression of the U rban Heat Island over Mountainous Terrain in Winter
51
1.40E-03 1.20E-03
beta total Pylône Bq/m3 alpha total Pylône Bq/m3
Bq m-3
1.00E-03
beta total Chorier Bq/m3 alpha total Chorier Bq/m3
8.00E-04 6.00E-04 4.00E-04 2.00E-04 0.00E+00 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
Fig. 3 Radon (Bq/m3) at CEA Pylône (black) and Chorier (grey)
20 18 16
mesure [Bq/m3]
14
modèle [ug/m3]
12 10 8 6 4 2 0 02/02 03/02 04/02 05/02 06/02 07/02 08/02 09/02 10/02 11/02 12/02 13/02 14/02
Fig. 4 Radon concentration at CEA from: measurements (black line) and METPHOMOD (grey line)
5. Conclusions and Prospects The Grenoble valley is a typical urban area. The Y-shaped valley is located over highly complex terrain at the junction of three valleys. Whatever the season, synoptic winds interact with local circulations, which are mainly initiated by thermal convection or radiative cooling. The orography forces this interaction by selecting scales and flow direction from typical valley width, depth and orientation. In addition, wintertime periods are more difficult to simulate because of stable conditions and rather weak turbulent mixing. Hence, the performance of the model is strongly
52
C. Chemel et al.
determined by its ability to detect changes in the stratification of the lowest layers of the atmosphere. Different flow regimes were observed with drastic consequences on pollutant dispersion. Basically two regimes were identified: either ‘free’ or ‘trapped’, depending upon the interaction of the valley wind system with the synoptic flow in the Southern part of the valley. Models are well known to underestimate residual mixing either at night or in winter and more generally under stable conditions, as illustrated for instance in Section 4. Improvement in subgrid-scale turbulence parameterisation is required to tackle this issue, especially for high-resolution simulations. Simulations are currently being conducted with finer vertical grid resolution near ground (a few meters) and horizontal resolution as fine as 300 m for the innermost domain. The urban heat island index was found to be in the order of 5 K at night and 3 K during the day, whether an urban canopy model or not is used in the simulations. Further research is needed to quantify the reduction in the stratification of the lower layers of the atmosphere by the UHI and mixing of surface-based pollutants, which may be trapped below a very shallow temperature inversion. Acknowledgments The authors thank GIERSA and ASCOPARG for making measurements and emission inventory available.
References Bornstein R, Lin Q (2000) Urban heat island and summertime convective thunderstorms in Atlanta: three case studies, Atmos. Environ. 34, 507–516. Chaxel E (2006) Photochimie et aérosol en région alpine: mélange et transport, Thèse de l’Université Joseph Fourier, Grenoble, France, Available at http://tel.archives-ouvertes.fr/ Chaxel E, Chollet JP (2007a) Ozone production from Grenoble city during the August 2003 heat wave: modelling and analysis, 6th International Conference on Urban Air Quality, Limassol, Cyprus, 27–29 March. Chaxel E, Chollet JP (2007b) Modelling of summer photochemistry and winter aerosols in Grenoble urban area in the French Alps, 11th International Conference on Harmonisation within Atmospheric Dispersion Modelling for Regulatory Purposes, Cambridge, UK, 2–5 July. Chemel C, Chaxel E, Chollet JP (2007) On the role of the Grenoble valley topography in vertical transport of mass and pollutants, 29th International Conference on Alpine Meteorology, Chambéry, France, 4–8 June. Couach O, Balin I, Jiménez R, Ristori P, Perego S, Kirchner F, Simeonov V, Calpini B, Van den Bergh H (2003) An investigation of ozone and planetary boundary layer dynamics over the complex topography of Grenoble combining measurements and modelling, Atmos. Chem. Phys. 3, 549–562.
On the Suppression of the U rban Heat Island over Mountainous Terrain in Winter
53
Couach O, Kirchner F, Jiménez R, Balin I, Perego S, Van den Bergh H (2004) A development of ozone abatement strategies for the Grenoble area using modeling and indicators, Atmos. Environ. 38, 1425–1436. Grell GA, Dudhia J, Stauffer DR (1995) A description of the Fifth-Generation Penn State/NCAR Mesoscale Model (MM5), NCAR Technical Note NCAR/TN-398 + STR, NCAR, Boulder, CO, USA. Martilli, A, Clappier A, Rotach MW (2002) An urban surfaces exchange parameterisation for mesoscale models, Bound.-Layer Meteorol. 104, 261–304. Perego S (1993) Metphomod – a numerical mesoscale model for simulation of regional photosmog in complex terrain: model description and application during Pollumet 1993 (Switzerland), Met. Atmos. Phys. 70, 43–69. Perrino C, Pietrogangelo A, Febo A (2001) An atmospheric stability index based on radon progeny measurements for the evaluation of primary urban pollution, Atmos. Environ. 35, 5235–5244. Runnalls KE, Oke, TR (2000) Dynamics and controls of the near-surface heat island of Vancouver, BC, Physical Geography, 21, 283–304. Sailor DJ, Lu L, Fan H (2003) Estimating urban anthropogenic heating profiles and their implications for heat island development, Proc. of the 5th International Conference on Urban Climate, Lodz, Poland, 1–5 September. Schmidt H, Derognat C, Vautard R, Beekmann M (2001) A comparison of simulated and observed ozone mixing ratios for the summer of 1998 in western Europe, Atmos. Environ., 35, 6277–6297. Skamarock WC, Klemp JB, Dudhia J, Gill DO, Barker DM, Wang W, Powers JG(2007) A description of the Advanced Research WRF Version 2, NCAR Technical Note NCAR/TN-468 + STR, NCAR, Boulder, CO, USA. Xue M, Droegemeier KK, Wong V (2000) The Advanced Regional Prediction System (ARPS) – a multi-scale non hydrostatic atmospheric simulation and prediction model. Part I: model dynamics and verification, Met. Atmos. Phys. 75, 161–193.
Discussion A. Vargas: About radon exhalation rate my question is how is considered the local scale since they use homogeneous value of this rate and in fact it is heterogeneous. J.-P. Chollet: Radon was measured at two locations, about 3 km away one from the other, and values were observed to be very close ones to the others at least during the cold episodes under consideration. So radon appears to be rather homogeneous at least in the extent of the town.
1.1 On-line Integrated Meteorological and Chemical Transport Modelling: Advantages and Prospectives Alexander Baklanov and Ulrik Korsholm
Abstract The strategy for developing new-generation integrated Meso-Meteorological (MetM) and Atmospheric Chemical Transport Model (CTM) systems is discussed and an overview of the European COST728 (http://www.cost728.org) integrated systems is given. Advantages and disadvantages of on-line integration versus the more common off-line coupling of MetMs and CTMs are mentioned using DMI-ENVIRO-HIRLAM (HIgh Resolution Limited Area Model) as a specific example. Current progress in the DMI-ENVIRO-HIRLAM system development and its urban on-line coupled modelling applications are considered. Several sensitivity tests of off-line versus on-line coupling in DMI-ENVIRO-HIRLAM as well as verification versus the ETEX experiment are considered, and results are discussed. Keywords Aerosol feedbacks, chemical weather forecast, climate change, ENVIRO-HIRLAM system, integrated models, on-line coupling
1. Introduction Historically air pollution forecasting and numerical weather predictions (NWP) were developed separately. This was plausible in the previous decades when the resolution of NWP models was too poor for meso-scale air pollution forecasting. Due to modern NWP models approaching meso- and city-scale resolution (due to advances in computing power) and the use of land-use databases and remote sensing data with finer resolution, this situation is changing. As a result the conventional concepts of meso- and urban-scale air pollution forecasting need revision along the lines of integration of meso-scale meteorological models (MetMs) and chemical transport models (CTMs). For example, a new Environment Canada conception suggests to switch from weather forecasting to environment forecasting. Some European projects (e.g. FUMAPEX, see: fumapex.dmi.dk) already work in this direction and have set off on a promising path. In case of FUMAPEX it is the Urban Air Quality Information and Forecasting Systems (UAQIFS) integrating
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
3
4
A. Baklanov, U. Korsholm
NWP models, urban air pollution (UAP) and population exposure models (Baklanov, 2005; Baklanov et al., 2007b) (see Figure 1). In perspective, integrated NWP-CTM modelling may be a promising way for future atmospheric simulation systems leading to a new generation of models for improved meteorological, environmental and “chemical weather” forecasting. Both, off-line and on-line coupling of MetMs and CTMs are useful in different applications. Thus, a timely and innovative field of activity will be to assess their interfaces, and to establish a basis for their harmonization and benchmarking. It will consider methods for the aggregation of episodic results, model down-scaling as well as nesting. The activity will also address the requirements of meso-scale meteorological models suitable as input to air pollution models. The current COST728 Action (http://www.cost728.org) addresses key issues concerning the development of meso-scale modelling capabilities for air pollution and dispersion applications and, in particular, it encourages the advancement of science in terms of integration methodologies and strategies in Europe. The final integration strategy will not be focused around any particular model, instead it will be possible to consider an open integrated system with fixed architecture (module interface structure) and with a possibility of incorporating different MetMs/NWP models and CTMs. Such a strategy may only be realised through jointly agreed specifications of module structure for easy-to-use interfacing and integration. The overall aim of working group 2 (WG2), ‘Integrated systems of MetM and CTM: strategy, interfaces and module unification’, is to identify the requirements for the unification of MetM and CTM modules and to propose recommendations for a European strategy for integrated meso-scale modelling capabilities. The first report of WG2 (Baklanov et al., 2007a) compiles existing state-of-the-art methodologies, approaches, models and practices for building integrated (off-line and online) meso-scale systems in different, mainly European, countries. The report also includes an overview and a summary of existing integrated models and their characteristics as they are presently used. The model contributions were compiled using COST member contributions, each focussing on national model systems.
2. Methodology for Model Integration The modern strategy for integrating MetMs and CTMs is suggested to consider air quality modelling as a combination of (at least) the following factors: air pollution, regional/urban climate/meteorological conditions and population exposure. This combination is reasonable due to the following facts: meteorology is the main source of uncertainty in air pollution and emergency preparedness models, meteorological and pollution components have complex and combined effects on human health (e.g., hot spots in Paris, July 2003), pollutants, especially aerosols, influence climate forcing and meteorological events (precipitation, thunderstorms, etc.).
On-line Integrated Meteorological and Chemical Transport Modelling
5
In this context, several levels of MetM and CTM coupling/integration can be considered: Off-line: x Separate CTMs driven by meteorological input data from meteo-preprocessors, measurements or diagnostic models x Separate CTMs driven by analysed or forecasted meteodata from NWP archives or datasets x Separate CTMs reading output-files from operational NWP models or specific MetMs at limited time intervals (e.g. 1, 3, 6 hours) On-line: x On-line access models, when meteodata are available at each time-step (possibly via a model interface as well) x On-line integration of CTM into MetM, where feedbacks may be considered. We will use this definition for on-line coupled/integrated modelling FUMAPEX UAQIFS: Meteorological models for urban areas Soil and Urban heat flux parametrisation sublayer models for urban areas
Module of feedback mechamisms: - Direct gas & aerosol forcing - Cloud condensation nuclei model - Other semidirect & indirect effects
Urban roughness classification & parameterisation
Usage of satellite information on surface
Meso- / City - scale NWP models
Interface to Urban Air Pollution models Mixing height and eddy diffusivity estimation
Down -scaled models or ABL parameteris ations
Estimation of additional advanced meteorological parameters for UAP
Grid adaptation and interpol ation, assimilatio n of NWP data
All 3D meteorological & surface fields are available at each time step
Urban Air Pollution models
Population Exposure models Populations/ Groups
Microenvironments
Outdoor concentrations Indoor concentrations
Exposure
Time activity
Fig. 1 Extended FUMAPEX scheme of Urban Air Quality Information & Forecasting System (UAQIFS) including feedbacks. Improvements of meteorological forecasts (NWP) in urban areas, interfaces and integration with UAP and population exposure models following the off-line or online integration (Baklanov, 2005; after EMS-FUMAPEX, 2005)
The main advantages of the on-line coupled modelling approach comprise: • Only one grid; no interpolation in space • No time interpolation • Physical parameterizations and numerical schemes (e.g. for advection) are the same; no inconsistencies • All 3D meteorological variables are available at the right time (each time step)
6
A. Baklanov, U. Korsholm
• No restriction in variability of meteorological fields • Possibility to consider feedback mechanisms, e.g. aerosol forcing • Does not need meteo- pre/post-processors However, not always the on-line approach is the best way of the model integration. For some specific tasks (e.g., for emergency preparedness, when NWP data are available) the off-line coupling is more efficient way. The main advantages of offline models comprise: • Possibility of independent parameterizations • More suitable for ensembles activities • Easier to use for the inverse modelling and adjoint problem • Independence of atmospheric pollution model runs on meteorological model computations • More flexible grid construction and generation for ACT models • Suitable for emission scenarios analysis and air quality management The on-line integration of meso-scale meteorological models and atmospheric aerosol and chemical transport models enables the utilisation of all meteorological 3D fields in CTMs at each time step and the consideration of feedbacks between air pollution (e.g. urban aerosols), meteorological processes and climate forcing. These integration methodologies have been realised by several of the COST action partners such as the Danish Meteorological Institute, with the DMI-ENVIROHIRLAM model (Chenevez et al., 2004; Baklanov et al., 2004; Korsholm et al., 2007) and the COSMO consortium with the Lokal Modell (Vogel et al., 2006; Wolke et al., 2003). These model developments will lead to a new generation of integrated models for: climate change modelling, weather forecasting (e.g., in urban areas, severe weather events, etc.), air quality, long-term assessments of chemical composition and chemical weather forecasting (an activity of increasing importance which is due to be supported by a COST action starting in 2007).
3. Overview of European On-line Integrated Models The experience from other European as well as non-European union communities will need to be integrated. On-line coupling was first employed at the Novosibirsk scientific school (Marchuk, 1982; Penenko and Aloyan, 1985; Baklanov, 1988), for modelling active artificial/anthropogenic impacts on atmospheric processes. Currently American, Canadian and Japanese institutions develop and use on-line coupled or on-line access models operationally for air quality forecasting and for research (GATOR-MMTD: Jacobson, 2005, 2006; WRF-Chem: Grell et al., 2005; GEM-AQ: Yeh et al., 2002; CFORS: Uno et al., 2003, 2004). Such activities in Europe are widely dispersed and a COST Action seems to be the best approach to integrate, streamline and harmonize these national efforts
On-line Integrated Meteorological and Chemical Transport Modelling
7
towards a leap forward for new breakthroughs beneficial for a wide community of scientists and users. Such model integration should be realized following a joint elaborated specification of module structure for potential easy interfacing and integration. It might develop into a system, e.g. similar to the USA ESMF (Earth System Modelling Framework, see e.g.: Dickenson et al., 2002) or European PRISM (PRogram for Integrating Earth System Modelling) specification for integrated Earth System Models: http://prism.enes.org/ (Valcke et al., 2006). Community Earth System Models (COSMOS) is a major international project (http://cosmos.enes.org) involving different institutes in Europe, in the US and in Japan, for the development of complex Earth System Models (ESM). Such models are needed to understand large climate variations of the past and to predict future climate changes. The main differences between the COST728 integrating strategy for meso-scale models and the COSMOS integration strategy regard the spatial and temporal scales. COSMOS is focusing on climate time-scale processes, general (global and regional) atmospheric circulation models and atmosphere, ocean, cryosphere and biosphere integration, while the meso-scale integration strategy will focus on forecast time-scales of one to four days and omit the cryosphere and the larger temporal and spatial scales in atmosphere, ocean and biosphere. The COST728 overview (Baklanov et al., 2007) shows a surprisingly large (at least 10) number of on-line coupled MetM and CTM model systems already being used in Europe: x x x x x x x x x x
BOLCHEM (CNR ISAC, Italy) DMI-ENVIRO-HIRLAM (DMI, Denmark) LM-ART (Inst. for Meteorology and Climatology, FZ Karlsruhe, Germany) LM-MUSCAT (IfT Leipzig, Germany) MCCM (Inst. of Environmental Atmospheric Research at FZ Karlsruhe, Germany) MESSy: ECHAM5 (MPI-C Mainz, Germany) MC2-AQ (York Univ, Toronto, University of British Columbia, Canada, and Warsaw University of Technology, Poland) GEM/LAM-AQ (York Univ, Toronto, University of British Columbia, Canada, and Warsaw University of Technology, Poland) WRF-CHem: Weather Research and Forecast and Chemistry Community modelling system (NCAR and many other organisations) MESSy: ECHAM5-Lokalmodell LM planned at MPI-C Mainz, Univ. of Bonn, Germany
However, it is necessary to mention, that many of the above on-line models were not built for the meso-meteorological scale, and several of them (GME, ECMWF GEMS, MESSy) are global-scale modelling systems, originating from the climate modelling community. Besides, at the current stage most of the on-line coupled models do not consider feedback mechanisms or include only simple direct effects of aerosols on meteorological processes (COSMO LM-ART and MCCM). Only
8
A. Baklanov, U. Korsholm
two meso-scale on-line integrated modelling systems (WRF-Chem and DMIENVIRO-HIRLAM) consider feedbacks with indirect effects of aerosols.
4. Feedback Mechanisms, Aerosol Forcing in MetMs In a general sense air quality and CTM modelling is a natural part of the climate change and MetM/NWP modelling. The role of greenhouse gases (such as water vapour, CO2, O3 and CH4) and aerosols in climate change has been highlighted as a key area of future research (Watson et al., 1997; IPCC, 2001; AIRES, 2001). Uncertainties in emission projections of gaseous pollutants and aerosols (especially secondary organic components) need to be addressed urgently to advance our understanding of climate forcing (Semazzi, 2003). In relation to aerosols, their diverse sources, complex physicochemical characteristics and large spatial gradients make their role in climate forcing particularly challenging to quantify. In addition to primary emissions, secondary particles, such as, nitrates, sulphates and organic compounds, also result from chemical reactions involving precursor gases such as SOx, DMS, NOx, volatile organic compounds and oxidising agents including ozone. One consequence of the diverse nature of aerosols is that they exhibit negative (eg sulphates) as well as positive (eg black carbon) radiative forcing characteristics (IPCC, 2005; Jacobson, 2005). Although much effort has been directed towards gaseous species, considerable uncertainties remain in size dependent aerosol compositional data, physical properties as well as processes controlling their transport and transformation, all of which affect the composition of the atmosphere (Penner et al., 1998; Shine, 2000; IPCC, 2001). Probably one of the most important sources of uncertainty relates to the indirect effect of aerosols as they also contribute to multiphase and microphysical cloud processes, which are of considerable importance to the global radiative balance (Semazzi, 2003). In addition to better parameterisations of key processes, improvements are required in regional and global scale atmospheric modelling (Semazzi, 2003). Resolution of regional climate information from atmosphere-ocean general circulation models remains a limiting factor. Vertical profiles of temperature, for example, in climate and air quality models need to be better described. Such limitations hinder the prospect of reliably distinguishing between natural variability (e.g. due to natural forcing agents, solar irradiance and volcanic effects) and human induced changes caused by emissions of greenhouse gases and aerosols over multidecadal timescales (Semazzi, 2003). Consequently, the current predictions of the impact of air pollutants on climate, air quality and ecosystems or of extreme events are unreliable (e.g. Watson et al., 1997). Therefore it is very important in the future research to address all the key areas of uncertainties so as provide an improved modelling capability over regional and global scales and an improved integrated assessment methodology for formulating mitigation and adaptation strategies. In this concern one of the important tasks is to develop a modelling instrument of coupled ‘Atmospheric chemistry/Aerosol’ and ‘Atmospheric Dynamics/Climate’
On-line Integrated Meteorological and Chemical Transport Modelling
9
models for integrated studies, which is able to consider the feedback mechanisms, e.g. aerosol forcing (direct and indirect) on the meteorological processes and climate change (see Figure 2). WP5
Emission databases, models and scenarios
Inverse methods and adjoint models
Atmospheric chemistry and transport models
Aerosol dynamics models
Radiative & optic properties models
Ocean dynamics model
Cloud condensation nuclei (CCN) model
General Circulation & Climate models
Ecosystem models
WP7
Integrated Assessment Model
Fig. 2 The integrated system structure for studies of the meso-scale meteorology and air pollution, and their interaction
Chemical species influencing weather and atmospheric processes include greenhouse gases which warm near-surface air and aerosols such as sea salt, dust, primary and secondary particles of anthropogenic and natural origin. Some aerosol particle components (black carbon, iron, aluminium, polycyclic and nitrated aromatic compounds) warm the air by absorbing solar and thermal-IR radiation, while others (water, sulphate, nitrate, most of organic compounds) cool the air by backscattering incident short-wave radiation to space. It is necessary to highlight, that effects of aerosols and other chemical species on meteorological parameters have many different pathways (direct, indirect, semidirect effects, etc.) and they have to be prioritised and considered in on-line coupled modelling systems. Following Jacobson (2002) the following effects of aerosol particles on meteorology and climate can be distinguished: • Self-feedback effect • Photochemistry effect • Smudge-pot effect • Daytime stability effect • Particle effect through surface albedo
A. Baklanov, U. Korsholm
10
• Particle effect through large-scale meteorology • Indirect effect • Semidirect effect • BC-low-cloud-positive feedback loop Sensitivity studies are needed to understand the relative importance of different feedback mechanisms. Implementation of the feedbacks into integrated models could be realized in different ways with varying complexity. The following variants serve as examples: One-way integration (off-line): x The chemical composition fields from CTMs may be used as a driver for Regional/Global Climate Models, including aerosol forcing on meteorological processes. This strategy could also be realized for NWP or MetMs. Two-way integration: x Driver and partly aerosol feedbacks, for CTMs or for NWP (data exchange with a limited time period); off-line or on-line access coupling, with or without the following iterations with corrected fields). x Full feedbacks included on each time step (on-line coupling/integration). For the realisation of all aerosol forcing mechanisms in integrated systems it is necessary to improve not only CTMs, but also NWP/MetMs. The boundary layer structure and processes, including radiation transfer, cloud development and precipitation must be improved. Convection and condensation schemes need to be adjusted to take the aerosol-microphysical interactions into account, and the radiation scheme needs to be modified to include the aerosol effects.
5. The On-line Coupled DMI-ENVIRO-HIRLAM System Currently the Danish Meteorological Institute (DMI) is developing a new version of the meteorological model HIRLAM which includes on-line coupled tracers (DMI-ENVIRO-HIRLAM) (based on Chenevez et al., 2004) and has implemented a versatile aerosol-cloud module and heterogeneous chemistry in their Atmospheric Chemical Transport Model ‘Cloud-Aerosol-Chemistry’ (CTM-CAC) (Gross and Baklanov, 2004). Implementation of the CTM-CAC in the DMI-ENVIRO-HIRLAM makes the inclusion of regional to urban scale feedbacks between the CTM-CAC and DMI-HIRLAM possible (see the red box and dashed arrows in Figure 1) (Baklanov et al., 2004, 2008; Korsholm et al., 2007). Simplified feedback mechanisms, including direct and indirect aerosol forcing, are implemented in the current version (Figure 3) of the DMI-ENVIRO-HIRLAM integrated model (Korsholm et al., 2007).
On-line Integrated Meteorological and Chemical Transport Modelling
11
-GEMS/TNO -EMEP
CAC-Aerosol Dynamics Modal approach model Log-normal modes: nuclei, accumulation, coarse Moment equations: Intra-modal coagulation, Inter-modal coagulation, condensation, nucleation
CHEM Gas-phase chemistry: RADM, RACM, CBMZ Aerosol dynamics: MOSAIC, SORGAM Photolysis: Madronich Cloud chemistry Convection Deposition Plumerise
Fig. 3 Current version of the DMI-ENVIRO-HIRLAM modelling systems, showing the components of a forecast
DMI-ENVIRO-HIRLAM is developing as an on-line integrated system with a possibility of the off-line coupling as well. The system realisation includes the following steps: (i) Nesting of models for high resolutions (ii) Improved resolution of boundary and surface layer characteristics and structures (ii) ‘Urbanisation’ of the model (iii) Improvement of advection schemes (iv) Implementation of chemical mechanisms (v) Implementation of aerosol dynamics (vi) Realisation of feedback mechanisms (vii) Assimilation of monitoring data The model is to be used for operational as well as research purposes and will comprise aerosol and gas transport, dispersion and deposition, aerosol physics and chemistry, as well as gas-phase chemistry. On-line versus off-line sensitivity tests have been performed with DMI-ENVIROHIRLAM along with verification versus the European Tracer Experiment and the Chernobyl accident (Korsholm et al., 2006, 2007). Preliminary tests (see e.g., Figures 4 and 5) show that on-line integration of meso-scale meteorological models, atmospheric aerosol and chemical transport models including feedbacks of air pollution (e.g. urban aerosols) on meteorological processes and urban climate is a promising way for future systems of atmospheric environment forecasting.
12
A. Baklanov, U. Korsholm
Fig. 4 One example of on-line versus off-line DMI-ENVIRO-HIRLAM runs for the ETEX experiment: normalized mean square difference (Korsholm et al., 2006) at ETEX station DK02. Simple test: only Advection part is differently coupled; on-line: each time step; off-line: 0.5, 1, 2, 4, 6, 12, 24 hours; typical meteorological conditions (ETEX-1 case)
Using the current version of DMI-ENVIRO-HIRLAM on-line versus off-line coupling was tested for the ETEX experiment. A simple test was considered where only the advection part was differently coupled: on-line (updated each time step) or off-line, with coupling intervals of 0.5, 1, 2, 4, 6, 12 or 24 hours. Figure 4 shows an example of the simulations; the normalized mean square difference (Korsholm et al., 2007). As we can see for this simple case, the difference becomes significant for a coupling interval of a few hours. Another test included feedbacks through the first indirect effect (Korsholm et al., 2006). Urban sulphate particle emissions were considered and the effect of including the aerosol feedback mechanism on deposition was examined (reference run without feedbacks, perturbed run including feedbacks). Figure 5 shows the difference fields (reference – perturbation) for the accumulated wet deposition (ng/m2). This preliminary case study suggested that indirect effects modulate dispersion by affecting atmospheric stability.
On-line Integrated Meteorological and Chemical Transport Modelling
13
Fig. 5 Another example of On-line versus Off-line DMI-ENVIRO-HIRLAM runs for urban aerosol forcing test (ref: no feedbacks, perturbation: feedback included): Difference (ref: perturbation) in accumulated wet deposition (ng/m2)
7. Concluding Remarks The on-line integration of meso-scale meteorological models and atmospheric aerosol and chemical transport models enables the utilisation of all meteorological 3D fields in CTMs at each time step and the consideration of the feedbacks of air pollution (e.g. urban aerosols) on meteorological processes and climate forcing. These on-line coupled model developments will lead to a new generation of integrated models for climate change modelling, weather forecasting (e.g., in urban areas, severe weather events, etc.), air quality, long-term assessment chemical composition and chemical weather forecasting. Main advantages of the on-line modelling approach include: x Only one grid; No interpolation in space x No time interpolation x Physical parameterizations and numerical schemes are the same; No inconsistencies x All 3D meteorological variables are available at the right time (each time step) x No restriction in variability of meteorological fields x Possibility to consider feedback mechanisms x Does not need meteo- pre/post-processors
14
A. Baklanov, U. Korsholm
While the main advantages of the off-line approach include: x Possibility of independent parameterizations x More suitable for ensemble activities x Easier to use for the inverse modelling and adjoint problem x Independence of atmospheric pollution model runs on meteorological model computations x More flexible grid construction and generation for ACT models x Suitable for emission scenarios analysis and air quality management The COST728 overview shows a quite surprising number of on-line coupled MetM and CTM model systems already being used in Europe. However, many of the online coupled models were not built for the meso-meteorological scale, and they (e.g. GME, ECMWF GEMS, MESSy) are global-scale modelling systems and first of all designed for climate change modelling. Besides, at the current stage most of the online coupled models do not consider feedback mechanisms or include only direct effects of aerosols on meteorological processes (like COSMO LM-ART and MCCM). Only two meso-scale on-line integrated modelling systems (WRF-Chem and ENVIRO-HIRLAM) consider feedbacks with indirect effects of aerosols. The realisation of the on-line integration was demonstrated using the DMIENVIRO-HIRLAM integrated system. Our preliminary tests of the on-line versus off-line integrated versions of DMI-ENVIRO-HIRLAM showed that the on-line integration of MetMs and ACTMs with consideration of feedbacks between air pollution (e.g. urban aerosols), meteorological processes and urban climate is a promising way for the development of future systems of atmospheric environment forecasting. Acknowledgments This study was supported by the COST Action 728 and the Copenhagen Global Change Initiative (COGCI). The authors are grateful to a number of COST728, FUMAPEX and DMI colleagues, who participated in the above-mentioned projects, for productive collaboration and discussions.
References AIRES (2001) AIRES in ERA, European Commission, EUR 19436. Baklanov A (1988) Numerical modelling in mine aerology, Apatity: USSR Academy of Science, 200 p. (in Russian). Baklanov A (2005) Meteorological advances and systems for urban air quality forecasting and assessments. Short Papers of the 5th International Conference on Urban Air Quality Valencia, Spain, 29–31 March 2005, CLEAR, pp. 22–25. Baklanov A, Gross A, Sørensen JH (2004) Modelling and forecasting of regional and urban air quality and microclimate. J. Comput. Technol., 9:82–97. Baklanov A, Fay B, Kaminski J, Sokhi R (2007a) Overview of existing integrated (off-line and on-line) mesoscale meteorological and chemical transport modelling systems in Europe. Joint Report of COST Action 728 and GURME, May 2007. WMO TD No. 1427, GAW Report No. 177. Available from http://www.cost728. org
On-line Integrated Meteorological and Chemical Transport Modelling
15
Baklanov A, Hänninen O, Slørdal LH, Kukkonen J, Bjergene N, Fay B, Finardi S, Hoe SC, Jantunen M, Karppinen A, Rasmussen A, Skouloudis A, Sokhi RS, Sørensen JH, Ødegaard V (2007b) Integrated systems for forecasting urban meteorology, air pollution and population exposure. Atmos. Chem. Phys., 7:855–874. Baklanov A, Korsholm U, Mahura A, Petersen C, Lindberg K, Gross A, Rasmussen A, Sørensen JH, Amstrup B, Chenevez J (2008) ENVIRO-HIRLAM on-line coupled modelling of urban meteorology and air pollution. Adv. Sci. Res., 2, 41–46. Chenevez J, Baklanov A, Sørensen JH (2004) Pollutant transport schemes integrated in a numerical weather prediction model: model description and verification results. Meteorol. Appl., 11(3):265–275. COSMOS: Community Earth System Models Integrating strategy. Web-site: http://cosmos.enes.org Dickenson RE, Zebiak SE, Anderson JL, Blackmon ML, DeLuca C, Hogan TF, Iredell M, Ji M, Rood R, Suarez MJ, Taylor KE (2002) How can we advance our weather and climate models as a community? Bull. Am. Met. Soc., 83:431–434. EMS-FUMAPEX (2005) “Urban Meteorology and Atmospheric Pollution”, Baklanov A, Joffre S, Galmarini S (Eds.). Special Issue of Atmospheric Chemistry and Physics Journal. Grell GA, Peckham SE, Schmitz R, McKeen SA, Frost G, Skamarock WC, Eder B (2005) Fully coupled “online” chemistry within the WRF model. Atmos. Environ., 39(37):6957–6975. Gross A, Baklanov A (2004) Modelling the influence of dimethyl sulphide on the aerosol production in the marine boundary layer. Int. J. Environ. Pollut., 22(1/2):51–71. IPCC (2001) Climate Change 2001, The Scientific Basis, Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change (IPCC), edited by L. Houghton et al., Cambridge University Press, Cambridge, United Kingdom/New York. IPCC (2005) IPCC Expert Meeting on Emission Estimation of Aerosols Relevant to Climate Change held on 2–4 May 2005, Geneva, Switzerland. Korsholm U, Baklanov A, Mahura A, Petersen C, Lindberg K, Gross A, Rasmussen A, Sørensen JH, Chenevez J (2006) ENVIRO-HIRLAM. An On-Line Coupled Multi-Purpose Environment Model. ACCENT/GLOREAM Workshop 2006 Proceedings. http://euler.lmd.polytechnique.fr/gloream/ Korsholm U, Baklanov A, Gross A, Sørensen JH (2007) On the importance of the meteorological coupling interval in air pollution modeling, submitted to Atm. Env.: Special Issue COST728, UAQ2007. Jacobson MZ (2002) Atmospheric Pollution: History, Science and Regulation. Cambridge University Press, New York. Jacobson MZ (2005) Fundamentals of Atmospheric Modeling, Second Edition. Cambridge University Press, New York, 813 pp.
16
A. Baklanov, U. Korsholm
Jacobson MZ (2006) Comment on “Fully coupled ‘online’ chemistry within the WRF model,” by Grell et al., Atmos. Environ., 39:6957–6975. Marchuk GI (1982) Mathematical modeling in the environmental problems. Nauka, Moscow. Penenko VV, Aloyan AE (1985) Models and methods for environment protection problemsNovosibirsk, Nauka (in Russian). Penner JE et al. (1998) Climate forcing by carbonaceous and sulphate aerosols. Clim. Dynam., 14:839–851. Uno I et al. (2004) Numerical study of Asian dust transport during the springtime of 2001 simulated with the Chemical Weather Forecasting System (CFORS) model. J. Geophys. Res., 109, D19S24, doi:10.1029/2003JD004222. Uno I et al., (2003) Regional chemical weather forecasting system CFORS: model descriptions and analysis of surface observations at Japanese island stations during the ACE-Asia experiment, J. Geophys. Res., 108 (D23), 8668, doi: 10.1029/2002JD002845. Semazzi F (2003) Air quality research: perspective from climate change modelling research. Environment International, 29:253–261. Shine KP (2000) Radiative forcing of climate change. Space Sci. Rev. 94:363–373. Valcke S, Guilyardi E, Larsson C (2006) PRISM and ENES: a European approach to Earth system modelling. Concurrency. Comput. Pract. Exp. 18:231–245. Vogel B, Hoose C, Vogel H, Kottmeier Ch (2006) A model of dust transport applied to the Dead Sea area. Meteorologische Zeitschrift, 14:611–624. Watson RT et al. (1997) The regional impacts of climate change: an assessment of vulnerability. Special Report for the Intergovernmental Panel on Climate Change. Wolke R, Hellmuth O, Knoth O, Schröder W, Heinrich B, Renner E (2003) The chemistry-transport modeling system LM-MUSCAT: description and CITYDELTA applications. Proceedings of the 26th International Technical Meeting on Air Pollution and Its Application. Istanbul, May 2003, 369–379. Yeh K-S, Cote J, Gravel S, Methot A, Patoine A, Roch M, Staniforth A (2002) The CMC-MRB global environmental multiscale (GEM) model. Part III: Nonhydrostatic formulation. Mon. Wea. Rev., 130, 2, 339–356.
Discussion D. Steyn: In all cases of integrated met/chem modelling, we first evaluate the veracity of meteorological fields, then the chemical fields (in the same way we did before we had integrated models). We usually do this by simply switching off the chemistry. Is there a less crude way of doing this within the capability of an integrated model?
On-line Integrated Meteorological and Chemical Transport Modelling
17
A. Baklanov: As the first step this way is also used for evaluations of integrated models. However, it is not enough to validate separately the meteorological model (with switched off ACTM) and ACTM driven by meteorological fields from NWP model without including aerosols and other precursors feedbacks. So, when we include the feedback mechanisms into integrated models we also have to validate additionnally both parts including the feedbacks in both directions. I have to say that not always the including aerosol feedback mechanisms will immediately improve the meteorological forecast, because in most of NWP and MetMs some effects of aerosols on clouds and radiation very roughly already indirectly included into experimental coefficients of parameterisations for cloud and radiation processes. So, in integrated models we should re-tune some constants to avoid double counting the aerosol effects, etc. Currently the COST Action 728 ‘Meso-meteorology for atmospheric pollution modelling’ makes an inventory of validation approaches and will build recommendations for the integrated model quality assurance procedure. P. Builtjes: You make the remark that inverse modelling/chemical data assimilation is easier for off-line than for on-line. Can you explain why that is? A. Baklanov: For on-line integrated system you have to consider the adjoint problem for both meteorological and ACTM equations, this is more difficult due to the nonlinear nature of atmospheric dynamic processes. For off-line models (e.g. for source-term estimation or chemical data assimilation) it is enough to consider the adjoint problem only for ACTM part.
1.8 Origin and Influence of PM10 Concentrations in Urban and in Rural Environments Andreas Kerschbaumer, Rainer Stern and Martin Lutz
Abstract In order to derive measures to effectively reduce human exposure to PM10 pollution it is vital to understand how different components of particulate matter (PM) are related to distinct sources and physico-chemical processes in the atmosphere. Secondary inorganic aerosols like sulphate, nitrate and ammonium are related to emission and transformation of gaseous species, SO2, NOx and NH3, respectively. Secondary organic aerosol is related to emission and transformation of VOC, while elemental carbon and primary organic compounds stem from direct emissions and the mineral fraction from dust entrainment. Thus, PM10 concentrations are determined by different sources at different distances from a considered receptor site. The aim of this study is to analyse the importance of an extended urban agglomeration for the pollution in the rural surrounding areas and, vice versa, the relevance of remote sources for PM levels in the urban background. It could be shown that PM10 concentration levels in urban environment are influenced by long range transport form remote source areas. However, the same holds for rural background concentration levels which are strongly determined by urban emissions. High resolution measurements have shown that long range transport is responsible for up to 70% of the city PM10 background. The Aerosol Chemistry Transport Model REM_Calgrid (RCG) has been used to simulate the impact of local and remote emissions on the PM10 concentration field. Berlin emissions have been switched off in order to estimate the local influence on the surroundings areas. It has been shown that two thirds of the urban background concentration in the Berlin’s centre and one third of the PM10 levels in the suburbs are due to Berlinspecific PM10 and PM-precursor emissions. While only 10% of the secondary inorganic and organic aerosol concentrations in the city can be traced back to cityrelated emissions, more than 60% of primary particles in the centre and only 30% at the outskirts come from city sources. On the other hand, city-related emissions influence homogenously the rural air-pollution concentrations. In conclusion, while abatement measures to curb primary PM10 emissions ought to be focused on urban agglomerations, large scale coordinated emission control of precursor components of secondary PM would contribute to an essential reduction of the widespread exposure of the urban and the rural population to PM10. Keywords Aerosol concentration, emission reduction potential, urban environment
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
72
Origin and Influence of PM10 Concentrations in Urban and in Rural Environments
73
1. Introduction High particulate matter concentrations are commonly observed in large European cities, causing concern over short-and long-term population exposure. Within the EU, particulate matter is currently regulated as 24-hour mean PM10 (50 µg m-3 not to be exceeded more than 35 times per year) and annual mean PM10 in air (40 µg m-3 limit value) (Council Directive 1999/30/EC). At present, the PM10 limit values do not discriminate between the many different anthropogenic and natural sources of particles. Consequently, routinely collected PM10 data are of limited use in developing air pollution control strategies. Effective PM10 control strategies require detailed knowledge of specific particle sources in a given environment. State-of-the-Art Chemistry transport models (CTM) simulate the composition of PM10 departing from anthropogenic and biogenic precursors and are able to reproduce the primary and secondary aerosol fractions. Furthermore, the spatial extension of Eulerian Grid Models allows to follow the pathways of remotely emitted species and to calculate the impact on locations of concern. Mathematical models allow effectively source-receptor calculations, where emissions for each emitter of one or more precursors are reduced. This technique is widely used for nation-wide transboundary considerations. In individual countries the emissions of primary particles will be switched off and the emissions of oxidised (NOx) and reduced nitrogen (NHx) and oxidised sulphur (SOx) will be reduced by approximately 15–20% (Fowler et al., 2005). This only small reduction is to account for the non-linearity in atmospheric pollution chemistry. The differences between the not-disturbed simulation and the scenario simulation give an indication of the impact of these species on the air pollution situation in all other countries. Typical spatial resolution is approximately between 30 and 100 km. We adopted the same philosophy on a local scale using a CTM with a horizontal resolution of 4 km.
2. Method 2.1. Model domain Approximately 3.4 Mio of inhabitants live in the 892 km² of the Berlin territory while only 2.6 Mio people live in the 29,479 km² of the Brandenburg region. In the inner Berlin circle with an extension of about 100 km² there live approximately 1.1 Mio people. The city is located in the Northern German. The main wind direction is from South-West to North-West (51%) with mean wind speed of 3.7 m/s. The industrial areas are mainly concentrated inside the Berlin city borders. Considering these characteristics, Berlin may be regarded as an excellent natural laboratory for analysing the impact of a big city on the mostly rural surrounding environment on relatively short distances. On the other side, the influence of remote sources on the Berlin air quality may be regarded as long range transport as there are no
74
A. Kerschbaumer et al.
significant pollution sources within a radius of approximately 300 km around the city.
2.2. Aerosol-chemistry-transport model REM_Calgrid (RCG) In order to calculate the source-receptor relationship between city-related PM10 precursors the chemical transport model REM_Calgrid (RCG) (Stern et al., 2003; Beekmann et al., 2007) has been used. PM10 is simulated considering different chemical fractions: PM10 = PMcoarse + PMprim25 + EC + OCprim + SOA+SO42- + NO3-+ NH4+ + Na+ + Cl(1) For efficiency reasons, only two sizes are considered: less than 2.5 µm and sizes between 2.5 and 10 µm. For PMcoarse is mainly mineral coarse particles between 2.5 and 10 µm diameter, sodium and chloride (sea salt). The equilibrium between solid, aqueous and gas phase concentrations for inorganic ions as a function of temperature and humidity is calculated on-line with the ISORROPIA thermodynamic module (Nenes et al., 1999). Production of secondary organic aerosol (SOA) from anthropogenic and biogenic VOC is treated with the SORGAM module (Schell et al., 2001). The aerosol scheme also includes resuspension of mineral aerosol as a function of friction velocity and the nature of soil; both the direct entrainment of small particles (Loosmore and Hunt, 2000) and saltation, i.e. the indirect entrainment due to large particles which fall back to the soil and entrain smaller particles (Claiborn et al., 1998) is taken into account. The sea-salt aerosol emissions (Na+, Cl–) are parameterized according to Gong et al. (1997) as a function of size and wind speed.
2.3. Simulations PM10 concentration were simulated for the meteorological year 2005 based on the most recent emission reports delivered by Berlin local authorities. Scenario runs were performed reducing and switching off totally the emissions in the only Berlin area. This means that an important emission-source-area, i.e. the highway ring around Berlin lying in the Brandenburg Area, has not been taken into account when deciding to reduce emissions. This decision has been taken because we wanted to examine the only contribution of the administrative city of Berlin to the air quality of the surrounding areas. We considered a total reduction of all Berlin emissions. For sensitivity purposes we also considered a 50% NOx-reduction of all cityrelated emitters. We analysed separately primary and secondary constituents.
Origin and Influence of PM10 Concentrations in Urban and in Rural Environments
75
3. Results Simulated Berlin aerosol burden, on a yearly average, is approximately 26 µg m-3 in the city-centre and ca. 15 µg m-3 in the suburbs. The rural environment exhibits a typical PM10 concentration of about 14 µg m-3. Compared to a typical local-scale pollutant like NOx, this spatial gradient is rather low. Different PM10 components, however, show very different urban-rural-behaviours: EC, e.g., is three times higher in the city than in the rural area (Figure 1)., mineral particles in the fine fraction are calculated to be four times higher in the city than in the back-country and primary mineral particles in the coarse fraction are simulated five times higher than in the rural background. On the other hand, secondary aerosols show no urbanrural gradient (Figure 1).
Fig. 1 Elemental carbon (left) and secondary aerosols (SAER) (right) spatial annual distribution. Isolines in µg m-³
Figure 2 shows the maximum Berlin PM10-concentration-reduction-potential, i.e. emissions in the Berlin city area have been switched off and differences between base-case and this emission scenario have been plotted.
Fig. 2 PM10 reduction potential in Berlin in percentages (isolines)
76
A. Kerschbaumer et al.
The radial PM10 reduction is evident. The city centre shows a reduction potential of approximately 33%, while the city edges would exhibit approximately 10% less PM10 concentrations if there were no Berlin emissions at all. The influence of Berlin emissions on the Brandenburg environment is more the 5% for ca. 20 km with longer range of influence toward the north-east and shorter range of influence in the south-west. This may be due to the meteorological situation: there is a higher frequency of western wind directions. On the other side, the Berlin high-way ring in the south of Berlin is completely in the Brandenburg region, whereas there is a part of this high-way lying inside the Berlin domain in the North-East.
Fig. 3 Primary PM25 (left) and elemental carbon (EC) (right) reduction potential in Berlin in percentages (isolines)
The Berlin influence is by far more important considering the only primary component in the PM10 agglomerate. Figure 3 shows the Berlin concentration reduction potential considering primary particles smaller than 2.5 µm which stem mainly from mineral components in the RCG-model. EC influence ranges more toward north east, which is due to the presence of the high-way in the Berlin domain in the north-east, while primary PM25 shows a more radial behaviour. This may be explained by the stronger influence of natural soils in the mineral PM10component. A 10% reduction of EC pollution in Brandenburg due to only Berlin emissions is experienced still in a 100 km distance from the northern edge of the city.
Origin and Influence of PM10 Concentrations in Urban and in Rural Environments
77
Fig. 4 Nitrate (left) and sulfate (right) reduction potential in Berlin in percentages (isolines)
Figure 4 shows the percentaged reduction of particle nitrate and sulfate due to total emission elimination in Berlin. For both components there is a higher reduction inside the city of Berlin than in the Brandenburg Area, the reduction, however, is even in the city centre, smaller than 7%. The urban suburbs experience only a reduction of approximately 3% with no Berlin emission sources. As already seen for the total secondary aerosol spatial distribution (Figure 1), the gradient between urban and rural environment is very low. Also a total elimination of the secondary aerosol precursors does not influence strongly the spatial distribution of the secondary PM components.
80 PPM10
PPM10
70
PPM10:Prim. Components PM10:Total PM10 SAER:Sec. Components
60 50 %
PPM10
40
PM10
PPM10
PM10 PPM10
30 20
PM10
PM10 PM10
10
SAER
SAER
SAER
SAER
SAER
0 Centre
Urban backgr.
North. Outskirt
North
South
Fig. 5 Nitrate (left) and sulfate (right) reduction potential in Berlin in percentages (isolines)
Figure 5 summarises the above findings: Berlin maximum concentration potential – via hypothetical complete elimination of Berlin emissions – would lead to an overall 35% PM10 reduction in the centre of Berlin and some 11–16% in the southern and northern suburbs. This reduction potential varies heavily considering
78
A. Kerschbaumer et al.
primary and secondary PM components: primary particles could hypothetically be reduced to approximately 75% in the inner city and still to 35–40% in the outskirts, while secondary PM components even in the inner part of Berlin could be reduced only by approximately 5%, if there were no Berlin-specific pollutants.
4. Discussion Primary particles like EC, primary OC and mineral particles do not undergo any chemical transformation in RCG. Thus, an estimate of the potential PMconcentration reduction due to Berlin emissions for the primary components can be done eliminating totally Berlin emissions. In this way, also the spatial range of influence can be established. Secondary inorganic and organic aerosol components building processes are not linear. Thus, reducing one aerosol-precursor (NOx, SO2, NH3, VOC) only disturbs the aerosol chemistry. The secondary inorganic aerosol composition strongly depends on the availability of ammonia, on the presence of sulphate and nitrate ions. As Berlin is a prominent source of NOx especially because of traffic, the reduction of this pollutant is more feasible than the other two precursors. Thus, reduction of NOx by 50% in the city of Berlin has been explored in more detail.
Fig. 6 Nitrate (left) and sulfate (right) concentration reduction due to 50% NOx emission reduction in Berlin. Dashed isolines: concentration reductions, full isolines: increments (in percentages)
Figure 6 shows a sensitivity study reducing 50% of NOx emissions in the Berlin area. Aerosol nitrate concentrations do not change more than 0.2% over the whole simulation area while aerosol sulphate concentrations augment for maximum 0.6% in the inner city of Berlin. The long-range transport seems to be by far more important in the local secondary aerosol composition than local sources.
Origin and Influence of PM10 Concentrations in Urban and in Rural Environments
79
5. Summary and Outlooks PM10 simulations for Berlin show spatial maxima in the centre of the city and lower concentrations on the edges of the city. This is very species-dependent as for example secondary aerosols are distributed almost uniformly over the city and also over the rural environment outside Berlin. Maximum concentration reduction potential has been analysed distinctively for primary and secondary aerosol components diminishing the only Berlin emission sources by maximum 100%. Primary aerosols can be explained by Berlin specific emissions for more than 70% in the inner city and 30–40% on the city-edges. 10% spatial influences of Berlin emissions range for primary particles for approximately 100 km. Secondary aerosols seem to be much less influenced by local precursors. The 100% Berlin-emission reduction scenario – although altering the aerosol chemistry – shows only a maximum 5% influence on the inner city and almost no concentration change in the rural environment.
References Beekmann M, Kerschbaumer A, Reimer E, Stern R, Möller D (2007) PM measurement campaign HOVERT in the Greater Berlin area: model evaluation with chemically specified particulate matter observations for a one year period, Atmos. Chem. Phys., 7, 55–68. Claiborn C, Lamb B, Miller A, Beseda J, Clode B, Vaughan J, Kang L, Nevine C (1998) Regional measurements and modelling of windblown agricultural dust: The Columbia Plateau PM10 Program, J. Geophys. Res., 103(D16), 19753– 19767. Fowler D, Muller J, Smith RI, Cape JN, Erisman JW (2005) Nonlinearities in source receptor relationships for sulfur and nitrogen compounds. AMBIO: J. Hum. Environ., 34(1), 41–46. Gong SL, Barrie LA, Blanchet J-P (1997) Modelling sea-salt aerosols in the atmosphere. 1. Model development, J. Geophys. Res., 102, 3805–3818. Loosmore GA, Hunt JR (2000) Dust resuspension without saltation, J. Geophys. Res., 105(D16), 20663–20671, doi:10.1029/2000JD900271. Nenes A, Pilinis C, Pandis SN (1999) Continued development and testing of a new thermodynamic aerosol module for urban and regional air quality models, Atmos. Environ., 33, 1553–1560. Schell B, Ackermann IJ, Hass H, Binkowski F, Ebel A (2001) Modelling the formation of secondary organic aerosol within a comprehensive air quality model system, J. Geophys. Res., 106(D22), 28275–28293. Stern R, Yamartino R, Graff A (2003) Dispersion modelling within the European community’s air quality directives: long term modelling of O3, PM10 and NO2. In: Proceedings of the 26th ITM Conference on Air Pollution Modelling and its Application, May 26–30, 2003, Istanbul, Turkey.
80
A. Kerschbaumer et al.
Discussion B. Fisher: Measures are shown to have little effect on the PM concentrations in Berlin. Have you investigated the effect of measures taken on a regional scale, that is elsewhere in Germany and elsewhere in Europe, to reducing concentration in Berlin? A. Kerschbaumer: Local emission reduction measures influence only the primary part of PM10 which has been shown to be only about 40–50% of the total PM10 concentration. This 40–50%, however, can be reduced drastically by emission reduction scenarios. Regional scale measures influence the secondary PM10 homogeneously, thus also Berlin. S. Aksoyoglu: A. Kerschbaumer:
Which species in PM10 are underestimated by the model? Mainly the primary mineral part of PM10 is underestimated.
2.2 A Regional Air Quality Model over the Kanto Region of Japan: The Effect of the Physics Parameterization on the Meteorological and Chemical Fields Masanori Niwano, Masayuki Takigawa, Hajime Akimoto, Masaaki Takahashi and Mitsuhiro Teshiba
Abstract The effects of physics parameterization on meteorological and chemical fields were examined over the Kanto region of Japan using an air quality (AQ) model, which consists of two chemistry transport models: the global model CHASER and the regional model WRF/Chem. For hindcast experiments without a chemical module for June–July in 2005 and 2006, two non-local planetary boundary layer (PBL) schemes (Yonsei University YSU and NCEP GFS) showed a deeper PBL height and a stronger sea-breeze related with warmer surface temperature than for the observation and the local PBL scheme (Mellor-Yamada level 2.5). Two experiments with a chemical module for July–August in 2005 clarified that NCEP GFS scheme transported O3 plumes to higher altitudes above 2 km and also produced a higher concentration near the surface than for Mellor-Yamada scheme. Keywords Air pollution, land-sea breeze, planetary boundary layer
1. Introduction A lot of studies have devoted to understanding and forecasting air pollution in the metropolitan region. Over Japan, seasonal cycle of surface ozone (O3) shows a maximum in spring by the long-range transport from Asian countries and a minimum in summer by the transport of marine clean air. On the other hand, photochemical O3 episodes frequently occur in the inland country-side of the Kanto plain during the summer season (Kawamura, 1979). This is due to the “extended” landsea breeze which is a combination of land-sea breeze (LSB) and valley-mountain winds in the Kanto plain surrounded by western-northern mountains, in addition to the evolution of planetary boundary layer (PBL) and the coastal complex meteorology. While a global chemical weather forecast has been operational (Lawrence et al., 2003; Takigawa et al., 2005), a regional air quality (AQ) forecast over Japan C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
109
110
M. Niwano et al.
has not been operational until recent (Uno et al., 2003) because of the requirement of huge computational facility. A one-way nested AQ forecasting system has been developed with full chemistry and with an “on-line” calculation of meteorological and chemical modules (Takigawa et al., 2007). This study examines the effects of physical parameterization on local meteorology and chemical transport in PBL using the AQ model to present the feasibility of AQ forecast over the summer central.
2. Model Description and Experimental Design The model used in this study is a one-way nested AQ model which is composed of a global chemical transport model (CTM), CHASER (CHemical Atmospheric general circulation model for Study of atmospheric Environment and Radiative forcing) (Sudo et al., 2002), and a three-dimensional non-hydrostatic model coupled with CTM, Weather Research and Forecasting/Chemistry model (WRF/ Chem) version 2.1.2 (Grell et al., 2005). The domain in the regional model consists of two domains: the outer domain (D1) with 84 × 84 grids at 27-km resolution, and the inner domain (D2) with 39 × 39 grids at 9-km resolution (Figure 1). Both domains have 30 layers from the surface to about 100 hPa (and 11 layers below 2 km). The details of the AQ model are described in Takigawa et al. (2007). Initial and boundary conditions for meteorology in D1 were obtained from the analysis data of meso-scale model (MSM) of Japan Meteorological Agency (JMA) with a 3hour interval. Initial and boundary data for chemical field in D1 were obtained from output data of daily global chemical weather forecasting using CHASER with a 3hour interval. Hindcast experiments for 24 hours were conducted every day. At 00 Universal Time (UT), only the meteorological fields were initialized by the JMA MSM analysis. As a default set of physics parameterizations, the height and vertical diffusion of PBL are calculated by the Mellor-Yamada level 2.5 scheme (Janjiü, 2002, referred to as MYJ), which has a local closure with the prediction of turbulent kinetic energy (TKE). The horizontal eddy coefficient is obtained by the Smagorinsky scheme with a 1st order closure. Deep convection is represented by using a GrellDovenyi ensemble (G-D) scheme (Grell and Devenyi, 2002). We performed two types of experiments; the first was a run without chemical module RADM2 to understand the effect of physics parameterization on local meteorology for June– July in 2005 and 2006 (Section 3), and the second was with RADM2 to understand the effect of PBL scheme on chemical transport for July–August 2005 when vertical O3 measurements by lidar were conducted over Tsukuba (Section 4). For the first experiments without the chemical module, seven kinds of experiments were designed (Table 1). In Exps. 2 and 3, the MYJ PBL scheme was exchanged with the Yonsei University (YSU) scheme (Hong et al., 2006) and the NCEP GFS scheme (Hong and Pan, 1996), respectively. Both schemes have a non-local effect for unstable condition. In Exp. 4, the coordinate of turbulent/mixing and eddy
A Regional Air Quality Model over the Kanto Region of Japan: The Effect of the Physics
111
Fig. 1 The regional model domains (D1 and D2) and terrain height in m (grey scales). Inset: Zoom-up of domain 2 and observation site, TSU: Tsukuba, MAE: Maebashi, KUM: Kumagaya, KOS: Koshigaya, YOK: Yokohama, MIT: Mito, and KAT: Katsuura. Surface meteorological AMeDAS stations(MAE, KUM, KOS, and YOK), VHF-band wind profiler(KUM, MIT and KAT), and O3 lidar(TSU)
coefficient was switched with those calculated in the geometric coordinate. In Exp. 5–7, the G-D cumulus parameterization was alternated by Kain-Fritsch new eta (K-F) scheme (Kain and Fritsch, 1990), Betts-Miller (BMJ) scheme (Janjiü, 1994) and simplified Arakawa-Shubert (SAS) scheme, respectively. Simulated data were compared with observations. Vertical profile of O3 over Tsukuba comes from observation during 27–29 July and 16–22 August 2005 by a Differential Absorption Lidar (DIAL) developed by Nakazato et al. (2007). The location of observational stations is summarized in Figure 1.
3. Results and Discussion The results from experiments without chemical module is compared with observations. Table 2 shows the skill scores for June–August in 2005 and 2006 based on four stations in the Kanto region. The obtained correlation coefficient r shows 0.9 for temperature and ~0.7 for wind vectors. The variability of correlation coefficient is very small with ~0.02 for temperature and ~0.03 for wind vectors. However, variability of root-mean-square error (RMSE) is relatively large with 0.3 K for temperature and 0.3 m s-1 for winds. Variability of mean biases is also large with 1.2 K and 0.4 m s-1 for temperature and wind speed, respectively. Compared with the results from Exp. 1 using a local MYJ PBL scheme, Exps. 2 and 3 using nonlocal PBL schemes present the largest impact on the RMSE and the mean biases. From the results, it is concluded that switching with non-local PBL schemes have an effect on reducing cold bias and larger wind speed bias near the surface, compared to switching with cumulus parameterizations and horizontal turbulence and diffusion scheme.
M. Niwano et al.
112
Table 1 Experimental design. MYJ is Mellor-Yamada level 2.5 scheme, YSU is Yonsei University scheme, andNCEP is NCEP GFS scheme. G-D is Grell-Dovenyi ensemble scheme, K-F is Kain-Fritsch new Eta scheme, BMJ is Betts-Miller scheme, and SAS is simplified Arakawa-Schibert scheme. Experiment number
PBL
Turbulent/mixing & eddy coefficient
Cumulus parameterization
Exp. 1
MYJ
2nd order & 2D Smag.
G-D
Exp. 2
YSU
2nd order & 2D Smag.
G-D
Exp. 3
NCEP
G-D G-D
Exp. 4 Exp. 5
MYJ MYJ
2nd order & 2D Smag. Evaluate 3D mixing stress & 1.5 order TKE 2nd order & 2D Smag.
Exp. 6
MYJ
2nd order & 2D Smag.
BMJ
Exp. 7
MYJ
2nd order & 2D Smag.
SAS
K-F
Table 2 Skill scores for surface data obtained from experiments and observationfor JuneΩ August 2005 and 2006. Mean bias is defined as WRF minus observation. Bold font indicates the best result in the seven experiments. Correlation Experiment number Temperature Wind vector 0.643
RMSE Temperature Wind vector (K) (m s–1) 1.926 3.50
Mean Bias Temperature Wind speed (K) (m s-1) – 0.83 1.27
Exp. 1
0.899
Exp. 2
0.881
0.629
2.188
3.34
0.01
1.11
Exp. 3
0.890
0.657
2.061
3.20
0.3
0.99
Exp. 4
0.896
0.640
1.960
3.55
– 0.89
1.29
Exp. 5
0.900
0.657
1.868
3.50
– 0.73
1.36
Exp. 6
0.902
0.624
1.875
3.59
– 0.82
1.36
Exp. 7
0.902
0.632
1.876
3.59
– 0.82
1.36
To reproduce the O3 enhancement and transport over the Kanto region in AQ model, land-sea breeze is a key issue. Diurnal cycle averaged on a period of 25 July–07 August 2006, when land-sea breeze and O3 episode frequently occurred, was generally in agreement with observation (not shown). However, simulated results exhibited a systematic difference in wind direction with observation when sea breeze was switched by land breeze. Horizontal maps of wind direction at 18 UT (03 Japanese Standard Time [JST = UT + 9h]) averaged during a period of 25 July–07 August 2005 (Figure 2) shows that land breeze dominates over the Kanto region, so that north-easterly land breeze is seen over the south-eastern region of D2, north-westerly land-breeze over the central region, and south-easterly land breeze over the north-western region. However, all the experiments cannot reproduce the remnant of south-easterly sea breeze in the Tokyo bay. This difference is consistent with cold bias at night in the all seven experiments. Since the cold region is located in the Tokyo metropolitan area, the implementation of an urban canopy model treating urban heat might revise the feasibility to simulate the landbreeze at night.
A Regional Air Quality Model over the Kanto Region of Japan: The Effect of the Physics
113
Fig. 2 Horizontal maps of wind direction in degrees (grey scales) with stream lines at 18 UT (03 JST)for 25 JulyΩ7 August 2006 for Exps. 1–7 and the observation
Vertical profiles of averaged diurnal cycle are compared with the result from wind profiler over Kumagaya (139.38o E, 36.15o N). Averaged diurnal cycles of wind direction and wind speed are calculated based on 00 UT 25 JulyΩ00 UT 08 August 2006. The observed wind direction presents sea breeze of south-easterly denoted by dark colours (Figure 3a). The sea breeze evolves up to 1.3-km ASL and continues until 17 UT (01 JST). Above the sea-breeze, wind direction drastically changes to north-westerly denoted by blight colours and the north-westerly continues until 06 UT (15 JST). All the 7 experiments can simulate observational features of wind direction. However, simulated south-easterly sea breeze was switched by north-westerly land-breeze until 16 UT (01 JST) earlier and was altered by counter flow above 1.3 km with smoother vertical gradient than observation. For wind speed (Figure 3b), similar features can be seen as a minimum wind speed layer at 0.5–1.3 km, and observation and simulation agrees with each other. In Exps. 2 and 3, sea breeze develops with deeper height and stronger wind speed than for those observed, in contrast to small errors in RMSE and mean bias for surface data (Table 2).
Fig. 3 Time-height sections of averaged diurnal cycle of wind direction in degrees (a) and wind speed in m s-1 (b) over Kumagaya (139.38° E, 36.15° N) from 00 UT to 00 UT and from 0 to 5 km height from seven experiments and wind profiler. The averaged diurnal cycle is calculated based on data from 00UT 25 July to 00UT 23 UT 7 August 2006
114
M. Niwano et al.
Next, a simulation was again performed from 25 July to 23 August 2005 with a chemical module RADM2 and the parameterization sets used in Exps. 3 and 5. The model and observation showed overall agreement in diurnal cycles and longer timescale variations of temperature, wind direction and O3 (Figure 4). On 10 August, the time series in temperature and wind direction showed a difference between the observation and simulation, because the cold front development was not simulated in the model. The model overestimated the observed O3 mixing ratio from the afternoon to the evening for the periods of 1–10 August, and 18–22 August, which is due to the underestimate in the development of convective system and of consequent strong convergent wind related to thunder storm in the northern Kanto region in the afternoon. The underestimate of the convergent wind led to the false O3 accumulation over the Kanto region for the model.
Fig. 4 Time series of (a) surface temperature in K, (b) wind direction in degrees, (c) ozone mixing ratio in ppbv over Tsukuba (140.125° E, 36.057° N) from the observation (black) and Exp. 5 (grey) from 00 UT on 25 July to 23 UT on 22 August 2005
Time-height sections of the simulated O3 concentration over Tsukuba (Figure 5) showed layer structures below 2 km and between 2 and 7 km during a period of July–August 2005. The layer structures are seen at 2–6 km for 26–29 July and 15– 18 August. Thin layer structures were simulated at 1–2 km for 28 July, 18 and 19 August, while vertically homogeneous enhancement appeared below 2 km for 20 August. No enhancement was observed below 2 km for 17 August. The simulated vertical and temporal variations are in consistent with those from the DIAL measurement during periods from 27 to 29 July and from 16 to 21 August (Niwano et al., 2007). Comparing the simulated O3 variations in Exp. 3 (Figure 5a) and Exp. 5 (Figure 5b), the layer with high O3 concentration appears at 1–2 km in both Exps. 3 and 5, but that is located at higher altitudes for Exp. 3. For example, high O3 plumes on 1 and 21 August are clearly extended to higher altitudes for Exp. 3. In consistence with the difference in the height of high O3 plumes, the height of PBL (white lines) is also higher for Exp. 3. The difference in the PBL heights can lead to the difference in the strength of vertical mixing, such that a deeper PBL height results in the lower O3 concentration at the surface. However, the Exp. 3 is accompanied by a higher O3 concentration at the surface level over Tsukuba. In fact,
A Regional Air Quality Model over the Kanto Region of Japan: The Effect of the Physics
115
Fig. 5 Time-height sections of O3 concentration in molec. m-3 (grey scales) over Tsukuba from the simulation of Exp.3 (a) and Exp. 5 (b) from 00 JST on 27 July to 23 JST on 22 August2005. The height of PBL is also plotted by white lines
the high O3 plume over Tsukuba is attributable mainly to the horizontal transport by sea breeze from the Tokyo metropolitan area. The other factor to produce the higher O3 concentration is the O3 chemistry depending on the difference in surface temperature. Actually, the surface temperature is higher in Exp. 3 than that in Exp. 5, and the nitrogen oxide concentration near the surface is also higher in Exp. 3.
4. Summary and Conclusion This paper examined the effects of physics parameterization on meteorological and chemical fields over the Kanto region of Japan using the global-regional AQ model. The AQ model consists of the global CTM CHASER and the regional CTM WRF/Chem. Hindcast experiments without a chemical module for June–July in 2005 and 2006 showed that two non-local PBL schemes (YSU and NCEP GFS) produced a deeper PBL height and a stronger sea-breeze related with warmer surface temperature than for the observation and the local PBL scheme (MellorYamada level 2.5). Two experiments using a chemical module for July–August in 2005 found that the NCEP GFS scheme transported O3 plumes to higher altitudes above 2 km and also resigned a higher concentration near the surface than for the Mellor-Yamada scheme. Acknowledgments The authors would like to appreciate Dr. G. Grell, Professor K. Sudo, and other members for developing of the WRF/Chem and the CHASER. The authors also acknowledge to those who engaged in the observation at air quality monitor stations. The present work has been supported by the internal special project fund of JAMSTEC.
116
M. Niwano et al.
References Grell GA, Peckham SE, Schmitz R et al. (2005) Fully coupled “online” chemistry within the WRF model. Atmos. Environ., 39, 6957–6975. Grell GA, Devenyi D (2002), A generalized approach to parameterizing convection combining ensemble and data assimilation techniques, Geophys. Res. Lett., 29(14), Article 1693. Hong S-Y, Noh Y, Dudhia J (2006) A new vertical diffusion package with an explicit treatment of entrainment processes, Mon. Wea. Rev., 134, 2318–2341. Hong S-Y, Pan H-L (1996) Nonlocal boundary layer vertical diffusion in a medium-range forecast model, Mon. Wea. Rev. 124, 2322–2339. Janjiü ZI (2002) Nonsingular implementation of the Mellor-Yamada level 2.5 scheme in the NCEP meso model, NCEP Office Note, No. 437, 61 pp. Janjiü, ZI (1994) The step-mountain eta coordinate model: further developments of the convection, viscous sublayer and turbulence closure schemes, Mon. Wea. Rev., 122, 927–945. Kain JS, Fritsch JM (1990) A one-dimensional entraining/detraining plume model and its application in convective parameterization, J. Atmos. Sci., 47, 2784– 2802. Kawamura T (1979) “Toshi no Taiki-Kankyo”. Todai Shuppan-kai, pp. 185 (in Japanese). Lawrence M et al. (2003) Global chemical weather forecasts for field campaign planning: predictions and observations of large-scale features during MINOS, CONTRACE, and INDOEX, Atmos. Chem. Phys., 3, 267–289. Nakazato M, Nagai T, Sakai T, Hirose Y (2007) Tropospheric ozone differentialabsorption lidar using stimulated Raman scattering in carbon dioxide, Appl. Opt., 46, 2269–2279. Niwano M, Takigawa M, Akimoto H et al. (2007) Evaluation of vertical ozone profiles simulated by WRF/Chem using lidar-observed data, SOLA, in revision. Sudo K, Takahashi M, Akimoto H (2002) CHASER: a global chemical model of the troposphere 1. Model description, J. Geophys. Res., 107, doi: 10.1029/2001JD001113. Takigawa M, Sudo K, Akimoto H et al. (2005) Estimation of the contribution of intercontinental transport during the PEACE campaign by using a global model, J. Geophys. Res., 110, D21313, doi:10.1029/2005JD006226. Takigawa M, Niwano M, Akimoto HO, Takahashi M (2007) The impact of meteorological field on the air quality forecasting, SOLA, 3, 31–84. Uno I, Carmichael GR, Streets DG et al. (2003) Regional chemical weather forecasting system CFORS: model descriptions and analysis of surface observation at Japanese island stations during the ACE-Asia experiment, J. Geophys. Res., 108 (D23), 8668, doi:10.1029/2002JD002845.
A Regional Air Quality Model over the Kanto Region of Japan: The Effect of the Physics
117
Discussion D. Steyn: It has been shown that aerosol layer depth is often greater than thermodynamically defined MLD. Could this be a possible explanation for the rather poor agreement on your scatter plot of MLD? M. Niwano: It is possible in the case of Mellor-Yamada scheme (aerosol layer depth > MLD), although a similar results are obtained even if horizontal advection of aerosol layers from mountainous region are omitted. In addition, MLD obtained from NCEP GFS scheme (a non-local scheme) was generally deeper than aerosol depth, in the opposite sense to your suggestion. This overestimate implies that MLD calculated by NCEP GFS scheme tends to be overestimated. A. Graff: At the beginning it was stated that ozone concentrations are increasing though NOx and VOC emissions are decreasing. (1) Can you please explain if this statement is related to peak ozone concentrations or at e.g. to average concentration levels. (2) Is this statement based on measurements? M. Niwano: 1. The increase in ozone concentration and the decrease in NOx and VOC concentrations are based on the time-mean (and spatialaverage) concentration sence. 2. Yes, it is based on the measurements (more than 100 monitoring stations in Japan). So, we think polluted air masses, in which ozone plumes has been produced from NOx and VOC whereas NOx and VOC have been converted to NOy and other oxidized species, comes to Japan from upwind regions taking a few days.
2.10 Air Pollution Modelling with Perturbational Downscaling Eugene Genikhovich, Mikhail Sofiev, Guy Schayes and Irene Gracheva
Abstract A technique of downscaling based on the reformulation of the governing equations in the perturbational form with arbitrary amplitude of the perturbbations had been introduced in our previous works and successfully applied in meteorological modelling. In particular, preliminary results of testing such a technique, built-in into the mesoscale met model TVM, have indicated that it has certain advantages as compared with conventional methods of downscaling based on the brute-force damping of computational errors in the strip of cells located near the boundaries of the downscaled domain. The present paper is devoted to application of this technique in the atmospheric dispersion modelling. Because the case of the continuous point source located inside the downscaled domain seems to be most critical, the problem of numerical approximation of the source term on nested grids is considered first of all. In distinction from NWP problems, the source term in dispersion modelling is extremely sensitive to the spatial resolution, and its approximations in the whole computational domain and in the downscaled domain should be in concert one with another. The efficiency of perturbational downscaling is tested first in the model case where the advection-diffusion equation (ADE) has an analytical solution. It is demonstrated that the perturbational downscaling in a given sub-domain provides the numerical solution of ADE in this sub-domain “of the same quality” as the direct numerical solution of ADE with the use of fine grid uniformly in the whole domain. Work is on-going towards the practical application of the perturbational downscaling using the TVM model as a meteorological driver for the SILAM dispersion model. A specific case of an accidental release in Europe is considered as a test scenario where the sensitivity of the concentration field to the spatial resolution in NWP and dispersion models is high enough to justify nesting efforts. The ETEX first release case is used for the trial application as a prominent example.
Keywords Atmospheric dynamics and composition modelling, nested modelling, perturbational downscaling
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
182
Air Pollution Modelling with Perturbational Downscaling
183
1. Introduction Downscaling (also known as one-way nesting) is one of the most important tools for obtaining the detailed information on meteorological features and chemical composition of the atmosphere over a limited domain. A general approach to the problem is to solve the set of governing equations twice – for large but coarse domain and for the smaller fine-resolution grid with boundary conditions taken from the coarse solution. This, however, causes disruptions near the border of the inner domain because the conditions at the border are defined in the coarse grid and thus do not satisfy the governing equations in the fine-grid domain which quite generally differ significantly from the coarse domain ones. One of the most common ways to limit the impact of this inaccuracy is the Davies relaxation algorithm (Davies, 1976; Miyakoda, Rosati, 1977), according to which the boundary conditions are gradually mixed into the inner domain within the limited-size band, e.g., five fine-grid cells along the border line. Such numerical trick reduces the problem but does not solve it leaving the near-border areas of the inner domain practically unusable. Genikhovich and Schayes (2007) and Genikhovich, Sofiev and Gracheva (2007) have introduced the alternative methodology based on variational consideration when the field in the high-resolution domain is represented as a sum of coarse-grid solution and fine-scale perturbations. Their sum provides the full fine-scale solution for the governing equations. Contrary to generally used perturbational formalism, no assumptions regarding the amplitude of the perturbations are made.
2. Implementation Using the TVM Model (ETEX Experiment) The original TVM (Thermal Vorticity Mesoscale) model is a three-dimensional non-hydrostatic anelastic mesoscale atmospheric model solving the dynamic equations in vorticity-mode (for details see Schayes et al., 1996; Thunis and Clappier, 2000). The general methodology of the perturbation formulation implementation in the TVM model was presented in a previous paper (Genikhovich and Schayes, 2007). This text focuses on the present modifications for the test case using the ETEX experiment of 1994. The variables U i ,T , P, U are separated in large scale ( L ) and meso-local scale ( M ) contributions (e.g. U i U iL U iM ) and the meso model only solves the meso contribution.
E. Genikhovich et al.
184
In the present formulation, the boundary conditions (BC) over any mesoscale perturbation variable ) M U i ,T is specified as follows:
Upper BC : perturbation ) M 0 Lateral BC : w) M / wxi 0 Bottom BC : for wind, we have UiM = 0; for temperature an ordinary ground energy balance is performed on T and then T M T T L
The model has been run to simulate the ETEX dispersion experiment that took place in Europe in October 1994. The domain extension is 90×75 grid points with a grid size of 15 km covering Northern France, the Benelux and Germany. There are 33 vertical levels extending up to 9,500 m and with the lowest one at 10 m. The topography data is extracted from the GTOPO and the land use is obtained from the CORINE data bases. In this version, the model uses a standard MOST formulation for the surface layer. The LS data are obtained from the ECMWF reanalyses ERA40 and are interpolated in space and time on all model grid points. From an example of the resulting fine-grid wind in Figure 1 (left-hand panel) and corresponding wind perturbation (right hand panel) it is seen that the wind perturbation is strongly related to the surface features and topography while the total fine-resolution wind is also largely driven by the synoptic-scale patterns. However, it is still very possible that the present way of data passing between the LS and mesoscale domain, as well as the BC, are not the optimal choice. Finally, it must be noted that in this exercise no intermediate (nesting) domain exists between the LS data (ECWMF) and the local scale TVM model. The scaling ratio between the coarse grid (ECMWF ERA-40, 1.125q resolution) and the fine grid (0.250 × 0.150 grid cell size) is 5 along the longitude and over 8 along the latitude. This is a ratio usually considered as impossible for standard ways of downscaling.
Fig. 1 An example of total and perturbation near-surface wind (note the different wind scales: maximum wind is 13.4 m/s in the left panel and 3.8 in the right panel)
Air Pollution Modelling with Perturbational Downscaling
185
3. Downscaling of Tracer Dispersion Fields Perturbational considerations for the tracer transport is somewhat simpler because there is only the advection-diffusion equation that is to be rewritten. From the other side, interactions of both large-scale and perturbational components of the total fine-grid concentration have to be taken into account. To verify the approach, a set of experiments with known analytical solution has been performed. These included: 2D dispersion (horizontal and vertical dimensions) with different horizontal eddy diffusivity Kx, dispersion of heavy aerosol with various relations of diffusion coefficient, wind and sedimentation velocity. Finally, a 2D dispersion was considered for downscaling the problem with fine-scale strong perturbation in the dry deposition intensity. We assumed presence of a highlyadsorptive surface area, which is fully unresolved in the coarse domain but wellrepresented in the fine grid. For simple cases the model results were within 5% from analytical solutions (measuring the maximum deviation of instant concentrations). The downscaled case with highly adsorptive fine-resolution area was compared with very high-resolution numerical simulations and revealed somewhat larger discrepancies at the edge of the adsorptive region, being otherwise again within 5% from the accurate solution. A simplification for the ETEX case is that the source is a single point that is inside the fine-scale domain. Consequently, the input from the outer grid is zero and the boundary conditions for the tracer become trivial. Then the source itself can be considered as a perturbation while the large-scale field is zero. These steps efficiently reduce the problem to only hydrodynamic perturbational downscaling. We utilised this simplification and used the current exercise to establish the interface between the CTM SILAM and ECMWF-TVM couple of NWP models. The SILAM modelling system (http://silam.fmi.fi) has a dual dynamic core that includes Lagrangian and Eulerian advection-diffusion routines. The Lagrangian version of the system is described by Sofiev et al. (2006), while the new Eulerian kernel is presented by Sofiev et al. (this issue). For the current exercise, we used passive tracer that is transported by the Eulerian dynamic kernel. SILAM has an operational interface for the ECMWF operational model output, which has higher resolution than the ERA-40 re-analysis. Therefore, for the comparison with the runs driven by ERA-TVM-downscaled meteorology, we used the data of operational archive of ECMWF. SILAM can perform the dispersion computations in the grid different from that of the meteorological driver, therefore we performed the control run with the same grid as that of the downscaled fields, which made the runs directly comparable. An example in the Figure 2 (left-hand panel for ECMWF reference meteorology, right-hand panel for TVM-downscaled fields) illustrates the obtained differences. Firstly, the small-scale fluctuations of the 3D wind field absent in coarse reference dataset are much better reflected in the downscaled wind pattern. It resulted in
186
E. Genikhovich et al.
much more significant dilution of the plume. Its more significant both upward and horizontal mixing made the concentrations inside it closer to the observed ones, while the values with the reference dataset are significantly higher. Secondly, the plume location is somewhat different: the TVM wind field has first transported cloud directly the east and even slightly southwards from the emission point. According to map of observations (Sofiev et al., 2006), this is probably not the entirely correct path. A simulation with over-scaled vertical wind has showed that this turn is partly related to the transport with higher-level winds. Since TVM generated stronger vertical exchange, the fraction of mass transported aloft has increased shifting the plume position. The second potential reason is somewhat too strong air-surface interaction in the TVM formulations. This hypothesis is facilitated by a stronger turbulent mixing (expressed via Kx,y,z) reported in the fine-scale fields. The SILAM meteo processor generated substantially lower values using the ECMWF profiles as an input. Detailed analysis is needed for understanding the reasons for such a move. In all cases, the specific behaviour of the plume is not connected with the downscaling methodology, which is the primary target of the current study. Thirdly, after about one day of transport, the fine-scale features of the concentration pattern show up in the downscaled simulations. They are entirely absent in the reference dataset and recall the discussion on whether the plume was actually split during the second and the third days of transport. According to Baklanov (personal communication), such split (correctly or not) can be observed from the simulations with very high resolution of the meteorological fields (about 1.5 km) where the local-scale processes dominate the transport near the surface. According to our fine-scale simulations, the full split has not been reached but the tendency is quite clear. Finally, as seen from the Figure 2, the downscaled wind has no disturbances near the border – the plume leaves the fine-scale computation domain without any noticeable distortion.
4. Summary and Conclusion In this paper, the perturbational downscaling algorithm with large perturbations suggested by Genikhovich and Schayes (2007) has been extended to the tracer transport problem and applied to a real case of the ETEX experiment. The TVM meteorological model modified to include the perturbational downscaling generated the fine-scale fields from the ECMWF ERA-40 reanalysis. The downscaling was performed in one step with the ratio between the resolutions of the coarse and fine grids of 5 and >8 in longitudal and latitudal directions, respectively. The downscaled meteorological fields were interfaced to the CTM SILAM, which used simplified downscaling formulations to perform the exercise.
Air Pollution Modelling with Perturbational Downscaling
187
Fig. 2 An example of reference (left-hand panels) and downscaled simulations (right-hand panels)
188
E. Genikhovich et al.
The results of the simulations showed that the downscaling procedure works as expected and creates no visible problems near the boundaries of the inner domain. The fine-scale wind pattern is more dynamic, which caused an extra dilution of the concentration fields and brought them closer to the observations. Position of the plume in the fine-resolution run, however, was somewhat shifted to the south in comparison with the reference run and the observations. Such a shift, however, probably reflects the specific features of the NWP driver used for the downscaling, and has no relation to the downscaling mechanism discussed in the paper. In-depth investigation is needed to understand the main reasons for such a shift. Acknowledgments The SILAM simulations have been partly supported by EUGEMS and ESA-PROMOTE project.
References Davies HC (1976) A lateral boundary formulation for multi-level prediction models. Quart. J. Roy. Met. Soc., 102, 405–418. Genikhovich EL, Sofiev M, Gracheva IG (2007) Interaction between meteorological and dispersion models at different scales. In: Air Pollution Modelling and Its Applications XVII (Ed. C Borrego, A-L Norman), Springer, New York, pp. 158–166. Genikhovich EL, Schayes G (2007) Perturbational downscaling and its applications in air pollution and meteorological problems. In Air Pollution Modelling and its Applications XVIII (Ed. C Borrego, E Renner), Elsevier, Amsterdam, The Netherlands, pp. 123–133. Miyakoda K, Rosati A (1977) One-way nested grid models: the interface conditions and the numerical accuracy. Quart. J . Roy. Met. Soc., 105, 1092–1107. Nayfeh AH (1973) Perturbation Methods, Wiley, New York. Schayes G, Thunis P, Bornstein R (1996) Topographic Vorticity-Mode MesoscaleE (TVM) Model. Part I: Formulation. J. Appl. Meteor. 35, 1815–1823. Sofiev M, Siljamo P, Valkama I, Ilvonen M, Kukkonen J (2006) A dispersion modelling system SILAM and its evaluation against ETEX data. Atmos. Environ., 40, 674–685, doi:10.1016/j.atmosenv.2005.09.069 Thunis P, Clappier A (2000) Formulation of a nonhydrostatic mesoscale vorticity model (TVM). Mon. Wea. Rev., 128, 3236–3251.
Air Pollution Modelling with Perturbational Downscaling
189
Discussion A. Baklanov: You mentioned in the presentation previous work of Davies (1976). However, you probably know, that the Novosibirsk school of G. Marchuk, L. Gutman and many their colleagues suggested and actively used such perturbation approach even earlier, from 1970, for both meteorological and pollution models downscaling. E. Genikhovich: The so-called Davies relaxation method introduced in 1976 is still one of the most widely used methods of harmonising the fields between the nested domains – with all its strengths and weaknesses. As for the works carried out by the Novosibirsk team, we are aware of these too. Indeed, they used the idea of separating the flow into its “background” – and “perturbed” parts but in corresponding studies they assumed, in particular, that some of the nonlinear terms describing the interaction between these two parts of the flow were negligible small and omitted them from the governing equations. Unfortunately, these terms are not necessarily small when solving the problem of downscaling. As a result, the governing equations introduced by the authors mentioned above are, generally speaking, not applicable for a general case of downscaling. Looking broader, the idea of separation of a solution of some nonlinear system of equations into its background and perturbed parts is for centuries at the core of numerous classical works in mathematics, mechanics and hydrodynamics. In particular, it was actively used in geophysical hydrodynamics, meteorology, and theory of the atmospheric diffusion. However, one of the key assumptions in practically all these works is that the perturbation part is either small by itself or has small average values or its non-linear interaction with the main field is small. In our work, these assumptions were avoided.
2.1 Contribution of Biogenic Emissions to Carbonaceous Aerosols in Summer and Winter in Switzerland: A Modelling Study ù. Andreani-Aksoyo÷lu, J. Keller, M.R. Alfarra, A.S.H. Prévôt, J.J. Sloan and Z. He
Abstract The MM5/CAMx model system was applied to the complex terrain of Switzerland for winter and summer periods. The focus in this paper is on the formation and transport of particulate matter (PM) and the contribution of biogenic sources to the aerosol formation. Both model results and measurements indicate that particulate nitrate and organic aerosols are the major components of the aerosol composition in winter in northern Switzerland. Organic aerosols dominate the aerosol composition in summer and they are mostly secondary. Measurements show that biogenic emissions in Zurich contribute about 60% and 27% to organic carbon (OC) in summer and winter, respectively. The model predictions of the biogenic contribution are very close to the measurements in summer. The biogenic precursors of secondary organic aerosols are mainly monoterpenes emitted from Norway Spruce forests in northern Switzerland. The model predictions suggest that biogenic emissions contribute predominantly to the secondary organic aerosols (SOA) in winter as well, although concentrations are lower than in summer. The fraction of biogenic SOA is much lower in the south, around the polluted region of Milan. Keywords Aerosols, biogenic emissions, SOA, CAMx, MM5
1. Introduction Organic aerosol (OA) components account for a large fraction of atmospheric particulate matter. They influence the physicochemical properties of aerosol particles and thus their effects on the atmosphere, climate, biosphere and human health. Primary organic aerosols (POA) are contained in combustion particles, biological particles and various other types of particles emitted directly into the atmosphere. Secondary organic aerosol (SOA) components are formed in the atmosphere by several pathways, such as partitioning of semi-volatile organic compounds (SVOC) from the gas phase onto existing particles and participation of SVOC in the formation of new particles (nucleation). Several recent studies indicate that low- or non-volatile C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
101
102
ù. Andreani-Aksoyo÷lu et al.
organic compounds (NVOC) can be formed by heterogeneous or multiphase reactions of VOC or SVOC on the surface or in the bulk of aerosol and cloud particles. These compounds enable SOA formation from VOC which had previously been assumed not to contribute to SOA formation (e.g. isoprene) by acid-catalysed or radical oligo- or polymerisation reactions (Claeys et al., 2004). Currently, uncertainties in Chemical Transport Model (CTM) representations of OA are large. A significant part of this uncertainty is due to the lack of understanding of SOA formation rates and their controlling factors. The relative contributions of anthropogenic and biogenic sources (mainly trees) to SOA formation is of great importance for emission reduction strategies. It is known that the biogenic contribution to SOA is large on a global scale, but the anthropogenic contribution can be important in polluted regions and depending on the aerosol sources, transport processes and atmospheric conditions, the composition of OA can be dominated either by POA or by SOA (Artaxo et al., 2002; Kanakidou et al., 2005). Recent studies suggest that current models underestimate SOA concentrations (e.g. Zhang et al., 2004) while the POA contribution has been overestimated (Robinson et al., 2007). It was shown that organic matter, sulfate and nitrate are the main contributors to the annual PM2.5 mass concentration in Switzerland (Hueglin et al., 2005). A recent study (Szidat et al., 2006) suggested that wood burning is an important source for POA in winter around Zurich, where it amounted to 41% of the OC. Lanz et al., (2007) showed that a high fraction (60– 69%) of the organic aerosol mass measured in summer was oxygenated organic aerosol (OOA), which was interpreted mostly as secondary organic aerosol (SOA). The large forest coverage in Switzerland increases the probability of high SOA formation from biogenic emissions.
2. Modelling Method In this study, the three-dimensional photochemical model CAMx (Comprehensive Air Quality Model with Extensions, version 4.40) was applied in a Lambert Conic Conformal coordinate system using three nested domains as shown in Figure 1 (Environ, 2006). The resolutions of the three domains were 27, 9 and 3 km, respectively. In the vertical dimension, there were 14 ı-layers in a terrain-following coordinate system, the first being about 40 m above ground. The model top corresponds to about 500 hPa. The winter simulations refer to the month of January 2006. The summer calculations were carried out for a short period in August 2003 and they are still going on for a longer period in summer 2006, for which detailed aerosol mass spectrometer (AMS) measurements are available. Meteorological data were calculated using the MM5 meteorological model (PSU/NCAR, 2004). MM5 was initialized by assimilated data of the Alpine Model (aLMo) of MeteoSwiss. The four-dimensional meteorological data assimilation was conducted using surface measurements, balloon soundings and aLMo upper level data. The emission inventory was prepared by compiling European and Swiss anthropogenic emissions
Contribution of Biogenic Emissions to Carbonaceous Aerosols in Summer and Winter
103
Fig. 1 Topography of three model domains with 27, 9 and 3 km resolutions, respectively
from various data sources, as described in Keller et al. (2007). Emissions of PM2.5 and PM10 of nine source categories were obtained from INFRAS and Meteotest for the reference year 2000. The spatial resolution of particulate emissions is 200 m. Using land use and meteorological data, biogenic emissions were calculated by means of temperature and irradiance dependent algorithms (Andreani-Aksoyoglu and Keller, 1995). In the biogenic emission inventory, the most abundant species are monoterpenes, which are emitted mainly by Norway Spruce and fir trees. Less abundant is isoprene, emitted by oak trees and pasture, mainly in the southern part of Switzerland. Initial and boundary conditions were obtained from the global model MOZART (Horowitz et al., 2003). Calculations of aerosols with d < 2.5 µm were performed with the fine/coarse option of the aerosol module.
3. Results and Discussion
3.1. Summer The highest aerosol mass concentrations in summer were predicted for the polluted area around Milan. This region has high emissions of primary particles and of gaseous precursors for secondary aerosols. As seen in Figure 2, the model is able to predict the relative contribution of particulate NO3, NH4 and EC to the total aerosol composition reasonably well. The model prediction for particulate SO4 is higher than the measurements. On the other hand, organic aerosol concentrations were underestimated. The difference in measured and modeled organic aerosol
104
ù. Andreani-Aksoyo÷lu et al.
Fig. 2 Relative contribution of components to the total PM2.5 aerosols in Zurich. Measurements (left) (Szidat et al., 2004, for August 2002) and model calculations (right) for August 2003
concentrations is believed to be due to the underestimation of secondary organic aerosols. The experimental evidence for oligomerization reactions in organic aerosols indicated the need to readdress the current assumptions in models about the partitioning of oxidation products in the gas and the particle phase (Kalberer et al., 2004). A recent model study by Morris et al. (2006) showed that including mechanisms that are not treated yet in current models, such as polymerization, SOA formation from isoprene and sesquiterpenes, led to increased SOA yields. These mechanisms will be considered in future. The model results suggest that the biogenic SOA dominates the total SOA concentrations in northern Switzerland while anthropogenic SOA is more important in northern Italy, around Milan (Figure 3). These results are supported by the measurements performed in Zurich by Szidat et al. (2006). The measurements suggest that about 60% of organic carbon has biogenic origin while the model predicts that 63% of organic aerosols are formed from biogenic precursors. The fraction of biogenic SOA to total SOA is about 80% in the north whereas it is substantially lower in southern Switzerland (40%) and northern Italy (15–25%).
Fig. 3 The modeled biogenic (left) and anthropogenic (right) SOA concentrations (µg/m3) on 7.8.2003 14:00–15:00
Contribution of Biogenic Emissions to Carbonaceous Aerosols in Summer and Winter
105
3.2. Winter Some meteorological parameters such as wind speed, precipitation and temperature during January 2006 are shown in Figure 4. Based on temperatures, this period can be divided into four sections: (1) during 1–6 January, the temperature was above zero and the wind speed was moderate; (2) the second period between 6 and 17 January was colder, with temperatures below zero and with low wind speed; (3) after this long period, some precipitation occurred, the temperatures increased to above zero (between 17 and 23 January) and the wind speed was stronger; and (4) the last period between 17 and 31 had mixed conditions with variable wind speed and temperatures. The modelled concentrations of organic aerosols were compared with the AMS measurements. The model performance was better for both inorganic (not shown) and organic aerosols (Figure 5) during the third period when the wind speed was stronger. On the other hand, the model seems to underestimate aerosol levels when the wind is weak and the temperature is low. The foggy period (period II) could not be reproduced well by the model. The results, therefore, are discussed for the third period. Both AMS measurements and model results suggest that the main components of the winter aerosols in Zurich are particulate nitrate and organic aerosols (Figure 6). The model prediction for organic aerosols is underestimated probably due to missing wood-burning emissions and underestimation of secondary organic aerosols. The model results suggest that secondary organic aerosols formed from the biogenic precursors (monoterpenes) can be more important than the SOA produced from the anthropogenic precursors in winter, although the absolute concentrations are much lower compared to summer (Figure 7). The anthropogenic SOA was predicted to be more important for the region of Milan.
Fig. 4 Wind speed (m/s), precipitation (mm) and temperature (C) measured at the NABEL station in Zurich during January 2006
106
ù. Andreani-Aksoyo÷lu et al.
Fig. 5 Comparison of modelled (PM2.5) and measured (PM1, by AMS) concentrations (µg/m3) of organic aerosols (POA + SOA) in Zurich during January 2006
Fig. 6 Relative contribution of components to the total PM2.5 aerosols in Zurich. Measurements (left) and model calculations (right) for January 2006
Fig. 7 The modeled biogenic (left) and anthropogenic (right) SOA (µg/m3) on 19.1.2006 14:00– 15:00
Contribution of Biogenic Emissions to Carbonaceous Aerosols in Summer and Winter
107
Acknowledgments This project is financially supported by the Federal Office for the Environment (FOEN) in Switzerland. The collaboration between PSI and WCAS was partly supported by the INTROP grant of the European Science Foundation (ESF). We thank EMPA, MeteoSwiss, Meteotest, FUB, TNO, INFRAS, ACCENT, and M. Schultz for providing various data or other support.
References Andreani-Aksoyoglu S, Keller J (1995) Estimates of monoterpene and isoprene emissions from the forests in Switzerland, J. Atmos. Chem., 20, 71–87. Artaxo P, Martins JV, Yamasoe MA, Procopio AS, Pauliquevis TM, Andreae MO, Guyon P, Gatti LV, Leal AMC (2002) Physical and chemical properties of aerosols in the wet and dry seasons in Rondonia, Amazonia, J. Geophys. Res., 107(D20) 8081, doi:10.1029/2001JD000666. Claeys M, Graham B, Vas G, Wang W, Vermeylen R, Pashynska V, Cafmeyer J, Guyon P, Andreae MO, Artaxo P, Maenhaut W (2004) Formation of secondary organic aerosols through photooxidation of isoprene, Science, 303, 1173–1176. Environ (2006) User’s Guide, Comprehensive Air Quality Model with Extensions (CAMx), Version 4.4, Environ International Corporation, California. Horowitz LW, Walters S, Mauzerall DL, Emmons LK, Rasch PJ, Granier C, Tie X, Lamarque J-F, Schultz MG, Tyndall GS, Orlando JJ, Brasseur GP (2003) A global simulation of tropospheric ozone and related tracers: description and evaluation of MOZART, version 2, J. Geophys. Res., 108, 4784, doi:4710.1029/ 2002JD002853. Hueglin C, Gehrig R, Baltensperger U, Gysel M, Monn C, Vonmont H (2005) Chemical characterisation of PM2.5, PM10 and coarse particles at urban, nearcity and rural sites in Switzerland, Atmos. Environ., 39, 637–651. Kalberer M, Paulsen D, Sax M, Steinbacher M, Dommen J, Prévôt ASH, Fisseha R, Weingartner E, Frankevic V, Zenobi R, Baltensperger U (2004) Identification of polymers as major components of atmospheric organic aerosols, Science, 303, 1659–1662. Kanakidou M, Seinfeld JH, Pandis SN, Barnes I, Dentener FJ, Facchini MC, Dingenen RV, Ervens B, Nenes A, Nielsen CJ, Swietlicki E, Putaud JP, Balkanski Y, Fuzzi S, Horth J, Moortgat GK, Winterhalter R, Myhre CEL, Tsigaridis K, Vignati E, Stephanou EG, Wilson J (2005) Organic aerosol and global climate modelling: a review, Atmos. Chem. Phys., 5, 1053–1123. Keller J, Andreani-Aksoyoglu S, Tinguely M, Flemming J, Heldstab J, Keller M, Zbinden R, Prevot ASH (2007) The impact of reducing the maximum speed limit on motorways in Switzerland to 80 km h-1 on emissions and peak ozone, Environ. Mod. Software, doi:10.1016/j.envsoft.2007.04.008. Lanz VA, Alfarra MR, Baltensperger U, Buchmann B, Hueglin C, Prevot ASH (2007) Source apportionment of submicron organic aerosols at an urban site by linear unmixing of aerosol mass spectra, Atmos. Chem. Phys., 7, 1503–1522.
108
ù. Andreani-Aksoyo÷lu et al.
Morris RE, Koo B, Guenther A, Yarwood G, McNally D, Tesche TW, Tonnesen G, Boylan J, Brewer P (2006) Model sensitivity evaluation for organic carbon using two multi-pollutant air quality models that simulate regional haze in the southeastern United States, Atmos. Environ., 40, 4960–4972. PSU/NCAR (2004) MM5 Version 3 Tutorial Presentations. Robinson AL, Donahue NM, Shrivastava MK, Weitkamp EA, Sage AM, Grieshop AP, lane TE, Pierce JR, Pandis SN (2007) Rethinking organic aerosols: semivolatile emissions and photochemical aging, Science, 315, 1259–1262. Szidat S, Jenk TM, Gäggeler HW, Synal H-A, Fisseha R, Baltensperger U, Kalberer M, Samburova V, Wacker L, Saurer M, Schwikowski M, Hajdas I (2004) Source apportionment of aerosols by 14C measurements in different carbonaceous particle fractions, Radiocarbon, 46, 475–484. Szidat S, Jenk TM, Synal H-A, Kalberer M, Wacker L, Hajdas I, Kasper-Giebl A, Baltensperger U (2006) Contributions of fossil fuel, biomass burning, and biogenic emissions to carbonaceous aerosols in Zurich as traced by 14C, J. Geophys. Res., 111, D07206, doi:07210.01029/02005JD006590. Zhang Y, Pun B, Vijayaraghavan K, Wu S-Y, Seigneur C, Pandis SN, Jacobson M Z, Nenes A, Seinfeld JH, Binkowski FS (2004) Development and application of the model of aerosol dynamics, reaction, ionization and dissolution (MADRID), J. Geophys. Res., 109, doi:10.1029/2003JD003501.
Discussion M. Kaasik: Do you have any estimates on emissions of biogenic VOC from non-tree species? Which methods you will use to estimate the emissions from wood burning? ù. Andreani:
We calculate isoprene and monoterpene emissions from pasture and crops as well as from trees, for our emission inventory. Our estimates suggest that emissions from pasture and crops are very low (less than 5%) compared to the forest emissions and they vanish outside the vegetation period. We are not doing wood burning estimations ourselves. This is planned and coordinated by the Federal Office of Environment.
2.7 Development and Applications of Biogenic Emission Term as a Basis of Long-Range Transport of Allergenic Pollen Pilvi Siljamo, Mikhail Sofiev, Tapio Linkosalo, Hanna Ranta and Jaakko Kukkonen
Abstract This study presents a birch pollen long-range transport forecasting system, which is developed at Finnish Meteorological Institute together with the Aerobiology Unit of the University of Turku, and the Department of Forest Ecology of the University of Helsinki. The forecasting system consists of several submodels. It is based on a numerical weather prediction (NWP) model. The use of good-quality NWP model is essential for the forecasting system, as the sub-models for the pollen emission, that is a dispersion model, a pollen release model and a phenological model for the starting date of flowering, are all sensitive to the quality of the weather data. Numerical forecasts of birch pollen concentration in springs have been done at FMI since 2005 and the model has been developed throughout these years. The latest version of the forecasting system gives realistic result, despite a tendency to underestimate pollen concentrations. Especially the timing of long-range transport episodes is well predicted. Keywords Biogenic emission term, birch, long-range transport, phenological model, pollen
1. Introduction Emission modelling of naturally produced pollutants, including allergenic pollen, is a major challenge for air quality models. Spring-time pollination during flowering is sensitive to a wide variety of factors, including prevailing weather, last-year meteorological conditions, biological condition of the plants, etc. Start of the flowering can vary from year to year up to a month on one site only, depending on temperature during the spring. Since pollen grains can be transported over hundreds of kilometres (e.g., Sofiev et al., 2006a, Siljamo et al., 2007a, b), a European-scale prediction of pollination season is an imperative for reasonable forecasts over any part of the continent.
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
154
Development and Applications of Biogenic Emission Term
155
Spring phenological events are commonly modelled and predicted using Thermal Time-type models, where temperature is integrated over time until the accumulated temperature sum meets a prescribed threshold value, manifesting the change of a phenological phase (e.g., start of flowering). In the current work, this model is, for the first time, parameterized using a compilation of multi-annual data from over 6,000 phenological stations and about 300 pollen monitoring stations across Europe. The results of application of the above emission term in operational pollen forecasting system, as well as its performance during springs 2006 and 2007 are presented below.
2. Materials and Methods Figure 1 shows a schematic picture of SILAM pollen long range transport modelling system. The system is based on input flow of meteorological information obtained from numerical weather prediction (NWP) models. Both phenological models for the starting and ending date of flowering and a pollen release model itself need information from the NWP model. This information controls also transport and sinks processes of pollen grains. Release of pollen grains, as presented in our model, is a complex system which contains a map of the fraction of birches in Europe (Sofiev et al., 2006a) as well as a starting date of flowering, number of catkins and a pollen release model.
2.1. Numerical weather prediction models and dispersion model Two main sources of meteorological information are used for the forecasts. The operational information comes from the HIRLAM model. It covers the whole Europe in space and goes up to 54 hours to the future time. The ECMWF’s ERA-40 analysis data going back in time for more than 40 years were used for long-term analysis of the phenological information and compilation of the Thermal-Time flowering model. The key parameter from the ERA-40 data was the temperature at 2 m above the ground. Resolution of ERA-40 analysis data is 6 hours and 1.125º. Pollen long-range transport forecasts are done with the dispersion modelling system SILAM (Sofiev et al., 2006b). It is capable both forward and inverse simulations, for which Lagrangian random-walk particle model and Eulerian dynamic cores are available. Current pollen simulations use SILAM v4.0.1 with Eulerian dynamics.
P. Siljamo et al.
156 Results: Birch pollen concentrations, temperature-sum, pollen left in the catkins
Pictures to www-page: http://pollen.fmi.fi
Dispersion modelling system SILAM Pollen release model (T, RH, V, rain)
Fraction of birch
Start and end of flowering (Phenological models) Relative number of catkins
Numerical weather prediction model (HIRLAM, ECMWF, ERA-40) Fig. 1 Schematic picture of SILAM birch pollen long-range transport forecast system
2.2. Phenological data, pollen observations and phenological models The starting time of flowering is forecasted using a Thermal Time-type phenological model (e.g. Häkkinen et al., 1998; Linkosalo, 2000). The temperature-sum based model is parameterized via fitting to both leaf unfolding and pollen observations of birch. Optimization method was gradient minimization for one, two and three fitted parameters. In one-parameter fitting, only the temperature-sum threshold for flowering was varying while the starting date of temperature-sum accumulation and temperature cut off limit were fixed. In two-parameter fitting, the starting date of temperature-sum accumulation was fixed and the cut off temperature and temperature-sum were varying. In three-parameters fitting, all three variables were varying. Phenological data were taken from the database collected within the POLLEN project of the Academy of Finland. The database contains information about bud burst, leaf unfolding and the first flowering day from 15 countries, for more than 6,500 stations and 60 000 data points. The main part of the data was collected after 1985 but the oldest observations go back to the mid-18th century, with the longest time series covering a time period of over 30 years.
Development and Applications of Biogenic Emission Term
157
Most of the available phenological data consists of dates of leaf unfolding. Fortunately the birch pollination starts within a few days of the leaf unfolding, so the latter can be used as substitute data also to estimate the timing of flowering. Phenological observations are in many cases made by amateurs, and are not done on daily basis, but usually twice or three times per week. This has created certain problems in the flowering model parameterization and introduced extra noise into the system. To minimize the influence of these uncertainties, the phenological analysis was combined with direct pollen concentration observations. Birch pollen observations were available from European Aeroallergen Network (EAN), which receives the data from about 35 countries and about 300 sites. Pollen observations began in 1974 but the bulk of data is after 1985. Pollen observations are always made by professionals and have the time resolution one day or better. However, the concentration observations cannot distinguish between the local and long-range transported pollen, thus reporting too early start of the season in northern Europe (Ranta et al., 2006) and too late end of the season in southern Europe. Pollen observation are also sensitive to weather. For example, the birches can be ready to flower, but they can not start it because it is too rainy or too humid, so pollen can not be observed despite the flowering stage has been actually reached. This was a case, e.g., in Finland in spring 2007.
3. Results
3.1. Phenological model (start of flowering stage) and pollen release model Both leaf unfolding phenological stage observations and pollen observations were used for the temperature-sum model in SILAM forecasts. Pollen observations were taken in Southern and Central Europe, while the phenological observation used in Northern Europe, where pollen long-range transport disturbs birch pollen observations. Since multi-parameter optimization did not improve results substantially, a temperature-sum model with fixed starting date of heat accumulation and temperature cut off limit was selected. Starting date of heat accumulation utilized was the 1st of March and temperature cut of limit was 3.5ºC. The temperature sum threshold resulting from the optimization procedure over Europe varied between 85 degreed days (dd) and 45 dd, but over large area it was around 52 dd. Because uncertainty in phenological data is large, a blurring filter with 20% span was applied to the threshold in the model. This means that, for instance, if optimal temperature-sum is 50 dd, the pollination gradually starts, when temperature-sum reaches 40 dd, with full-flowering reached by 60 dd. The pollen release model describes an intensity of flowering at each time moment and the end of pollination. The end of the season is determined from the principle of the “open pocket”, which is based on two quantities: total amount of
158
P. Siljamo et al.
pollen accumulated in catkins and integrated release rate. The rate is dynamic and driven by meteorology. It continues until all pollen grains have been used up. Temperature, wind speed, humidity and precipitation affect the release rate of pollen grains. If temperature is below the cut off limit, pollen release is suppressed. Also precipitation and high humidity restrain pollen grains falling out of the catkins. Significantly below the corresponding thresholds, these variables do not affect the release rate. In particular small precipitation is allowed, because it is not possible to get short-time, small-scale convective precipitation cell from the model. Such patchy rain still allows the trees to emit pollen grains from the dry parts of the grid cell area. Wind, to the contrary, promotes the pollen release. In case of no wind but strong convection, turbulence-induced wind may be sufficient to kick-start the release. When it is windier, wind promotes the release by picking the grains from open catkins, but the rate increases until the forcing wind speed reaches the saturation level of 5 m/s.
3.2. Predicted birch pollen concentrations in Finland April 2006 was cold in Finland and birch flowering started in southern Finland on the beginning of May. After long, cold period, temperature rapidly increased. Thus, the pollen season was short but intensive and pollen counts increased up to thousands of grains per cubic-meter (Figure 2a). At the end of April and beginning of May, strong pollen long-range transport episode from Russian forests dominated pollen observations in Finland (while local flowering was about to start) as well as in the Northern Germany and Denmark (where local flowering was almost over) and even in Iceland, where all observed birch pollen was long-range transported. As is shown in Figure 2a, the shape of the pollen concentration in Turku, Finland, is well predicted, but the SILAM-predicted birch pollen concentrations are below observed. A reason for that can be the very short and intensive pollen season, when all grains fell away from catkins almost simultaneously and the model was not able to catch the situation. Winter 2006–2007 was warm in Europe and also March 2007 was warmer than average. This resulted in birches to be almost ready to flower very early. However, after warm March, April came colder and rainy. Birches did not start their flowering in the cold, humid or rainy conditions. When the flowering finally started, further rainy and humid days suspended pollination, so the pollen season was long but not that intensive (Figure 2b). Birch pollen concentrations forecasted by SILAM were not as accurate in Finland in 2007 than in 2006. A reason for this is a negative bias (about 1.5ºC) in HIRLAM’s 2 m temperature forecasts in northern latitudes in April, which resulted in very late prediction of pollination in the model. While the birches started to flower in Turku on the 19th of April and again after a rainy period on the 25th of April, the model reached the temperature sum threshold only by 10th of May.
Development and Applications of Biogenic Emission Term
a)
159
b)
Fig. 2 Daily averages of birch pollen counts and SILAM pollen concentration in Turku: (a) April 15–May 15, 2006. Low pollen counts <10 grains/m³, moderate 10–100 grains/m³, abundant >100 grains/m³; (b) April 15–May 14, 2007
However, a long-range transport episode during the 15th–18th of April was predicted correctly, and when the flowering in the model had started, the predictions quickly got correct again (Figure 2b).
3.3. Daily pollen concentration forecasts Daily birch pollen long-range transport forecast have been done at FMI in springs since 2005. The system has contained SILAM dispersion modelling system, a map of the fraction of birch in Europe, and a numerical weather prediction model and a simple release and starting time of flowering, but in 2007 the forecasts contained the first time temperature-sum based starting and ending dates of flowering, manually compiled forecast of the relative number of catkins compare to average year and a flowering model which is inside the SILAM-model. The forecasts were presented at the project web site http://pollen.fmi.fi.
4. Discussion Birch pollen forecasting system is a combination of many models and all its compounds have strong and weak points. The whole system is driven by NWP model and strongly depends on its quality. Thus, the temperature-sum model for the starting date of flowering is very sensitive to any bias in the NWP-model, because accumulation period before reaching heatsum can be two to three months and already a temperature bias of 0.5ºC has a major effect on the accumulated temperature sum. To the contrary, zero-mean fluctuations are very efficiently integrated out during this long accumulation time.
160
P. Siljamo et al.
Birch flowering is also sensitive to actual weather parameters, such as precipitation or humidity, which can delay the flowering. Boundary layer height and winds are important to the transport module of the dispersion model. Other input data, such as the fraction of birch or the relative number of catkins, also significantly affect model accuracy and inherit their own uncertainties. There is still a lack of knowledge of the total amount and distribution of birches in Europe, and the total amount of grains, based on the estimates of catkins accumulated from previous year is an educated guess of professional aerobiologists. Despite all these uncertain elements, pollen concentration forecasts are useful, especially if users understand the system and at least in basics of its submodels. Experience shows that the accuracy of the pollen concentration forecasts tends to decrease when the local flowering season starts, because the local influence of nearby sources is more irregular and stochastic than long-range transported plumes, which are better mixed in space. Secondly, majority of pollen grains released tend to remain in the vicinity of the source because of the large size of grains. Thus, with only minor disagreement of model temperature-sum computations or threshold values and local peculiarities in the vicinity of the station, the forecasted concentrations might get dramatically different. In the case of long-range transported pollen, this effect is smoothed out due to synchronising effect of synoptic-scale meteorology and larger source areas, so that the uncertainties in the heat-sum computations are not that harmful. For the pollen forecasts, also low concentrations are important. Already concentrations as low as 10 grains/m³ can harm the most sensitive birch allergic persons. While pollen counts can exceed thousands of grains/m³, already 100 grains/m³ is considered abundant. The importance of low concentrations makes modeling a particularly challenging task because it implies that the model should catch tails of several days-long long-distance episodes and also the start- and end of the local flowering (when just a fraction of trees participates in pollinating). Experience of last two years shows that the new system manages this task quite well. Thus, in 2006, after April 22, SILAM pollen forecasts were in the right range in 16 cases out of 20 despite the pollen concentration was sometimes factor of ten too low (Figure 2).
5. Conclusions Since spring 2007, SILAM long-range transport forecast system for birch pollen has consisted of several sub-models covering the whole range of processes responsible for development, release, transport, and removal of pollen grains. The dispersion model SILAM represents a core of the birch pollen forecasting system, with the phenological models for starting and ending dates of flowering and the pollen release model linked to or embedded in it. The main input parameters are meteorological fields, maps of birch distribution over Europe and the forecasted relative number of catkins.
Development and Applications of Biogenic Emission Term
161
Starting date of flowering is forecasted using Thermal Time -type phenological model, where the temperature-sum accumulation starts on the 1st of March and cut-off temperature is 3.5ºC. Temperature-sum threshold value is about 52 degreedays over large areas in Europe. Forecasting the temperature-sum model for starting date of flowering is very sensitive to the near-surface temperature in the NWP model. Its bias can cause large errors in predicted starting date of pollination. The pollen release model is also sensitive to several other weather parameters. Rain and high humidity can suspend the pollination, while higher temperature and wind promote it by peeling the pollen grains from catkins. Birch pollen concentration forecasts have found to be particularly useful in forecasting the long-range transported pollen before or after the local flowering season. Simulations for 2006 and 2007 showed that the current model setup tends to underestimate pollen concentrations but provides very accurate timing of the long-range transport episodes in most cases. Acknowledgments The current study is a part of a project of Academy of Finland “Evaluation and forecasting of atmospheric concentrations of allergenic pollen in Europe” (http://pollen.fmi.fi), and the ESA PROtocol MOniToring for the GMES Service Element: Atmosphere (PROMOTE) project (http://www.gse-promote.org).
References Häkkinen R, Linkosalo T, Hari P (1998) Effects of dormancy completion and the environmental factors on timing of bud burst in Betula pendula, Tree Physiology 18, 707–712 Linkosalo T (2000) Mutual regularity of spring phenology of some boreal tree species: predicting with other species and phenological models, Canadian Journal of Forest Research 30, 667–673 Ranta H, Kubin E, Siljamo P, Sofiev M, Linkosalo T, Oksanen A, Bondestam K (2006) Long distance pollen transport cause problems for determining the timing of birch pollen season in Fennoscandia by using phenological observations, Grana 45(4), 297–304 Siljamo P, Sofiev M, Ranta H (2007a) An approach to simulation of long-range atmospheric transport of natural allergens: an example of birch pollen in Air Pollution Modeling and Its Application XVII (edit. C Borrego and A-L Norman), Springer, New York, pp. 331–339 Siljamo P, Sofiev M, Severova E, Ranta H, Polevova S (2007b) On influence of long-range transport of pollen grains onto pollinating seasons in Air Pollution Modeling and Its Application XVIII (edit. C Borrego and E Renner), Developments in Environmental Science, Vol 6, Elsevier, Amsterdam, The Netherlands, pp. 708–716 Sofiev M, Siljamo P, Ranta H, Rantio-Lehtimäki A (2006a) Towards numerical forecasting of long-range air transport of birch pollen: theoretical consideration and a feasibility study, International Journal of Biometeorology 50, 392–402
P. Siljamo et al.
162
Sofiev M, Siljamo P, Valkama I, Ilvonen M, Kukkonen J (2006b) A dispersion system SILAM and its evaluation against ETEX data, Atmospheric Environment, 40, 674–685
Discussion S.-E. Gryning: How did you model the deposition? P. Siljamo: Dry deposition is computed via standard resistance analogy, wet deposition is based on scavenging coefficient and distinguishes between snow and rain, in- and sub-cloud scavenging. Y.P. Kim: 1. Average residence time of pollens in the air. 2. How important is the resuspension of pollens in the overall processes? 2. There exists resuspension of pollen grains and after long dry period, like in spring 2006, it might be important. Usually it is not considered important and it is not involved in our pollen dispersion model. P. Siljamo: 1. Because birch pollen grain is big (20 ȝm in diameter), most of observed pollen comes from local sources. On the other hand, birch pollen grain is light (800kg/m³) and thus it can stay in the atmosphere a few days and transport over 1,000 km if weather conditions are suitable. Despite average residence time of birch pollen is not very long (half-time about one day), the most lucky grains can transport long distances. Already small amount of pollen can be significant to allergic people. When pollen counts can be thousand or even more than 10,000 grains/m³ near source area, already 10 grains/m³ can cause symptoms to the most sensitive people and 100 grains/m³ is considered as abundant. G. Kallos: Why you model only birch pollen and not from other trees? You call it European model but you do not cover pollen sources from other trees, especially from Southern and Central Europe, this is not true. P. Siljamo: We chose birch as the first species, because its pollen is one of the most important allergen carrier in Central and Northern Europe. Despite the area of birch distribution is mainly limited to Northern and Central Europe, the modelling covers the whole of Europe and, e.g., Spain and Italy do get birch pollen every year – both in observations and in the model results. Numerous scientific and technical problems take time and make the work exciting but comparatively slow from the user’s point of view, but expanding is on-going to include grass and olive pollens and, possibly, others.
2.13 Development and Verification of TAPM Peter Hurley
Abstract TAPM consists of an inline, nested, prognostic meteorological and air pollution model that solves the fundamental equations of atmospheric flow, thermodynamics, moisture conservation, turbulence and dispersion, wherever practical, for horizontal modelling domains of up to 1,500 km in size and horizontal grid spacing typically from 30 km down to 300 m for meteorology, with optionally even finer grid spacing for air pollution. The meteorological component of the model is nested within large-scale analyses/forecasts, which drive the model at the boundaries of the outer grid. The use of integrated modules for plume rise, Lagrangian particle, building wake, and Eulerian grids with condensed chemistry schemes, allows industrial and urban air pollution to be modelled accurately at fine resolution for long simulations. TAPM has been verified for a number of Australian and international datasets, and results from these studies have shown good model performance for both meteorology and air pollution predictions, particularly for the study of annual extreme (high) concentrations important for environmental impact assessments. In this paper, the development and verification of TAPM V3 will be summarised, along with current development and verification studies that will eventually be available in a future version of TAPM. These new developments will include options for more complete land surface schemes and global datasets for soil and vegetative canopies, enhanced turbulence schemes, and the addition of several comprehensive meteorology and turbulence and air pollution verification datasets. Keywords Air pollution, meteorology, modelling, verification, TAPM
1. Introduction Air pollution models that can be used to predict hourly pollution concentrations for periods of up to a year, as often required for regulatory purposes, are generally semi-empirical/analytic approaches based on Gaussian plumes or puffs, and usually either ignore chemistry or treat it very simply. They typically use either a simple surface-based meteorological file or a diagnostic wind field model based on available observations. However the required meteorological information is not always available or not available in sufficient detail, which can make application of these models difficult. Moreover, for short-range dispersion under complex flow and diffusion conditions (e.g. in coastal regions or complex terrain) these types of C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
208
Development and Verification of TAPM
209
models are either not applicable or the simple assumptions and extensions used therein lack generality. An alternative approach is to use airshed models and/or dispersion models coupled to prognostic meteorology from a mesoscale model. However, due to long run times, model simulations are generally restricted to case studies for most of these types of models. The model presented in this paper, TAPM, is a fast prognostic model that uses the complete equations governing the behaviour of the atmosphere and the dispersion of pollutants. TAPM uses large-scale weather information (analyses or forecasts) typically available at a horizontal grid spacing of 100 km, as boundary conditions for the model outer grid. TAPM then ‘zooms-in’ to model local scales at a finer resolution using a nested approach, predicting local-scale meteorology. A prognostic approach eliminates the need to have site-specific meteorological data to drive the dispersion model, but allows assimilation of observations if desired. All input data sets, except emissions, accompany the model and are easily transferred through a graphical user interface to nested grids for the region of interest. This paper presents an overview of the model and summarises model performance for a number of verification studies for both TAPM V3.0 and for a new version (V3.5) currently under development.
2. Model Overview TAPM is a PC-based, nestable, prognostic meteorological and air pollution model driven by a Graphical User Interface (GUI). It was designed to be easy to use and fast to run, but also to be based on comprehensive science. Datasets of important inputs needed for meteorological simulations accompany the model, allowing model set up for any region, although user-defined databases can be connected to the model if desired. The only user-supplied data required for air pollution applications are emission information. The model outputs can be examined easily and quickly, with various types of output processing options provided. TAPM uses the fundamental equations of atmospheric flow, thermodynamics, moisture conservation, turbulence and dispersion, wherever practical. For computational efficiency, it includes a nested approach for meteorology and air pollution, with the pollution grids optionally able to be configured for a sub-region and/or at finer grid spacing than the meteorological grid, which allows the user to ‘zoom-in’ to a local region of interest quite rapidly. The meteorological component of the model is nested within large-scale analyses/forecasts that drive the model at the boundaries of the outer grid. The coupled approach taken in the model, whereby meteorological fields are passed to the pollution module every 5 minutes, allows pollution modelling to be done accurately during rapidly changing conditions. The use of integrated plume rise, Lagrangian particle, building wake, and Eulerian grid modules, allows industrial plumes to be modelled accurately at fine resolution for long simulations. Similarly, the use of a condensed chemistry scheme also allows nitrogen dioxide, ozone, and particulate mass to be modelled for long periods.
210
P. Hurley
2.1. TAPM V3.0 TAPM V3.0 was released in May 2005 (Hurley et al., 2005c). A complete technical description of the model equations, parameterisations, and numerical methods are described by Hurley (2005), and a summary of some verification studies is given in more detail by Hurley et al. (2005a). A summary of the characteristics of the meteorological component of TAPM is: x 3D Eulerian grid, nestable, shallow, incompressible and optionally nonhydrostatic x Predicts winds, temperature, pressure, water vapour, cloud water/ice, rain, snow and turbulence for regions less than 1,500 km in size x Surface scheme for water, soil, vegetation and urban categories x Radiation and cloud microphysics schemes x Wind data assimilation and user-defined databases options x Forecasting option – need access to synoptic-scale weather forecasts A summary of the characteristics of the air pollution component of TAPM is: x 3D Eulerian grid, nestable, shallow and incompressible x Prognostic concentration mean and (optionally) variances x Emissions from point, line, area/volume or gridded sources x Tracer mode (up to four tracer groups) x Multiple pollutant mode (PM10, PM2.5, NOX, NO2, O3, SO2) x Dust mode (PM30, PM20, PM10, PM2.5) x Lagrangian particle module (point sources) x Advanced numerical plume rise module and building wake module x Photochemistry (GRS mechanism), aqueous chemistry and deposition x Particle gravitational settling x Peak-to-mean ratios using concentration mean and variance x Optional offline mode to allow pollution to run from saved meteorology x Option to configure and run CSIRO’s CTM for urban applications that may require more complex chemistry (e.g. carbon bond).
2.2. Enhancements for TAPM V3.5 TAPM continues to develop in response to new science and the needs of the various applications that the model is used for, subject to available resources within CSIRO. This section outlines two of the main areas of model development in V3.5, namely, improving land surface and turbulence and convection processes. 2.2.1. Land Surface Scheme
The TAPM land surface scheme is described in detail in Hurley (2005) and Hurley et al. (2005c). It consists of a single-level force-restore equation for soil temperature and moisture content, a surface energy balance approach for vegetation
Development and Verification of TAPM
211
temperature and moisture, and a force-restore approach for urban surface temperature. The equations are solved for surface temperature and moisture, and the surface fluxes of these variables are computed using Monin-Obukov surface layer similarity. The various contributions of the variables for each surface type are weighted by the fractional coverage of each type within a grid cell to produce surface temperature and moisture, and momentum, temperature and moisture fluxes. The TAPM V3.5 land surface scheme uses a different approach to calculate the surface temperature, moisture and fluxes that overcomes a number of limitations of the old surface scheme for soil and vegetation. The main features of the scheme are: x Vegetation now vertically overlays the soil, rather than treating the soil and vegetation as separate horizontal fractional patches that make up the total horizontal surface. This allows more realistic interaction between the vegetation and soil, including for radiation and fluxes, and eliminates the need for a value of the vegetation fraction to characterise each type. x Soil temperature and moisture content are now calculated using a 15-level three-dimensional numerical solution of the heat diffusion equation for temperature and Richards’ equation for soil moisture content (e.g. see the formulation in Pielke (2002)). This approach is more realistic than the force-restore equations used in the old surface scheme, and allows the soil temperature and moisture to vary with depth and evolve over time in response to changes in surface fluxes, rainfall, and soil and vegetation processes. The scheme also allows thirteen soil types to be used (note that the old scheme was an approximate fit to the types of equations used in the new scheme for only three soil types). Because of these improvements the deep soil input parameters are now only used as initial conditions. A zero-plane displacement height has also been included in the new scheme. This should help to improve wind speed predictions in forested or urban areas with high canopies. In addition to these changes to the model, two extra default global databases for soil type and Leaf Area Index (LAI) are provided with the model. 2.2.2. Turbulence and Convection Scheme
The turbulence scheme in TAPM V3.0 is described in detail in Hurley (2005) and Hurley et al. (2005c), and consists of two prognostic equations for turbulence kinetic energy and eddy dissipation rate, which feed into an eddy diffusivity for momentum, heat and scalars. There is also a simple (constant) counter-gradient flux that is included for heat. The Lagrangian Particle Module (LPM) for near-source pollution dispersion also needs the horizontal and vertical velocity variances that are determined diagnostically from the mean meteorological variables and the two prognostic turbulence variables. In TAPM V3.5, this approach has been modified to use the eddy-diffusivity/ mass-flux (EDMF) approach based on Soares et al. (2004) to replace the (constant) countergradient term with one based on a mass-flux scheme. This approach gives a more general framework to include the effect of the large eddies in the convective boundary layer (CBL) through the mass-flux term while using the same prognostic turbulence approach as before for the smaller scale turbulence. An extension of this
P. Hurley
212
approach to include a mass-flux component to the vertical velocity variance and eddy dissipation rate are used to improve the turbulence representation in the upper-half of the CBL for use in the LPM in TAPM. The new formulation has been shown by Hurley (2007) to represent mean and turbulence fields in the CBL very well for dry convective conditions and, as demonstrated by Soares et al. (2004), has the potential to unify the treatment of CBL turbulence and moist shallow convection within the one approach.
3. Model Verification 3.1. TAPM V3.0 results As described in Hurley et al. (2005a, c), TAPM V3.0 has been run for a number of verification studies including laboratory experiments of concentration mean and variance and dispersion within building wakes, for Kalgoorlie SODAR upper-level wind comparisons, for the Kincaid and Indianapolis international tracer studies (SF6), and for the following annual evaluation studies: Anglesea 2002 and 2003 (SO2); Kwinana 1997 (SO2); Bowline 1981 (SO2); Lovett 1988 (SO2); Westvaco December 1980–November 1981 (SO2); Pilbara 1999 (NOX, NO2, O3); and Melbourne July 1997–June 1998 (NOX, NO2, O3, PM10, PM2.5). Table 1 Observed and predicted RHC (robust highest concentration) for each study. For Kincaid and Indianapolis, RHC are calculated for the arc-wise maximum concentrations (scaled by the emission rate) at several distances downwind of the source for all hours, with corresponding RMSE (root mean square error) over all distances/hours. For all other (annual) datasets, the monitoring site concentrations of annual RHC are averaged over all monitoring sites, with corresponding RMSE over all sites. Kincaid and Indianapolis results are shown for Q2&3 data (highest two classes of data quality) and Q3 data (highest class of data quality). Dataset
#Sites
OBS
TAPM
RMSE
Kincaid Q2&3 SF6 (scaled)
n/a
291
364
–
Kincaid Q3 SF6 (scaled)
n/a
275
246
– –
Indianapolis Q2&3 SF6 (scaled)
n/a
1,248
1,367
Indianapolis Q3 SF6 (scaled)
n/a
1,175
1,345
–
Bowline SO2 (µg m-3)
4
439
325
192 74
Lovett SO2 (µg m-3)
9
228
191
Westvaco SO2 (µg m-3)
11
1,852
1,455
691
Anglesea SO2 (µg m-3)
4
595
625
103
Kwinana SO2 (µg m-3)
6
139
137
43
Melbourne NOX (ppb)
8
483
408
156
Melbourne NO2 (ppb)
8
75
77
13
Melbourne O3 (ppb)
9
97
88
17
Melbourne 24-hour PM10 (µg m-3)
5
49
45
8
Melbourne 24-hour PM2.5 (µg m-3)
3
34
29
6
Development and Verification of TAPM
213
Table 2 Observed and predicted MAX (maximum concentration) for each study. See Table 1. Dataset
#Sites
OBS
TAPM
Kincaid Q2&3 SF6 (scaled)
n/a
319
394
56
Kincaid Q3 SF6 (scaled)
n/a
319
269
50
Indianapolis Q2&3 SF6 (scaled)
n/a
1,154
1,289
262
Indianapolis Q3 SF6 (scaled)
n/a
1,068
1,289
302
Bowline SO2 (µg m )
4
441
299
224
Lovett SO2 (µg m-3)
9
244
246
76
Westvaco SO2 (µg m-3)
11
1,735
1,762
426
Anglesea SO2 (µg m-3)
4
525
610
136
Kwinana SO2 (µg m-3)
6
137
143
28
Melbourne NOX (ppb)
8
480
407
151
Melbourne NO2 (ppb)
8
76
76
8
Melbourne O3 (ppb)
9
93
90
15
Melbourne 24-hour PM10 (µg m-3)
5
48
44
7
Melbourne 24-hour PM2.5 (µg m-3)
3
34
30
7
-3
RMSE
Results for extreme (high) concentrations are summarised in Tables 1 and 2 for a number of these studies. The results show that TAPM performs very well for predicttion of extreme concentrations for both intensive field campaigns (Kincaid and Indianapolis) for arc-wise maximum concentration at various distances from the source and for annual extreme concentration statistics in the other studies. The statistics for the annual studies presented here consist of situations such as building wake affected dispersion (Bowline), sea breeze fumigation (Kwinana), flow and dispersion in complex terrain such as Anglesea, Lovett and Westvaco, and for meteorology and photochemistry in the complex terrain of urban Melbourne. The results give confidence in the use of TAPM for Environmental Impact Assessments in any region, even without the use of local meteorological data to drive the model. The only inputs not provided with TAPM are pollutant emissions. A model evaluation and inter-comparison of AUSPLUME, CALPUFF and TAPM has also been made by Hurley and Luhar (2005) for the Kincaid and Indianapolis datasets and by Hurley et al. (2005b) for the Anglesea and Kwinana datasets. A further model evaluation and inter-comparison of AUSPLUME, AERMOD and TAPM was made by Hurley (2006) for seven field datasets of point source dispersion, including Kincaid, Indianapolis, Anglesea, Kwinana, Bowline, Lovett and Westvaco. Results from these inter-comparison studies have shown that TAPM performs comparatively very well for these datasets.
3.2. TAPM V3.5 results Tables 3–5 show TAPM V3.5 annual meteorological performance statistics for: x The Kwinana (Western Australia) 1997 10 m tower dataset x The Cardington (England, UK) 2005 50 m tower dataset x The Bowline (New York, USA) 1981 100 m tower dataset
P. Hurley
214
TAPM was configured for each of these datasets using nested grids of 25 × 25 × 25 points down to a resolution of 1,000 m spacing for meteorology, and used default model settings and databases (except Bowline which used NCEP re-analyses). Table 3 Kwinana (Australia) Annual Meteorological Statistics for Observed (OBS) and for TAPM V3.5 Modelled (MOD) 10 m level wind speed (WS) and components of the wind (U and V), and temperature (T). AVG is the annual mean; STD is the annual standard deviation; RMSE is the root mean square error; IOA is the Index of Agreement (a value of 0.00 means no agreement and 1.00 means perfect agreement).
WS (m/s) U (m/s)
OBS
MOD
OBS
MOD
AVG
AVG
STD
STD
RMSE
4.0
3.9
1.7
1.5
1.32
0.81
–0.2
–0.6
3.2
3.2
1.72
0.93
IOA
V (m/s)
1.1
0.4
2.8
2.5
1.71
0.89
T (ºC)
18.2
17.8
5.2
5.2
1.95
0.96
Table 4 Cardington (UK) Annual Meteorological Statistics for Observed (OBS) and for TAPM V3.5 Modelled (MOD) 50 m level wind speed (WS) and components of the wind (U and V), and temperature (T). See the Table 3 caption for more information on the statistics. OBS
MOD
OBS
MOD
AVG
AVG
STD
STD
RMSE
WS (m/s)
5.8
5.8
2.9
2.5
1.46
0.92
U (m/s)
2.3
2.2
4.1
3.8
1.54
0.96
IOA
V (m/s)
1.0
1.0
4.4
4.4
1.66
0.96
T (ºC)
10.9
10.5
5.9
6.1
1.61
0.98
Table 5 Bowline (USA) Annual Meteorological Statistics for Observed (OBS) and for TAPM V3.5 Modelled (MOD) 100 m level wind speed (WS) and components of the wind (U and V), and temperature (T). See the Table 3 caption for more information on the statistics. OBS
MOD
OBS
MOD
AVG
AVG
STD
STD
RMSE
WS (m/s)
4.7
4.5
2.8
2.2
2.06
0.81
U (m/s)
1.7
1.9
3.6
2.7
2.15
0.87
IOA
V (m/s)
–1.0
–0.8
3.7
3.6
2.13
0.91
T (ºC)
11.4
10.7
9.9
9.9
2.47
0.98
Model results show excellent performance for these datasets, including good annual average and standard deviation, low Root Mean Square Error (RMSE) and high Index of Agreement (IOA) (for detailed definitions of the statistics see Hurley et al., 2005a). Compared to TAPM V3.0 results for these datasets, TAPM V3.5 shows improved average wind speed, improved temperature standard deviation, lower RMSE by approximately 0.3 m/s for wind and 0.5ºC for temperature, and slightly higher IOA.
Development and Verification of TAPM
215
4. Conclusions It has been illustrated here that TAPM V3 performs well for a number of model evaluation and inter-comparison studies, using quality datasets where both hourlyvarying emission rates and air quality monitoring data are well known. The results from these and other studies show that TAPM exhibits consistently good performance across multiple studies for meteorology and air pollution, even when local meteorological data are not assimilated into annual simulations. A new more physically realistic land surface scheme and turbulence scheme have been included in TAPM V3.5. Preliminary results show that the model performs very well at several sites in several countries around the world. Acknowledgments The encouragement and suggestions from CSIRO and external users of TAPM since it was first released in 1999 are gratefully acknowledged. Thanks to those organisations who have allowed access and use of datasets with TAPM – please see references for each study. Thanks also to the following organisations for provision and use of meteorological data: DoE Western Australia for Kwinana; UK Met Office for Cardington; and to the US EPA for Bowline.
References Hurley P (2005) The Air Pollution Model (TAPM) Version 3. Part 1: Technical Description, CSIRO Atmospheric Research Technical Paper No. 71. Hurley P (2006) An evaluation and inter-comparison of AUSPLUME, AERMOD and TAPM for seven datasets of point source dispersion, Clean Air, 40, 45–50. Hurley P (2007) Modelling mean and turbulence fields in the dry convective boundary layer with the eddy-diffusivity/mass-flux approach, Bound.-Layer Meteorology, 125, 525–536. Hurley P, Luhar A (2005) An evaluation and inter-comparison of AUSPLUME, CALPUFF and TAPM. Part I: The Kincaid and Indianapolis field datasets, Clean Air, 39, 39–45. Hurley P, Physick W, Luhar A, Edwards M (2005a) The Air Pollution Model (TAPM) Version 3. Part 2: Summary of some verification studies, CSIRO Atmospheric Research Technical Paper No. 72. Hurley P, Hill J, Blockley A (2005b) An evaluation and inter-comparison of AUSPLUME, CALPUFF and TAPM. Part II: Anglesea and Kwinana annual datasets, Clean Air, 39, 46–51. Hurley P, Edwards M, Physick W, Luhar A (2005c) TAPM V3 – Model Description and Verification, Clean Air, 39, 32–36. Pielke R (2002) Mesoscale Meteorological Modelling, International Geophysics Series, Vol. 78, Academic, San Diego, CA, 676 pp.
216
P. Hurley
Soares P, Miranda P, Siebesma A, Teixeira J (2004) An eddy-diffusivity/mass-flux parameterisation for dry and shallow cumulus convection, Quart. J. Roy. Met. Soc., 130, 3365–3383.
Discussion B. Fisher: I have heard criticisms of the GRS chemical scheme that is too simple. How do you respond to that? P. Hurley: The GRS chemical scheme used in TAPM is a relatively simple approach compared to those used in other CTMs. However, the scheme has been shown to work very well for prediction of Criteria pollutants such as NO2, O3 and PM10, and, in fact, it works as well as the more complex mechanisms for these species (e.g. see documentation and associated papers on the TAPM web page at www.cmar.csiro.au/research/tapm, particularly the Melbourne (urban) and Pilbara (industrial) verification studies). The highly condensed, but computationally fast, nature of the scheme allows good prediction of these pollutants without the overhead of carrying many extra species that often are not needed for regulatory purposes, and allows TAPM to be able to run for year-long regulatory applications to simulate the above pollutants at fine resolution. For applications that require a more complex chemical mechanism (e.g. to simulate species not contained in the GRS scheme, such as PAN or speciated VOCs), then we offer a TAPM-CTM option that uses the same TAPM meteorology, but with a more typical urban-scale pollution model containing the Carbon Bond Mechanism (e.g. IV, 99, 05), however this configuration is more complex to use and slower to run.
2.14 Development of Fire Emissions Inventory Using Satellite Data Biswadev A. Roy, George A. Pouliot, J. David Mobley, Thompson G. Pace, Thomas E. Pierce, Amber J. Soja, James J. Szykman and J. Al-Saadi
Abstract There are multiple satellites observing and reporting fire imagery at various spatial and temporal resolutions and each system has inherent strengths and limitations. In this study, data are acquired from the Moderate Resolution Imaging Spectro-radiometer (MODIS) aboard the National Aeronautics & Space Administration’s (NASA’s) Earth Observing System satellites. The MODIS-equipped satellite is polar orbiting with one daytime equatorial crossing and a 1-km2 resolution product. Fire-counts are obtained from two MODIS instruments aboard two different satellites having 10:30 AM and 1:30 PM equatorial crossing time, respectively. Here, a general methodology of processing the MODIS data is provided. An effective area burned estimate, obtained using the MODIS fire count product, is compared with fire occurrence and area burned estimates obtained independently from a 2002 ground-based fire database. Successful development and application of this technique for characterizing fire emissions in the United States (U.S.) could enhance the development of techniques for characterization of fire emissions for air quality modeling and its applications. Keywords Air quality, fire emissions, satellite observations of fires
1. Introduction Quantitative information about the spatio-temporal distribution of wildfires is indispensable to fire-ecology, wildlife and forestry management, as well as air quality management. Vegetation fires emit substantial amounts of carbon dioxide (CO2), carbon-monoxide (CO), methane (CH4), and nitrogen oxides (NOx). Emissions from biomass burning affect ambient concentrations of PM2.5, ozone (O3), air toxics, and regional haze. Fires contributed about 30% of the PM2.5 primary emissions (Pace and Pouliot, 2007) in the U.S. Environmental Protection Agency’s (EPA’s) 2002 National Emissions Inventory (NEI). Prior to 2002, characterization of fires in emissions inventory developmental activities was not time- and locationspecific. Annual emissions at the State level were apportioned to county-level (Pace, 2007). Inadequate spatial and temporal characterization of fires in the emissions inventory could lead to potentially ineffective plans for improving air quality. C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
217
218
B.A. Roy et al.
Hence, we strive in this project to improve the emission estimation methods for fires in the emission inventory to ensure that most appropriate air quality management decisions are made and are supported effectively by air quality modeling activities. Wildfire detection from space has been discussed by Cahoon et al. (2000) and Kaufman et al. (1990a, b). This project evolved from an initial study on wildfire emissions modeling that was initiated using the Blue-sky modeling framework (Pouliot et al., 2005). We enhanced the initial study by making use of the two sets of MODIS fire observations from sensors aboard the Terra satellite at 10:30 AM equatorial crossing time and Aqua satellite with 1:30 PM equatorial crossing time over the continental United States (CONUS). The combined map is produced on a daily basis for emissions inventory purposes. In this project, the fire count data product was obtained from the MODIS sensors using the rapid response contextual algorithm (Giglio et al., 2003), the mid-infra red, and the thermal infra-red bands (Justice et al., 2002). An attempt was made previously to incorporate MODIS fire counts using a technique (Justice et al., 2002) to reallocate the year 2001 NEI emissions estimates and to check the model performance in terms of its prediction of total carbon and particulate fine-mass in the surface layer. We have already seen an improvement in chemical transport model’s performance with MODIS-enhanced temporal and spatial resolution of the emissions from wildfires (Roy et al., 2007). In the present study, we have advanced these techniques by generating biomassburning emissions directly using the MODIS fire counts after establishing a reliable relationship between the MODIS pixel counts and the ground observed area burned estimates from the 2002 emission inventory (U.S. EPA, 2007). In 1990, the United States Congress amended the Clean Air Act (CAA) to require the EPA to address the regional haze problem, which is caused by emission of air pollutants from numerous sources located over a wide geographic region. As result of the Regional Haze Rule, five Regional Planning Organizations were formed across the United States to coordinate the affected states and tribes. The Regional Planning Organizations conducted an extensive effort to locate the size and timing of fires across the United States (U.S.). The results of this effort were incorporated into the 2002 National Emission Inventory (US EPA, 2007). The 2002 NEI provided groundtruth for the assessment and application of the MODIS data. Since the MODIS data from both Aqua and Terra instruments were only available for the August– December 2002 period, we concentrated our efforts on data for this period. A suitable clustering technique is applied to produce daily maps of clustered fire counts over the CONUS before establishing the relationship between fire count and area burned. In order to develop a fire inventory, we consider the rapid response fire count product from the two MODIS instruments. The MODIS is a cross-track sensor and, because of the bow-tie effect, the dimensions of the remotely sensed pixel increases with swath angle; hence, the modified scan and track dimensions exceed the 1 km2 regular pixel size which occurs only at the nadir. We have noticed that the pixel area sensed by MODIS increases monotonically with the total number of pixels clustered for each fire location. The minimum detectable fire size with 100% probability of detection is 100 m2 when the satellite is directly over the fire (nadir view); whereas for the non-nadir
Development of Fire Emissions Inventory Using Satellite Data
219
view, the probability of detection is 50% and the minimum detectable size is 50 m2. Section 2 of the paper describes how MODIS fire counts are related to the groundobserved, and area burned estimates. The area burned per pixel count for various fire categories is described as well as the methodology of using the area burned per pixel information to obtain the emission rates in the inventory. In Section 2, we also provide a case study using the large Biscuit fire that occurred during the summer period of year 2002 in the state of Oregon. Section 3 provides a summary of this newly developed method and also provides some suggestions about potential improvement using multiple satellite products.
2. Relationship Between Fire Count and Area Burned 2.1. Creation of clustered fire count map for estimation of burned area per MODIS pixel Our interest is the whole domain of the CONUS; hence, we need to allocate both the MODIS and NEI data at the 12 × 12 km horizontal grids. For the purpose of clustering, we have developed a scheme that first accumulates the pixel counts over all the grid cells that contain at least one candidate fire event based on the 2002 NEI, and then we integrate the daily MODIS pixel counts over all the days of the fire. This is represented by the following formula where Np is the accumulated pixel count, L is the life-time of the fire, and n is the total number of pixels registered over a period of one day at the fire location.
Np
L
n
¦ ¦N
(1)
fireday 1 1
If MODIS pixel found, count as Np (Eq. 1).
1
2
3
8
Candidate pixel (CP) in NEI 6
4
-1 DAY
7
Step 1: If MODIS pixel not found in CP, search adjoining cells 1 through 8 and count if found. +1 DAY
5
Step 2: If MODIS pixel not found in adjacent cells, search MODIS detect within all 9 cells for 1 day delay and if still not found extend the search for 1 day advanced and count if found.
Fig. 1 The scheme for MODIS clustering using Eq. (1) and match-up with ground reports using the 12 km grid
220
B.A. Roy et al.
There is an adjacency test done for each observed fire pixel (candidate fire in the 2002 NEI). MODIS data are searched first within the target area; if not found, then a search is done for detects in all the adjoining eight pixels as Step 1; then, if not found, Step 2 is exercised by repeating Step 1 but using a map for the previous and/or the next day. This is done to assure all appropriate fire counts are considered. The counting method is summarized in the schematic as shown below: For processing purposes, we define three types of fires – “large” fires for size >10,000 acres per month; “medium” fires for sizes between 1,000 to 10,000 acres per month; and “small” fires for sizes <1,000 acres per month. In order to combine the ground based data with the MODIS data, we check the NEI to locate “large,” “medium,” and “small” fires and remove the duplicate detects occurring on the same day and over same location from the MODIS data set. For large fires, we match MODIS pixels with an NEI fire using the Federal Information Processing Standards (FIPS) code and the date. For small and medium fires, we match MODIS pixels with the NEI fire based on a 12 km gridded domain and then follow the test for obtaining a “match” of the fire following the method described in Figure 1. Then, we examine the land-use data in the area of the unmatched MODIS pixels. If the land-use is greater than 25% agriculture, then we assign the fire as an agricultural fire. This process results in the fire inventory being divided into five groups: (a) “large” fires with matching MODIS pixels and NEI acreage, (b) “small” and “medium” fires with matching MODIS pixels and NEI acreage, (c) “unmatched” NEI fires, (d) “unmatched” MODIS pixels that are assigned as agricultural fires, and (e) “unmatched” MODIS pixels that are not agricultural burning. The PM2.5 emissions and acres burned for each category of fire are shown for all months of year 2002 in Figure 2 below. There are many unmatched cases in the MODISNEI data set. Soja et al. (2007) found 50% coincidence in MODIS and NEI data in Oregon and 20% coincidence in Arizona (coincidence criteria: 3 km distance between areas, time expanded four days on the beginning and end of NEI data range). However, because larger fires are detected, the area burned within the 50 and 20% of the number of fires is 97 and 69% of the wildland fires in Oregon and Arizona, respectively. There are several reasons for these discrepancies. First, MODIS instruments are overhead only 1 per day (2 instruments + 1 nighttime overpass = 4 total overpasses per day), although there is some overlap, particularly at higher latitudes. The instruments can not capture fires that burn when they are not overhead, particularly short-lived and/or agricultural fires. Additionally, cloud cover, high background temperatures, solar reflectance and thick canopy cover can obscure the instruments ability to detect fire (Flannigan et al., 1986; Robinson, 1991; Soja et al., 2004). Secondly, there are imperfections in the NEI ground-based data, which include errors in the dates and locations of recorded fire wildfires that vary by state. Some states have excellent records, others did not require precise record keeping, and others recorded many fires at the county centroid. Additionally, the location for non-federal rangeland burning is not correct, although the area burned estimates are correct (NEI reference). Nonetheless, we believe the NEI data is the best area burned data available in the US, which when paired with satellite data can help improve emission estimates.
Development of Fire Emissions Inventory Using Satellite Data
221
There are particular discrepancies in prescribed and managed fires in the Southeastern (SE) states that require additional detailed analysis. Hence, for the SE region, we classify the unmatched area burned to prescribed and managed burns. For the ‘unmatched’ cases for other regions, we consider them as ‘wildfires’ For the agricultural burns, we use the fire related NEI emissions reallocation using the MODIS pixel counts as described in Roy et al. (2007).
Fig. 2 The burned area in acres per month and corresponding PM2.5 emissions (tons) for the different groups of the NEI fire inventory
Fig. 3 (a) Scatter plot for NEI burned area (in acres) versus MODIS cumulative pixel count (Np) by considering all fires in August–December 2002, (b) same as 4(a) but, scatter plot for the fires having Source Classification Codes (SCC) as ‘wildfire’ in the inventory
Using the matching methodology described above, the acres burned per pixel by region and month are estimated for the large, medium, and small fires, respectively. For the year 2002, we have analyzed the August–December 2002 NEI and obtained the burned area in acres per pixel on a per fire basis during the whole period. Figure 3 displays the relationship between the area burned from NEI data and MODIS pixel counts for all fires for the five month period.
B.A. Roy et al.
222
Figure 3a shows the linear-fit between the NEI burned area versus MODIS pixel count and Figure 3b shows a similar relationship for the select fires that have the source classification codes as “wildfire” in the NEI. The correlation improves for the wildfire case alone, and the acres per pixel changes from 63 for all fires to 77 for wildfires. We could improve the correlation further by doing the computation on a per fire basis and by excluding the fires that are “partially” detected by MODIS, i.e., the fires whose lifetime as detected by MODIS do not match up with the NEI data even if the locations match up well. For small fires (<200 acres), it is difficult to establish the relationship (as shown in Figure 3a) since the MODIS frequently loses its detection capability due to its restricted field of view (Giglio et al., 2003). Hence, we will rely mostly on the ground reports for these small size fires. For the entire CONUS there were 65% of MODIS pixels that matched up with the ground reports, and in terms of area wild fire related area burned, MODIS matched up with approximately 91% of the total burned area acreage reported in the NEI.
2.2. Using burned area per MODIS pixel for emissions estimate Once Np is ascertained for each ground reported fire, then the daily area burned, A(i,t), in a spatial region (labeled by index ‘i’) and during a fixed time period (labeled by index ‘t’) could be calculated using the following equation where Į is the constant of proportionality (acres burned per MODIS pixel count):
A(i, t ) DN p (i, t )
(2)
Fig. 4 Left panel shows the MODIS imagery captured on August 12 (Image courtesy: NASA http://veimages.gsfc.nasa.gov/3448/California.A2002224.jpg), 2002, and the right panel shows the fine mass (PM2.5) emissions obtained from the Biscuit fire in Oregon. The circles and the squares represent respective fire clusters that match with the imagery
Comparing Figure 3a with b showing the linear fit between the NEI derived area burned and MODIS pixel counts, we can see that the correlation between area burned and MODIS fire count has improved when handling only wildfires. All fire
Development of Fire Emissions Inventory Using Satellite Data
223
cases include the prescribed (managed) and agricultural burns for which MODIS does not detect as well. The emission rate (in grams per second) can then be estimated using the following equation: Emission = Emission factor × Fire Area × Fuel loading
(3)
Emissions are calculated in grams per day while using emission factor is grams pollutant emitted per kilogram (kg) of material burned, fire size burned expressed as km2 per day, and fuel-loading expressed in the units of kilograms per square kilometers. By applying these techniques, an emission inventory of fire emissions can be developed with better spatial and temporal resolution and with better estimates of emissions than has ever been compiled before. Using A(i,t) as calculated based on Eq. (2) for a large fire case (Biscuit fire, July 14–31, 2002), we are able to compute the fine mass (PM2.5) emission rates using Eq. (3). PM2.5 emissions from the Biscuit fire in tons per day (the right panel in Figure 4) is shown along with the MODIS image acquired on the same day. Cumulative area for each cluster (C1 and C2 respectively) are used for PM2.5 emission rates. Note that the PM2.5 emissions from cluster C2 is higher than cluster C1. This is because the integrated area burned per day for the cluster C2 is higher than the cluster C1.
3. Summary We have attempted to explore the relationship between the fire areas burned (including wild fires, prescribed fires, managed fires, and agricultural burns) with the MODIS fire count product and the 2002 NEI. From this relationship, we can use the MODIS fire count product to improve the NEI reporting for other years. We initially processed the MODIS data so that the data are clustered on a per fire basis. For this procedure, we identified a reasonably robust technique after determining that we have high confidence in the MODIS rapid response product. We illustrated that MODIS fire detect information can improve spatial and temporal allocation of emissions from large fires with a high degree of confidence. The NEI is suspected to have some inconsistencies especially for small fires. Nevertheless, the relationship can be applied to provide a temporal and spatial estimate of emissions from all fires, including small fires. Thus, we have developed a technique to improve the characterization of fires in the emission inventory which will enable improvements in air quality modeling for fine particles, regional haze, ozone, air toxics, and other applications. Acknowledgments The authors acknowledge Ms. Minnie Wong and Ms. Diane Davies of the MODIS Rapid Response team at the University of Maryland, providing the fire product. The research presented here was performed under the Memorandum of Understanding between the U.S. Environmental Protection Agency and U.S. Department of Commerce’s National Oceanic and Atmospheric Administration (NOAA) under Interagency Agreement Number DW13921548.
224
B.A. Roy et al.
This work constitutes a contribution to the NOAA Air Quality Program. Although it has been reviewed by EPA and NOAA and approved for publication, it does not necessarily reflect their policies or views.
References Cahoon DR, Jr, Stocks BJ, Alexander ME, Baum BA, Goldammer JG (2000) Wildland fire detection from space: theory and application, In: Biomass Burning and Its Inter-relationship with the Climate System, Advances in Global Change Research Series, M Beniston (senior ed.), Kluwer, Dordretch/Boston, MA, pp. 151–169. Flannigan MD, Vonder Haar TH (1986) Forest fire monitoring using NOAA satellite AVHRR, Can. J. Forest Res., 16, 975–982. Giglio L, Descloitres J, et al. (2003) An enhanced contextual fire detection algorithm for MODIS, Remote. Sens. Environ., 87 (2–3), 273–282. Justice CO, Giglio L, Korontzi S, Owens J, Morissette JT, Roy D, Descloitres J, Alleaume S, Petitcolin F, Kaufman Y (2002) The MODIS fire products, Remote Sens. Environ., 83, 244–262. Kaufman YJ et al. (1990a) Remote sensing of biomass burning in the tropics, In: Fire in the Tropical Biota: Ecosystem Processes and Global challenges, JG Goldammer (ed.), Springer-Verlag, Berlin, pp. 371–399. Kaufman YJ et al. (1990b) Remote sensing of biomass burning in the tropics, J. Geophy. Res., 95(D), 9927–9939. Pace TG (2007) Wildland fire National Emissions Inventory – Past, Present and Future, 16th Annual International Emissions Inventory Conference – “Emission Inventories: Integration, Analysis, Communication”, Raleigh, NC, May 14–17, 2007. Pace TG, Pouliot GA (2007) EPA’s perspective on wildland fire emission inventories-past, present and future, Air and Waste Management Association Annual Conference and Exhibition, Pittsburg, PA, June 2007. Pouliot G, Pierce T, Benjey W, O’Neill SM, Ferguson SA (2005) Wildfire emission modeling: integrating BlueSky and SMOKE, 14th International Emission Inventory Conference, Las Vegas, NV, April 11–14. Robinson JM (1991) Fire from space: Global fire evaluation using infrared remote sensing, Int. J. Remote Sens., 12(1), 3–24. Roy Biswadev GA, Pouliot A, Gilliland T, Pierce S, Howard PV, Bhave, Benjey W (2007) Refining fire emissions for air quality modeling with remotely sensed fire counts: A wildfire case study, Atmos. Environ., 41, 655–665. Soja A, Al-Saadi J, Pierce B, Kittaka C, Szykman J, Giglio L, Randall D, Raffuse S, Roy B, Williams DJ, Pace T, Kordzi J, Pierce TE, Moore T (2007) A methodology for estimating area burned using satellite-based data in near-realtime in Oregon and Arizona, 16th Annual International Emissions Inventory Conference – “Emission inventories: Integration, Analysis, Communication”, Raleigh, NC, May 14–17, 2007.
Development of Fire Emissions Inventory Using Satellite Data
225
Soja AJ, Sukhinin AI, Cahoon DR Jr, Shugart HH, Stackhouse PW Jr (2004) AVHRR-derived fire frequency, distribution and area burned in Siberia, Int. J. Remote Sens., 25(10), 22. US Environmental Protection Agency (US EPA) (2007) Technology Transfer Network, Clearinghouse for Inventories & Emissions Factors, Available online at http://www.epa.gov/ttn/chief/net/index.html
Discussion A. Venkatram: J. Mobley:
Could you comment on the uncertainty in the emission factor for PM2.5 from fires? Additional work on emission factors for PM2.5 and other species is needed. Specifically, the PM2.5 emission factors for most source categories, including fires, need improvement. Thus, I would classify the uncertainty as relatively high. Nevertheless, I am gratified that we now have the capability to estimate emissions from fires on a near real time basis and can utilize this information in emissions as well as air quality modeling activities. With application of this data, we can determine the importance and priority for better emission estimates which can help justify the resources for improving emission factors for fires.
2.11 Forest Fires Impact on Air Quality over Portugal A.I. Miranda, A. Monteiro, V. Martins, A. Carvalho, M. Schaap, P. Builtjes and C. Borrego
Abstract The main purpose of this work is to estimate the air pollution effects of 2003 forest fires through the application of two air quality modelling systems (CHIMERE and LOTOS-EUROS) over Portugal and its intercomparison. Forest fire emissions were estimated based on specific southern European emissions factors, on type of vegetation and area burned, and incorporated in the emission input data of both modelling systems. Results showed a significant performance improvement when forest fires are taken into account. PM10 and O3 values can reach differences in the order of 30%, showing the importance and the influence of this type of emissions from local to regional air quality. The different results of the two models may give an indication of the uncertainty associated by using different models to investigate the impact of forest fires. Historical datasets of area burned, number of fires and air quality data were evaluated from 1995 to 2005 aiming to investigate a potential relationship between forest fire activity and air pollutants concentrations. The obtained results point to statistically significant correlations between fire activity in Portugal and PM10 and O3 levels in the atmosphere. Keywords Air quality modelling, forest fire emissions, model performance 1. Introduction Smoke has to be considered one of the several disturbing effects of forest fires. Its impacts on air quality and human health can be significant, because large amounts of pollutants, like particulate matter (PM), carbon monoxide (CO), volatile organic compounds (VOC) and nitrogen oxides (NOx), are emitted to the atomsphere (Reinhardt et al., 2001). The effects of these emissions are felt at different levels: from the contribution to global climate change to the occurrence of local atomspheric pollution episodes (Miranda et al., 1994; Miranda, 2004; Simmonds et al., 2005). Currently, there is a growing awareness that smoke from wildland fires can expose individuals and populations to hazardous air pollutants. Forest fire emissions depend on multiple and interdependent factors like forest fuels characteristics, burning efficiency, burning phase, fire type, meteorology and geographical location. Fuel type and load are some of the most important factors C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
190
Forest Fires Impact on Air Quality over Portugal
191
affecting fire emissions. Burning efficiency is also a significant fire emissions parameter, which is usually defined as the ratio of carbon released as carbon dioxide (CO2) to total carbon present in the fuel. In laboratorial and field experiments, the burning efficiency can be expressed as the fraction burned related to the total biomass available. Variations in fuel characteristics and consumption may contribute to 30% uncertainties in estimates of wildfires emissions (Peterson, 1987). In order to understand and to simulate forest fire effects on air quality, several questions should be analysed and integrated: fire progression, fire emissions, atomspheric flow, smoke dispersion and chemical reactions. Most research on this subject is performed in the United States of America, Canada and Australia and there are several numerical tools in development, some of them already available, aiming to estimate the dispersion of smoke from forest fires (e.g. Wang et al., 2006; Hodzic et al., 2007). The main purpose of this paper is to evaluate the effects of forest fires emissions on the air quality applying two air quality modelling systems to the 2003 Portuguese forest fires: MM5-CHIMERE and LOTOS-EUROS. As a 1st approach a statistical analysis was performed to investigate a potential relationship between forest fire activity and air pollutants concentrations in the atmosphere.
2. Statistical Analysis The statistical analyses was based on the concentration values of PM10 and ozone (O3) measured at the Portuguese air quality monitoring network between 1995 and 2005, and on the area burned and the number of fires, for the same period. This analysis was focused in three different periods: annual, June to September (JJAS) and August. The daily area burned and the number of fires were correlated with the daily maximum O3 concentration and the daily averaged PM10, registered in each air quality station, by district. Only background stations with the required acquisition efficiency were considered in the analysis. The Spearman correlation coefficients between the pollutants concentrations and the area burned and the number of fires were estimated. All results are statistically significant at a 0.05 significance level. The relationship between O3 daily maximum concentrations and fire activity presents higher correlation coefficients with the number of fires comparatively to the area burned. Porto district presents the highest correlations with the area burned and the number of fires reaching 0.70 and 0.72, respectively. Concerning PM10 daily average, the best correlations are obtained for the number of fires and for August. The northern coastal districts present the highest correlation coefficients between the PM10 daily average and the number of fires, reaching 0.88 for Porto. For the period under analysis (1995–2005) these districts registered the highest number of forest fire occurrences accounting for 43%, of the total number. The relationship between PM10 daily average and area burned is not so
A.I. Miranda et al.
192
high. The best correlations were obtained for August and the maximum value was also attained in Porto (0.85).
3. Case Study In 2003 Portugal faced the worst fire season ever recorded. There were 4,645 fires burning 8.6% of the Portuguese forest area (EC, 2004). The highest values of area burned were registered in the central eastern districts with a total of 126,589 ha of forest consumed. A large number of fires occurred in August, with 86% of total area burned.
3.1. Forest fire emissions estimation Emissions from forest fires can be estimated using a simple methodology, which include emission factors, burning efficiency, fuel loads and area burned. Generically, emissions can be estimated by: Ei = EFi × ȕ × B × A where, Ei – emission of compound i (g); EFi – compound i emission factor (g.kg-1); ȕ – global burning efficiency; B – fuel load (kg.m-2); A – area burned (m2). Specific values for Portugal were selected based on data from the National Forest Inventory, on the characteristics of the consumed forest type and shrubs, and fire data like ignition point and time and area burned. Emission factors for CO2, CO, methane (CH4), non-methane hydrocarbons (NMHC), PM2.5, PM10, and NOx come from a bibliographic review that allowed to select the most suitable for south European ecosystems, namely for the Portuguese land use types (Miranda, 2004). Table 1 summarises the chosen parameters. Table 1 Fuel load, combustion efficiency and emission factors for Portuguese conditions. Fuel Shrubs Resinous Deciduous Eucalyptus
Fuel load (kg m-2)
Combustion efficiency
1.00 8.60 1.75 3.90
0.80 0.25
Emission factor (g kg-1) CO2 1 477 1 627 1 393 1 414
CO 82 75 128 117
CH4 4 6 6 6
NMHC PM2.5 9 9 5 10 6 11 7 11
PM10 10 10 13 13
NOx 7 4 3 4
The described approach allowed developing an algorithm to calculate forest fire emissions. They were estimated for the year 2003, for each large fire, and compared to anthropogenic emissions (Table 2) aiming to understand their potential contribution to the total values.
Forest Fires Impact on Air Quality over Portugal
193
Table 2 Comparison between anthropogenic and forest fires emissions in Portugal, for 2003. Source Forest fires Transports Industry and Services Forest fires/total emissions (%)
Estimated emissions 2003 (t) CH4 NMHC PM2.5 25 773 31 616 25 773 35 660 62 847 9 849
CO2 22 167 772 19 472 820
CO 456 858 315 265
30 919 120
357 701
2 760
120 887
30.6
40.4
40.2
14.7
PM10 53 440 9 877
NOx 21 194 130 109
80 372
106 365
140 371
22.2
31.5
7.2
Forest fires can represent a significant percentage of the total emissions, reaching 40% for CO and CH4 and 30% for CO2 and PM10.
3.2. Modelling application The air quality modelling applications were performed using two different air quality modelling systems. One is composed by the chemistry-transport model CHIMERE, forced by the MM5 meteorological fields (Monteiro et al., 2007) and the other one is the LOTOS-EUROS system (Schaap et al., 2008). Both are operational 3D chemistry transport models aimed to simulate air pollution in the lower troposphere. They were applied first at a continental scale (with 50 × 50 km2 resolution – CHIMERE, and with 35 × 25 km2 – LOTOS-EUROS) and then to mainland Portugal domain, using the same physics and a one-way nesting technique, with 10 × 10 km2 horizontal resolution (see Figure 1).
Fig. 1 European (a) and Portuguese (b) domains used by the air quality modelling systems
CHIMERE was specifically developed for simulating gas-phase chemistry, aerosol formation, transport and deposition at European and urban scales. The model simulates the concentration of 44 gaseous species and 6 aerosol chemical compounds. The gas-phase chemistry scheme, derived from the original complete mechanism MELCHIOR, has been extended to include sulfur aqueous chemistry, secondary organic chemistry and heterogeneous chemistry of HONO and nitrate. The aerosol model accounts for both inorganic and organic species, of primary or secondary origin, as SOA. The vertical domain in the CHIMERE model was divided into eight layers with an extension of 3,000 m. It was used the most updated annual emission inventory (2003 year) developed by the Portuguese Agency for the Environment. Simpson et al. (1999) methodology was adopted to calculate
194
A.I. Miranda et al.
biogenic emissions with the CHIMERE model. Time disaggregation was obtained by application of monthly, weekly and hourly profiles from the University of Stuttgart. For the European simulation emission data from the inventory of EMEP Program (Co-operative Programme for Monitoring and Evaluation of the Longrange Transmission of Air Pollutants in Europe) for 2001, were used. The LOTOS-EUROS model (Schaap et al., 2008) includes the O3 chemistry using a modified CBM4-mechanism. The model incorporates primary (combustion) particles (elemental carbon (EC), organic carbon (OC)), sea salt and secondary inorganic aerosols (SIA: SO4, NO3 and NH4). Crustal matter (CM) and SOA are not incorporated. In the vertical the model has four layers up to 3,500 m following the dynamic mixing layer approach. The option to zoom in a factor of 2 over Portugal using a one-way nest with boundary conditions obtained from simulation over the full domain was used. Anthropogenic emissions were obtained from a Europe wide emission inventory made at TNO, with total emissions identical for EMEP. Time patterns and biogenic emissions were treated as in CHIMERE. This and previous studies show that the model underestimates the PM10 concentrations significantly, mainly due to the non-modelled fractions and large uncertainties in the modelling of the carbonaceous components. For a detailed discussion on the model performance for PM, O3, and its components we refer to Schaap et al. (2004, 2008). In this study we have coupled the fire emission data to both models through the previously developed algorithm. Simulations were performed for August 2003, regarding gaseous and particulate pollutants. A baseline simulation (BS) was performed, including “conventional” emissions, and a forest fire simulation (FS), which also considered emissions from forest fires larger than 100 ha. These simulations were the first applications of LOTOS-EUROS model over Portugal. Hence, forest fire emissions values based on the methodology previously described, were added to the anthropogenic and biogenic grid emissions, according to the fire location and assuming a uniform fire spread and a constant injection altitude above 50 m in the MM5-CHIMERE system and in the dynamic mixing layer in the LOTOS-EUROS model.
4. Results and Discussion Modelling results were compared to background monitoring data from the national air quality network. Some statistical parameters were used to evaluate the simulations results: root mean square error (RMSE), systematic error (BIAS) and correlation coefficient (r). Tables 3 and 4 present the statistical analysis for both simulations (with and without forest fire emissions), and for both modelling systems, for PM10 and O3, respectively. Results are analysed considering the averages for each district. Both models performance increase substantially when forest fire emissions are included. In average RMSE decreases 10% for CHIMERE and 9% for LOTOSEUROS. CHIMERE and LOTOS-EUROS present a tendency to underestimate the PM10 values for both simulations. Considering fire emissions the correlation
Forest Fires Impact on Air Quality over Portugal
195
coefficients increase from 0.72 to 0.78 and from 0.42 to 0.63 for CHIMERE and LOTOS-EUROS, respectively. Table 3 Statistical analysis of models performance for BS and FS, . for PM 10 District
CHIMERE -3
RMSE (µg m ) BS
LOTOS-EUROS -3
BIAS (µg m )
FS
BS
-3
RMSE (µg m )
r
FS
BS
FS
BS
FS
BIAS (µg m-3) BS
FS
r BS
FS
Aveiro
35.89
33.66
28.18
26.70
0.67
0.73
16.43
15.23
41.28
34.57
0.45
0.47
Coimbra
52.14
48.74
43.01
39.46
0.75
0.77
18.72
15.77
53.11
37.75
0.48
0.73
Leiria
34.21
32.03
14.90
14.02
0.72
0.78
16.95
14.48
41.17
29.30
0.48
0.67
Lisboa
17.36
15.82
5.62
5.15
0.77
0.82
13.27
12.74
24.94
19.99
0.39
0.65
Porto
43.84
34.27
–11.28
–6.38
0.73
0.75
17.01
16.37
43.85
40.74
0.51
0.67
Santarém 41.27
40.00
28.62
26.66
0.74
0.87
17.28
16.64
24.56
23.06
0.27
0.51
Setúbal Average
22.74 35.35
21.01 32.22
–3.19 15.12
–2.39 14.75
0.67 0.72
0.71 0.78
14.87 16.36
13.89 15.02
33.78 37.53
27.07 30.35
0.37 0.42
0.68 0.63
RMSE =
ࢴ(O
N
i
BIAS
- Mi )2
i =1
1 N ¦ ( Oi M i ) N i
N is the number of samples, Oi are observations and Mi are model predictions
Table 4 Statistical analysis of models performance for BS and FS, for O 3 District
CHIMERE RMSE (µg m-3)
Aveiro
BS
FS
42.29
40.50
LOTOS-EUROS
BIAS (µg m-3)
RMSE (µg m-3)
r
BS
FS
BS
FS
BS
FS
–25.90
–23.27
0.75
0.75
30.56
BIAS (µg m-3)
r
BS
FS
BS
FS
30.85
–15.24
–20.21
0.71
0.72 0.53
Cast Branco
38.50
32.90
9.48
5.14
0.72
0.79
31.50
30.85
10.00
5.94
0.49
Coimbra
39.83
23.23
25.38
14.14
0.79
0.92
29.49
28.64
–9.77
–14.48
0.72
0.76
Lisboa
36.26
31.01
23.27
9.65
0.72
0.77
31.34
31.29
7.55
4.96
0.51
0.54
Porto
29.70
30.19
–9.64
–10.86
0.58
0.67
28.44
29.00
–13.94
–16.48
0.67
0.66
Santarém Setúbal Average
68.29 32.29 41.02
55.39 25.61 34.12
56.59 –8.77 10.06
42.90 –4.45 4.75
0.79 0.69 0.72
0.86 0.75 0.79
36.64 29.25 31.03
37.16 29.28 31.01
23.14 –2.60 –0.12
21.67 –5.52 –3.45
0.46 0.60 0.59
0.43 0.62 0.61
In what concerns O3, both models performance also increase when forest fire emissions are included. In average RMSE values decrease 20% with CHIMERE, but the improvement with LOTOS-EUROS is small. Estimated biases do not present a clear trend, spatially and among both modelling systems results. In average there is a trend to underestimate O3 with CHIMERE, and overestimate it with LOTOS-EUROS. LOTOS-EUROS presents lesser RMSE and correlation coefficients. Nevertheless, regarding O3, there is a substantial number of districts where the impact of forest fires is overestimated, and the inclusion of their emissions did not improve model performance. This could be explained by the presence of high concentration of aerosols, coming from forest fires, that significantly alter atomspheric radiative properties and decrease photolysis rates (Hodzic et al., 2007), which was not taken into account in the simulations. According to the Framework Directive and its Daughter Directives, the uncertainty for modelling is defined as the maximum deviation of the measured and
A.I. Miranda et al.
196
calculated concentration levels, over the period for calculating the appropriate threshold, without taking into account the timing of the events. The quality objectives defined for PM10 (annual mean) and O3 (hourly mean) are 50%. Instead of this maximum deviation proposed by the Directives we decided to evaluate uncertainty based on the RMSE. This quality indicator improves considerably when forest fires are considered, reaching differences in the order of 21% (CHIMERE) and 16% (LOTOS-EUROS) for PM10 and 42% (CHIMERE) and 3% (LOTOSEUROS) for O3 depending on the district. In average, uncertainty based on RMSE is always under 50% for both modelling systems results and both pollutants. In order to contribute to the analysis of the spatial impact of forest fire emissions on the air quality Figure 2 shows the spatial differences between both simulations (FS-BS), obtained for both modelling systems, for one of the most critical days (August 3) concerning daily values for PM10. Both modelling system indicate a severe degradation of PM10 and O3 levels due to forest fires, which can achieve PM10 daily averages higher than 200 µg m-3. For this specific day, the impact of forest fires is higher at the central inland part of Portugal, where monitoring air quality stations are not available. CHIMERE results
LOTOS-EUROS results
para o dia 3 Agosto 2003
diff PM10 (µg.m-3)
diff PM10 -3 (µg.m )
350 325 300 275 250 225 200 175 150 125 100 75 50 25
Fig. 2 Spatial differences (µg m-3) between simulation with (FS) and without (BS) forest fire emissions, for PM10 daily averages on August 3, 2003
5. Conclusions This work investigated the relationship between forest fire activity and air quality in Portugal. The 1995–2005 period was analysed, at district level, and significant correlation coefficients were obtained. The O3 daily maximum concentrations are highly correlated to the area burned and the number of fires, reaching 0.70 and 0.72, respectively. PM10 daily average also presents significant correlation coefficients especially in August. The inclusion of forest fire emissions in both modelling systems substantially improved their performance, with better statistical quality indicators. Both models presented the same kind of behaviour concerning PM10, but for O3 the pattern is slightly different. LOTOS-EUROS O3 results are not so sensitive to the inclusion of forest fire emissions. Also, this modelling system have smaller values of RMSE (less uncertainty), but lesser correlation coefficients, when compared to CHIMERE results.
Forest Fires Impact on Air Quality over Portugal
197
Future work should concentrate in the comparative spatial analysis of results, which is very important because the presented comparative evaluation is based on the available monitoring stations that are located preferentially along the coast, faraway from the larger fire occurrences. Moreover, the analysis of joint results from both models could to reduce the uncertainty of simulations. Acknowledgments The authors thank the Portuguese Foundation for Science and Technology for the Ph.D. grants of A. Monteiro (SFRH/BD/10922/2002) and A. Carvalho (SFRH/BD/10882/2002) and for the Project INTERFACE (POCI/ AMB/60660/2004). ACCENT Network of Excellence (GOCE/CT/2004/ 505337) is also acknowledged. SAS Portugal is also acknowledged for the free software availability.
References EC – European Commission, Schmuck G, San-Miguel-Ayanz J, Barbosa Camia, A, Kucera J, P Libertà G (2004) Forest Fires in Europe – 2003 fire campaign. Official Publication of the European Communities, SPI.04.142.EN. Hodzic A, Madronich S, Bohn B, Massie S, Menut L, Wiedinmyer C (2007) Wildfire particulate matter in Europe during summer 2003: meso-scale modeling of smoke emissions, transport and radiative effects. Atmos. Chem. Phys. Discuss., 7, 4705–4760. Miranda AI (2004) An integrated numerical system to estimate air quality effects of forest fires. Int. J. Wildland Fire 13, 217–226. Miranda AI, Coutinho M, Borrego C (1994) Forest fires emissions in Portugal: a contribution to global warming? Environ. Pollut. 83, 121–123. Monteiro A, Miranda AI, Borrego C, Vautard, Ferreira J, Perez AT (2007) Longterm assessment of particulate matter using CHIMERE model. Atmos. Environ., doi:10.1016/j.atmosenv.2007.06.008. Peterson J (1987) Analysis and reduction of the errors of predicting prescribed burn emissions. Ph.D. thesis, University of Washington, Seattle. Reinhardt E, Ottmar R, Castilla C (2001) Smoke Impacts from Agricultural Burning in a Rural Brazilian Town. J. Air Waste Manage. Assoc. 51, 443–450. Schaap M, Timmermans RMA, Sauter FJ, Roemer M, Velders GJM, Boersen GAC., Beck JP, Builtjes PJH (2008) The LOTOS-EUROS model: description, validation and latest developments. Int. J. Environ. Pollut., Vol. 32, No. 2, pp. 270–290. Schaap M, van Loon M, ten Brink HM, Dentener, FD, Builtjes PJH (2004) Secondary inorganic aerosol simulations for Europe with special attention to nitrate, Atmos. Phys. Chem., 4, 857–874. Simmonds PG, Manning AJ, Derwent RG, Ciais P, Ramonet M, Kazan V, Ryall D (2005) A burning question. Can recent growth rate anomalies in the greenhouse gases be attributed to large-scale biomass burning events? Atmos. Environ. 39, 2513–2517.
198
A.I. Miranda et al.
Simpson D, Winiwarter W, Börjesson G, Cinderby S, Ferreiro A, Guenther A, Hewitt CN, Janson R, et al. (1999) Inventorying emissions from nature in Europe. J. Geophys. Res. 104, 8113–8152. Wang J, Christopher SA, Nair US, Reid JS, Prins EM, Szykman J, Hand JL (2006) Mesoscale modeling of Central American smoke transport to the United States: 1. “Topdown” assessment of emission strength and diurnal variation impacts, J. Geophys. Res., 111, D05S17, doi:10.1029/2005JD006416.
Discussion P. Kishcha: Could it be that the synoptic conditions for forest fires in Portugal are dominated by hot air from the Western Sahara? If it is the case, PM10 concentrations can be affected by desert dust rather than by black carbon from forest fires? A. Carvalho: It is not wise to answer this question without a deep study on the dust events and its identification on the aerosol measurements over Portugal. Typically, large fire events over Portugal are related to an easterly flow at surface coming from the Iberian Peninsula centre. We can also state that it is not probable that the high correlation between number of fires and PM10 could be due to an artefact of simultaneity dust events and fire events, because dust events spread throughout the season, and fire events do not. S. Napelenok: Are there many different species that are burnt in Portugal forest fires, and if so, did you account for differences in emission between species? A. Carvalho: In 2003 (year for which this work is based on) 66% of the burnt area was in forest stands and the shrubland accounted for 34%. The main affected forest species were the Maritime Pine (Pinus pinaster) and the Eucalypt (Eucalyptus globulus) but minority species (deciduous, others resinous) were also considered. The forest fire emissions estimation was based on variables that take into account the percentage of the forested burnt area and the typical characteristics of the Mediterranean vegetation like fuel load, burning efficiency and emission factor.
2.8 High Resolution Nested Runs of the AURAMS Model with Comparisons to PrAIRie2005 Field Study Data Paul A. Makar, Craig Stroud, Brian Wiens, SunHee Cho, Junhua Zhang, Morad Sassi, John Liggio, Michael Moran, Wanmin Gong, Sunling Gong, Shao-Meng Li, Jeff Brook, Kevin Strawbridge, Kurt Anlauf, Chris Mihele and Desiree Toom-Sauntry
Abstract The PrAIRie2005 campaign took place in the summer of 2005 in the city of Edmonton, Alberta. The measurement campaign was designed and led by air-quality modellers with the scientific objective of determining the extent to which air pollution events in the city are the result of locally emissions versus long-range transport. A nested version of the AURAMS model was constructed for postcampaign simulations and evaluation against the measurement data. The nested model runs at different resolutions, the highest of which is a 3 km horizontal resolution centered on the urban area. The high resolution model results show good agreement with observations, with side-by-side sampling through the real and model atmosphere. The comparison shows the same features for particle composition (Aerodyne AMS measurements compared to speciated PM1), for airborne gases (continuous NO, NO2, O3) and ground-based measurements of gases, particle composition and particle layering structure. The simulations reveal that air-quality in the Edmonton area is complex, largely due to multiple local sources, with occasional long-range-transport events Keywords Air-pollution, high-resolution, nesting, field study
1. Introduction The city of Edmonton in the Canadian Province of Alberta (population ~1 million) is of interest from an air-quality perspective due to factors relating to its geographical location and the makeup of its emissions. Edmonton is located approximately 300 km east of the Rocky Mountains (downwind, in terms of the prevailing synoptic meteorology). The proximity to the mountains means that the city may be subjected to topography – induced meteorological effects, such as “Chinook” winds (synoptic scale subsidence events downwind of the mountain range, sometimes lasting several days), and anabatic and katabatic winds (diurnally varying up and
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
163
P.A. Makar et al.
164
downslope winds). The meteorology may have a complex effect on the local airquality, as will be shown below. Edmonton also has high emissions for a city of its population, due to the proximity of natural resource extraction and processing in the region. The greater Edmonton area has the second highest number of petrochemical processing facilities in a municipal area in North America (second to Houston, Texas). Approximately 8,000 point sources (stacks from industrial facilities) may be found within 150 km of the city. Sixty kilometers to the west of the city, near-surface seams of sub-bituminous coal have led to the construction of coal-fired power-plants, one of which is the largest in Canada. A major petrochemical extraction facility, Canada’s Oil Sands, is located several hundred kilometres to the north-north-east. This combination of factors led to the PrAIRie2005 study, a small scoping study designed to examine the following hypothesis: Air-pollution events in Edmonton are the result of locally emitted pollutants (Table 1). A secondary hypothesis to the primary hypothesis is that air-pollution events in Edmonton are associated with meteorological conditions that allow local pollutants to have a long residence time. In order to examine these hypotheses, Environment Canada’s A Unified Regional Air-quality Modelling System (AURAMS) was used in 48-hour forecast mode to direct several ground-based mobile laboratories (CRUISER, MAML) and plan aircraft flight paths, during the period August 17–September 7, 2005. Table 1 Platforms and measurements during PrAIRie2005. Platform
Species measured
CRUISER – ground-based mobile chemical speciation laboratory.
Continuous CO, NO, NOy/NOx, SO2, O3, VOCs (PTRMS), black carbon, PM mass, HNO3, ultrafine particle counts, 30–1,000 nm particle size and composition (AMS)
RASCAL – ground-based scanning LIDAR
LIDAR: backscattered light from airborne particles; distribution/location of particle layers, dual wavelength (1,064/532 nm), 3 m resolution along beam path
MAML – ground-based chemical speciation laboratory (Alberta Environment)
Continuous O3, CO, NOx, NH3, SO2, H2S, THC,
Cessna 207 – Instrumented aircraft
NO, NOy, O3, particle size and composition (AMS), particle counter FSSP, filters, meteorology
Cessna 188 – Instrumented aircraft
Meteorology, particle number and size (FSSP, PCASP), O3, VOC canisters
CAMML – Mobile monitoring station
PM (three instruments), NO NOy, O3, H2S
Edmonton East – monitoring station enhanced with PILS
Particle-in-liquid-sampling (PILS) ; particle ion composition
PAHs, TSP, PM10, PM2.5
2. AURAMS Simulations of PrAIRie2005 Modelling efforts subsequent to the measurement campaign have focused on the use of AURAMS to explain some of the main features of air-quality events during the study period. Two aspects of that work will be described here: (1) the use of the
High Resolution Nested Runs of the AURAMS Model with Comparisons to PrAIRie2005
165
model in regional mode to explain the likely nature of a particulate matter event taking place between the 25th and 27th of August, and (2) the effect of highresolution nesting on simulation accuracy, relative to observations. AURAMS is a regional air-quality modelling system that has been described in previous ITM/NATO conferences (Bouchet et al., 2004, 2007; Gong et al., 2004, 2007; Makar et al., 2004, 2007). Post-campaign AURAMS simulations were carried out at two different resolutions. AURAMS is an off-line model, being driven by a regional weather forecast model simulation (GEM; Cote et al., 1998), with the additional recentparameterization for anthropogenic heat islands (Makar et al., 2006). For the simulations shown here, GEM 3.2.0 was used in a regional configuration (global domain with 15 km resolution in a “core” centered over North America), which in turn was used to drive a higher resolution GEM run at 2.5 km resolution. The 15 km resolution GEM simulation was used to drive a 21 km AURAMS simulation covering north-western North America, Both this low resolution AURAMS simulation and the high resolution GEM simulation were used to drive a 3 km resolution AURAMS simulation.
2.1. Analysis of PM2.5 events: regional versus local sources During the period August 26–29, 2005, Alberta Environment monitoring sites around Edmonton recorded an air pollution event in which the PM2.5 concentration twice exceeded 30 Pg/m3. The first event occurred between 8 pm August 26 and 2 pm August 27, and the second event occurred between midnight and 7 am on August 29. By evening on August 29, a frontal system with rain showers passed through the area, with a subsequent drop in particle concentrations. Figure 1a and b show the monitoring site measurements of total PM2.5, and measurements of PM2.5 sulphate and of SO2 at the CRUISER mobile laboratory, located 80 km downwind of Edmonton. PM2.5 Measurements at CRUISER PM2.5 Measurements at Alberta Environment's Edmonton Sites
Concentration (ug/m3)
Concentration (ug/m3)
Edmonton Central Edmonton East Edmonton NW
40 30 20
5
40 4 30
3
20
2
10
10
1
0
0 8/24/05 0:00
6
8/25/05 0:00
8/26/05 0:00
8/27/05 0:00
Time (Local)
8/28/05 0:00
8/29/05 0:00
8/30/05 0:00
AMS SO4 (ug/m3), SO2 (ppbv)
50 50
CRUISER PM2.5 CRUISER AMS SO4 (ug/m3) CRUISER SO2 (ppbv)
0
8/24/05 0:00 8/25/05 0:00 8/26/05 0:00 8/27/05 0:00 8/28/05 0:00 8/29/05 0:00 8/30/05 0:00
Time (Local)
Fig. 1 (a) Alberta environment PM2.5 monitoring, August 23–30, ȝg/m 3 ; (b) CRUISER PM2.5, AMS SO4, and SO2. Units: ȝg/m 3 except SO2 (ppbv)
P.A. Makar et al.
166
The PM2.5 measurements between August 24 and 27 suggest a large regional event, with all city monitoring stations (Figure 1a) showing similar time series, also present in the measurements 80km from the city (Figure 1b, thin line). Spikes in the latter are likely due to local particulate emissions in a campground near the CRUISER measurement site. On August 29, elevated levels of PM2.5 were observed at the Edmonton Northwest and East monitoring sites, but no matching peak was observed in Edmonton Central. Edmonton and CRUISER PM2.5 peaks are timecoincident in the first event, while the CRUISER PM2.5 observations on the 29th are time-delayed by approximately 12 hours relative to those in downtown Edmonton. Equivalent AURAMS 21km (GEM 15 km) time series during the same period are shown in Figure 2, below. AURAMS 21km PM2.5, PM2.5 SO4, and SO2 @ CRUISER
PM2.5, Edmonton, 21km Resolution
6
Concentration (ug/m3)
Concentration (ug/m3)
Edmonton Central 40 30 20 10
PM2.5 PM1.0 SO4 SO2
4
30
3
20
2
10
1
0
0 8/24/05 0:00 8/25/05 0:00 8/26/05 0:00 8/27/05 0:00 8/28/05 0:00 8/29/05 0:00 8/30/05 0:00
5
40
8/24/05 0:00
8/25/05 0:00
8/26/05 0:00
8/27/05 0:00
8/28/05 0:00
8/29/05 0:00
PM2.5 SO4 (ug/m3), SO2 (ppbv)
50
50
0
8/30/05 0:00
Local Time
Local Time
Fig. 2 (a) AURAMS 21 km PM2.5 at Edmonton Central (peak value 75ȝg/m3; (b) AURAMS 21 km PM2.5, PM1.0 SO4, SO2 at CRUISER. Compare to Figure 1
The 21 km Edmonton model values (Figure 2a) have a similar timing as (Figure 1a), with a main event roughly centred on midnight on the 27th, and a second event of lower magnitude following after midnight on the 29th. The 21 km model misses the earlier rise in PM2.5 mass on the 25th observed in the measurement record in down-town Edmonton (Figure 1a). At CRUISER, the overall PM2.5 mass is underpredicted (Figure 2b, thin, line, compare to Figure 1b). The first three peaks in PM1.0 SO4 are similar in timing to the AMS SO4 observations, but the final peak on the 29th is missed in the simulation. The SO2 concentrations are over-predicted, aside from the observed peak value at 18:00 on August 26th (compare thick solid lines, Figures 1b and 2b). In the higher resolution nested run, the model predictions for SO2 and PM1 SO4 at CRUISER are considerably improved relative to the observations, although the net PM2.5 concentration is still low (Figure 3a, b). AURAMS 3km, PM1 SO4, and SO2, and CRUISER observations AURAMS 3km PM2.5, PM1 SO4, and SO2, at CRUISER
6 6
AURAMS 3 km PM2.5 AURAMS 3 km PM1 SO4 AURAMS 3 km SO2
4 30
3
20
2
10
1
0
0
8/24/05 0:00
8/25/05 0:00
8/26/05 0:00
8/27/05 0:00
Local Time
8/28/05 0:00
8/29/05 0:00
8/30/05 0:00
PM1 SO4 (ug/m3), SO2 (ppbv)
40
5
5 PM1 SO4 (ug/m3), SO2 (ppbv)
PM2.5 Concentration (ug/m3)
50
AURAMS 3 km PM1 SO4 AURAMS 3 km SO2 Observed PM1 SO4 Observed SO2
4 3 2 1 0 8/24/05 0:00
8/25/05 0:00
8/26/05 0:00
8/27/05 0:00
8/28/05 0:00
8/29/05 0:00
8/30/05 0:00
Local Time
Fig. 3 (a) AURAMS 3 km values for PM2.5, PM1 SO4, SO2 (compare to Figures 2b, 1b); (b) comparison of observed and modelled (3 km) PM1 SO4 and SO2 concentrations
High Resolution Nested Runs of the AURAMS Model with Comparisons to PrAIRie2005
167
The 3 km run results show a similar behaviour to the observations (Figure 3b), though the timing of the two highest peaks in the measurement record are not matched in the model. Model concentration maps from both low and high resolution simulations can be used to suggest the causes for the main features in the simulation, along with reasons for the differences between modeled and measured results. Figure 4a–d shows AURAMS 21 km predicted values of PM1 SO4 at midnight local time on August 25, 27, and 29, and at 15:00 local time on August 29. These correspond to the times of the observed peak sulphate concentrations in Figure 3b.
Fig. 4 A portion of the 21 km AURAMS domain, showing PM1 SO4 at (a) August 25, 0:00 local time; (b) August 27, 0:00; (c) August 29, 0:00; (d) August 29, 15:00. Times shown on the figure are UTC
Figure 4a shows the PM1 sulphate concentrations predicted by the low resolution version of the model on August 25 at 0:00 local time. The city of Edmonton is shown as a small white circle in the south-centre of the map. Just to the south-east of Edmonton are two small lakes, which mark the position of CRUISER at Miquelon Lakes Provincial Park. The main feature shown on the map is a high sulphate zone extending from the location of Alberta’s Oil Sands (a very large petrochemical extraction site) south-south-west towards Edmonton. This suggests that at least part of the high concentrations observed on the 25th of August may have resulted from long-range transport from this site.
168
P.A. Makar et al.
Figure 4b also shows the plume from the oil sands, but the wind direction has shifted the plume away from the city. High concentrations in the Edmonton area on the 27th appear to be due to local sources, with the highest concentrations to the west of the city, coincident with the location of the coal-fired power-plants described above. This may explain the high SO2 concentrations observed at this time in the measurement record (Figure 3b). Figure 4c, 0:00 August 29th, shows a very complex flow pattern; the main source close affecting the CRUISER site is a coal-fired power-plant to the south-east of the city (Battle River power-plant), though a small high concentration area just south of and associated with the city of Edmonton is also visible. Similarly, later on same day (Figure 4d), the flow pattern is also complex, with both city and power-plant sources being possible causes of observed high sulphate and SO2. The complex nature of the local flow patterns suggests that some of the sources may be better described at high resolution (Figure 5).
Fig. 5 The 3 km AURAMS nested domain, showing PM1 SO4 at (a) August 25, 0:00 local time; (b) August 27, 0:00; (c) August 29, 0:00; (d) August 29, 15:00. Times shown on the figure are UTC
Figure 5a shows the large regional event on the 25th to the east of the city, with the plume from the more local power-plants to the east. It seems likely that the observed high particulate values on this day are related to long-range transport. CRUISER was located at the lakes just to the south-east of the city. Figure 5b shows that the high sulphate and SO2 observed at CRUISER on the 27th was likely the result of the urban plume mixing with the plume from the power-plants, the latter arching south then west from that source. At 0:00 local time on August 29th (Figure 5c), the only source of PM1 SO4 near the city of Edmonton seems to be on the east side of the city itself. This is the location of the petrochemical processing facilities, and may explain why high particle concentrations are observed in the east and north-west monitoring stations, but not in the city centre (Figure 1a). Later on
High Resolution Nested Runs of the AURAMS Model with Comparisons to PrAIRie2005
169
the 29th (Figure 5d, the wind has shifted direction to come from the north. The only plume close to the CRUISER location at this time is from the city of Edmonton itself. The rapid change in flow from Figures 5c to d highlights the complexity of the wind fields at this time. A small error in the forecast winds (e.g. for Figure 5c, a 10 degree shift to the west in the Battle River plume would cause it to impact the CRUISER site.
2.2. High-and low-resolution air-quality models: urban versus downwind Comparisons with other data can be used to show the relative accuracy of high and low resolution versions of the model at various locations. Figure 6a, below, compares measured and modeled NO at CRUISER for both high and low resolution versions of the model; the high resolution simulation gives a better match to peak NO values at this downwind location from the city. Central Edmonton NO observations, AURAMS 21km and 3km simulations.
CRUISER observations of NO versus AURAMS 21km and 3km simulations
800
4.0 3.5
Observed NO
AURAMS 3km NO 2.5 2.0 1.5 1.0 0.5
Concentration (ppbv)
Concentration (ppbv)
3.0
Observed NO
700
AURAMS 21km NO
AURAMS 21 km NO AURAMS 3km NO
600 500 400 300 200 100
0.0 8/24/05 8/25/05 8/25/05 8/26/05 8/26/05 8/27/05 8/27/05 8/28/05 8/28/05 8/29/05 8/29/05 8/30/05 8/30/05 12:00 0:00 12:00 0:00 12:00 0:00 12:00 0:00 12:00 0:00 12:00 0:00 12:00
Local Time (MDT)
0 8/24/05 12:00
8/25/05 12:00
8/26/05 12:00
8/27/05 12:00
8/28/05 12:00
8/29/05 12:00
8/30/05 12:00
Local Time (MDT)
Fig. 6 (a) Comparison of modelled and measured NO at CRUISER; (b) comparison of modelled and measured NO at Edmonton Central
The accuracy of the high resolution version is limited in the downtown urban core, however – observed and high and low resolution NO are compared for this location in Figure 6b. Concentrations are over-predicted relative to the observations for both resolutions, with the high resolution being worse than the low resolution model at this location. Despite the addition of an anthropogenic heat island effect and other boundary layer changes, the extent of turbulence in the urban core is being underestimated in the current modelling setup. These parameterizations are actively being investigated.
3. Remarks: Conclusions and Ongoing Investigations The factors contributing to the observed particulate matter peaks in the Edmonton area during August 25–29 were complex. The model simulations suggest that the start of the period was affected by long-range transport, possibly from the Oil Sands refining area to the North. Subsequent peaks in PM levels on the 27th and 29th of August seem to be more associated with local sources, notably the coal-
170
P.A. Makar et al.
fired power-plants to the west of the city (27th), the petrochemical facilities on the east side of the city (morning of the 29th), and both the city/petrochemical plume and possibly the Battle River power-plant plume (afternoon of the 29th). The impact of model horizontal resolution on simulation accuracy varies with location; high-resolution being more accurate relative to observations, except in the urban core. Improved turbulent mixing parameterizations in urban regions are required to further improve high resolution model accuracy. Several other investigations are underway using the PrAIRie2005 data set. Due to space limitations in the ITM manuscript, these will be mentioned only briefly here. One of us (Stroud) has examined the diurnal pattern of local flow during the above study period, and has found a regular cycle likely linked to anabatic/katabatic winds in the nearby Rockies. A detailed examination of the power-plants and the accuracy of reported/estimated emissions is currently underway (Cho). Comparisons of different versions of the weather forecast model, with and without heat islands and boundary layer changes, has suggested that these modifications do not have a very strong impact on simulation results, but have a significant impact on forecast boundary layer heights in comparison to LIDAR measurements (Makar, Strawbridge). Analysis of PrAIRie2005 data using Principal Component Analysis (Liggio) and Positive Matrix Factorization (Stroud) suggests that in conjunction with the high resolution modelling results, these methodologies may be used to infer local sources from airborne measurement data. The experiment was successful: combined measurements and modelling are capable of determining the extent to which local sources impact air-quality in this region.
References Bouchet VS, Ménard S, Gaudreault S, Cousineau S, Moffet R, Crevier L-P, Gong W, Makar PA, Moran MD, Pabla B (2004) “Real-time regional air quality modelling in support of the ICARTT 2004 campaign”. Proc. 27th NATO/CCMS ITM on Air Pollution Modelling and Its Application, Banff, Canada, October 25–29. [See also Air Pollution Modeling and Its Application XVII, 2007, C Borrego and A-L Norman, Eds., Springer, New York]. Cote J, Gravel S, Methot A, Patoine A, Roch M, Staniforth A (1998) Mon. Wea. Rev., 126, 1373–1395. Gong W, Bouchet VS, Makar PA, Moran MD, Gong S, Leaitch WR (2004) “Cloud processing of gases and aerosols in a regional air quality model (AURAMS): evaluation against aircraft data”. Proc. 27th NATO/CCMS ITM on Air Pollution Modelling and Its Application, Banff, Canada, October 25–29. [See also Air Pollution Modeling and Its Application XVII, 2007, C Borrego and A-L Norman, Eds., Springer, New York]. Makar PA, Bouchet VS, Gong W, Moran MD, Gong S, Dastoor AP, Hayden K, Boudries H, Brook J, Strawbridge K, Anlauf K, Li S-M (2004) “AURAMS/ Pacific2001 Measurement Intensive comparison”. Proc. 27th NATO/CCMS ITM on Air Pollution Modelling and Its Application, Banff, Canada, Oct. 25–29.
High Resolution Nested Runs of the AURAMS Model with Comparisons to PrAIRie2005
171
[See also Air Pollution Modeling and Its Application XVII, 2007, C Borrego and A-L Norman, Eds., Springer, New York]. Makar PA, Gravel S, Chirkov V, Strawbridge KB, Froude F, Arnold J, Brook J, (2006) Anthropogenic heat flux, urban properties, and regional weather, Atmos. Environ., 40, 2750–2766.
Discussion A. Venkatram: The high resolution model should yield better results than the low resolution model in principle. So if the low resolution model performs better, you might want to examine the formulation of processes in the high resolution model. W. Gong: If the relevant processes are not well represented in the model (at either resolution), high-resolution model simulations do not guarantee better performance. There is an indication that the current parameterization in the model for boundary layer mixing is inadequate particularly over urban areas. The large model grid size at low resolution defacto mixes fresh urban emissions over a larger area than high resolution, which is why the low resolution model performs “better” over urban areas. However, the downwind performance of the low resolution model is worse than the high resolution model. Both resolutions made use of the same boundary layer mixing processes, which do not adequately vertically mix pollutants over urban areas. The problem exists in both resolutions – the high resolution model lacks the compensating error of emissions being distributed over a larger area, and hence the problem becomes more noticeable at high resolution. S. Hanna: Your conclusions regarding model comparison with observations were qualitative (e.g., “looks good”). Do you have plans to do quantitative comparisons and perhaps use acceptance criteria for deciding whether the model is performing good or not? W. Gong: Yes. These comparisons are underway and will be submitted for publication in future work. The aircraft and ground-based high time resolution observations have been time averaged to the high resolution AQ model’s 2 minute time step, and a quantitative modelmeasurement statistical comparison has been made for all flights, using standard measures of model error. By comparing individual flights, we have been able to contrast “good performance” days to “poor performance” days, and link the latter to simulated turbulence conditions. For example, airborne correlation coefficients for ozone
172
P.A. Makar et al.
are as high as 0.8 on a good performance day, but anticorrelated on a poor performance day. M. Mircea: Is the 24 August 2005 the first day of the simulations or you used another spin-up period for the model? W. Gong: The air-quality model runs shown here started at August 22; a two day spin-up was employed for both high and low resolution models. The meteorology was run with consecutive 12 hour simulations starting from data-assimilated (4DVAR) initial states; six hours of spin-up followed by six hours of AQ-model input meteorology. This methodology was used to constrain the meteorology to the best possible forecast. For the high resolution air-quality runs, the domain is sufficiently small that the boundary condition crosses the model domain within 12 hours.
2.5 Long-Term Simulations of Surface Ozone in East Asia During 1980Ω2020 with CMAQ and REAS Inventory Toshimasa Ohara, Kazuyo Yamaji, Itsushi Uno, Hiroshi Tanimoto, Seiji Sugata, Tatsuya Nagashima, Jun-ichi Kurokawa, Nobuhiro Horii and Hajime Akimoto
Abstract Long-term simulations of surface ozone (O3) over East Asia during 1980–2020 were conducted using the regional scale chemical transport model (CMAQ) and the newly developed year-by-year emission inventory in Asia (REAS). The CMAQ with the REAS could reproduce the spatial and seasonal variations of the observed surface O3 concentrations in 2000 and 2001. The historical simulation from 1980 to 2003 demonstrates that an annually-averaged concentration of surface O3 over the Central East China (CEC) and Japan increases about 12 ppbv (1% year-1) and 5 ppbv (0.4% year-1) during a quarter century, respectively. This simulated trend in Japan generally agrees with the observed trend measured at monitoring stations and is correlated with the trend of Chinese NOx and NMVOC emissions. The future emissions up to 2020 were projected based on three emission scenarios (PSC, REF, and PFC). In 2020, the Chinese NOx emissions in each scenario are expected to increase by Ω1% (PSC), +40% (REF), and +128% (PFC) from 2000, respectively. The worst scenario (PFC) shows that the East Asian NOx emissions almost double between 2000 and 2020. We find that the surface O3 concentrations in East Asia will increase significantly in the near future due to projected increases in NOx emissions.
Keywords CMAQ, East Asia, long-term simulation, REAS, surface ozone 1. Introduction Recently it was reported that tropospheric ozone (O3) levels observed over Japan have been rising over the last three decades, likely as a consequence of increasing emissions of NOx from Asia (Naja and Akimoto, 2004). Asian emissions also have a potential impact on air quality over the United States, and on widespread O3 pollution in the Northern Hemisphere through intercontinental transport (Wild and Akimoto, 2001). The O3 levels will be enhanced in future, particularly over Asia where NOx emissions are estimated to increase most severely (Akimoto, 2003). In C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
136
Long-Term Simulations of Surface Ozone in East Asia during 1980–2020 with CMAQ
137
the present study we report the results of trend analysis and future projection of surface O3 over East Asia using by the CMAQ model and new developed Asian emission inventory (REAS).
2. CMAQ Simulation and REAS Emission Inventory The three-dimensional regional-scale CTM used in this study has been developed jointly by Kyushu University and National Institute for Environmental Studies (NIES) (Tanimoto et al., 2005; Uno et al., 2007) based on the Models-3 CMAQ modeling system version 4.4 released by the US EPA (Byun and Ching, 1999). The model is driven by meteorological fields calculated by RAMS, the Regional Atmospheric Modeling System version 4.3 (Pielke et al., 1992), with NCEP/NCAR 2.5q × 2.5q reanalysis data sets at six hour intervals. The spatial domain for CMAQ and RAMS (shown in Figure 1) is 6,240 × 5,440 km2 (inside domain) and 8,000 × 5,600 km2 (outside domain) on rotated polar stereographic map projection centered at 25q N, 115q E with 80 × 80 km2 grid resolutions, respectively. For vertical resolution, the CMAQ has the ı-z coordination system up to 23 km and 14 vertical layers. In this study, the SAPRC-99 scheme is applied for gas-phase chemistry, and AERO3 module is used for aerosol calculation. 90E
120E 105E
150E 135E
45N
45N
30N
30N
15N
15N
CMAQ DOMAIN 20N
20N
90E
120E
RAMS DOMAIN
Fig. 1 Model domain for the RAMS/CMAQ simulation. Small black circles indicate the locations of Japanese remote monitoring sites of EANET (Rishiri, Tappi, Sadoseki, Happo, Oki, Tsushima, Ebino, Hedo, and Ogasawara) and WDCGG (Yonaguni)
We developed a new emission inventory for Asia (Regional Emission inventory in ASia (REAS) Version 1.1) for the period 1980–2020. REAS is the first inventory to integrate historical, present, and future emissions in Asia on the basis of a consistent methodology (Ohara et al., 2007). We estimated historical emissions for 1980– 2003, and projected emissions for 2010 and 2020 of NOx, SO2, CO, NMVOC, black carbon (BC), and organic carbon (OC) from fuel combustion and industrial sources and those of NOx, N2O, NH3 and CH4 from agricultural sources. These have been developed as grid data with 0.5q × 0.5q resolution.
138
T. Ohara et al.
Future projection of Asian emissions was performed on the basis of emission scenarios and emissions for 2000. Three emission scenarios for China have been developed for the years 2010 and 2020. The socioeconomic indices, such as population, urbanization, and GDP, are almost the same under these scenarios. The first scenario was termed the Policy Failed Case scenario, or PFC. This is a pessimistic scenario with high emission rates caused by continuation of the current energy structure, increased energy consumption, and the slow deployment of new energy technologies and new emission control technologies. The second scenario was termed the Reference scenario, or REF. This was a sustainable scenario with moderate emission rates caused by the suppression of energy consumption through energy conservation, a change to clean energy, and the moderate deployment of new energy technologies and new emission control technologies. We considered that this represented our “best guess” as to what emissions in Asia will be in 2010 and 2020. The third scenario was termed the Policy Succeed Case scenario, or PSC. This was the optimistic case, with low emission rates owing to the implementation of strong energy and environmental policies and the fast deployment of new energy technologies and new emission control technologies. Each concept – PSC, REF, and PFC – resembled that of the B1 scenario, the B2 scenario, and the A2 scenario of IPCC, respectively. In this study, a twenty four-full-year simulation was conducted for 1980–2003. For this CMAQ modeling system, all emissions were obtained from 0.5q × 0.5q resolution of the REAS database. The initial fields and monthly averaged lateral boundary condition for most chemical tracers are provided from a global chemical transport model (CHASER; Sudo et al., 2002). This fixed lateral boundary condition is used for the twenty four-full-year simulation (i.e., no interannual variation of lateral conditions is assumed). Two sets of numerical experiments were conducted. Series EyyM00 simulations used the fixed meteorology for 2000 with year-by-year emissions. Series EyyMyy simulations used both year-by-year emissions and meteorology. These two experiments were set to elucidate the sensitivity to changes of meteorology and emissions.
3. Recent Trends of Emissions and Surface Ozone The reproducibility of surface O3 concentration simulated by CMAQ/REAS was confirmed by comparing with observation data in 2000 and 2001 at Japanese remote monitoring sites from the Acid Deposition Monitoring Network in East Asia (EANET) and from the WMO World Data Centre for Greenhouse Gasses (WDCGG) (see Figure 1). Our model system can catch the observed O3 concentration levels and the day-to-day variations with correlation coefficients based on daily-averaged O3 ranging between 0.61 and 0.85. More comprehensive and systematic validation has been conducted in Yamaji et al. (2006, 2008). Total emissions of NOx in Asia (Figure 2) showed a monotonic increase between 1980–2003. The emissions increased by 2.8 times between 1980 and 2003, with values of 10.7 Mt in 1980 and 29.5 Mt in 2003. In particular, Chinese NOx
Long-Term Simulations of Surface Ozone in East Asia during 1980–2020 with CMAQ
139
emissions increased dramatically by 3.8 times from 1980 to 2003 – an annuallyaveraged growth rate of 6%, with the highest growth after 2000 (by 1.3 times over only three years). Recently, these NOx trends in the period 1996–2003 over China were validated by comparison with column NO2 data from the GOME (Global Ozone Monitoring Experiment) satellite by Akimoto et al. (2006) and Uno et al. (2007). For NMVOC in Asia, the fundamental features of year-by-year variations were similar to those for NOx. Asian NMVOC emissions increased by 2.1 times between 1980 and 2003 (2.5 times in China).
Fig. 2 Temporal evolution of NOx emissions in Asia between 1980 and 2003
Figure 3 shows the annually-averaged concentrations of CMAQ simulated surface O3 in 1980 and 2000. Surface O3 concentrations increased over East Asia between 1980 and 2000. Especially, rapid O3 growth is simulated in the regions of Central East China (CEC), Korea, and Japan. In these regions, the annuallyaveraged concentrations of O3 are over 45 ppbv.
Fig. 3 Spatial distributions of annually-averaged surface O3 (ppmv) in 1980 (left) and 2000 (right)
Figure 4a shows the change of annually-averaged concentrations of surface O3 measured at air quality monitoring stations (597 stations) over Japan. Average O3 concentrations measured every five years during the 20 years of 1985–2004 are 21.1
140
T. Ohara et al.
Fig. 4 Interannual variations of annually-averaged vale of (a) observed and CMAQ simulated O3 concentrations over Japan, (b) NOx and NMVOC emissions in China and observed NOx and NMHC concentrations over Japan, and (c) CMAQ simulated O3 concentrations over Japan from CNTL (EyyMyy), EyyM00, and EyyMyy – EyyM00
ppbv, 22.7 ppbv, 24.2 ppbv, and 24.9 ppbv, increasing at an increment of about 0.25 ppbv year-1 (1% year-1). Moreover, the O3 concentrations measured at remote site in Japan also indicate an upward tendency during and after 1990. The average O3 concentrations during the ten years of 1992–2002 at Happo (mountainous area) and Ryori (rural coastal area) reveals increments of 9.0 ppbv (2% year-1) at Happo and of 7.3 ppbv (2% year-1) at Ryori. On the other hand, the surface concentrations of NOx and NMHC, the precursors of O3, have decreased over Japan as shown in Figure 4b. Figure 4a also shows the year-by-year changes of simulated surface O3 between 1980 and 2003. The simulated O3 concentration increases at the rate of about 0.22 ppbv year-1 during 1980–2003, which corresponds to the observed O3 growth. The upward tendency of surface O3 concentrations in Japan and that of Chinese emissions (see Figure 4b) are quite similar. It is indicated that the annually-averaged concentration of surface O3 in the CEC region increased by 1 ppbv with the annual increase of a million ton of the NOx emissions in China. Consequently, the average concentration in summer increased by about 8 ppbv during 1996–2003. Additionally, the O3 concentration in Japan is also enhanced by the increasing of Chinese
Long-Term Simulations of Surface Ozone in East Asia during 1980–2020 with CMAQ
141
emissions and its increasing rate is 30–50% of that in China. This means that the O3 produced by the increased emissions from the Asian Continent caused the rising of the surface O3 concentration in Japan by the transboundary air pollution.
4. Future Projection We will briefly summarize the future emissions for NOx and NMVOC in the years of 2010 and 2020. Figure 5 shows the change of Asian regional emissions from 2000 to 2010 or 2020. In the 2020REF scenario, NOx emissions in China (15.6 Tg) will increase by 40% from 2000 (11.2 Tg). Regional NOx emissions from other East Asia and Southeast Asia will increase by 25% from 2000 (4.4 Tg) to 2020 (5.5 Tg) and by 53% from 2000 (5.8 Tg) to 2020 (3.8 Tg), respectively. In the 2020PSC scenario, the NOx emissions in China have a little decrease of 1% from 2000 to 2020. In the 2020PFC scenario, NOx emissions emitted in China will increase by 128% from 2000. In the 2020REF scenario, NMVOC emissions in China (35.2 Tg) will increase rapidly by 128% from 2000 (14.7 Tg). Regional NMVOC emissions from other East Asia and Southeast Asia will increase by 70% from 2000 (3.7 Tg) to 2020 (6.3 Tg) and by 53% from 2000 (11.1 Tg) to 2020 (19.1 Tg), respectively. In the 2020PSC scenario, the NMVOC emissions emitted in China have a large increase of 97% from 2000. In the 2020PFC scenario, NMVOC emissions emitted in China will increase by 163% from 2000. The control technologies and environmental policies for anthropogenic NMVOC emissions will be behind to those for NOx emission in many Asian countries, therefore the growth of Asian NMVOC emissions is expected to be greater in either emission scenario. 50
NOx (Mt yr-1)
40
China
Japan
Other East Asia
Southeast Asia
India
Other South Asia
30 20 10
Fig. 5 Asian NOx emissions in 2000, 2010, and 2020
0
2000
PSC
REF
2010
PFC
PSC
REF
2020
PFC
Future projections of surface O3 over East Asia were conducted using the CMAQ/REAS under the meteorological conditions for the year 2000 (Yamaji et al., 2008). Figure 6 shows the spatial distributions of annually-averaged O 3 concentration changes from 2000 to 2020 under three emission scenarios, PSC, REF, and PFC. Firstly, we demonstrate the increase of O3 concentrations from 2000 to 2020 under the REF scenario. The annually-averaged O3 concentration increase is relatively small (<5 ppbv) over the northeast Asia (CEC, the Korean peninsula, and
T. Ohara et al.
142
2020PSC-2000
2020REF-2000
2020PFC-2000 45N
45N
30N
30N
15N
15N
YEAR
YEAR
YEAR 0
5 ppbv
10
Fig. 6 Spatial distribution of surface O3 increase from 2000 to 2020 under three scenarios (PSC, REF, and PFC) for annually-averaged concentrations
Japan), while a high O3 increase are projected over the southern area of the model domain (latitude lower than approximately 35° N). Especially, in the latitude belt of 20–35° N (the coast and seashore of the southeast China), the 2020REF scenario shows an increase of nearly 5 ppbv in the O3 concentration compared to the 2000 level. These increases are reflected by the increases of NOx and NMVOC emissions in the costal area in China between 20–40° N from 2000 to 2020. Another important feature of O3 increases form 2000 to 2020 is that the O3 growth as well as the latitudinal zone indicating the maximum O3 enhancement are strongly dependent on the season, due to the seasonal variations of meteorology in East Asia (Asian monsoon). It is important to compare the spatial distribution of O3 concentrations in three emission scenarios, REF, PSC, and PFC, for the year 2020. The spatial distribution of ¨PSC (=2020PSC–2000) is quite different from the others and shows a little decrease of O3 concentrations over northeast parts of China. This is affected by a decrease of NOx emissions in this area. Meanwhile, due to the high NOx emission growth in some mega-cities (Beijing, Tenjing, Shanghai, and HongKong), the increases of O3 concentrations around these mega-cities are predicted even in PSC scenario. While, the feature of spatial distribution in ¨PFC(=2020PFC–2000) is close to that in ¨REF (=2020REF–2000) although the O3 growth rates are different between these scenarios (for example, 6–12 ppbv in ¨REF and 18–24 ppbv in ¨PFC over the north China plain). We conclude that the future O3 concentrations show a high sensitivity to the change of Chinese NOx emissions under future emission scenarios. Additionally, it is projected that in the Central Japan, O3 concentration will increase by 5 ppbv from 2000 to 2020PFC.
Long-Term Simulations of Surface Ozone in East Asia during 1980–2020 with CMAQ
143
5. Summary and Conclusions Long-term simulations of surface O3 over East Asia during 1980–2020 were conducted using the chemical transport model (CMAQ) and the newly developed year-by-year emission inventory in Asia (REAS). The CMAQ with the REAS could reproduce the spatial and seasonal variations of the observed surface O3 concentrations in 2000. The historical simulation from 1980 to 2003 demonstrates that an annually-averaged concentration of surface O3 over the CEC and Japan increases about 12 ppbv (1% year-1) and 5 ppbv (0.4% year-1) during a quarter century, respectively. This simulated trend in Japan generally agrees with the observed trend measured at monitoring stations and is correlated with the trend of Chinese emissions. The future emissions up to 2020 were projected based on the emissions for 2000 and three emission scenarios. In 2020, the Chinese NOx emissions in each scenario are expected to increase by Ω1% (PSC), +40% (REF), and +128% (PFC) from 2000. The PFC scenario shows that the East Asian NOx emissions almost double between 2000 and 2020. We find that the surface O3 concentrations in East Asia will increase significantly in the near future due to increase in NOx emissions.
References Akimoto H (2003) Global air quality and pollution. Science 302:1716–1719 Akimoto H, Ohara T, Kurokawa J, Horii N (2006) Verification of energy consumption in China during 1996–2003 by satellite observation. Atmos Environ 40:7663–7667 Byun DW, Ching JKS (eds.) (1999) Science algorithms of the EPA Models-3 community multi-scale air quality (CMAQ) modeling system. NERL, Research Triangle Park, NC EPA/ 600/R-99/030 Naja M, Akimoto H (2004) Contribution of regional pollution and long-range transport to the Asia-Pacific region: analysis of long-term ozonesonde data over Japan. J Geophys Res 109: D21306, doi:10.1029/2004JD004687 Ohara T, Akimoto H, Kurokawa J, Horii N, Yamaji K, Yan X, Hayasaka T (2007) An Asian emission inventory of anthropogenic emission sources for the period 1980–2020. Atmos Chem Phys 7:4419–4444 Pielke RA, Cotton WR, Walko RL, Tremback CJ, et al. (1992) A comprehensive meteorological modeling system – RAMS. Meteor Atmos Phys 49:69–91 Sudo K, Takahashi M, Kurokawa J, Akimoto H (2002) CHASER: a global chemical model of the troposphere – 1. Model description. J Geophys Res 107 (D17):4339, doi:10.1029/2001JD001113 Tanimoto H, Sawa Y, Matsueda H, Uno I, Ohara T, Yamaji K, Kurokawa J, Yonemura S (2005) Significant latitudinal gradient in the surface ozone spring maximum over East Asia. Geophys Res Lett 32:L21805, doi:10.1029/ 2005GL023514
144
T. Ohara et al.
Uno I, He Y, Ohara T, et al. (2007) Systematic analysis of interannual and seasonal variations of model-simulated tropospheric NO2 in Asia and comparison with GOME-satellite data. Atmos Chem Phys 7:1671–1681 Wild O, Akimoto H (2001) Intercontinental transport of ozone and its precursors in a three-dimensional global CTM. J Geophys Res 106:27729–27744 Yamaji K, Ohara T, Uno I, Tanimoto H, Kurokawa J, Akimoto H (2006) Analysis of seasonal variation of ozone in the boundary layer in East Asia using the Community Multi-scale Air Quality model: what controls surface ozone level over Japan? Atmos Environ 40:1856–1868 Yamaji K, Ohara T, Uno I, Kurokawa J, Pochanart P, Akimoto H (2008) Future prediction of surface ozone over east Asia using Models-3 Community Multiscale Air Quality Modeling System and Regional Emission Inventory in Asia, J Geophys Res 113: D08306, doi:10.1029/2007JD008663
Discussion A. Venkatram: Could you comment on the inevitable uncertainties in the NOx and VOC emissions from China on your comparison between model estimated and observed trends of ozone in Japan? T. Ohara: China’s emissions are thought to be quite uncertain. Streets et al. (2003, An inventory of gaseous and primary aerosol emissions in Asia in the year 2000, Journal of Geophysical Research, 108, 8809, doi:10.1029/2002JD003093) demonstrated the uncertainty in China’s emissions was ±23% (NOx) and ±59% (NMVOC), measured as 95% confidence intervals. These uncertainties influence the modelsimulated O3, which is transported from continent to Japan. In particular, a trend of China’s emissions is an important factor in the O3 trend in Japan. Uno et al. (2007, Systematic analysis of interannual and seasonal variations of model-simulated tropospheric NO2 in Asia and comparison with GOME-satellite data, Atmospheric Chemistry and physics, 7, 1671–1681) and He et al. (2007, Variation of the increase trend of tropospheric NO2 over Central East China during the past decade, Atmospheric Environment, 41, 4865– 4876) pointed out that the NOx emission trends for China based on the REAS inventory underestimated the rapid growth of emissions based on the GOME/SCIAMACHY satellite observations. This means that an increasing rate of O3 in Japan caused by China’s emissions may be underestimated in this study. Therefore, the reduction of the uncertainties in China’s emissions and those trends is very important task in future.
2.4 Modelling the Deposition of Reduced Nitrogen at Different Scales in the United Kingdom Anthony J. Dore, Mark R. Theobald, Maciej Kryza, Massimo Vieno, Sim Y. Tang and Mark A. Sutton
Abstract A national scale atmospheric transport model (FRAME) was applied to the UK to generate maps of ammonia concentration and the wet and dry deposition of reduced nitrogen. The model was found to give satisfactory agreement with measurements from the national monitoring networks of ammonia and ammonium aerosol concentrations and wet deposition of ammonium. Scatter in the model correlation with measurements of ammonia concentration was observed due to the strong sub-model grid (5 km) variation in ammonia concentration. Application of a local dispersion model to calculate ammonia concentrations emitted from a typical poultry unit showed that areas of high concentration were restricted to a 1 km2 area surrounding the point source. FRAME showed good overall agreement with the EMEP model for national average levels of ammonia concentration and reduced nitrogen deposition. However, a difference in spatial patterns of wet deposition occurred due to the different treatment of precipitation in the two models. It was concluded that finer resolution national scale modelling of reduced nitrogen is required in order to improve estimates of critical loads exceedance for acid deposition and nutrient nitrogen deposition. Keywords Ammonia, reduced nitrogen, dry deposition, wet deposition, atmospheric transport model
1. Introduction Emissions of NH3 in the UK have fallen by 19% since 1990. Further decreases of 10% are forecast by the year 2010. Much larger decreases in emissions of SO2 and NOx have occurred in the UK (by 88% and 43% respectively in the last 35 years). Further decreases of 55% and 38% are forecast by the year 2020. As a result of these changes, levels of acid deposition and nitrogen deposition have decreased. However the relative contribution of ammonia to nitrogen deposition and to acid deposition (resulting from in-soil oxidation of ammonia) is increasing. In addition to efforts to nationally monitor the levels of ammonia concentration (Sutton et al., C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
127
128
A.J. Dore et al.
2001; Tang et al., 2001) and deposition of reduced nitrogen (Fowler et al., 2005), numerical models have been developed to estimate nationally the concentrations of ammonia and ammonium aerosol and the deposition of reduced nitrogen. Models have the advantage that they are able to give good spatial coverage, where measurement data may not be available, as well as the ability to simulate future emissions scenarios. Below we present the results of the Fine Resolution Atmospheric Multipollutant Exchange model (FRAME) developed by Singles et al. (1998), Fournier et al. (2005a, b), Vieno (2005) and Dore et al. (2007).
2. Overview of the FRAME Model The main features of the model can be summarised as: x 5 × 5 km2 resolution over the British Isles (incorporating the Republic of Ireland); grid dimensions: 244 × 172. x Input gas and aerosol concentrations at the edge of the model domain are calculated with FRAME-Europe, using European emissions and run on the EMEP 150 km scale grid. x Air column divided into 33 layers moving along straight-line trajectories in a Lagrangian framework with a 1o angular resolution. The air column advection speed and frequency for a given wind direction is statistically derived from radio-sonde measurements (Dore et al., 2006a). Variable layer thickness from 1 m at the surface to 100 m at the top of the mixing layer. x Emissions of NH3 are gridded separately for cattle, pigs, poultry, sheep, fertiliser and non-agricultural sources and mixed into the model at heights of 0–6 m, 0–1 m, 2–10 m, 0–1 m, 0–1 m and 0–1 m, respectively. x Vertical diffusion in the air column is calculated using K-theory eddy diffusivity and solved with the Finite Volume Method. x Wet deposition is calculated using a diurnally varying scavenging coefficient depending on mixing layer depth and a ‘constant drizzle’ approximation. A precipitation model is used to calculate wind-directiondependent orographic enhancement of wet deposition. x Dry deposition for NH3 is ecosystem specific and includes five land classes: forest, moorland, grassland, arable, urban and water. A canopy resistance parameterisation is employed including an optional canopy compensation point module for representation of bi-directional exchange of NH3. x The model chemistry includes gas phase and aqueous phase reactions of oxidised sulphur and oxidised nitrogen and conversion of NH3 to ammonium sulphate and ammonium nitrate aerosol. x The modelled chemical species treated include: NH3, NH4+ aerosol, NO, NO2, HNO3, PAN, NO3– aerosol, SO2, H2SO4 and SO42- aerosol. x Current model run time: 25 minutes on CEH Edinburgh Beowulf cluster using 100 processors.
Modelling the Deposition of Reduced Nitrogen at Different Scales in the United Kingdom
129
3. Results of the Model The output from the model includes maps of annual average surface concentration of NH3 (Fig. 1a) which may be used to assess exceedance of the critical level. Maps of annual vegetation-specific dry deposition and wet deposition of reduced nitrogen (Fig. 1b and c) are used for calculation of exceedance of critical loads for acid deposition and nitrogen deposition
a)
b)
c)
d)
Fig. 1 UK FRAME model prediction for 2002: (a) NH3 surface concentration (ȝg m-3), (b) NHx dry deposition (kg N ha-1), (c) NHx wet deposition (kg N ha-1), and (d) NHx wet deposition with the EMEP model (kg N ha-1)
A.J. Dore et al.
130
Assessment of the accuracy of FRAME in estimating atmospheric concentrations and deposition rates of reduced nitrogen was made by comparison with measurements. For this purpose, data from the UK national ammonia monitoring network were employed comprising 94 DELTA samplers and ALPHA samplers (http://www.cara.ceh.ac.uk/nh3network). The network uses monthly sampling from the CEH DELTA system, (DEnuder for Long Term Atmospheric sampling; Sutton et al., 2001). ALPHA samplers are passive diffusion samplers, developed for long term monitoring and suitable for use in remote areas with low ammonia concentrations (Tang et al., 2001). Wet deposition data were obtained from the secondary acid precipitation monitoring network, comprising fortnightly collections of precipitation from 38 sites with ion concentrations analysed by ion chromatography (NEGTAP, 2001).
b)
a)
c) Fig. 2 Correlation of modelled: (a) NH3, (b) NH4+ aerosol concentrations and (c) NH4+ wet deposition with measurements from the national monitoring network for the year 2002. The continuous line shows the linear regression. The dashed line represents a 1:1 relationship
Modelling the Deposition of Reduced Nitrogen at Different Scales in the United Kingdom
131
Fig. 2a–c illustrate the correlation of the model with measurements. The correlation of modelled concentrations of NH3 with measurements (Fig. 2a) shows considerable scatter. The principal reason for this is the highly localised nature of NH3 emissions, such that the modelled average concentration from a 5 × 5 km2 model grid cell may differ significantly from that measured at a specific location within the grid cell (Dragosits et al., 2002). The graph shows evidence that, particularly at low concentrations, the model overestimates NH3 surface concentrations. There is a need for finer scale national modelling of ammonia concentrations, preferably at a 1 km resolution, in order to perform a more accurate modelmeasurement comparison. A better correlation is observed between modelled and measured NH4+ concentrations (Fig. 2b) and wet deposition (Fig. 2c). This is due to the more slowly changing pattern in NH4+ aerosol concentrations, which are not expected to vary on a scale smaller than the 5 km model grid resolution. Fig. 2b shows that the model generally underestimates NH4+ aerosol concentrations which may indicate either an underestimate in the rate of production of NH4+ aerosol from NH3 gas or in the import of aerosol at the model boundaries.
4. Comparison with a Local Scale Dispersion Model FRAME was run with the model set up to consider only emissions of NH3 from a theoretical typical poultry farm. The poultry unit was assumed to contain 40,000 birds, each with an annual emission of 0.05 kg NH3, comprising a total of 2 Mg NH3 per year. The unit was assumed to be side ventilated with emissions in the height range 1–2 m. The simulation was reduced to one of simple transport, diffusion and dry deposition by switching off both the model chemical scheme and washout from precipitation. A neutral atmospheric thermal stratification was assumed. The results from FRAME were compared with those obtained from ADMS, a local dispersion model. Two model runs were undertaken with both FRAME and ADMS, firstly with local land cover assumed to be grassland and secondly with land cover assumed to be forest. Both models assumed a deposition velocity of 5 mm s-1 for grassland and 40 mm s-1 for forest. For the ADMS simulation representing grassland, the model was run both with emissions evenly distributed across a 5 × 5 km2 area, similar to FRAME, and with emissions located in a single 50 × 50 m2 square, more typical of a real poultry unit. The results of comparing FRAME with ADMS for evenly distributed emissions are given in Table 1. Table 1 Comparison of ammonia concentration and NHx dry deposition modelled with FRAME for a single 5 km grid square and for ADMS with a distributed 5 × 5 km source. Average concentration (Pg m-3) Average deposition (kg N ha-1)
FRAME 0.044 0.056
ADMS 0.039 0.050
A.J. Dore et al.
132
Close agreement in estimates of concentration and deposition between the two models was found despite the different approaches adopted in calculating vertical diffusion. For poultry farms and other intensive farming techniques, the even distribution of NH3 emissions over a 25 km2 area is clearly physically unrealistic. In reality, emissions may be confined to a small area (e.g. a single building or group of buildings). This is better represented with the local dispersion model by allocating emissions to a single 50 by 50 m grid square as illustrated in Figure 3a and b for grass and forest land cover respectively. The use of the fine scale local dispersion model shows that the areas of high concentration are restricted mostly to the 1 × 1 km square at the centre of which is located the point of emissions. Higher concentrations are located to the north east of emissions source due to the predominance of south-westerly winds. The presence of forest land cover (Figure 3b) and its associated higher turbulence and deposition velocity is clearly seen to restrict the area of high concentrations to a smaller area. Across the 5 × 5 km2 domain, in the presence of forested vegetation, average concentrations with ADMS were found to be 3.3 times lower and NHx deposition 4.3 times higher than with the grassland scenario. An earlier study with local scale dispersion model (Dragosits et al., 2002) found that enhanced NH3 air concentrations resulting from a large emissions source were estimated to extend to an area with ~2.5 km radius before becoming from NH3 concentrations over agricultural fields.
(a)
(b)
Fig. 3 Ammonia concentration from a single point source (Pg m-3) modelled with ADMS: (a) grassland; (b) forest
With the local dispersion model, it is clearly seen that ammonia concentrations associated with a single point source emitter vary by over an order of magnitude on the scale associated with a single 5 km FRAME grid cell. This gives the clear message that the current 5 km resolution of national scale assessment of nitrogen deposition will have major uncertainties associated with it in particular areas, depending on the nature of the emissions source. This may result in an overestimation of ammonia concentrations at sites more than 1 km from point sources. This emphasises the need to develop national modelling capabilities (i.e. with FRAME) at a finer 1 km resolution.
Modelling the Deposition of Reduced Nitrogen at Different Scales in the United Kingdom
133
5. Comparison with the EMEP European Model The wet deposition of reduced nitrogen modelled with the EMEP European scale model (Tarrasón et al., 2003) is illustrated in Figure 1d. The larger 50 km grid size is required to give European coverage in the model domain. At this grid spacing, however, it is apparent that the effect of fine scale orography on precipitation is not fully captured by the meteorological model so wet deposition is less strongly correlated to the hill areas of the UK than with FRAME, Figure 1c. The UK budgets for dry and wet deposition of reduced nitrogen and average NH3 concentrations in the UK are compared in Table 2. Overall, despite differences in spatial patterns, a reasonable agreement between FRAME and EMEP for the national deposition budgets is apparent. Interpolation of wet deposition from the 38 sites in the national monitoring network, however, results in a somewhat higher value for the wet deposition budget. In part this may be due to the inclusion of orographic enhancement of wet deposition by the seeder-feeder effect (Dore et al., 2006b) in the mapping procedure. Orographic enhancement of washout is also included in FRAME, but not in the EMEP model. Interpolation from the National Ammonia Monitoring Network leads to an average value of NH3 concentrations in the UK of 1.7 ȝg m-3. Reasonable agreement is found with FRAME (1.8 ȝg m-3) and EMEP (1.4 ȝg m-3). The under-estimate with EMEP may be due to the deep (90 m) lowest model layer. Table 2 Comparison of UK national NHx deposition budgets and mean NH3 concentrations for FRAME, EMEP and data interpolated from the national monitoring networks. NHx wet deposition (Gg N) NHx dry deposition (Gg N) NH3 concentration (ȝg m-3)
FRAME 84 68 1.8
EMEP 73 61 1.3
Measurements 107 65 1.7
Acknowledgments This work was funded by the UK Department of the Environment, Food and Rural Affairs and the Natural Environment Research Council.
References Dore AJ, Vieno M, Fournier N, Weston KJ, Sutton MA (2006a) Development of a new wind rose for the British Isles using radiosonde data and application to an atmospheric transport model. Quart. J. Roy. Met. Soc., 132, 2769–2784. Dore AJ, Mousavi-Baygi M, Smith RI, Hall J, Fowler D, Choularton TW (2006b) A Model of Annual Orographic Precipitation and Acid Deposition and its application to Snowdonia. Atmos. Environ., 40, 3316–3326. Dore AJ, Vieno M, Tang YS, Dragosits U, Dosio A, Weston KJ, Sutton MA (2007) Modelling the atmospheric transport and deposition of sulphur and nitrogen over the United Kingdom and assessment of the influence of SO2 emissions from international shipping. Atmos. Environ., 41, 2355–2367.
134
A.J. Dore et al.
Dragosits U, Theobald MR, Place CJ, Lord E, Webb J, Hill J, ApSimon HM, Sutton MA (2002) Ammonia emission, deposition and impact assessment at the field scale: a case study of sub-grid spatial variability. Environ. Pollut., 117, 147–158. Fournier N, Weston KJ, Dore AJ, Sutton MA (2005a) Modelling the wet deposition of reduced nitrogen over the British Isles using a Lagrangian multi-layer atmospheric transport model. Quart. J. Roy. Met. Soc., 131, 703–722. Fournier N, Tang YS, Dragosits U, Kluizenaar Y, Sutton MA (2005b) Regional atmospheric budgets of reduced nitrogen over the British Isles assessed using an atmospheric transport model. Water, Air Soil Pollut., 162, 331–351. Fowler D, Smith RI, Muller JBA, Hayman G, Vincent KJ (2005) Changes in the atmospheric deposition of acidifying compounds in the UK between 1986 and 2001. Environ. Pollut., 137, 15–25. NEGTAP (2001) Transboundary Air Pollution: Acidification, Eutrophication and Ground Level ozone in the UK. Report of the National Expert Group on Transboundary Air Pollution, DEFRA, London. Singles RJ, Sutton MA, Weston KJ (1998) A multi-layer model to describe the atmospheric transport and deposition of ammonia in Great Britain. Atmos. Environ., 32, 393–399. Sutton MA, Tang YS, Miners B, Fowler D (2001) A new diffusion denuder system for long-term, regional monitoring of atmospheric ammonia and ammonium. Water, Air Soil Pollut.: Focus, 1, 145–156. Tang YS, Cape JN, Sutton MA (2001) Development and types of passive samplers for monitoring atmospheric NO2 and NH3 concentrations. TheScientificWorld, 1, 513–529. Tarrasón L, Fagerli H, Eiof Jonson J, Klein H, van Loon M, Simpson D, Tsyro S, Vestreng V, Wind P, Posch M, Solberg S, Spranger T, Cuvelier K, Thunis P, White L (2003) Transboundary Acidification, Eutrophication and Ground Level Ozone in Europe. PART I Unified EMEP Model description. EMEP Status Report 2003. Vieno M (2005) The use of an Atmospheric Chemistry-Transport Model (FRAME) over the UK and the development of its numerical and physical schemes. Ph.D. thesis, University of Edinburgh, Edinburgh.
Modelling the Deposition of Reduced Nitrogen at Different Scales in the United Kingdom
135
Discussion R. Uni: Did you look at the effect of enhanced surface roughness on the concentrations of ammonia modelled over grassland and forest as they could explain lower modelled concentrations? A.J. Dore: The local scale dispersion model included the influence of surface roughness on turbulent dispersion so the lower ammonia concentrations over forest, as compared to grassland, were caused by a combination of more rapid vertical mixing and increased removal rate from the atmosphere by dry deposition to vegetation. The relative importance of these two effects was not tested in this study but changes in surface roughness are considered to be significant in leading to more rapid dispersion of ammonia. M. Schaap: The wet deposition map shows different features between model and interpolated means. Rain fall is a strong function of air mass origin. Does your constant drizzle assumption affect the distribution much? Would a different strategy to account for rain as function of trajectory improve your results? And how would you do that? A.J. Dore: The constant drizzle approach does not account for the episidocity of precipitation events. It is possible to run a combination of dry and wet trajectories although earlier tests showed that this did not lead to an improved correlation with measurements of wet deposition. The rate of constant drizzle does however include a directional dependence (derived from a directional orographic precipitation model). In practice therefore washout of material from the atmosphere is higher for ‘rainy directions’ such as south-westerly flow. The sum of the directional precipitations was normalised to give a total precipitation for each model grid square in line with measurements of annual precipitation.
3.8 A Suggested Correction to the EMEP Database, Regarding the Location of a Major Industrial Air Pollution Source in Kola Peninsula Marko Kaasik, Marje Prank, Jaakko Kukkonen and Mikhail Sofiev
Abstract For tracing the sources of aerosol fractions measured in the campaign at Värriö, Finnish Lapland, spring 2003, the SILAM model was applied in two modes: (1) inverse (adjoint) run, with measured aerosol concentrations as the sensitivity source function; and (2) forward run with EMEP database of sulphur dioxide and particulate matter emissions. One of coarse aerosol (0.1–10 µm) peaks resulted in total aerosol concentration exceeding 20 µg/m3 that is very high for polar latitudes. The inverse model for that peak points at the border area between Russia and Norway, but forward run failed to reproduce it. It was found that the well-known metal industries of Apatity-Kirovsk and Monchegorsk, Kola Peninsula, Russia, are represented adequately in EMEP database, but no significant emissions were found at the site of Nikel (a major metal smelting industry), 7 km from the border with Norway. Very high emissions were originated from about 100 km to the east instead. Then the database used to run SILAM was corrected on the basis of satellite images of the region. When running the SILAM model using the revised database, the agreement of measured and modelled peak concentrations was substantially better. A proposal for database correction is made to EMEP. The correction is supposed to improve the quality of air pollution transport calculations. Keywords Air pollution transport, EMEP, industrial emissions, Lapland
1. Introduction This paper is based on an unexpected finding when studying the possible sources of aerosol peaks measured during the field campaign at Värriö, Finnish Lapland, spring 2003 (Ruuskanen et al., 2007). In general, the EMEP database of emissions, based on the reports of European countries and estimations of EMEP experts, is believed to be a reliable source of information for atmospheric transport modelling. Emissions, including both anthropogenic and natural, are given for each year with space resolution 50 km. C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
331
332
M. Kaasik et al.
The SILAM model (Sofiev et al., 2006) with the EMEP emission database is regularly used by Finnish Meteorological Institute to estimate the concentrations and deposition loads of sulphates, PM10 and PM2.5 in the whole of Europe. On the basis of model runs for recent years it was found that concentrations for northern Finland are seriously underestimated with respect to measurements at monitoring stations. It is a well-known fact that main pollution sources for Lapland are the large metallurgy plants located at Kola Peninsula, Russia. Previously it has been assumed that the reason for this underestimation is either deficiency of modelling techniques or incomplete data on emissions.
2. Methods An atmospheric aerosol measurement campaign was carried out at Värriö, Finnish Lapland, 67o 46’ N, 29o 35’ E (about 10 km from the Russian border), in April– May 2003 (Ruuskanen et al., 2007). Aerosol spectra in the size range of 3 nm–10 µm were measured with time step of 10 minutes, using electric aerosol spectrometer EAS (Tammet et al., 2002). For tracing the sources of aerosol fractions, the SILAM model (Sofiev et al., 2006) was applied in two modes: (1) inverse (adjoint) run, with measured aerosol concentrations as sensitivity source function; and (2) forward run with EMEP database of sulphur dioxide and particulate matter emissions, reference year 2001. For forward modelling the well-tested Lagrangian kernel and new Eulerian kernel (test version 4.1 released in 2007, presently under evaluation) of SILAM were applied with both ECMWF and HIRLAM (FMI) meteorological datasets, thus producing ensemble of four runs. Parameters of a highly buoyant plume from high stack were assigned: source height distributed from 200 to 1,000 m. SILAM calculates the SO2 to SO4 conversion via a single-reaction transformation scheme after Sofiev (2000). The meteorological fields from short-term forecasts of HIRLAM (Finnish Meteorological Institute) and ECMWF were used (Kaasik et al., 2006). It was assumed that at peak time the accumulation mode (0.1– 1 µm) particles consist mainly of sulphur dioxide. SILAM takes sea salt emissions into account, according to the estimations by Mårtensson et al. (2003). SILAM was run with 20 km grid resolution. Two types of aerosol concentration peaks were found: (1) bursts of nanoparticles, the origin of which was identified as natural (phytoorganic or marine); and (2) particles originated from industrial sources, within the size range from 0.1 to 10 µm. We have investigated in this study the highest concentration peak that occurred at night of May 2–3 (Figure 1).
A Suggested Correction to the EMEP Database, Regarding the Location of a Major Industrial 333
Concentration, Pg/m3
20 18
1 - 10 micrometres
16
0.1 - 1 micrometres
14 12
Peak under consideration
10 8 6 4 2 0 4/28 4/29 4/30
5/1
5/2
5/3
5/4
5/5
5/6
5/7
5/8
5/9
5/10 5/11 5/12
Date
Fig. 1 Measured surface-level concentrations of aerosol particles with diameter 0.1–1 and 1–10 µm in Värriö, 2003
3. Results and Discussion
3.1. The location of source and suggested correction A model computation in order to reproduce the peak of May 2–3 using the original EMEP database was unsuccessful (Figure 2) – none of pollution plumes from the major sources in the Kola Peninsula reached the measurement station at this time. The inverse model computation that was performed in order to clarify the probable source area pointed at a small region located directly north of the measurement site at Värriö (Figure 3). The town of Nikel and a large metallurgy plant are situated in this area, in Russia immediately near the border with Norway. Looking at the exact location of Nikel in the EMEP database (69º 20’ N, 30º 04’ E), we see that emissions in this grid cell do not differ from the sparsely inhabited tundra around (Figure 4). However, an extremely strong source of sulphur dioxide and particulate matter is placed about 100–150 km eastwards, in the Murmansk area. As there are no known major point sources and urban emissions cannot include such amounts of sulphur dioxide, we expect that the emissions of the Nikel metallurgy plant were misplaced near Murmansk. According to these results, we have made a correction in our EMEP-based database of sources. The emissions from EMEP sector S1 (large combustion in energy and transformation industry) in Murmansk area were moved to the grid cell of Nikel location. Also, other sectors, expected to contribute to the infrastructure of a large factory, where reduced in estimated proportions. Thus, the computations using SILAM with that correction is expected to produce commonly narrow pollution plumes from Nikel.
M. Kaasik et al.
334
Concentration, g/m 3
35
ECMWF
30
HIRLAM
25
Obsered aerosol
20 15 10 5 0 428
429
4- 5-1 5-2 5-3 5-4 5-5 5-6 5-7 5-8 5-9 530 10 Date
511
512
Fig. 2 Observed PM10 concentrations compared with model calculations (SO4, primary PM10 and sea salt included) with ECMWF and HIRLAM meteorological fields, original EMEP database of emissions (Lagrangian SILAM)
+a
b Norway
+Nikel +
+ Finland
Värriö
Russia
Fig. 3 (a) Sensitivity distribution (footprint) of accumulation mode (0.1–1 Pm) aerosol, May 3 at 0–6 a.m. Inverse run of SILAM with ECMWF meteorological data. (b) Map of RussianNorwegian-Finnish border areas with exact location of Nikel (Google Earth)
3.2. Results with correction and uncertainties related to the meteorological input Model runs with the Lagrangian kernel produced significantly different results (Figure 5a): the HIRLAM meteorological fields resulted in a good agreement with
A Suggested Correction to the EMEP Database, Regarding the oLcation of a Major Industrial
335
measurement, whereas ECMWF fields produced as poor results as without correction. Both Eulerian runs show the peak, but slightly delayed and concentrations are overestimated (Figure 5b). Reasons of such behaviour are seen in Figures 6 and 7: x The Lagrangian plume with ECMWF data narrowly missed the monitoring station passing eastwards x The Lagrangian plume with HIRLAM data matched the monitoring station, predicted concentrations and timing were rather similar to measured ones x The Eulerian plume with ECMWF data matched the monitoring station with slight delay, predicted concentrations were overestimated x The Eulerian plume with HIRLAM data matched the monitoring station with slight delay, predicted concentrations were highly overestimated In the Lagrangian run with ECMWF meteorological data the plume propagated from Nikel directly to south, missing the Värriö site by a narrow margin. HIRLAM predicted slightly different wind directions – plume was moved first to south-southwest and then, after wind turned, to south-east over Värriö. In general, the Eulerian scheme produced much wider horizontal spread than the Lagrangian one despite the higher concentration peaks.
Fig. 4 EMEP emissions of SOx applied for forward runs
336
M. Kaasik et al.
Fig. 5 Concentrations of sulphate aerosol during the May 2–3 event, computed with corrected EMEP database of emissions, applying SILAM with (a) Lagrangian and (b) Eulerian kernel. Comparison with measured concentrations of aerosol of diameter below 1 µm (PM1) included
4. Conclusions We have compared the results of inverse and forward computations of an advection-diffusion model with the concentration data measured at a station located in the Finnish Lapland. In addition, we have used the satellite images of the region and information on actual location of major sources in Kola Peninsula. As a result, we could suggest a substantial improvement in the emission database of EMEP. Also, the current exercise highlights the objective difficulties in predicting the short-term episodes originated from a single nearly-point source and observed at a single site. It is, however, evident that an ensemble of four partly independent chemical-weather predictions appeared able to show both high probability of the episode, its stochastic features and a potential range of uncertainties in the results of simulations.
A Suggested Correction to the EMEP Database, Regarding the oLcation of a Major Industrial
337
Fig. 6 Surface-level concentration fields of sulphate (µg/m3), 0 GMT at May 3, 2003, calculated with Lagrangian SILAM: (a) with ECMWF meteorological fields, (b) with HIRLAM (FMI) meteorological fields. Arrows – wind at height of 25 m
Fig. 7 Surface-level concentration fields of sulphate (µg/m3), 0 GMT at May 3, 2003, calculated with Eulerian SILAM: (a) with ECMWF meteorological fields, (b) with HIRLAM (FMI) meteorological fields. Arrows – wind at height of 25 m
338
M. Kaasik et al.
Acknowledgments This study was supported by the Estonian Science Foundation, research grant 7005 and NetFAM, a network cooperation program for meso-scale atmospheric modelling under Nordic Research Board (NordForsk).
References Kaasik M, Sofiev M, Prank M, Ruuskanen T, Kukkonen J, Kulmala M (2006) Model-delineated origin and growth of particles during the nucleation events observed in Värriö campaign in 2003, in Proceedings of BACCI, NEC and FcoE activities in 2005. Report Series in Aerosol Science, 81, 221–226. Mårtensson EM, Nilsson ED, de Leeuw G, Cochen LH, Hansson, H-C (2003) Laboratory simulations of the primary marine aerosol production. J. Geophys. Res., 108, D9, 4297, doi:10.1029/2002JD002263. Ruuskanen TM, Kaasik M, Aalto PP, Hõrrak U, Vana M, Mårtensson EM, Yoon YJ, Keronen P, Mordas G, Ceburnis D, Nilsson ED, O’Dowd C, Noppel M, Alliksaar T, Ivask J, Sofiev M, Prank M, Kulmala M (2007) Concentrations and fluxes of aerosol particles during the LAPBIAT measurement campaign in Värriö field station. Atmospheric Chemistry and Physics, 7, 3683–3700. Sofiev M (2000) A model for the evaluation of long-term airborne pollution transport at regional and continental scales. Atmospheric Environment 34, 15, 2481–2493. Sofiev M, Siljamo P, Valkama I, Ilvonen M, Kukkonen J (2006) A dispersion modelling system SILAM and its evaluation against ETEX data, Atmospheric Enviroment, 40, 674–685. Tammet H, Mirme A, Tamm E (2002) Electrical aerosol spectrometer of Tartu University. Atmos. Res. 62, 315–324.
Discussion S.-E. Gryning: Is there an explanation why the two models (HIRLAM and ECMWF) predict so different wind directions? M. Kaasik: Most probably due to different treatment of surface-layer wind shear. But this is not clear yet, we will study that issue further.
3.5 Air Quality Forecasting During Summer 2006: Forest Fires as One of Major Pollution Sources in Europe Mikhail Sofiev, Pilvi Siljamo, Ari Karppinen and Jaakko Kukkonen
Abstract This paper considers an influence of wild-land fires on air quality in Europe during spring and summer seasons of 2006 and discusses the experience of its forecasting by the SILAM modelling system that considers anthropogenic, natural and fire-related emission sources. The emission of anthropogenic pollutants was based on the EMEP database, while the fire emission was deducted from the near-real-time MODIS satellite retrievals, which were assimilated daily by the emission pre-processor. The system also included allergenic birch pollen as described by Siljamo et al. (this volume). We analysed several episodes and compared an impact of fires with contribution from anthropogenic sources. For example, exceptionally high concentrations of nearly all pollutants were detected in Central, Eastern and Northern Europe in April and May of 2006. Simulations showed that this episode was formed by contributions of all three main sources: major wild-land fires in Russia, substantial amounts of anthropogenic air pollutants accumulated for a few days over Eastern Europe, and intensive birch flowering in Russia. A synchronization role was played by meteorology. Specific weather conditions promoted formation of the multi-component pollution cloud, which was then transported across most of Central and Northern Europe causing widespread allergic symptoms and other illnesses associated to poor air quality. Several other episodes were more localised. Thus, August of 2006 was particularly difficult in Portugal and Spain, but also in Finland. The model predictions were compared with available information from ground-based monitoring sites and satellites retrievals. The agreement was fairly good, especially for timing of rise and fall of concentrations, while the absolute levels were usually under-estimated by the model. That pointed to necessity to extend the list of sources of, in particular, atmospheric aerosols to include wind-blown dust, sea salt, nitrates, SOA, etc, and also to refine the parameterization of emission from biomass burning. Keywords Atmospheric dynamics and composition modelling, air quality forecasting, forest fires
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
305
M. Sofiev et al.
306
1. Introduction The intensive wild-land fires in spring in western Russia are known to take place nearly every year when the remnants of the last-year grass get sufficiently dry and ready to be burned. However, in 2006 the fumes of the fires were transported in unusual direction – towards the west – and the time of fires coincided with poorventilation conditions over Central Europe and with intensive birch flowering. As shown by Saarikoski et al. (2007) and Stohl et al. (2006), at the end of April, species emitted by fires were transported to north-west heavily affecting the air quality in Finland and reaching Spitsbergen. The second hit, even more powerful, was in westward direction when the plumes reached Atlantic Ocean and Iceland and were registered by satellites (e.g., EUMETSAT, 2006 and Figure 2). Existing air quality forecasting systems in Europe commonly address anthropogenic pollutants, such as NOx and O3, sometimes including other species, such as SOx, PM or NH3 (Kessler et al., 2001; Bessagnet et al., 2004). The forecasts are usually available for horizon of 24–48 hours ahead in time. The air quality modelling system SILAM (Sofiev et al., 2006a, b) is the only model in Europe that also provides the forecasts of fine particles (PM2.5) originated from satelliteobserved wild-land fires and birch pollen at a continental scale (these are available at http://silam.fmi.fi). This paper provides a retrospective analysis of the SILAM forecasts for the spring and summer of 2006 and analyses the origins, evolution and contributions of each of the source types to the above allergenic air pollution episode in Europe.
2. Materials and Methods The forecasting system. Since late winter of 2006, the regional forecasting system of FMI covers three major types of sources (Figure 1): anthropogenic emission of sulphur oxides and primary particulate matter PM 2.5 and PM 10, biological sources of birch pollen and satellite-retrieved real-time information about the wildland fires (based on hot-spots counts from MODIS instrument onboard NASA Aqua and Terra spacecrafts). The SILAM model. The operational version of SILAM used in 2006 was based on a Lagrangian dispersion core that applies an iterative advection algorithm and a Monte Carlo random-walk diffusion representation (Sofiev et al., 2006b). The system can directly utilize the meteorological data from the HIRLAM and ECMWF numerical weather prediction models, as well as their archives. The operational forecasts included the following pollutants and source categories:
– –
Primary PM2.5 originated from biomass burning Sulphur oxides originated from anthropogenic and volcanic sources
Air Quality Forecasting During Summer 2006
– –
307
Primary particulate matter originated from anthropogenic sources (both fine particulate matter with diameter below 2.5 Pm, PM2.5, and the coarse fraction with diameter from 2.5 to 10 Pm, PM2.5–10) Birch pollen originated from birch forests during the flowering season
Fig. 1 A structure and main items of the regional AQ forecasting system of FMI
Input emission data for the anthropogenic pollutants and Etna volcano are based on the database of European Monitoring and Evaluation Programme EMEP (http://www.emep.int). The near-real-time information on active biomass burning is extracted from the observations of the MODIS instrument onboard the NASA Aqua and Terra satellites (http://modis.gsfc.nasa.gov) with a spatial resolution of 1 × 1 km2. The emission fluxes of PM2.5 are based on the daily temperature anomalies (the temperature anomaly is defined as the difference of the observed and the long-term average temperatures) following the procedure described in Saarikoski et al. (2007). The start day of the release is additionally adjusted to the conditions of the specific year using the near-real-time pollen observations of the European Aeroallergen Network (EAN, http://www.univie.ac.at/ean/). Specifics of the pollen grains as an atmospheric pollutant are taken into account in the parameterizations of the dry and wet deposition. The accuracy of pollen predictions has been evaluated against several historical episodes (Sofiev et al., 2006a). During pre-operational SILAM evaluation its results were compared with the data of the European Tracer Experiment (Sofiev et al., 2006b), Chernobyl and Algeciras accidental releases. For the air quality-related species, the system output was compared with the European air quality networks for the period of 2000–2002 and showed quite good agreement.
308
M. Sofiev et al.
3. Results Numerical results and evaluation for April–May 2006. During the episode in April and May 2006, the pollutants originated from three sources: major wild-land fires in Russia, anthropogenic emission, and birch pollen from flowering trees in Eastern and Northern Europe and Western Russia. We used PM2.5 as a tracer for both anthropogenic and fire plumes.
Fig. 2 A pattern of fire-originated PM 2.5 concentrations (upper left panel, Pg PM m-3), anthropogenic primary PM 2.5 (upper right panel, Pg PM m-3), and total-aerosol plumes (yellow-grey haze over ocean and continents, EUMETSAT, 2006) as seen from the space. All pictures are for 8 May 2006
Since middle of April the daytime ambient air temperatures were quite high – close to or above +20qC – over most of Eastern Europe and the European part of Russia. Together with small amount of precipitation, they promoted an intensive birch flowering and contributed to the drying of remnants of the last-year grass and leaves increasing risk of wild-land fires. The prevailing anti-cyclonic conditions were also associated with fairly low wind speeds limiting the dilution of
Air Quality Forecasting During Summer 2006
309
anthropogenic pollutants over Central and Eastern Europe. At the beginning of May the resulting pollution cloud was blown towards the west leading to strong deterioration of air quality in most of Central and Northern Europe – up to Iceland. An example of predicted concentrations of pollen grains and fine particles (originated from both anthropogenic sources and wild-land fires), during 8 May are presented in Figure 2, which shows that by that day the pollution was distributed across most of Central and Northern Europe and large areas over Atlantic Ocean. The reported patterns are qualitatively very similar to total aerosol column presented in the EUMETSAT images (EUMETSAT, 2006). From Figure 2, it is seen that the spatial distributions of the concentrations of anthropogenic and fire-born PM2.5 resemble strong similarities, especially over the northern Atlantic. Despite the different source areas, the specific properties of the atmospheric flows have synchronised the pollution motion over large territory and created the multi-substance plume. The predicted pollution cloud was very wide (approximately 1,000 × 2,000 km) and substantial concentrations occurred as far as over Iceland and Spitsbergen (the northward transport took place 25 April–2 May and was studied by Saarikoski et al. (2007) with SILAM and Stohl et al. (2006) with FLEXPART models, the latter also involving the satellite retrievals of the aerosol optical depth and carbon monoxide total columns). Total PM 2.5 at Helsinki, urban background 120
100
ug/m3
80 Measured
60
Modelled
40
20
ay
ay
ay 20 -M
15 -M
11 -M
6M ay
2M ay
27 -A pr
23 -A pr
18 -A pr
14 -A pr
9Ap r
5Ap r
0
Fig. 3 Comparison of predicted PM2.5 from fires and anthropogenic sources with total-PM2.5 observed at Kumpula station (urban background in Helsinki)
A comparison of the modelled PM2.5 concentrations originated from anthropogenic sources and the wild-land fires with the observations in Helsinki is presented in Figure 3. Qualitatively, the model has succeeded fairly well in predicting the timing of most of the highest concentrations. It has largely missed, however the second rise of the concentrations during 3–5 May. It happened, most probably, due to inaccurate source reporting (visibility was partly obscured by clouds during these days).
310
M. Sofiev et al.
Evaluation of the episode during August 2006. The late part of the fire season in 2006 was characterised by many hot spots in the south of Europe (Figure 4, lefthand panel), which in the south-east (Ukraine and south of Russia) formed wide burning areas covering significant parts of the country’s territories. A set of comparatively small but strongly polluting fires appeared in the Karelia.
Fig. 4 Fire hot spots in August 2006 (left-hand panel) and predicted PM2.5 plumes from fires for 21 August 2006
As seen from the right-hand panel of Figure 4, the main transport of plumes from the Ukrainian and Russian fires was eastward, which reduced their impact on European countries. Northern fires, to the opposite, were under the influence of westward flow. As a result of their influence, the hourly concentrations in the south of Finland and in Helsinki were episodically exceeding 200 Pg PM m-3 with sharp reduction of visibility. Unfortunately, the model simulations were unable to reproduce the absolute levels of the peaks. The timing was nearly perfect (error of the plume arrival and departure in Helsinki, for instance, was less than 1–2 hours) but absolute under-estimation was as strong as about 20 times at some moments. This happened due to strong under-estimation of the emission flux by the fire assimilation system. The emission factors used in this system during 2006 were calibrated to spring episode of open-grass burning. In August, however, the burning biomass mainly consisted of bog and mosses, with fires going on under the tree crowns. Such burning produced much stronger PM emission than comparable grass fires in spring but left much weaker fingerprint in the satellite data.
Air Quality Forecasting During Summer 2006
311
4. Conclusion We have investigated two major fire periods during 2006 that have strongly affected several European countries. The episode in April and May was characterised by simultaneous high concentrations of anthropogenic, fire-born, and natural pollutants, with meteorology playing a major synchronization role. The August episode was caused by several burning areas, which appeared to affect different regions. The most noticeable effect in Northern Europe was caused by small but strongly emitting fires in Karelia. We utilised the SILAM modelling system for analysing the cases, as its forecasts cover all relevant source categories. Previously, no modelling forecasts have covered pollen together with other pollutants, despite the mounting evidence that one should take them all into account analysing the adverse health effects of air pollution (Behrendt and Becker, 2001; D’Amato et al., 2001). The model predictions were compared with the EAN birch pollen counts, groundbased observations of PM2.5 and the satellite data. The predictions agree fairly well with the measured data, especially in terms of timing of the plume arrival. Absolute levels, however, were underestimated, sometimes strongly, which highlighted the needs for better emission description of these complicated phenomena. The findings were used in development of the next generation of the SILAM model (Sofiev et al., this volume) and for on-going refinement of the fire assimilation system. Acknowledgments The study was performed within the scope of EU-GEMS project (Global Earth-System Monitoring using satellite and in-situ data) and the ESA PROtocol MOniToring for the GMES Service Element: Atmosphere (PROMOTE) project (http://www.gse-promote.org).
References D’Amato G, Liccardi G, D’Amato M, Cazzola M (2001) The fole of outdoor air pollution and climatic changes on the rising trends in respiratory allergy. Respir. Med. 95, 606–611. Behrendt H, Becker W-M (2001) Localization, release and bioavailability of pollen allergens: the influence of environmental factors. Curr. Opin. Immunol., 13(6), 709–715. Bessagnet B, Hodzic A, Vautard R, Beekmann M, Cheinet S, Honoré C, Liousse C, Rouil L (2004) Aerosol modeling with CHIMERE – preliminary evaluation at the continental scale, Atmos. Environ., 38, 2803–2817 EUMETSAT (2006) European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) press release, http://www.eumetsat.int/ idcplg? IdcService=SS_GET_PAGE&nodeId=442&l=en
M. Sofiev et al.
312
Kessler Ch, Brücher W, Memmesheimer M, Kerschgens M, Ebel A (2001) Simulation of air pollution with nested models in North Rhine-Westphalia, Atmos. Environ, 35(Suppl. 1), S3-S12. Saarikoski S, Sillanpää M, Sofiev M, Timonen H, Saarnio K, Teinilä K, Karppinen A, Kukkonen J, Hillamo R (2007) Chemical composition of aerosols during a major biomass burning episode over northern Europe in spring 2006: experimental and modelling assessments. Atmos. Environ., 41, 3577–3589. Sofiev M, Siljamo P, Ranta H, Rantio-Lehtimäki A (2006a) Towards numerical forecasting of long-range air transport of birch pollen: theoretical considerations and a feasibility study, Int. J. Biometeorol., doi:10 1007/s00484-006-0027-x. Sofiev M, Siljamo P, Valkama I, Ilvonen M, Kukkonen J (2006b) A dispersion modelling system SILAM and its evaluation against ETEX data, Atmos. Environ, doi:10.1016/j.atmosenv.2005.09.069. Stohl A, Berg T, Burkhart JF, Fjæraa AM, Forster C, Herber A, Hov Ø, Lunder C, McMillan WW, Oltmans S, Shiobara M, Simpson D, Solberg S, Stebel K, Ström J, Tørseth K, Treffeisen R, Virkkunen K, Yttri KE (2006) Arctic smoke – record high air pollution levels in the European Arctic due to agricultural fires in Eastern Europe, Atmos. Chem. Phys. Discuss., 6, 9655–9722.
Discussion M. Kaasik: Perhaps you need a map of areas to identify the Fires? M. Sofiev: To some extent, yes, that would reduce the uncertainty in emission speciation. However, more significant variability originates from the type of fire. Variation in emission fluxes between flames and smouldering is much larger than e.g. between the same-type samestrength fires in different forests. And, unfortunately, this dimension is much more difficult to account for. The second problem with the land use is that in Europe very strong changes can occur at a scale of a few tens of metres: cultivated areas neighbour small wild-land fields, forest patches mix with small bog-fields, etc. Fires tend to group in the areas where more easy-to-burn fuel is available. Thus, pit-fire is more probable and longer-lasting than the burning of a nearby forest, especially in case of not very dry weather, etc. Therefore, the land use has to come with very high resolution to be valuable.
3.3 An Observing System Simulation Experiment (OSSE) for Aerosols Renske Timmermans, Martijn Schaap, Arjo Segers, Hendrik Elbern, Richard Siddans, Stephen Tjemkes, Robert Vautard and Peter Builtjes
Abstract The hope to obtain real-time comprehensive two-dimensional or threedimensional analysis or forecasts of air quality pollutants may be fulfilled by combination of model simulations and observations. Space-borne observations can be of particular interest for e.g. aerosols. Following consultation with representatives of the operational meteorological and air chemistry/air quality community, requirements have been documentted for Aerosol Optical Depth (AOD) derived from space borne observations used for operational air quality applications. The users requested an AOD product in two broad layers (planetary boundary layer and free troposphere) with a time and horizontal space resolution of 0.25–1 hour and 0.5–5 km respectively. The objective of this study is to determine whether these requirements are necessary to have an impact on the forecast and analysis of PM2.5 levels over Europe and to investigate if AOD measurements with more relaxed requirements in time and space will lead to noticeably less impact. To this end, an Observing System Simulation Experiment (OSSE) has been designed. OSSEs are commonly used to quantify the impact of observations from future observation systems such as satellite instruments or groundbased networks on e.g. weather forecasts. In this study we apply such an OSSE to AOD measurements from two future satellite instruments using the LOTOS-EUROS chemistry transport model and the ensemble Kalman filter data assimilation method. Assimilation of synthetic AOD measurements from an imager type instrument providing total AOD in some cases improves the analysis of PM2.5 concentrations. The level of improvement depends on a.o. the vicinity of simultaneously assimilated groundbased measurements. In this paper the set-up of the study is explained and some first results are shown. Keywords Aerosols, AOD, data assimilation, OSSE, satellite measurements
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
287
288
R. Timmermans et al.
1. Introduction In Europe, particulate matter (expressed as aerosol mass concentrations at the surface, PM) is the most important air pollutant responsible for loss of human health. To protect against these negative health effects, the EU has set limit values for PM10. Recently, as a result of the Clean Air For Europe (CAFE) programme, new limit values for PM2.5 (Particulate Matter with a diameter less then 2.5 um) are being proposed by the European Commission, to be attained by 2010. Besides the limit values, the member state have an obligation to inform the public on the air quality situation. Further, many countries provide forecast of air pollution levels for the next day(s). Hence, as identified within for example the ESA-project PROMOTE, there is a need for accurate near real time assessments of air quality and forecasts. In contrast to ground based monitoring sites, satellite measurements of aerosol optical depth (AOD) provide full spatial coverage (although only during daylight and in cloudfree conditions) and – in principle – consistent data for the whole European region. This suggests that satellite measurements may be useful to improve the insight in PM distributions in Europe in combination with models and ground based measurements. Various studies in the U.S. have reported good correlations between satellite derived AOD and PM2.5 surface concentration measurements in parts of the U.S. and Europe (Wang and Christopher, 2003; Chu et al., 2003; Hutchison, 2003; Koelemeijer et al., 2006). In general, promising correlations are found between one-month time-series of AOD and PM2.5 for many stations in the Eastern and Midwest U.S. Other stations, however, particularly in the Western US, show hardly any correlation (Engel-Cox et al., 2004). Variations in local meteorological conditions, occurrence of multiple aerosol layers, and variations in aerosol chemical composition likely play an important role in determining the strengths of such correlations. Hence, this suggests that AOD observations may be useful to analyse PM distributions over Europe and could improve the initial conditions for forecasting applications. For the emerging operation air quality and air chemistry applications requirements on the resolution and accuracy of satellite measurements of aerosol optical depth have been formulated. In general terms, the requirements are a time resolution of 30 minutes, a vertical resolution of 2 km, and a horizontal resolution of about 10 × 10 km2. Although user consultations (e.g. Lelieveld, 2003) as well as the projects Promote and Capacity (Goede, 2005) have led to these requirements, a quantification of the impact of these requirements has been lacking so far. The aim of this study is to determine, in an as far as possible quantitative manner, whether the above formulated requirements for a new aerosol instrument are really necessary to have a substantial impact relative to the impact of ground level observations of PM2.5. It has been investigated if aerosol information with more relaxed requirements in time, 1–4 hour time resolution, and space, total vertically integrated AOD, will lead to noticeable less information content. The main model tool used in this study is the TNO CTM LOTOS-EUROS (Schaap et al., 2005) and the data assimilation method ensemble Kalman filter.
An Observing System Simulation Experiment (OSSE) for Aerosols
289
These two building blocks are used to create the system by which the impact has been quantified. The concept of the so-called Observing System Simulation Experiments (OSSEs), which are regularly performed in meteorology, has been followed in this study. An initial assessment of the added value of satellite measurements of AOD at different resolutions for the forecasting of ground level PM concentrations has been performed in the study: “Operational Use of satellite derived aerosol information” (Timmermans et al., 2006). Review of that work led to recommendations for continuation of the work and recommendations for improvements in the set-up of the OSSE. This paper describes the set-up of the follow-up study currently being performed together with some first results.
2. Set Up of OSSE To determine the impact of existing or future instruments, methodologies have been developed and used within the meteorological community. Observing System Experiments (OSEs) are used to assess and compare existing operational observing systems. Observing System Simulation Experiments (OSSEs) are used to analyse proposed instruments. Targeted Observations are used to determine the impact of (detailed) observations at critical places or/ and critical times. In this study an OSSE for AOD from future instruments, The Flexible Combined Imager (FCI) and the MTG Oxygen A-band sounder, has been carried out. The OSSE for aerosols contains the following elements: x Carrying out a so-called nature run or reference run, which is supposed to simulate the “true” state of the atmosphere. This run has been done with the CHIMERE model with a high resolution, and state-of-the-art descriptions of the relevant processes. x From the nature run both simulated groundbased observations at selected locations are taken, as well as simulated vertical profiles of AOD (Siddans et al., 2007). In this way pseudo-observations are created which reflect the “true” state. x An assimilation model: we use the LOTOS-EUROS model in combination with an ensemble Kalman filter assimilation system. The differences in calculated concentrations by the assimilation model and calculated concentrations by the nature run should ideally approximate the differences between a “state-of-the-art” model and the real atmosphere. x The assimilation model is used to assimilate the pseudo ground level observations. Subsequently pseudo AOD observations are added in the assimilation, in which AOD observations with more, or less detail in time resolution and vertical resolution are used. x Starting with the assimilated fields from above assimilation runs, forecast runs are performed in which the model runs free and no assimilation of observations is performed.
R. Timmermans et al.
290
The model results are analysed over an extended period for the different assimilation runs and over shorter periods for the forecast runs. The results are verified against the nature run to obtain a quantitative estimate of the impact of the proposed observing systems for AOD. For the study two different monthly periods have been chosen in 2003 corresponding to a mostly clear scenario and a partly cloudy scenario. The runs are performed at 12 × 12 km resolution over central European domain (42.5–60q N; –5q W–30q E) and 3 × 3 km resolution over a smaller domain around Paris. To investigate the advantage of vertically resolved aerosol measurements to detect high altitude aerosol loads, a temporary wild fire dust source is added to the nature run.
x
3. Model System The model used in this study is the LOTOS-EUROS model, a 3D Eulerian chemistry transport model of intermediate complexity. The model covers Europe and is aimed to describe air pollution in the lower troposphere (up to 5 km above sea level). In the vertical the system has five layers using the dynamic mixing layer approach. The standard horizontal resolution is 0.5° longitude x 0.25° latitude, approximately 25 × 25 km2, with the possibility to increase the resolution up to a factor 8. The model contains all relevant processes, although mostly in a parameterised way to avoid exceedingly long computing times. Prior applications of the model to aerosols have been documented in literature (Schaap et al., 2004a, b; Robles Gonzalez et al., 2003). A description of the model including latest developments is presented in Schaap et al. (2008). Data assimilation consists of making a best estimate of the state of the atmosphere on the basis of observations and a model prediction of the atmospheric state both of which have associated errors. Data assimilation basically defines a new atmospheric state by making a weighted average of the observed and modelled state in an intelligent and statistically sound way. Hence, if a model value is more uncertain than an observed value, more weight will be put on the observation, and the assimilated value will tend to get closer to the observed value and vice versa.In this study we used an ensemble Kalman filter to assimilate the AOT retrievals and PM measurements within LOTOS-EUROS. The uncertainties involved with the modelled and retrieved AOT/PM values determine the weights that are put on the measured and calculated values. With a Kalman filter there is no need to specify the model uncertainties as they are determined by the range of modelled states of the ensemble members. Hence, the specification of the noise influences the weights and therewith results of the procedure. We used an ensemble of 12 members and the random noise was added to the emissions of NOx, SOx, NH3 and particles. The noise factors were kept constant between analyses.
An Observing System Simulation Experiment (OSSE) for Aerosols
291
4. First Results Here we will show some first results of the follow-up study. Figure 1 shows the mean PM2.5 concentrations before and after assimilation of either synthetic groundbased PM2.5 observations or synthetic total AOD observations from the FCI instrument averaged over a one-month period.
Fig. 1 PM2.5 concentrations at ground level for 15 July-15 August 2003 from the nature run (top left), the LOTOS-EUROS model without assimilation (top right), the LOTOS-EUROS model with assimilation of synthetic groundbased PM2.5 measurements (middle left), the LOTOS-EUROS model with assimilation of halfhourly total synthetic AOD measurements from the Imager (middle right)
Assimilation of the synthetic groundbased PM2.5 measurements brings the analysed PM2.5 distribution closer to the nature run results (‘truth’) in the vicinity of the measurement locations. Assimilation of synthetic total AOD measurements from the Imager certainly has an influence on the analysed fields. For example over the south of Germany and France the concentrations are increased and closer to the nature run values. However, the increase over the southern part of the North Sea and Denmark is too large. This can be explained by looking at the differences in AOD between the model and the Imager (not shown here). We have higher retrieved Imager AOD values over the North Sea, France and South of Germany than calculated by the LOTOS-EUROS model. Naturally assimilation of these
292
R. Timmermans et al.
higher AOD values leads to an increase in analysed PM2.5 concentrations at the ground although over the North Sea this means a deterioration of the results. The impact of the Imager AOD values seems small when looking at the average fields over one month. However the impact can be larger when looking at just one specific time. Figure 2 shows the effect of assimilation of hourly groundbased PM measurements and additional assimilation of satellite AOD measurements on the analysed fields of PM2.5 for one specific time (17th of July 2003 at noon).
Fig. 2 PM2.5 concentrations at ground level for 17 July 2003 noon from the nature run (top left), the LOTOS-EUROS model without assimilation (top right), the LOTOS-EUROS model with assimilation of synthetic groundbased PM2.5 measurements (middle left) and the LOTOSEUROS model with assimilation of synthetic groundbased PM2.5 measurements and halfhourly total synthetic AOD measurements from the Imager (middle right)
We have to note that for this figure we used AOD measurement errors of 20% instead of the real simulated errors which can be larger during the day. It can be seen that when halfhourly total AOD observations from the FCI instrument are assimilated next to groundbased PM2.5 measurements, the satellite measurements have an additional positive impact on the analysed fields. The PM 2.5 fields are improved e.g. over South-Eastern Europe.
An Observing System Simulation Experiment (OSSE) for Aerosols
293
5. Summary and Conclusion A working OSSE system for aerosols has been set up that performs the assimilation of both groundbased PM observations as satellite based total or vertically resolved AOD measurements in the CTM LOTOS-EUROS. An initial assessment clearly shows the benefit of assimilating in-situ measurements of PM and to a lesser extent of satellite AOD observations. The work is ongoing and next to the benefit from the MTG Oxygen A-band sounder we will further investigate the effect on PM2.5 forecasts. A challenge in an OSSE for aerosols is the quantitative evaluation of the results. In meteorological OSSE’s the performance of runs are often evaluated by looking at the spatial correlation. Disturbances in a meteorological model will in general cause the runs to diverge from the nature run and, consequently, lead to lower spatial correlation in time. Contrarily, chemistry transport models are stable systems because of the continuous input of emissions and the meteorology as driving forces. The system has a tendency to converge as found in previous studies on ozone forecasts (Elbern et al., 2006; Blond and Vautard, 2004). Because of this convergence the spatial correlation will not necessarily decrease in time. This highlights the need for a common framework, different from meteorological practice, to evaluate forecast experiments for air quality.
References Blond N, Vautard R (2004) Three-dimensional ozone analyses and their use for short-term ozone forecasts, J. Geophys. Res., 109, D17303, doi:10.1029/ 2004JD004515. Chu DA, Kaufman YJ, Zibordi G, Chern JD, Mao J, Li C, Holben BN (2003) Global monitoring of air pollution over land from the Earth Observing SystemTerra Moderate Resolution Imaging Spectroradiometer (MODIS), J. Geophys. Res., 108, 4661, doi:10.1029/2002JD003179. Elbern H, Strunk A, Schmidt H, Talagrand O (2006) Emission rate and chemical state estimation by 4-dimensional variational inversion, Atmos. Chem. Phys. Discuss., 7, 1725–1783. Engel-Cox JA, Holloman CH, Coutant BW, Hoff RM (2004) Qualitative and quantitative evaluation of MODIS satellite sensor data for regional and urban scale air quality, Atmos. Environ., 38, 2495, doi:10.1016 /j.atmosenv.2004.01.039. Goede APH (2005) CAPACITY User Requirements Document, UR 001, WP1000. Hutchison KD (2003) Applications of MODIS satellite data and products for monitoring air quality in the state of Texas, Atmos. Environ., 37, 2403–2412. Koelemeijer RBA, Homan CD, Matthijsen J (2006) Comparison of spatial and temporal variations of aerosol optical thickness and particulate matter in Europe, Atmos. Environ. 40, 5304–5315.
294
R. Timmermans et al.
Lelieveld J (2003) Geostationary satellite observations for monitoring atmospheric composition and chemistry applications, Max Planck Inst Mainz. Final Report EUMETSAT contract EUM/CO/02/1015/SAT (available from www.eumetsat.int). Robles González C, Schaap M, de Leeuw G, Builtjes PJH, van Loon M (2003) Spatial variation of aerosol properties over Europe derived from satellite observations and comparison with model calculations, Atmos. Chem. Phys., 3, 521–533. Schaap M, van Loon M, ten Brink HM, Dentener FJ, Builtjes PJH (2004a) Secondary inorganic aerosol simulations for Europe with special attention to nitrate, Atmos. Phys. Chem., 4, 857–874. Schaap M, Denier Van Der Gon HAC, Dentener FJ, Visschedijk AJH, van Loon M, ten Brink HM, Putaud J-P, Guillaume B, Liousse C, Builtjes PJH (2004b) Anthropogenic black carbon and fine aerosol distribution over Europe, J. Geophys. Res., 109, D18201, doi:10.1029/2003JD004330. Schaap M, Roemer M, Sauter F, Boersen G, Timmermans R, Builtjes PJH (2005) ‘LOTOS-EUROS Documentation’, TNO report B&O 2005/297, TNO, Apeldoorn, The Netherlands. Schaap M, Timmermans RMA, Sauter FJ, Roemer M, Velders GJM, Boersen GAC, Beck JP, Builtjes PJH (2008) The LOTOS-EUROS model: description, validation and latest developments, Int. J. Environ. Pollut., Vol. 32, No. 2, pp. 270–290. Siddans R, Latter BG and Kerridge BJ (2007) Study to Consolidate the UVS Mission Requirements for the Oxygen A-band, Simulation of measurements for an aerosol OSSE, EUMETSAT report contract EUM/CO/05/1411/SAT (available from www.eumetsat.int). Timmermans RMA, Schaap M, Builtjes PJH, Siddans R, Elbern H, Vautard R (2006) The operational use of satellite derived aerosol information to assess fine particulate matter concentrations in Europe, TNO-Report, 2006-A-R0309-B. Wang J, Christopher SA (2003) Intercomparison between satellite-derived aerosol optical thickness and PM2.5 mass: implications for air quality studies, Geophys. Res. Lett., 30(21), 2095, doi:10.1029/2003GL018174.
An Observing System Simulation Experiment (OSSE) for Aerosols
295
Discussion M. Mircea: Do you test the impact of the model resolution on AOD? R. Timmermans: No we have not explicitly tested the impact of the model resolution, but we are planning to perform the experiment on two different resolutions/domains. The results shown at the ITM where for a domain covering central Europe at a resolution of approximately 12 × 12 km. In the near future we will do the same experiment on a smaller domain covering the Paris Bassin area at a resolution of approximately 3 × 3 km. From this we will be able to investigate the influence of different resolutions.
3.2 Comparison of Data Assimilation Methods for Assessing PM10 Exceedances on the European Scale Bruce Denby, Martijn Schaap, Arjo Segers, Peter Builtjes and Jan Horálek
Abstract Two different data assimilation techniques have been applied to assess exceedances of the daily and annual mean limit values for PM10 on the regional scale in Europe. The two methods include a statistical interpolation method (SI), based on residual kriging after linear regression of the model, and ensemble Kalman filtering (EnKF). Both methods are applied using the LOTOS-EUROS model. Observations for the assimilation and validation of the methods have been retrieved from the Airbase database using rural background stations only. The LOTOS-EUROS model is found to underestimate PM10 concentrations by a factor of 2. This large model bias is found to be prohibitive for the effective use of the EnKF methodology and a bias correction was required for the filter to function effectively. The results of the study show that both methods provide significant improvement on the model calculations when compared to an independent set of validation stations. The most effective methodology is found to be the statistical interpolation method. Keywords Data assimilation, air quality, uncertainty, ensemble Kalman filter, kriging
1. Introduction The aim of this study is to compare and assess two different data assimilation methods for calculating the spatial distribution of PM10 exceedances on the European scale, as defined in EU legislation (EC, 1999). Assessment of PM10 on this scale is required for estimating European wide health impacts and risk assessments suitable for policy development at the national and European level. Historically such assessments have been carried out with the use of ground based monitoring data only. This data is spatially inhomogeneous, providing information only at monitoring sites and as such cannot be used directly to calculate exceedances over the entire domain.
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
278
Comparison of Data Assimilation Methods for Assessing PM10 Exceedances
279
Chemical transport models (CTMs) may be used to calculate the spatial and temporal distribution of chemical compounds. For the case of PM10 regional scale CTMs tend to severely underestimate the measured concentrations (van Loon et al., 2004). As such, CTMs by themselves are not useful tools for assessing limit value exceedances of PM10 in Europe, however, they do provide important spatial information for the assessment. By combining such models with observations using data assimilation techniques the quality of the assessment can be significantly improved. There is a range of statistical interpolation techniques that can be used to assimilate or combine data from different sources to create spatial concentration fields. These include traditional optimal interpolation methods, kriging and residual kriging methods, regression and multiple regression techniques, e.g. Blond et al. (2003), Horálek et al. (2005) and Kassteele and Velders (2006). These statistical techniques are most often applied ‘off-line’ in that they take model output and postprocess the results. There are also a number of ‘on-line’ assimilation methodologies that actively interact with the model. These include the variational methods of 2d, 3d and 4d-var as well as ensemble methods such as Ensemble Kalman filters. The aim of all these methods is to combine various data sources to provide the best estimate of the spatial distribution of a particular pollutant. This paper presents the combined results of work conducted during the Air4EU FP6 project (www.air4eu.nl) into the application of assimilation methods on the European scale. Two different data assimilation methods are compared and assessed. The first uses statistical interpolation techniques, specifically residual kriging and linear regression. The second method makes use of the ensemble Kalman filter methodology. For both techniques the LOTOS-EUROS model is applied together with the assimilation of ground based PM10 observations taken from the Airbase database.
2. Methodology To make as fair a comparison as possible the datasets used for both assimilation methodologies are the same, i.e. model calculations with the LOTOS-EUROS model and observational data from Airbase for the year 2003 are used for the study. The region of Europe under focus is central Europe (Figure 1). To compare the different assimilation methodologies the Airbase data is split into an ‘assimilation’ and a ‘validation’ dataset.
Fig. 1 Positions of the assimilation (grey circles) and validation (black squares with station ID) stations used in the study. The size of the circle indicates the temporal coverage of the station ranging from 100% to 50%
280
B. Denby et al.
The result of the assimilations are analysed through the assessment of the daily mean concentrations, the annual mean concentrations and the number of exceedance (NOE) days at the validation stations. This is carried out using the statistical parameters of root mean square error (RMSE), mean absolute error (MAE), coefficient of determination (R2) and linear regression parameters of intercept and slope to indicate bias. Maps of the resulting exceedance fields are also provided for intercomparison.
2.1. Monitoring data The monitoring data used is taken from the Airbase database (Airbase, 2007). Only monitoring sites described as rural background are used for the assimilation. A further selection of spatial and temporal coverage is made, including all monitoring sites with temporal coverage >50% for assimilation stations and >75% for validation stations, all sites within the region bordered by –5° E–25° E and 42.5° N–62.5° N, and all sites below an altitude of 700 m. In total 127 stations are available after this selection. These are split, by random draw, into assimilation (102 stations) and validation (25) stations. This split has been carried out with the condition that validation stations are not within 100 km of each other. On average 91 stations were available each day for the assimilation.
2.2. The LOTOS-EUROS model The LOTOS-EUROS model (Schaap et al., 2008) is used to calculate the PM10 distributions over Europe for 2003. The LOTOS-EUROS model is an operational 3D chemistry transport model aimed to simulate air pollution in the lower troposphere. In the vertical the model has four layers up to 3.5 km following the dynamic mixing layer approach. The horizontal resolution of the model is 0.5° × 0.25° (approximately 35 × 25 km in Europe). The model incorporates primary (combustion) particles (EC, OC), sea salt and secondary inorganic aerosols (SIA: SO4, NO3, NH4). Crustal matter (CM) and secondary organic aerosols (SOA) are not incorporated yet, due to a lack of solid knowledge on emission strengths for CM and formation routes for SOA. This and previous studies show that the model underestimates the PM10 concentrations significantly, mainly due to the nonmodelled fractions and large uncertainties in the modelling of the carbonaceous components. For a detailed discussion on the model performance for PM and its components we refer to Schaap et al. (2004, 2008).
Comparison of Data Assimilation Methods for Assessing PM10 Exceedances
281
2.3. Residual kriging with linear regression Previous to this study a number of statistical interpolation methods have been assessed for calculating PM10 fields from model and observational datasets (Denby, 2007). The results from those studies have shown that residual kriging with linear regression is the most effective methodology. In this method the linear regression of the daily mean model fields is used as a basis for the interpolation and the residuals, observed minus regression model, are spatially interpolated using ordinary kriging (Cressie, 1993) using the nearest 50 stations to the interpolation point. This is similar to methods using kriging with external drift, e.g. Kassteele and Velders (2006), and has been extensively investigated in other studies, e.g. Horálek et al. (2005) and Denby (2007). For the kriging, the variogram parameters of nugget, sill and range are determined by fitting the sill and then optimising the other parameters to obtain the minimum cross-validation RMSE (Denby, 2007).
2.4. Ensemble Kalman filter The basic idea behind the ensemble filter is to express the probability function of the state in an ensemble of possible states or modes. Hence, with an EnKF there is no need to specify the model uncertainties as they are determined by the range of modes. The assimilation itself is performed at run time when the model reaches a point for which new measurements are available. The application of the EnKF is computationally expensive since for each ensemble member the model has to be evaluated. An ensemble (N = 15) was created by applying noise (with a mean of 1 and a standard deviation of 0.25) to seven parameters: the emissions of SOX, NOX, VOC, NH3, primary PM2.5 and PM10 as well as the aerosol dry deposition speed. The uncertainty in the measurements is defined using a fixed relative standard deviation of 12% in combination with an upper limit of 10 ȝg m-3. For a description of the implementation of EnKF with LOTOS-EUROS see van Loon et al. (2000). The application of the assimilation system to PM10 is hampered by a bias between modeled and measured PM10 concentrations. The current status of PM10 modelling is that the models, such as LOTOS-EUROS, underestimate the observed concentrations severely due to missing knowledge on processes and emissions to incorporate all components in the models. Hence, for PM10 a bias correction has been applied of a factor of 2. The application of this bias correction is a highly pragmatic approach and the results presented in this paper must be seen as preliminary in regard to the application of the EnKF for PM10.
3. Results The results of the residual kriging and EnKF assimilations are presented by direct comparison with the 25 validation stations in Figures 2–4. In these figures the annual mean, the daily mean correlation coefficient and the NOE days are given.
282 40
-3
Annual mean concentration (Pg.m )
B. Denby et al.
35 30
Model Residual kriging EnKF assimilation Observations
25 20 15 10 5
AT
01 0 AT 1 A 01 CH 62 A 00 2 CZ 4A 00 4 CZ 1A 00 DE 77A 00 DE 02R 00 0 DE 9R 00 1 DE 2R 00 3 DE 5R 00 3 DE 9R 07 3 DE 7A 11 2 DE 2A 12 5 DE 3A 12 94 DE A 13 0 DE 3A 13 8 DE 5A 13 8 DE 8A 13 8 FR 9A 00 FR 80A 01 1 FR 4A 05 7 FR 2A 07 5 FR 0A 10 G 88A B0 03 NO 6R 00 01 R
0
Fig. 2 Annual mean concentrations for the model, residual kriging and EnKF calculations. Shown are the 25 validation stations used in the study 1 0.9
2
Correlation (R )
0.8
Model Residual kriging EnKF assimilation
0.7 0.6 0.5 0.4 0.3 0.2 0.1
AT
01 0 AT 1 A 01 CH 62 A 00 2 CZ 4A 00 CZ 41A 00 DE 77A 00 0 DE 2R 00 09 DE R 00 1 DE 2R 00 3 DE 5R 00 3 DE 9R 07 3 DE 7A 11 22 DE A 12 DE 53A 12 9 DE 4A 13 0 DE 3A 13 8 DE 5A 13 8 DE 8A 13 8 FR 9A 00 8 FR 0A 01 1 FR 4A 05 FR 72A 07 5 FR 0A 10 G 88A B0 03 NO 6R 00 01 R
0
Fig. 3 Coefficient of determination of the daily mean concentrations for the model, residual kriging and EnKF calculations. Shown are the 25 validation stations used in the study
Number of exceedance days
80 70 60
Model Residual kriging EnKF assimilation Observations
50 40 30 20 10 0
AT
A 2 A 4A A 7A 2R 9R 2R 5R 9R 7A 2A 3A 4A 3A 5A 8A 9A 0A 4A 2A 0A 8A R R 41 7 01 6 5 1 7 8 8 0 00 01 03 03 73 12 25 29 30 38 38 38 2 3 6 001 01 T01 00 Z0 0 Z0 0 00 0 0 E0 E0 E0 E0 E1 E1 E1 E1 E1 E1 E1 R0 0 R0 1 R0 5 R0 7 R1 0 00 E H E B O A C C F F F F F D C D D D D D D D D D D D D N G
Fig. 4 Number of exceedance days for the model, residual kriging and EnKF calculations. Shown are the 25 validation stations used in the study. The dashed line indicates the EU exceedance level of 36 days
Comparison of Data Assimilation Methods for Assessing PM10 Exceedances
283
Table 1 Summary of the statistical performance of the assimilation methods applied in the study, when tested against the validation dataset, for the daily mean PM10 concentrations (all days and all validation stations) and the number of exceedance days (all validation stations). Validation statistic RMSE (µg m-3) MAE (µg m-3) Correlation (R2) Intercept (µg m-3) Slope RMSE (day) MAE (day)
Model Daily mean concentration 16.7 11.8 0.21 12.0 0.83 Number of exceedance days 21.2 16.6
Residual kriging
EnKF
9.2 6.4 0.65 2.0 0.85
13.5 9.3 0.41 9.3 0.58
11.1 8.1
14.9 10.4
In addition to these figures Table 1 provides the statistical results of the model and assimilation methods for all days and all validation stations. Both the residual kriging and the EnKF methods reduce the bias of the validation stations (Figure 2) significantly. Absolute bias is reduced in all but 4 of the 25 validation stations using both the residual kriging and the EnKF methods. The MAE for all days and all validations stations, given in Table 1, indicates that residual kriging provides the lowest bias estimate. The correlation (R2) is dramatically improved (Figure 3) in comparison to the model results, with use of residual kriging. Only two stations show a reduced correlation and R2 is above 0.7 for more than half of the validation stations. EnKF also shows improved correlation at all but one of the validation stations, though this improvement is not as significant as it is for the residual kriging (Table 1). It was found that the temporal correlation in the noise factors used in the EnKF has a large influence on the results and requires optimisation. The NOE days are significantly improved (Figure 4), in comparison to the model results, with use of residual kriging where only four stations show a worsened estimate for NOE. EnKF gives an improved estimate of the NOE days for all but seven of the validation stations. Both RMSE and MAE (Table 1) indicate that residual kriging gives the best performance of the two assimilation methods for NOE. Maps showing the annual mean concentrations and the number of daily mean exceedances are shown in Figures 5 and 6 for the residual kriging method and the EnKF assimilation. The maps resulting from the two assimilation methods have a very similar structure showing regions where exceedances occur, namely the Po valley region, the Netherlands, Eastern Europe, the North Sea and the Paris and London agglomerates. The absolute values, however, do differ between the methods. These maps can be compared to the initial model calculations (not shown) that indicate no exceedances of PM10 on the regional scale in Europe.
284
B. Denby et al.
Fig. 5 Map showing the annual mean concentration (left) and number of exceedance days (right) as calculated using residual kriging and linear regression. The annual mean EU limit value is 40 µg m-3 and the maximum number of exceedance days (daily mean > 50 µg m-3) is 36
Fig. 6 Map showing the annual mean concentration (left) and number of exceedance days (right) as calculated using EnKF assimilation with the LOTOS-EUROS model. The model domain for the EnKF calculations was slightly smaller to reduce calculation time
4. Conclusions The results presented here show an assessment of two different methodologies for combining ground based observations with CTM calculations to provide regional scale assessment of PM10 in Europe. Residual kriging in combination with linear regression is shown to be a very effective methodology for post-processing model fields. Based on the statistical analysis the uncertainty in the CTM calculations can be reduced by a factor of 2 using this method. The results presented for the EnKF methodology are in some ways preliminary. Optimisation of the filter parameters is still required and further testing and assessment of the methodology will likely lead to improved results. Further work is still required on the following points: x Improvement of CTMs in general, including LOTOS-EUROS, and their description of PM processes and emissions.
Comparison of Data Assimilation Methods for Assessing PM10 Exceedances
x x x
285
Further testing and optimalisation of the EnKF methodology and its application to PM10 calculations. Further assessment of the method for determining the variance used in the statistical interpolation methodologies. Further work to improve and communicate uncertainty in the methodologies in the form of both numbers and maps.
Acknowledgments This research was funded by the EU FP6 project Air4EU, contract no. 503596.
References Blond N, Bel L, Vautard R (2003) Three-dimensional ozone data analysis with an air quality model over the Paris area, J. Geophys. Res., 108(D23), 4744. Cressie N (1993). Statistics for Spatial Data (Revised Edition). Wiley, New York Denby B (2007) Basic data assimilation: application to the European scale. Air4EU case study report no. 13. http://www.air4eu.nl/reports_products.html EC (1999) Directive 1999/30/EC of the European Parliament and of the Council relating to limit values for sulphur dioxide, nitrogen dioxide and oxides of nitrogen, particulate matter and lead in ambient air, European Commission, Brussels. Horálek J, KurFürst P, Denby B, De Smet P, De Leeuw F, Brabec M, Fiala J (2005) Interpolation and assimilation methods for European scale air quality assessment and mapping, Part II: development and testing new methodologies, ETC/ACC Technical Paper 2005/8. http://air-climate.eionet.europa.eu/reports/ ETCACC_TechnPaper_2005_8_Spatial_AQ_Dev_Test_Part_II Kassteele van de J, Velders GJM (2006) Uncertainty assessment of local NO2 concentrations derived from error-in-variable external drift kriging and its relationship to the 2010 air quality standard. Atmos. Environ., 40(14), 2583– 2595 Schaap M, Denier Van Der Gon HAC, Dentener FJ, Visschedijk AJH, Van Loon M, Ten Brink HM, Putaud J-P, Guillaume B, Liousse C, Builtjes PJH (2004) Anthropogenic black carbon and fine aerosol distribution over Europe, J. Geophys. Res., 109, D18201, doi:10.1029/2003JD004330 Schaap M, Timmermans RMA, Sauter FJ, Roemer M, Velders GJM, Boersen GAC, Beck JP, Builtjes PJH (2008) The LOTOS-EUROS model: description, validation and latest developments. Int. J. Environ. Pollut., 32, 270–290. doi:10.1504/ IJEP.2008.017106 Van Loon M, Builtjes PJH, Segers A (2000) Data assimilation of ozone in the atmospheric transport chemistry model LOTOS. Environ. Mod. Software 15, 603–609. Van Loon M, Roemer M, Builtjes P (2004) Model intercomparison in the framework of the review of the Unified EMEP model. TNO-Rep. R 2004/282
286
B. Denby et al.
Discussion S.-E. Gryning: How does the temporal representativeness influence the results? B. Denby: The daily mean PM10 concentrations were used for both the EnKF and statistical interpolations. In the case of EnKF it is absolutely necessary to use daily, or hourly, concentrations for the assimilation calculations. For the statistical interpolations annual statistics may be used instead of daily, for both observations and model, with generally only a small increase in error for the resulting fields. This is true for indicators such as annual mean but there can be significant differences between the use of daily and annual statistics when interpolating the number of exceedances.
3.6 Comparison of Methods to Generate Meteorological Inputs for Modeling Dispersion in Coastal Urban Areas Akula Venkatram, Wenjun Qian, Tao Zhan and Marko Princevac
Abstract The paper compares the performance of a semi-empirical method with that of a comprehensive model in generating meteorological inputs for a dispersion model applicable to sources located in coastal urban areas. The model performed well when meteorological inputs are generated from the semi-empirical method. Outputs from the comprehensive model yielded adequate concentration estimates for surface releases, but it was less successful for elevated releases because it was unable to simulate the growth of the Thermal Internal Boundary Layer (TIBL) with distance from the shoreline. Keywords Coastal urban areas, comprehensive model, dispersion, meteorological inputs, semi-empirical method, thermal internal boundary layer (TIBL)
1. Introduction As part of a program to develop dispersion models that provide reliable assessment of exposures to air toxics released from shoreline sources, we conducted two field experiments in Wilmington, a suburb of Los Angeles in 2004 and 2005. The first field study, Wilmington 2004, was conducted during eight days in the period 26 August–10 September 2004. The tracer, SF6, was released at a height of 3 m from the power plant site on the shoreline, and the concentrations of the tracer were sampled on five arcs at 100, 400, 1,000, 3,000, and 5,000 m from the source during 6 hours of the day starting at 7 a.m. Concurrent meteorological measurements were made using sonic anemometers and mini sodars at the release site as well as at Los Angeles County Sanitation District’s Joint Water Pollution Control Plant (JWPCP), which is located approximately 4,000 m downwind of the source (LADWP). A microwave temperature sounder was used to determine the vertical temperature profile at JWPCP. A second field study, Wilmington 2005, which focused on elevated tracer releases, was conducted between June 24 and 28, 2005. Two types of releases were conducted: non-buoyant releases 3 m below the top of the 67 m stack, and releases into the buoyant stack gases. The data from these field studies were used to develop and evaluate a dispersion model that incorporated the entrainment C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
313
A. Venkatram et al.
314
of the elevated tracer plume by the growing thermal internal boundary layer (TIBL). The model performed well when meteorological inputs were constructed from onsite meteorological measurements. Because such measurements are not available at most sites, we examined the feasibility of using a semi-empirical method and a comprehensive meteorological model to generate meteorological inputs for this dispersion model. The performance of the dispersion model in explaining ground-level concentrations was used to compare these two methods of generating meteorological inputs.
2. Semi-empirical Dispersion Model 2.1. Model development The dispersion model used here to interpret the field data is similar to that developed by Van Dop et al. (1979), and improved by Misra (1980). We have modified the model to incorporate the measurements of turbulence made in the Wilmington experiment. Misra’s (1980) model is based on the following physical picture. As the elevated plume is transported above the internal boundary layer, it grows both horizontally and vertically due to atmospheric turbulence and turbulence generated by plume buoyancy. Because atmospheric turbulence is small above the TIBL, plume buoyancy generates most of the plume growth. This growing plume is entrained by the TIBL, whose height increases with distance from the shoreline. The entrained plume material is rapidly mixed to ground-level by the vigorous convective motions within the internal boundary layer. Thus, for releases above the TIBL, ground-level concentrations are sensitive to the rate of growth of the TIBL with distance from the shoreline. The thermal internal boundary height, zi, is computed using the expression (Venkatram, 1977): 1/2
zi
§ Q (x xo ) · a¨ o ¸ UȖ © ¹
(1)
where Qo is the average kinematic heat flux over land, x is the distance from the shoreline, U is the boundary layer averaged wind speed, and Ȗ is the potential temperature gradient above the TIBL. The parameter xo is the distance of the effective shoreline from the release. Taking xo = 100 m yielded the best agreement between modeled concentrations and observations corresponding to elevated releases. The parameter, a, is empirically determined to be 2.
Comparison of Methods to Generate Meteorological Inputs for Modeling Dispersion
315
2.2. Evaluation of dispersion model using onsite meteorology Model performance is quantified in terms of the geometric mean, mg, of the ratio of the estimated, Cp, to observed concentrations, Co, and the spread of observations about a model estimate is quantified using the geometric standard deviation, sg, of the ratio about the mean. Then if the observed concentrations are lognormally distributed about the model estimate the 95% confidence of the ratio of the observed to the estimated concentration is given by the interval mgsg1.96 to mgsg-1.96. r2 is the correlation coefficient between the logarithms of Co and Cp. Figure 1 compares observed arc maximum concentrations with model estimates obtained using onsite meteorological inputs. The left panel corresponds to elevated non-buoyant releases. Most of the model estimates are within a factor of two of the observed values, and the model explains 70% of the variance of the observations. The right panel indicates that model performance for elevated buoyant releases is also adequate, with all the observations lying within a factor of two of the model estimates; the model explains 80% of the observed variance.
Fig. 1 Comparison of measured arc maximum concentrations with model results for non-buoyant and buoyant releases for Wilmington 2005 study
3. Methods to Generate Meteorological Inputs The application of the shoreline dispersion model described above requires onsite meteorological measurements to construct model inputs. Because such measurements are unlikely to be available at most sites, there is a need for methods to estimate meteorological inputs from routinely available meteorological observations. We evaluated two types of methods to construct the meteorological inputs.
3.1. Semi-empirical model The semi-empirical method tested here is based on the Meteorological Preprocessor (AERMET) in AERMOD (Cimorelli et al., 2005). AERMET is based on a onedimensional boundary layer model that cannot be applied to shorelines where surface properties vary sharply across the water-land interface. In our approach, we
A. Venkatram et al.
316
first compute the TIBL height using Eq. (1). We then estimate the convective turbulence generated within the TIBL, which is then added to the turbulence levels generated by AERMET using rural information. Assuming that the vertical velocity fluctuations generated in the TIBL are dominated by free convection, the standard deviation, ıw, of these fluctuations can be written as: 1/ 3
Vw
§ g · D ¨ Qo zi ¸ © To ¹
(2)
where Į = 0.6 (Stull, 1988). Then, an expression for the convectively generated component of turbulence in the TIBL can be obtained by combining Eqs. (1) and (2) 1/3
ı wc
ĮQo1/2
§ g · ¨ ¸ © To ¹
1/6
§ 4x · ¨ ¸ © UȖ ¹
(3)
Then V wc is added to the shear generated components of turbulence as follows:
ıw where ı ws
ı
1.3u* and ı ws
3 wc
3 ı ws
1/3
and ı v
ı
3 wc
ı vs3
1/3
(4)
2.5u* . The surface heat flux, Qo, required to compute
the TIBL height is estimated through a surface energy balance over land.
3.1.1. Model Evaluation Figure 2 compares model estimates of friction velocity and surface heat flux with corresponding observations. The estimated friction velocities agree well with the corresponding observations, but the estimated heat fluxes are generally a factor of two higher than the observations.
Fig. 2 Comparison of estimated surface friction velocities and heat fluxes with observations at LADWP site for Wilmington 2004 study. The lines above and below the one-to-one correspond to factor of two interval
Comparison of Methods to Generate Meteorological Inputs for Modeling Dispersion
317
Fig. 3 Comparison of estimated surface and 50 m height turbulence levels with observations made at the LADWP site for Wilmington 2004 study. ‘BL’ denotes for ‘boundary layer’, referring to height of 50 m here. The lines above and below the one-to-one correspond to a factor of 2 interval
Figure 3 compares estimated turbulence levels at the surface and at height of 50 m with observations made with the sonic anemometer and the sodar. The estimated values of turbulent velocities are within a factor of two of the observations. We found that for the ground-level releases of the Wilmington 2004 study, model performance using estimated meteorological inputs was comparable to that based on measured inputs. Figure 4 indicates that for the elevated buoyant and nonbuoyant releases of the Wilmington 2005 study, model performance corresponding to estimates of meteorological inputs is comparable to that obtained using measured inputs, except for the 1,000 m arc at the second hour of buoyant release when the model severely underestimates the concentrations.
Fig. 4 Comparison of measured arc maximum concentrations with model results for elevated buoyant releases (left panel) and non-buoyant releases (right panel) for Wilmington 2005 study
3.2. TAPM The comprehensive meteorological model used here is version 3 of The Air Pollution Model (TAPM), a prognostic model, developed by CSIRO (Commonwealth Scientific and Industrial Research Organization) Atmospheric Research (Hurley, 2005) – Three cases of wind assimilation were used in the simulations: (1) without assimilation, (2) wind information from only the LADWP site, and (3) wind measurements from both the LADWP and JWPCP sites. Model estimates of
318
A. Venkatram et al.
meteorology as well as concentration are compared with corresponding observations for these three cases.
3.2.1. Evaluation of Meteorological Outputs from TAPM Figure 5 compares the horizontal wind speeds, horizontal and vertical turbulent velocities estimated by TAPM at the JWPCP site with the corresponding observations during the Wilmington 2005 field study. The estimates for wind speed are well correlated with the observed values, but they are overestimated by factors of 3–4. Assimilation of local wind information decreases the degree of overestimation of the wind speed. The standard deviation of the vertical velocity fluctuation, ıw, is overestimated by almost a factor of two by TAPM, presumably because it overestimates the mean wind speeds. Assimilation of winds appears to increase the overestimation slightly. TAPM produces acceptable estimates of ıv for cases with and without wind assimilation; the model estimates show little bias and are within a factor of two of the observations.
Fig. 5 Comparisons of mean wind speeds, horizontal turbulent velocities, vertical turbulent velocities at JWPCP site estimated by TAPM with observed values from Wilmington 2005 field experiment. (a), (d) and (g) are without wind assimilation; (b), (e) and (h) are with wind assimilated at LADWP site only; (c), (f) and (i) are with wind assimilated at both LADWP and JWPCP sites
Comparison of Methods to Generate Meteorological Inputs for Modeling Dispersion
319
3.2.2. Evaluation of Concentration Estimates Figure 6 shows the observed arc maximum concentrations on the three arcs at 1,000, 3,000 and 5,000 m are plotted against estimates from the shoreline dispersion model using meteorological inputs estimated with TAPM. For elevated nonbuoyant releases, the dispersion model yields acceptable estimates of most of the arc maximum concentrations when meteorological inputs are generated with TAPM. However, for the elevated buoyant releases, model estimated arc maximum concentrations are uncorrelated with observations when meteorological inputs are generated from TAPM with wind assimilated only at the LADWP site. Model performance improves to some extent when the wind is assimilated from both sites, although the arc maximum concentrations are still overestimated at the 1,000 m and 3,000 m arcs. TAPM’s poor performance for elevated releases is related to its inability to resolve the growth of the boundary layer height with distance from the shoreline.
Fig. 6 Comparison of observed arc maximum concentrations with estimates from Wilmington model with meteorological inputs predicted from TAPM. Elevated non-buoyant releases include the scenarios on June 26 and 28 2005, and elevated buoyant releases correspond to the cases on June 27 2005. (a) and (b), wind assimilated at LADWP in TAPM; (c) and (d), wind assimilated at LADWP + JWPCP in TAPM
4. Conclusions This paper compares the performances of two methods to generate meteorological inputs for a shoreline dispersion model that performs well with onsite measurements. The first semi-empirical method that combines a one-dimensional surface energy balance with a two-dimensional TIBL model yielded concentration estimates that were comparable to those from onsite measurements. The second method used a comprehensive meteorological model, TAPM, to generate meteorological variables in the shoreline TIBL. TAPM outputs resulted in adequate concentration estimates for near surface releases, although model performance was not as good as that corresponding to the semi-empirical method for generating shoreline meteorology.
320
A. Venkatram et al.
TAPM was less successful for elevated buoyant releases because it was unable to predict the growth of the coastal internal boundary layer as a function of distance from the coastline. This feature is critical to estimating concentrations associated with elevated sources located on the coast. Acknowledgments The research was funded by California Air Resources Board (CARB), and California Energy Commission (CEC), and the National Science Foundation under grant ATMOS 0430776.
References Cimorelli JA, Perry GS, Venkatram A, Weil CJ, Paine JR, Wilson BR, Lee FR, Peters DW, Brode WR (2005) AERMOD: a dispersion model for industrial source applications. Part I: general model formulation and boundary layer characterization. J. Appl. Met., 44, 682–693. Hurley P (2005) The Air Pollution Model (TAPM) Version 3. Part 1: Technical Description. Australia. CSIRO Atmospheric Research Technical Paper No. 71. Misra PK (1980) Dispersion from tall stacks into a shore line. Atmos. Environ., 14,397–400. Stull RB (1988) An Introduction to Boundary-Layer Meteorology. Kluwer, Dordrecht. van Dop H, Steenkist R, Nieuwstadt FTM (1979) Revised estimates for continuous shoreline fumigation, J. Appl. Met. 18, 133–137. Venkatram A (1977) A model of internal boundary-layer development, Bound.Layer Meteorol. 11, 419–437.
Discussion S.-E. Gryning: The mechanical turbulence is known to be important for IBL growth in urban areas, why did you use a model for IBL that do not consider the effect of u*? A. Venkatram: We assumed that TIBL growth was dominated by surface heat flux. Including the effects of u* would not have improved the estimates of TIBLheight because we were overestimating by almost a factor of two with uj st the surface heat flux.
Comparison of Methods to Generate Meteorological Inputs for Modeling Dispersion
321
D. Stein: TAPM did a poor job of capturing TFBL height in the coastal zone, and as you pointed out, the basin is subject to very high subsidence velocities in TAPM anywhere near those observed? A. Venkatram: TAPM does not appear to have provided realistic estimates of subsidence because TAPM estimates of TIBL height did not compare well with observations.
3.7 Developing a Method for Resolving NOx Emission Inventory Biases Using Discrete Kalman Filter Inversion, Direct Sensitivities, and Satellite-Based NO2 Columns Sergey L. Napelenok, Robert W. Pinder, Alice B. Gilliland and Randall V. Marin
Abstract An inverse method was developed to integrate satellite observations of atmospheric pollutant column concentrations with specie concentrations and direct sensitivities predicted by a regional air quality model in order to discern biases in the emissions of the pollutant precursors. Using this method, the emission fields were analyzed using a “top-down” approach with an inversion performed by Discrete Kalman Filter (DKF) and direct sensitivities calculated using the Decoupled Direct Method in 3D (DDM-3D) embedded in the Community Multiscale Air Quality (CMAQ) model. The system was tested through an experiment focusing on NO2 concentrations and emissions of NOx in the southeastern United States. The method reproduced the expected NOx emission fields from initially perturbed starting values. Responses to different parameters in the system, including assumptions for uncertainties in the emission fields and satellite observations, were also tested. The method is readily extendable to other pollutants. Keywords DDM-3D, direct decoupled method, emissions, inverse modeling, Kalman filter, satellite, sensitivity, NO2, NOx, 1. Introduction Current regional air quality models rely on well-developed emission inventories with high spatial and temporal resolution. While much work has been done in the development of such inventories, uncertainties still exist. At the same time, retrieval techniques for satellite data have improved and several datasets are available for observations of NO2, CO, and some hydrocarbons recorded by several satellites in orbit. A method was developed for using satellite NO2 column observations to check for biases in current emission inventories of NOx with Discrete Kalman filter (DKF) inversion and sensitivities calculated by the Decoupled Direct Method in three dimensions (DDM-3D), which had been previously integrated (Cohan et al., C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
322
Developing a Method for Resolving NOx Emission Inventory Biases Using Discrete Kalman 323
2005; Napelenok et al., 2006) into the Community Multiscale Air Quality (CMAQ) model (Byun and Schere, 2006). The method was tested using a pseudodata scenario representing hypothetical satellite observations. A base-case CMAQ simulation acted as the true representation of the relationship between NOx emissions and NO2 column concentrations in the domain. Ground-level NOx emissions were then adjusted in pre-defined geographic regions within the modeling domain to mimic possible biases. Finally, the inverse procedure was applied to attempt to arrive back at the base-case emissions taking into account uncertainties in transport and chemistry. The method has proved to converge robustly at the correct solution in only a few iterations for various spatially distributed emission biases. Integration of satellite observations of NO2 with regional air quality modeling efforts can potentially reduce uncertainty in emission inventories. Retrieval algorithms for NO2 column densities have been developed for several satellites, including GOME (Richter and Burrows, 2002), SCIAMACHY (Sioris et al., 2004), and more recently, OMI (Bucsela et al., 2006). Inverse modeling of NOx emissions has been applied previously, but typically on a global scale (Martin et al., 2003; Muller and Stavrakou, 2005) with some efforts on a continental scale (Quelo et al., 2005; Konovalov et al., 2006). In finer scale NO2 inverse modeling, the difficulties arise from the importance of resolving the nonlinearities in chemistry and transport, which are overcome in this exercise with the aid of direct sensitivities.
2. Method 2.1. Discrete Kalman filter Inverse modeling of the NOx emissions field was performed using Discrete Kalman Filter. DKF is an optimization technique used to estimate discrete time series and states that are governed by sets of linear differential equations. It has seen frequent use in inverse modeling of emissions on both the global scale and regional scales for various gaseous and particulate species (Hartley and Prinn, 1993; Chang et al., 1996; Haas-Laursen et al., 1996; Gilliland et al., 2003). Since chemical transport models also parameterize nonlinear processes, the linearity assumption is overcome by applying DKF iteratively. This method is also attractive for inverse modeling, because it allows for the use of uncertainty information in both the emissions fields and the observed pollutant values. A brief overview of DKF is presented here, while more detailed explanation is available elsewhere (Gilliland and Abbitt, 2001). DKF evolves the emission vector,
E t ,k 1
E t , according to the following:
obs
mod
E t ,k G t ,k F t F t
(1)
At iteration k+1 and time t, the emissions vector is altered based on the gain obs
matrix, G t , k , and the difference between the vectors of observations, F t
, and
S. Napelenok et al.
324 mod
modeled values, F t
. The gain matrix is defined in terms of the matrix of partial
derivatives of the change in concentration with respect to emissions, Pt, the matrix of the covariance of the error in the emissions field, Ct,k, and the noise matrix, Nt, such that:
G t ,k
Ct ,k PtT Pt Ct ,k PtT N t
1
(2)
The covariance of error matrix also evolves with subsequent iterations according to: Ct ,k 1 Ct ,k G t ,k Pt Ct ,k (3) The covariance functionally determines the degree to which the emissions vector is allowed to deviate from its initial values. As iterations are progressed, the covariance is reduced according to Eq. (3) and subsequent differences between E t ,k 1 and E t ,k are smaller in a mathematically stable system. In this application, the initial covariance of the error in the integrated emissions estimates, Ct,k = 0, was based on an estimate of the normalized uncertainty in the emissions, UE, according to the following:
U
C jj C jk , j z k
Ej
2
E, j
(4a)
U U E , k E j Ek § ¨¨ 0.1 E , j 2 2 ©
· ¸¸ ¹
2
(4b)
Similarly, the noise matrix was based on the estimated normalized uncertainties in the observations, Uobs, according to:
N jj
>
Max 0.5, U obs , j F obs j ,t N jk , j zk
@ 2
0.0
(5a) (5b)
Theoretically, the noise matrix, Nt, can account for both errors in observations, as it does here, and also errors in the modeling system. The minimum value of 0.5 (1015 molecules/cm2)2 was imposed to prevent mathematical instability.
2.2. Decoupled direct method in 3D The relationship between precursor emissions and resulting pollutant concentrations was represented using sensitivities calculated using the Decoupled Direct Method in 3D. DDM-3D is an efficient and convenient way to calculate responses in the outputs of an air quality model to perturbations in various combinations of input parameters (Dunker, 1981; Yang et al., 1997). DDM-3D propagates sensitivities using some of the same algorithms that are in place to solve the atomspheric diffusion equation:
wCi wt
uCi KCi Ri Ei ,
(6)
Developing a Method for Resolving NOx Emission Inventory Biases Using Discrete Kalman 325
where Ci is the concentration of species i, u is the fluid velocity, K is the diffusivity tensor, R is rate of chemical generation, and E is the emissions field. An analogous equation is developed to calculate sensitivities:
wS ij wt
uS ij KS ij J i S ij Ei' ,
(7)
where Ji is the ith row vector in the Jacobian matrix J, which represents the chemical interactions between species ( J ij wRi / wC j ), and Si is defined as the change of a pollutant i in space, x , and time, t, in respect to a perturbation in some model parameter, (emission rate, initial condition, etc.):
S ij x, t
wCi x, t . wp j
(8)
Implementation of DDM-3D for the CMAQ model has been evaluated previously for both gaseous and particulate species and has been shown be suitable for producing NO2 sensitivities to NOx emissions as compared to discrete difference sensitivity methods (Cohan et al., 2005; Napelenok et al., 2006).
3. Pseudodata Analysis To evaluate the inverse modeling system, a pseudodata scenario was developed in a sample domain centered on the southeastern United States. Emissions source regions were defined based on similar spatial emissions patterns of ground-level NOx during the summer months of 2004 and included the urban areas of Memphis, TN; Nashville, TN; Birmingham, AL; Atlanta, GA; and Macon, GA, as well as the rural areas approximately covered by the states of Tennessee, Mississippi, Alabama, and Georgia (Figure 1). A 144 km wide margin was left around the source regions in order to completely resolve sensitivity fields originating from the defined regions.
Fig. 1 Source region definitions and average groundlevel hourly NOx emissions (moles/s) on August 1, 2004
CMAQ with DDM-3D simulated base concentration fields of NO2 and sensitivities to NOx emissions from each source region. Sensitivities to the emissions from the surrounding “border” region and to the boundary conditions were also
S. Napelenok et al.
326
calculated and were found to have negligible impact on NO2 column densities in the inner region (Figure 2).
Fig. 2 Fractional contribution of total NO2 column sensitivity to NOx emissions only from (a) 144 km “border” region and (b) boundary conditions. The total NO2 column sensitivity is the sum of sensitivities to each source in addition to (a) and (b)
The emission rates in each source region were then arbitrarily adjusted by factors ranging between 0.6 and 1.7 and the simulation repeated with the assumption that the emissions were homogeneous within the region. NO2 concentrations and sensitivities from each simulation were aggregated to column values to more closely mimic the type of available satellite data. The perturbed emissions vector and the corresponding gridded NO2 column values became the a priori estimate for the mod
inverse method ( E t and F t
), while the base-case NO2 columns were used as obs
the representation of the “truth” in the inverse ( F t
). DKF was then applied
iteratively, recalculating concentration and sensitivity fields for each k. During this exercise, the uncertainties in the emissions were set to be relatively high (UE = 2.0) to allow a large range of deviation from the a priori emissions vector in the subsequent estimations. The uncertainty in observations was set low (Uobs = 0.1) to allow the modeled values to closely approach the observations. This combination of uncertainty parameters allows for the best test of the robustness of the system at arriving at the correct solution. The application of this pseudodata scenario revealed that the proposed inverse method was able to reproduce the original base-case emissions vector within only a few iterations (Figure 3). The corresponding NO2 fields were also nearly completely corrected (within 1%) after just four iterations (Fig. ). 2.0
Fig. 3 Aggregated regional emissions after each iteration in the pseudodata scenario normalized by the corresponding base-case values
Emissions Ratio to Base Value (Ek/Ebase )
1.8 1.6 1.4 1.2 1.0 ATLANTA MACON BIRMINGHAM MEMPHIS NASHVILLE GA AL TN MS
0.8 0.6 0.4 0.2 0.0 0
1
2 iteration (k )
3
4
Developing a Method for Resolving NOx Emission Inventory Biases Using Discrete Kalman 327 Fig. 4 Comparison of expected and modeled NO2 column values in each grid cell for (a) initial perturbed simulation and (b) the result of the inverse after four iterations (all values in 1015 molecules cm-2)
The response of the system to uncertainty assumptions was also evaluated by analyzing the predicted regional emissions adjustment after the first DKF iteration (k = 1). In this pseudodata scenario, the acceptable solution was found immediately after the first iteration (Figure 1); thus it was not necessary to carry the solution further for this analysis. As expected, large uncertainties in the observations that lead to larger values in the noise matrix (Nt) do not allow for large adjustments to the emissions fields, while large uncertainties in the a priori emissions estimates allow for larger adjustments through increasing the values in the initial covariance of error matrix Ct,k = 0 (Figure 5). Actual uncertainties in the NO2 column density measurements from the SCIAMACHY and GOME satellite have been shown to be approximately 0.5 × 1015 molecules cm-2 +30% from various assumptions in the retrieval algorithms (Martin et al., 2002; Boersma et al., 2004) suggesting values at the lower end of the tested range. It is more difficult to arrive at estimates of emissions uncertainties in specific geographic regions, but it was shown that if there are no mathematical instabilities, larger values should be selected to arrive at a solution in fewer iterations.
Fig. 5 Regional emissions after the first DKF iteration normalized by their corresponding basecase values as a function of uncertainties in observations and uncertainties in emissions. White areas show immediate near perfect prediction. Initial perturbations are noted next to region names
4. Discussion The proposed method was successful at reproducing correct regional emissions values in the pseudodata exercise. The translation to actual satellite measurements is likely to be more complex due to the limited coverage of the observations. When retrieval is not disrupted by heavy cloud cover that the instruments are unable to penetrate, the geographical extent of what is observed is relatively small. In
328
S. Napelenok et al.
the pseudodata test, all grid cells in the domain were allowed to represent an “observation point,” thus avoiding difficulties that can arise with real data resulting in the system being mathematically ill-posed and therefore unconstrained. Other regional scale inverse modeling attempts have used fairly long averaging periods to allow for enough satellite observations to populate the domain (Konovalov et al., 2006). The more recently available OMI satellite has significantly better overpass frequency than older instruments and should help alleviate this problem. Furthermore, the presented method assumes that the discrepancies in the modeled and observed NO2 concentrations are due solely to estimates of emissions at the ground level. Uncertainties in the chemical processes, emissions aloft from lightning sources and airplanes, and meteorological predictions also contribute to differences between modeled concentrations of NO2 and satellite observations. These uncertainties should be quantified and included in the noise matrix. The presented pseudodata analysis tests the reliability of the method before adding these complexities. Overall, the method is computationally efficient due to the ability to calculate sensitivities directly using DDM-3D and the fact that the matrix operations required by DKF are computationally insignificant. It promises to be directly applicable to NOx emissions inventory analysis and extendable to other species. Acknowledgments The authors would like to thank for contributions and advice from Rynda Hudman and Robin Dennis and insightful comments from all reviewers. Work at Dalhousie University was supported by the Natural Sciences and Engineering Research Council of Canada. A portion of the research presented here was performed under the Memorandum of Understanding between the U.S. Environmental Protection Agency (EPA) and the U.S. Department of Commerce’s National Oceanic and Atmospheric Administration (NOAA) and under agreement number DW13921548. This work constitutes a contribution to the NOAA Air Quality Program. Although it has been reviewed by EPA and NOAA and approved for publication, it does not necessarily reflect their policies or views.
References Boersma KF, Eskes HJ, Brinksma EJ (2004) “Error analysis for tropospheric NO2 retrieval from space.” Journal of Geophysical Research 109: D04311. Bucsela EJ, Celarier EA, Wenig MO, Gleason JF, Veefkind JP, Boersma KF, Brinksma EJ (2006) “Algorithm for NO2 vertical column retrieval from the ozone monitoring instrument.” IEEE Transactions on Geoscience and Remote Sensing 44(5): 1245–1258. Byun DW, Schere KL (2006) “Review of the governing equations, computational algorithms, and other components of the Models-3 Community Multiscale Air Quality (CMAQ) modeling system.” Applied Mechanics Reviews 59: 51–77. Chang ME, Hartley DE, Cardelino C, Chang W-L (1996) “Inverse modeling of biogenic isoprene emissions.” Geophysical Research Letters 23: 3007–3010.
Developing a Method for Resolving NOx Emission Inventory Biases Using Discrete Kalman 329
Cohan DS, Hakami A, Hu YT, Russell AG (2005) “Nonlinear response of ozone to emissions: source apportionment and sensitivity analysis.” Environmental Science and Technology 39(17): 6739–6748. Dunker AM (1981) “Efficient calculation of sensitivity coefficients for complex atmospheric models.” Atmospheric Environment 15: 1155–1161. Gilliland AB, Abbitt PJ (2001) “A sensitivity study of the discrete Kalman filter (DKF) to initial condition discrepancies.” Journal of Geophysical Research 106(D16): 17939–17952. Gilliland AB, Dennis RL, Roselle SJ, Pierce TE (2003) “Seasonal NH3 emission estimates for the eastern United States based on ammonium wet concentrations and an inverse method.” Journal of Geophysical Research 108(D15): 4477. Haas-Laursen DE, Hartley DE, Prinn RG (1996) “Optimizing an inverse method to deduce time-varying emissions of trace gases.” Journal of Geophysical Research 101(D17): 22823–22831. Hartley DE, Prinn RG (1993) “Feasibility of determining surface emissions of trace gases using an inverse method in a three-dimensional chemical transport model.” Journal of Geophysical Research 98: 5183–5197. Konovalov IB, Beekmann M, Richter A, Burrows JP (2006) “Inverse modelling of the spatial distribution of NOx emissions on a continental scale using satellite data.” Atmospheric Chemistry and Physics 6: 1747–1770. Martin RV, Chance K, Jacob DJ, Kurosu TP, Spurr RJD, Bucsela E, Gleason JF, Palmer PI, Bey I, Fiore AM, Li Q, Yantosca RM, Koelemeijer RBA (2002) “An improved retrieval of tropospheric nitrogen dioxide from GOME.” Journal of Geophysical Research 107: 4437. Martin RV, Jacob DJ, Chance K, Kurosu TP, Palmer, Evans MJ (2003) “Global inventory of nitrogen oxide emissions constrained by space-based observations of NO2 columns.” Journal of Geophysical Research 108(D17): 4537. Muller J-F, Stavrakou T (2005) “Inversion of CO and NOx emissions using the adjoint of the IMAGES model.” Atmospheric Chemistry and Physics 5: 1157– 1186. Napelenok SL, Cohan DS, Hu YT, Russell AG (2006) “Decoupled direct 3D sensitivity analysis for particulate matter (DDM-3D/PM).” Atmospheric Environment 40(32): 6112–6121. Quelo D, Mallet V, Sportisse B (2005) “Inverse modeling of NOx emissions at regional scale over northern France: preliminary investigation of the secondorder sensitivity.” Journal of Geophysical Research 110: D24310. Richter A, Burrows JP (2002) “Tropospheric NO2 from GOME measurements.” Advances in Space Research 29: 1673–1683. Sioris CE, Kurosu TP, Martin RV, Chance K (2004) “Stratospheric and tropospheric NO2 observed by SCIAMACHY: first results.” Advances in Space Research 34: 780–785. Yang YJ, Wilkinson JG, Russell AG (1997) “Fast, direct sensitivity analysis of multidimensional photochemical models.” Environmental Science and Technology 31(10): 2859–2868.
330
S. Napelenok et al.
Discussion B. Fisher: A major uncertainty in NOx emission inventories is the fraction emitted as NO2, the so-called primary NO2, which has increased in European cities in recent years following the introduction of particle traps to control particulate matter. Would it be possible to treat the primary NO2 as a further variable to be assimilated using the Kalman filter? S. Napelenok: The way that the inverse method is currently formulated allows treatment of sources of NOx from separate sectors, for example, primary NO2. A few difficulties are introduced with adding more variables. A minor one is the need for more computational resources, as each source of NOx (which can be an emission type, a geographical area, or a combination of the two) requires sensitivity fields resolved in time and space. A more challenging issue is assuring that the variables are linearly independent so that there is a unique solution to the system. Some preliminary results on a domain in the United States indicated that the independence condition is not necessarily always true for the tested sectors of NOx emissions that included mobile, point, area, nonroad, and biogenic. A. Venkatram:
S. Napelenok:
NO2 is perhaps a poor choice for a variable that can be used to correct NOx emissions using inverse modelling because it is oxidized quickly. The agreement between model estimates of NO2 column and observations is not surprising because you are essentially performing a linear regression. A true last of the value of the estimates would be to examine their impact on species such as ozone and particularly NOy. The fact that NO2 is oxidized quickly is one of its major attractions for global scale inverse modellers, because impacts of transport can be essentially ignored. The method demonstrated here is performed using a regional scale model and does account for transport. It is true that the improvement in agreement between modelled and observed NO2 columns is not surprising. However, that is the minimum one hopes to achieve and we have done better here by showing improvements in the comparison for an independent dataset of ground-level NO2 measurement at SEARCH. The comparison of NOy was also done, but not shown, because it followed the same pattern as NO2.
3.9 Fusing Observations and Model Results for Creation of Enhanced Ozone Spatial Fields: Comparison of Three Techniques Edith Gégo, P.S. Porter, V. Garcia, C. Hogrefe and S.T. Rao
Abstract This paper presents three simple techniques for fusing observations and numerical model predictions. The techniques rely on model/observation bias being considered either as error free, or containing some uncertainty, the latter mitigated with a Kalman filter approach or a spatial smoothing method. The fusion techniques are applied to the daily maximum 8-hour average ozone concentrations observed in the New York state area during summer 2001. Classical evaluation metrics (mean absolute bias, mean squared error, correlation, etc.) show that fused predictions are not better than a simple interpolation of observations. However, fused maps better reproduce the spatial texture of the model predictions. Keywords Data fusion, high-resolution maps, ozone
1. Introduction Because of their adverse effects on human health, a diverse suite of air contaminants is routinely monitored in the United States. Ozone concentrations, for instance, are recorded at more than 1,000 locations. In addition to the rich data base of observations, air contaminant concentrations can be reproduced by photochemical simulation models, numerically transcripting our scientific understanding of atmospheric chemistry and transport processes. The shortcomings of both observational and numerical information are well known. While considered unbiased, ozone measurements are spatially sparse point estimates. Extrapolation of observational information to unmonitored locations leads to smooth spatial images. Maps derived from numerical models, to the contrary, are spatially continuous and detailed but biased. In this context, we compare three simple, flexible and easy to implement techniques for fusing observations and numerical model outputs for the purpose of producing detailed spatial/temporal air contaminant concentration fields that are consistent with both observational and numerical information. Among the potential users of detailed air contaminant information is the public health research community that will be able to refine the exposure fields currently derived from air contaminant observations. The methods are applied to the daily maximum 8-hour average ozone concentrations in a 660 × 828 km rectangular domain centered on C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
339
340
E. Gégo et al.
the state of New York, and on the three month period from June 1 to August 31, 2001. Figure 1a shows the domain of interest and the monitoring sites.
2. Ozone Information 2.1. Numerical model estimates The ozone predictions used in this study were produced by the EPA photochemical simulation system CMAQ (Byun and Schere, 2006). For the application utilized (Appel et al., 2007), CMAQ was set to simulate most of the Eastern United States from January 1–December 31, 2001 with a horizontal grid size of 12 km. Only the predictions for the June 1–August 31 period in the domain of interest, i.e., a block of 55 model rows and 69 columns are utilized here.
2.2. Observations Measurements collected at 191 sites located within the model domain or in its immediate vicinity (36 km wide strip surrounding the domain) were retrieved from the U.S. EPA Air Data or the NAPS (National Air Pollution Surveillance program of Canada) data bases. The hourly concentrations were used to calculate the daily maximum 8-hour average ozone concentrations for each of the 92 days of interest and each site. Clustered data from sites that fall in the same model cell (15 pairs of sites) were averaged prior to any further operation, leaving a total of 176 ‘unclustered’ locations. Synthetizing the observation and model information, Figure 1 shows the mean of the daily maximum 8-hour average ozone concentration calculated for each site in the domain (panel a) and by CMAQ (panel b) from June 1 to August 31, 2001. Panel c presents the time series of the spatially-averaged averaged 8-hour mean daily maximum concentrations in the domain for the observations and CMAQ estimates. It appears that CMAQ underestimates ozone levels on episodic days, illustrating the need to intregrate model results and observations.
3. Methods Three simple fusion techniques are considered for this study, all of which aim at determining spatial/temporal bias fields which, when applied to CMAQ predictions, will result in unbiased maps having spatial texture approaching that of CMAQ.
Fusing Observations and Model Results for Creation of Enhanced Ozone Spatial Fields
341
Fig. 1 Site specific (panel a) and CMAQ (panel b) mean daily maximum 8-hour average ozone concentrations during the June 1st to August 31st , 2001 period; Daily mean of daily maximum 8-hour average ozone concentrations (panel c) measured at all 176 locations during the June 1–August 31 , 2001 period (blue line) and CMAQ estimate (red line)
3.1. Method 1: Inverse Distance Weighted (IDW) of bias fields Method I considers ozone observations error free. Bias is simply defined, therefore, as the difference between model prediction and observation at the 176 monitoring sites. Inverse distance weighting (IDW) method is used to produce spatially continuous 12 × 12 km resolution (CMAQ resolution) bias fields. Fused maps equal CMAQ minus the computed bias field. Preliminary investigation (not included here) showed that the IDW technique led to results as reliable as those obtained by kriging the bias but without the burden of identifying variograms for each simulated day. It also showed that utilizing the ten observations nearest the location being estimated was sufficient to obtain precise estimates.
3.2. Method II: Inverse Distance Weighted (IDW) of Kalman-smoothed bias fields Method II uses the Kalman filter algorithm (Kalman, 1960) to create an optimal estimate of the true ozone state from observations and model outputs. Primarily designed for time domain applications, the Kalman algorithm recursively estimates a state variable (in this case, model bias) at discrete time increments based on a state equation that describes the temporal evolution of the state variable, and on a series of measurements. Both the state equation and ozone measurements are assumed to be uncertain with respect to true concentrations. The Kalman filter interweaves model and observation uncertainty (varKand varH, respectively) to produce the best linear estimate of the state variable.
342
E. Gégo et al.
The relative strength of the state and observation uncertainties, varK and varHѽҏ respectively, is a key element of a Kalman filter application. If the measurements are believed more precise than the state estimates, the state estimate will be modified to fit the measurement more closely. Conversely, if the uncertainty of the state equation is insignificant in comparison to that of the measurement, the latter is ignored. In this study, the state variable defines the ‘true’ correction to be brought to model estimates and its temporal evolution is defined as a random walk. Two scenarios characterize uncertainty. In the first, the ratio of state to measurement variance (varKvarHҏҗҏ is fixed at 0.06, a value found by Kang et al. (2007) to be optimal for reducing the ozone forecast error of a photochemical model. In the second, varKvarHҏ is fixed at 1. The second setting therefore strengthens the correspondence between the estimated state and the measurements while the first setting mostly trusts the state equation. The filter is successively applied to smooth the 92-day time series of observed biases at each measurement location. Following temporal smoothing, fused (spatial) maps are created using the IDW scheme, as in Method 1.
3.3. Method III: spatially-smoothed bias fields In this scenario, observation/model bias is computed at spatial scales on the order of 200 km. The regional signals are extracted from the original signal using an iterated moving average scheme. More specifically, the calculation window is a 60 × 60 km window (5 × 5 model cells) progressively moved throughout the domain; the averaging process is repeated three times. This method can be seen as the spatial equivalent of the KZ filter originally defined for time series analysis (see Rao et al., 1997 for details). Like the KZ filter, the method can be applied to fields having empty cells, i.e., grid cells that do not contain a monitoring site, an interesting property for treatment of the observations. The correction to be brought to the original model fields are calculated by the difference between the spatially averaged model predictions and the spatially averaged observations.
4. Evaluation of Fused Fields Comparison of the fused maps generated by the three methods presented above is performed from two perspectives. First, a cross-validation exercise is performed to assess the similarities between the fused data and the original ozone observations. Second, the textures (relief) of the fused ozone fields are compared to those of the initial CMAQ results.
Fusing Observations and Model Results for Creation of Enhanced Ozone Spatial Fields
343
4.1. Calculation of evaluation metrics The comparability of the fused data with the original ozone observations is assessed using a cross-validation scheme whereby information is omitted one site at time and re-estimated from the remaining data. The procedure is successively repeated for each observation site. The cross validation set up is also utilized to assess the adequacy of ozone concentrations estimated with the IDW method applied to the measurements, hereafter referred to as ‘IDW observations’. Strictly speaking, IDW observations are not ‘fused’. However, they characterize the quality of maps that are obtained without utilizing model predictions. Mean absolute bias (MAB), root mean squared error (RMSE), and the squared correlation coefficient (R2) were the metrics of choice in comparing observations and fused data. These statistics were calculated for the ozone predictions generated with the three fusion techniques, as well as for the IDW observations and the original CMAQ estimates, the two latter being good benchmarks.
4.2. Texture comparison of fused maps and CMAQ maps The evaluation statistics described earlier allow assessment of the quality of ozone predictions at observation sites. However, they do not inform on the texture (relief) of the fields produced, an important feature of spatial information. ‘Relief’ was defined as the standard deviation of a 60 × 60 km window (5 × 5 grid cells) progressively moved (Increment: 12 km or 1-cell increment) to cover the entire domain. Because of the limited spatial extent utilized for its calculation, the standard deviation computed in this manner is referred to as the local standard deviation.
5. Results 5.1. Evaluation statistics Table 1 shows the evaluation metrics (MAE, RMSE, R2) calculated from the 16,192 (92 days x 176 locations) ‘estimate-observation’ pairs available for each prediction method. The best and poorest values for all three evaluation statistics are found for the IDW observations and CMAQ predictions, respectively. The performances of all fusion methods except Method II utilized with the error ratio varKҏҡvarH = 0.06 are quite similar to that of the IDW observations. Table 1 indicates that, although not detrimental, fusion techniques do not lead to better correspondence between predictions and observations than that obtained with IDW of observations only.
E. Gégo et al.
344
Table 1 Evaluation statistics for CMAQ, the interpolated observations (IDW) and the fused predictions. Statistic
CMAQ
IDW observ.
Fused estimates Method I
MAE RMSE R2
9.47 12.49 0.55
5.81 8.07 0.81
5.97 8.49 0.80
Method II varKvarH=1 7.07 9.62 0.75
varKvarH=0.06 8.67 11.52 0.62
Method III 6.54 8.35 0.84
A detailed inspection in time and in space of the evaluation statistics indicates that (not shown): (1) spatial-outliers, i.e., locations where measurements stand apart from their neighbours, are least well reproduced, whatever the estimation techniques; and (2) for all techniques, high ozone concentrations days are also high MAE and RMSE days.
5.2. Spatial texture Figure 2 displays maps of the maximum 8-hour average ozone concentrations predicted by CMAQ and the IDW observations for August 2, 2001 (day chosen at random). Focusing on the texture of each map, one may see that high concentration zones are more sharply delineated by CMAQ map than by the IDW interpolation (Figure 2c and d).
a
b
ppb
c
d
ppb
Fig. 2 Map of the predicted maximum 8-hour average ozone concentrations (ppb) for August 2, 2001 by CMAQ (panel a), and by IDW observations (panel b); local standard deviation of CMAQ (panel c) and of the IDW observations (panel d)
Figure 3 shows maps the maximum 8-hour average ozone concentrations predicted by the three fusion techniques for the same day as displayed on Figure 2 (2 August 2001, with method II varKvarHҏҏ = 1) (panels a–c), and the corresponding local
Fusing Observations and Model Results for Creation of Enhanced Ozone Spatial Fields
345
standard deviations (panels d–f). All ozone concentration maps (panels a–c) display the relief seen in the original CMAQ estimates combined with the effect of fusing to better fit the observations. High variability zones on panels d and e mostly match those of CMAQ (panel c of Figure 2), meaning that modification of CMAQ values to create fused estimates does not fundamentally alter the basic relief features. It appears that fusing may even have accentuated that relief, probably by establishing ozone levels similar to the observations. Note that the dark spots in panel c correspond to ‘empty cells’ for which method III did not lead to a numerical value, because these cells are too spatially isolated from any observation site.
Fig. 3 Map of the predicted maximum 8-hour average ozone concentrations (ppb) for August 2, 2001 obtained by application of fusion method I (panel a), method II (panel b) and method III (panel c) and corresponding local standard deviations (panels d–f) Generalizing the texture concepts illustrated in Figures 2 and 3 to the entire period, Table 2 presents the overall average (all days and cells included) of the local standard deviations characterizing CMAQ, the interpolated observations and the three fusion techniques, as well as their ratios to the CMAQ mean local standard deviation. The results presented clearly show that spatially interpolating the observations leads to smoother relief than that modeled by CMAQ. To the contrary, the three fusion techniques have a tendency to slightly accentuate CMAQ relief, with Method 2 and CMAQ average relief being remarkably similar. Table 2 Average local standard deviations of CMAQ, IDW observations and the three fusion techniques and their ratios to the CMAQ mean local standard deviation. CMAQ
IDW observ.
Fused results Method 1
Method 2
Method 3
Mean
2.44
1.54
2.76
2.62
3.34
Ratio
1.00
0.63
1.13
1.06
1.36
346
E. Gégo et al.
5. Summary and Conclusion The objective of this paper is to present three simple techniques for fusing observations and numerical model estimates. Spatial fields were obtained by: (1) spatially interpolating the observed biases with the inverse distance weights method (10 neighbours), (2) spatially interpolating smoothed biases, the Kalman filter being used for the smoothing, and (3) calculating spatially-smoothed bias fields with an iterative moving average technique. The methods were applied to the observed and modeled daily maximum 8-hour average concentrations in a 660 × 828 km domain centered on New York States from June 1 to August 31, 2001. The model estimates were obtained with CMAQ. The fusion techniques were able to maintain the texture of CMAQ estimates while reducing observation/model bias. In terms of classical comparative metrics (mean absolute bias, the root mean square error and the coefficient of determination between the predicted values and the corresponding observations), fused predictions are not better than simply interpolated observations (IDW method). However, the texture of fused maps is comparable to that of CMAQ in contrast to the smooth nature of interpolated observations. Disclaimer The research presented here was performed under the Memorandum of Understanding between the U.S. Environmental Protection Agency and the U.S. Department of Commerce’s National Oceanic and Atmospheric Administration and under the agreement number DW13921548. This work constitutes a contribution to the NOAA Air Quality Program. Although it has been reviewed by EPA and NOAA and approved for publication, it does not necessarily reflect their policies or views.
References Appel KW, Gilliland AB, Sarwar G, Gilliam RC (2007) Evaluation of the Community Multiscale Air Quality (CMAQ) model version 4.5: sensitivities impacting model performance: Part I – ozone, accepted for publication in Atmos. Environ. Byun D, Schere KL (2006) Review of the governing equations, computational algorithms, and other components of the Models-3 Community Multiscale Air Quality (CMAQ) modeling system. Applied Mechanics Reviews, 59, 51–77. Kalman RE (1960) A new approach to linear filtering and prediction problems, Journal of Basic Engineering, 82, 35–45. Kang D, Mathur R, Rao ST, Yu S (2007). Bias-adjustment techniques for improving ozone air quality forecasts, Journal of Geophysical Research, in review. Rao ST, Zurbenko IG, Neagu R, Porter PS, Ku JY, Henry RF (1997) Space and time scales in ambient ozone data, Bulletin of the American Meteorological Society, 78, 2153–2166.
3.4 Modelling of Benzo(a)pyrene Depositions over North Sea Coastal Areas: Impact of Emissions from Local and Remote Areas Ines Bewersdorff, Armin Aulinger, Volker Matthias and Markus Quante
Abstract The impact of benzo(a)pyrene (BaP) ship emissions on concentration and deposition distributions over Europe and in particular over North Sea coastal areas in comparison to land-based emissions is subject of this study. The carcinogenic BaP belongs to the group of polycyclic aromatic hydrocarbons (PAHs) and is here used as lead substance. A dataset with the ship emissions obtained from a bottom-up approach using ship movement data for the year 2000 derived from Lloyds Marine Intelligence Unit and average engine loads together with BaP emission factors is generated to serve as input for the chemistry transport model. Model runs with datasets including and excluding ship emissions are performed with the Community Multiscale Air Quality modelling system (CMAQ) which is set up on a 54 × 54 km2 grid for Europe. Provisional results show little influence of the BaP ship emissions on deposition and concentration distributions over sea and coastal areas. Keywords Benzo(a)pyrene, chemical transport modelling, persistent organic pollutants, ship emissions
1. Introduction The ship exhaust emissions of interest comprise mainly CO2, NO x, SO2 , CO, hydrocarbons, and particulates due to their impact on the environment as e.g. greenhouse gas (CO2 ), contribution to acid rain (NO x, SO 2), and/or due to their impact on human health (particulates) (Lloyd’s Register Engineering Services, 1995). Micropollutants affecting the human health as e.g. a variety of polycyclic aromatic hydrocarbons (PAHs) with their carcinogenic potential to humans and animals (ATSDR, 1995) have been neglected so far. This study focuses on the carcinogenic benzo(a)pyrene (BaP) which is the best investigated member of the group of PAHs. Here it is used as an indicator for PAHs. These are semivolatile, lipophilic persistent organic pollutants (POPs) which originate primarily from incomplete combustion of organic matter like fuel oil. Their persistence leads to long-range transport and therefore to their widespread distribution across the earth (Pacyna and Oehme, 1988). Hence, vessels emitting PAHs at sea may influence the PAH concentrations and depositions ashore. C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
296
Modelling of Benzo(a)pyrene Depositions over North Sea Coastal Areas
297
Previous studies have shown that especially NO x and SO2 emitted by ships contribute significantly to air pollution in coastal areas (Isakson et al., 2001; Tsyro and Berge, 1997). Furthermore, Corbett and Fischbeck (1997) have estimated that worldwide the annual ship emissions are within the same magnitude as the largest energy consuming nations. A new approach by Corbett and Koehler (2003) with actual fuel consumption by internationally registered ships together with fuel statistics have even shown the NO x and SO 2 ship emissions to be significantly greater than previously considered. In the study presented here, a bottom-up approach with emission factors in combination with engine power and fleet numbers is applied to investigate the contribution of BaP ship emissions to air pollution in coastal areas depending on season and meteorological conditions. Especially in summer, when residential combustion, one of the major PAH emission sectors, is negligible a significant BaP ship emission contribution is conceivable. The chemical transport modelling was performed with the Models-3 Community Multiscale Air Quality modelling system (CMAQ) (Byun and Ching, 1999; Byun and Schere, 2006) that is set up on a 54 × 54 km2 grid for Europe. The model is driven by meteorological fields that were calculated with the Fifth-Generation Pennsylvania State University/National Center for Atmospheric Research (NCAR) Mesoscale Model (MM5) (Grell et al., 1995) and adapted to the chemistry transport model with the Meteorology-Chemistry Interface Processor (MCIP, Otte, 1999). CMAQ has been modified by GKSS to additionally consider the transport of PAHs, in particular of BaP (Aulinger et al., 2007). Matthias et al. (2008) have shown that the model is applicable to North Sea coastal regions by evaluating model results by means of measured air concentrations.
2. Input Data 2.1. Ship database The ship database was purchased from Lloyds Marine Intelligence Unit (LMIU). It includes commercial vessels equal to or greater than 100 gross tonnages (GT) and provides information on ship movements for the year 2000 in parts of Europe (riparian states of the North and Baltic Sea, Atlantic coast of France, Spain and Portugal) and vessel characteristics together with engine characteristics. The ship movement database consists of previous departure, arrival, departure and next arrival places and dates of 15,625 ships including a total of 651,825 movements. These movements are covered by 58,324 different routes. The ship density derived from the routes of this database on the one hand in Europe and on the other hand focusing on the North Sea area is shown in Figure 1. The ship characteristics comprise amongst others information on vessel age, vessel type, engine type, engine year of manufacture, number of engines, fuel type, speed at the crankshaft, power and maximum speed. Approximately 9% of the ship movements could not be considered because of missing data in the LMIU data base. Thus, the
298
I. Bewersdorff et al.
actual compilation of ship tracks must be considered as preliminary and concepts to complete the data are currently applied.
Fig. 1 Total ship density of all ships 100 GT in Europe (left) and focused on the North Sea area (right) for the year 2000. Mediterranean movements not included
2.2. Ship emission factors Emission factors (power-based in g/kWh) used in this study are obtained from Cooper and Gustafsson (2004) who derived these emission factors for Sweden’s international reporting duties. The emissions from ships are affected by several characteristics of the ships as well as the fuel used. The engine type determines the combustion conditions and therefore the emission of some pollutants. Factors affecting emissions from ships are presented in detail in Whall et al. (2002). Due to a very limited dataset for determining the emission factors of the different PAH species, separate engine and fuel specific factors are not provided. Thus, it has to be considered that the BaP emission factor is highly uncertain. Furthermore, the BaP ship emission factor provided is similar to BaP heavy-duty diesel vehicle emission factors (Doel et al., 2005), although the PAH content of the fuel influences the BaP emissions (Doel et al., 2005). The extreme poor quality of the residual oil which is used as fuel by the majority of vessels suggests a higher ship emission factor. As a first approach the solely available BaP ship emission factors published by Cooper and Gustafsson (2004) were used.
2.3. Land-based emissions The database used for land-based emissions was derived from TNO (Denier van der Gon et al., 2005). The data were provided as annual bulk emissions and spatially distributed on the 50 × 50 km2 stereographic EMEP (Environmental Monitoring Programme) grid. To serve as input data for CMAQ, the emissions were interpolated to the 54 × 54 km2 grid. Furthermore, a temporal variation of the data meeting the seasonal dependence of the emissions with a minimum during summer
Modelling of Benzo(a)pyrene Depositions over North Sea Coastal Areas
299
months and a maximum during winter months was applied by linearly correlating emissions from residential combustion, the dominant source of BaP emissions, to the ambient air temperature. Figure 2 presents monthly averaged land-based and marine BaP emissions. The dominating sector for land-based BaP emissions (Figure 2, left), residential combustion, shows a strong seasonal variation depending on the temperature outside. By contrast, the ship frequency is more or less independent of the season (Corbett et al., 1999; Whall et al., 2002) and so are, therefore, the ship emissions (Figure 2, right). Marine emissions
July
January
Land-based emissions
Fig. 2 Average land-based (left) and marine (right) BaP emissions for January (top) and July (bottom) 2000 (Mind the different scales)
3. Methodology The outline of the procedure followed in this study to obtain the BaP deposition and concentration distributions over Europe as well as over North Sea coastal areas caused by ship emissions is depicted in Figure 3. For computing the routes of the vessels, it is assumed that each ship takes the shortest route between two ports at sea. Since the database provides only a daily time resolution for the movements, it is further assumed that the ships arrive and depart at 6 a.m. in the morning if they are not leaving the same day. If they are leaving the same day they leave the port at 6 p.m. in the evening or earlier if they call at another port the same day. The travelling time is distributed equally to all passed grid boxes on their route. Together with the engine power and the emission factors, the total emissions per grid box are calculated.
I. Bewersdorff et al.
300
The compiled emission dataset serves as input into the Eulerian air quality modelling system CMAQ which computes the concentration and deposition distribution over Europe. To evaluate the influence of the ship emissions, model runs with different emission files are performed. The emission files applied to CMAQ comprise on the one hand land-based emissions separately and on the other hand a combination of land-based and marine emissions. The focus is put on the months January and July as representatives for winter and summer. Vessel movement data
Basic vessel data
Route Time spent in grid box
Emission factor [g/kWh]
Total ship emissions per grid box Chemical transport modelling with CMAQ Deposition and concentration distribution over Europe Fig. 3 Procedure outline for quantifying BaP ship emissions and modelling depositions and concentrations
4. Results The results of the model runs presenting BaP concentration and deposition distribution plots are shown in Figures 4 (concentration) and 5 (wet deposition). In January, over wide areas of Europe the monthly mean concentration is higher than 1 ng/m3 (Figure 4a) which is the prospective annual target value for BaP set in the directive 2004/107/EC of the European Parliament and of the Council of 15 December 2004 relating to arsenic, cadmium, mercury, nickel and polycyclic aromatic hydrocarbons in ambient air. In summer, concentration peaks appear only close to the emission sources so that in general the impact on the coastal areas is small. Figure 4b shows that the impact of the ships on the concentration level in marine and coastal areas is marginal. In January over the North Sea area the concentration increase amounts to 0.3% at most, whereas in July the contribution of the ships to the concentration level adds up to 0.5% (Figure 4c). Due to minor emissions originating from residential combustion in summer the impact of the ships is slightly higher.
Modelling of Benzo(a)pyrene Depositions over North Sea Coastal Areas Difference
Difference in %
July
January
Average Concentration
301
a)
b)
c)
Fig. 4 Modelled average BaP concentrations with land-based and marine emissions (a), the concentration difference of included ship emissions and land-based emissions only (b), relative change of the concentration distribution caused by ship emissions (c) for January (top) and July (bottom) 2000 (Mind the different scales) Difference
a)
b)
Difference in %
July
January
Monthly Deposition
c)
Fig. 5 Modelled monthly BaP wet deposition with land-based and marine emissions (a), the wet deposition difference of included ship emissions and land-based emissions only (b), relative change of the wet deposition distribution caused by ship emissions (c) for January (top) and July (bottom) 2000 over the North Sea area (Mind the different scales)
Figure 5 displays the monthly BaP wet depositions. Comparable with the BaP concentration distributions a pronounced seasonal variation of depositions is evident (Figure 5a) because of the increased BaP emissions in winter caused by residential combustion. The wet deposition of BaP resulting from ships is
302
I. Bewersdorff et al.
approximately three orders of magnitude smaller than the contribution from emission sources (Figure 5b). Although the absolute BaP wet deposition is smaller in July than in January the relative impact in summer (up to 0.35%) is higher than in winter (Figure 5c) where it amounts to only 0.2% of the total BaP wet deposition over the North Sea area.
5. Conclusion As NO x and SO2 emitted by ships contribute significantly to air pollution in coastal areas the contribution of BaP ship emissions to pollution levels was studied as well. It was found that BaP ship emissions contribute to the BaP concentration and deposition distribution over North Sea coastal areas at most to 0.5% in summer and to a lesser extent in winter. A distinctive relative seasonal impact is seen due to increased residential combustion in winter. The contribution of the BaP ship emissions could be explained by on the one hand an underestimated ship density due to missing data in the LMIU database and on the other hand uncertain emissions. Further work to complete the data matrix is going on. After including all ship movements the contribution of the BaP ship emissions may increase noticeably. Additionally, the high uncertainty of the BaP ship emission factor and its indicated underestimation lead to the assumption of a further underestimation of the BaP ship emissions which is difficult to define precisely. Additional measurements would be desirable. Acknowledgments TNO Netherlands is acknowledged for providing emission data of BaP. We thank Sebastian Bewersdorff for the support concerning the programming of the shipping route routine.
References ATSDR (Agency for Toxic Substances and Diseases) (1995) Toxicological Profile for Polycyclic Aromatic Hydrocarbons (PAHs), Public Health Service, US Department of Health and Human Services, Atlanta, GA, USA. Aulinger A, Matthias V, Quante M (2007) Introducing a Partitioning Mechanism for PAHs into the Community Multiscale Air Quality Modelling System and Its Application to Simulating the Transport of Benzo(a)pyrene over Europe, Journal of Applied Meteorology and Climatology 46 (11), 1718–1730. Byun D, Ching JKS (1999) Science Algorithms of the EPA Models-3 Community Multiscale Air Quality Modelling System, EPA report, EPA/600/R-99/030, Office of Research and Development, Washington, DC 20406. Byun D, Schere KL (2006) Review of the Governing Equations, Computational Algorithms, and Other Components of the Models-3 Community Multiscale Air Quality (CMAQ) Modeling System, Applied Mechanics Reviews 59(2), 51–77.
Modelling of Benzo(a)pyrene Depositions over North Sea Coastal Areas
303
Cooper DA, Gustafsson T (2004) Methodology for Calculating Emissions from Ship: 1. Update of Emission Factors, SMHI Swedish Metrological and Hydrological Institute, Norrköping, Sweden. Corbett JJ, Fischbeck PS (1997) Emissions from Ships, Science 278(5339), 823– 234. Corbett JJ, Fischbeck PS, Pandis SN (1999) Global Nitrogen and Sulphur Inventories for Oceangoing Ships, J. Geopys. Res. 104 (D3), 3457. Corbett JJ, Koehler HW (2003) Updated Emissions from Ocean Shipping, Journal of Geophysical Research-Atmospheres 108 (D20), 4650. Denier van der Gon HAC, van het Bolscher M, Visschedijk AJH, Zandveld PYJ (2005) Study of Effectiveness of UNECE Persistent Organic Pollutants Protocol and Cost of Possible Additional Measures. Phase I: Estimation of Emission Reduction Resulting from the Implementation of the POP Protocol, TNO-Report, B&O-A R 2005/194, Appeldoorn, The Netherlands. Doel R, Jørgensen R, Lilley LC, Mann N, Rickeard DJ, Scorletti P, Stradling R, Zemroch PJ (2005) Evaluation of Automotive Polycyclic Aromatic Hydrocarbon Emissions. Prepared for the CONCAWE Automotive Emissions Managements Group, Report No. 4/05, Brussels. Grell GA, Dudhia J, Stauffer DR (1995) A Description of the Fifth-Generation Penn State/NCAR Mesoscale Model MM5, NCAR Technical Note, NCAR/TN-398 + STR, Boulder, CO, 138 pp. Isakson J, Persson TA, Selin Lindgren E (2001) Identification and Assessment of Ship Emissions and their Effects in the Harbour of Göteborg, Sweden, Atmospheric Environment 35(21), 3659–3666. Matthias V, Aulinger A, Quante M (2008) Adapting CMAQ to Investigate Air Pollution in North Sea Coastal Regions, Environmental Modelling and Software 23, 356–368. Otte TL (1999) Developing Meteorological Fields, in Byun DW and Ching JKS Science Algorithms of the EPA Models-3 Community Multiscale Air Quality Modeling System, EPA/600/R-99/030, US Environmental Protection Agency, Office of Research and Development, Washington, DC Pacyna JM, Oehme M (1988) Long-range Transport of Some Organic Compounds to the Norwegian Arctic, Atmospheric Environment 22(2), 243–257. Tsyro SG, Berge E (1997) The Contribution of Ship Emission from the North Sea and the North-eastern Atlantic Ocean to Acidification in Europe. EMEP/ MSCW Note 4/97. EMEP, Meteorological Synthesizing Centre – West, Norwegian Meteorological Institute, Oslo, Norway. Whall C, Cooper D, Archer K, Twigger L, Thurston N, Ockwell D, McIntyre A, Ritchie A (2002) Quantification of Emissions from Ships Associated with Ship Movements between Ports in the European Community, Report for the European Commission, Entec UK Limited, Northwich, Great Britain.
304
I. Bewersdorff et al.
Discussion J. Kushta: What was the background field applied to your model and boundary conditions? I. Bewersdorff: The model run was started with seven days spin-up time. The background field for BaP was set zero in the beginning of the model run. Prior studies have shown that this spin up time is sufficient to eliminate the influence of the background field on the model results. The boundary conditions for BaP were kept always equal to the boundary grid cells within the domain to prevent the formation of a gradient at the boundaries. S.-E. Gryning: What is the impact of the marine environment from ship emissions? I. Bewersdorff: The impact of BaP ship emissions in terms of wet depositions on coastal areas was focus of this study. As it was shown the contribution of the ship emissions are very little and in the North Sea area account for at most 0.5%. I cannot make a statement about the impact on the marine environment. This could be studied by a subsequent ecosystem modelling system which could use our data as input for their model run.
3.1 Rapid Data Assimilation in the Indoor Environment: Theory and Examples from Real-Time Interpretation of Indoor Plumes of Airborne Chemical Ashok Gadgil, Michael Sohn and Priya Sreedharan
Abstract Releases of acutely toxic airborne contaminants in or near a building can lead to significant human exposures unless prompt response measures are identified and implemented. Commonly, possible responses include conflicting strategies, such as shutting the ventilation system off versus running it in a purge (100% outside air) mode, or having occupants evacuate versus sheltering in place. The right choice depends in part on quickly identifying the source locations, the amounts released, and the likely future dispersion routes of the pollutants. This paper summarizes recent developments to provide such estimates in real time using an approach called Bayesian Monte Carlo updating. This approach rapidly interprets measurements of airborne pollutant concentrations from multiple sensors placed in the building and computes best estimates and uncertainties of the release conditions. The algorithm is fast, capable of continuously updating the estimates as measurements stream in from sensors. As an illustration, two specific applications of the approach are described. Keywords Bayesian updating, data fusion, real time source reconstruction
1. Introduction and Motivation Airborne acutely toxic contaminant releases in or near a building can lead to significant human exposures unless prompt response measures are taken. However, possible responses can include conflicting strategies, such as shutting the ventilation system off versus running it in a purge mode, or having occupants evacuate versus sheltering in place. The right choice depends in part on knowing the source locations, the amounts released, and the likely future dispersion routes of the pollutants. Determining this information is complicated by the complex nature of airflows typically found in multi-room, multi-story buildings. Merely detecting an airborne pollutant from one or more sensors placed in the building, may not reveal the location or strength of the source. The sensor measurements must be interpreted to estimate the source characteristics and quantify the uncertainties. For effective C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
263
264
A. Gadgil et al.
decision making, the measurements must also be interpreted quickly and continuously as data stream in from the sensors. Traditional algorithms for data interpretation generally use an inverse modeling approach (e.g., optimization and Gibbs sampling) to fit an indoor airflow and pollutant transport model to measurements of airborne pollutants. The fit is usually achieved by iteratively adjusting model input parameters until they reasonably predict the data. For on-line, real-time, sensor data interpretation, these approaches are too slow. They (i) wait to execute computationally intensive fate and transport models until data are first obtained; (ii) execute the models repeatedly as new or successive sensor data become available; and (iii) require a considerable amount of data before the algorithm finds a unique solution or estimates the uncertainty in the calibrated parameters. Lastly, the computational burdens required by these algorithms can be so great that using them for pre-event planning, such as to determine optimal monitoring locations, sampling plans and sensor performance criteria, can be impractically cumbersome. Many of these problems can be solved using a technique called Kalman filtering (see e.g., Grewal and Andrews, 2001). However, it is best used for linear systems with well-conditioned input-to-output parameter covariance matrices and strong observability between the internal state variables and the model outputs. For several reasons (skipped here for brevity), Kalman filtering is not well suited to the current problem. We describe an alternative algorithm called Bayes Monte Carlo updating, based on Bayesian statistics. This approach succeeds where traditional methods fail because it (1) decouples the simulation of predictive fate and transport models from the interpretation of measurements, and (2) incorporates uncertainty analysis in all parts of the framework. Thus, we can compute the time-consuming airflow and pollutant transport predictions and uncertainty estimates – without requirements on linearity of the models – prior to a pollutant release event, and interpret sensor data in real time during an event. The technique may be used to estimate the location, magnitude, and duration of the release, to characterize any unknown or variable building or weather conditions, and to predict future pollutant transport in the building. Initial estimates are provided as soon as a sensor detects a pollutant, and can be updated as each new measurement arrives. Bayes Monte Carlo updating has been applied to assess environmental health risk (e.g., Brand and Small 1995; Pinsky and Lorber 1998), analyze groundwater monitoring data (e.g., Wolfson et al., 1996; Sohn et al., 2000), and conduct environmental value-of-information analyses (e.g., Dakins et al., 1996). However, the research problems described in these papers are distinct from the current work in one important feature. The papers describe applications to interpret data well after they were collected, when interpretation and response were not time-critical. In the present work, we recognize and exploit a feature of Bayes Monte Carlo updating not previously recognized: modeling and data analysis can be decoupled, which allows for data to be interpreted while they stream in during a pollutant release event. This is a significant advance over previous uses of Bayes Monte Carlo methods. Furthermore, our research group has been the leader in application of this general approach to indoor air pollutant source characterization and airborne pollutant transport predictions.
Rapid Data Assimilation in the Indoor Environment: Theory and Examples from Real-Time
265
In this keynote paper, we (i) elucidate the Bayesian algorithm for interpreting sensor data in real time, and (ii) demonstrate the approach with two selected examples. In the first example, we successfully detect and characterize a pollutant release in a hypothetical five-room building. In this illustrative application, we generated synthetic data for two data collection scenarios: (1) concurrent sampling, and (2) sequential sampling. We also examine degradation of the predictive results with increasingly noisy data. In the second example, we apply the algorithm to characterize a tracer gas release in a three storey building. In this case study, not only are the sensors noisy, but they are trigger-type sensors, i.e., they only report a “yes” or a “no” regarding whether the local tracer gas concentration exceeds the pre-set trigger level.
2. Approach Since this is a diverse audience, we take the time to explain the approach assuming that not all readers are familiar with Bayesian Monte Carlo updating. The Bayesian data interpretation approach is divided into two stages. First, in the pre-event or simulation stage, the practitioner selects a fate and transport model, builds a computer model of the building, characterizes uncertainties of the model inputs, and simulates many hypothetical airflow and pollutant transport scenarios. These time-consuming tasks are completed before a pollutant release occurs. In the second stage, during a pollutant release event, the agreement between each of the model simulations and sensor data is rapidly evaluated using a technique called Bayesian updating (see e.g., Brand and Small, 1995; Sohn et al., 2000). This stage is quick and is conducted as data stream in from the sensors.
2.1. Pre-event computations Before a release event, the practitioner develops a model of the building’s indoor airflow and pollutant transport. Best estimates for model inputs are generated from, for example, previous building characterization exercises, tracer gas flow experiments and modeling, published literature, and professional judgment. Any uncertain model input parameter (e.g., outdoor temperature) or variable building characteristic (e.g., width of a window crack) is assigned an uncertainty distribution that describes the probabilistic range of possible values. Pollutant description uncertainties, e.g., the location, duration, and amount of pollutant released in an incident, are also assigned uncertainty distributions. Generally, wide distributions are assigned due to the limited prior information, particularly for describing the pollutant characteristics. The practitioner next generates a library of model simulations by sampling the distributions of the model input parameters using Monte Carlo or other sampling technique, and predicting airflow and pollutant transport for each set of parameters.
266
A. Gadgil et al.
Each model simulation represents a possible building configuration and pollutant release scenario. At this stage, each simulation is equally likely to occur. Thus, sufficient sampling of the uncertainty distributions is essential to represent the full range of possible building and pollutant release characteristics. One method for testing sufficiency of sampling is by increasing the sample size until changes in summary statistics (e.g., means, variances, coefficients of variation) of model predictions are negligible. The resulting library of simulations may consist of several thousand simulations (also called ‘realizations’). Since this stage is not time critical, a large library of simulations is not difficult to develop with the advances of fast personal computers and inexpensive data storage devices. It is important that the parameter ranges sampled in the uncertainty characterization and Monte Carlo sampling are wide enough to contain the parameter values of the actual event (to be diagnosed in real-time). Otherwise, the method will fail to indicate (i.e., assign a high likelihood to) any single set of parameter values as the right one. Of course such failure still provides some useful information: that the actual event parameters are not within the ranges sampled, or there is a model-misfit to the actual event.
2.2. During-event data interpretation During a release event, the algorithm compares data streaming in from sensors to each realization in the library of model simulations using a structured probabilistic method referred to as Bayesian updating. Bayes’ rule allows the practitioner to quickly estimate and update the level of agreement between the observed data and each model simulation (i.e., the pollutant transport predictions). To summarize the process, each realization in the library is compared to the data to quantitatively assess the likelihood that the realization describes the event in progress. A high likelihood estimate will result for the realization(s) (i.e., the model simulation(s) with predictions) that fit the sensor data well. This in turn suggests that the model inputs used to generate that realization in the pre-event simulation stage have high probability of describing the event in progress. By rapidly evaluating and comparing the relative fits for each realization using Bayesian statistics, the practitioner estimates the best-fitting suite of model inputs and the associated uncertainty. The difference between the data and the predictions could arise from measurement error, spatial and temporal averaging or correlations, and imperfect model representation. These are all considered when quantitatively estimating the data-tomodel agreement. The probability of each model simulation before and after assessing the agreement is termed the prior and posterior probability, respectively. Suppose the library of Monte Carlo simulations contains “N” simulations (or realizations), and each one is indexed with a number, “k”. With each realization, there is associated a set of output parameters (as a functions of time), and also input parameter values. “Updating” involves updating the probability assigned to each of the N realizations in the library by comparing the freshly obtained data (say, at a given time) with the corresponding prediction for that time by each of these N
Rapid Data Assimilation in the Indoor Environment: Theory and Examples from Real-Time
267
realizations. Before the updating, their probabilities are called “prior probabilities” and the updating process results in the “posterior probability” after that freshly obtained data is assimilated into the assigned probability for each realization. The posterior probability of the kth Monte Carlo simulation making prediction Yk given the sensor measurements, or observations, O, is denoted as p(Yk|O). The prior probability of the same kth Monte Carlo simulation making prediction Yk is denoted as p(Yk). Before onset of data comparison, each of the model realizations are usually assumed equally likely (i.e., p(Yk) = 1/K). As data streams in, for each time step, we compare the observations, denoted by “O”, with predictions “Yk” made by each of the N realizations, where the subscript k refers to the predictions for that time made by the kth realization. Using Bayes’ rule, p(Yk|O) is calculated using Eq. (1) (Brand and Small, 1995):
p (Yk | O)
p O | Yk p Yk 6 p O | Yi p Yi
(1)
N i 1
where p(Yk|O) is the posterior probability; p(O|Yk) is the likelihood of observing measurements O given model prediction Yk; p(Yk) is the prior probability of the kth Monte Carlo simulation; and N is the number of Monte Carlo simulations. The posterior probability, p(Yk|O), describes the probability of all of the model assumptions and predictions associated with the kth realization. Thus, the prior uncertainty of each model input parameter (e.g., source location or building characteristic) and model output (e.g., airborne concentration prediction) is updated according to how well model predictions in the prior uncertainty distribution agree with the sensor data. The posterior probability (Bayes Factor) of each simulation is applied to each parameter associated with that simulation, and thus a re-estimation is obtained for all input uncertainties. The posterior mean is calculated for each input parameter and a weighted estimate of the value of the parameter is obtained (Brand and Small, 1995). N
¦V p
Mean : PV
i
i
(2)
i 1
The likelihood function, p(O|Yk), in Eq. (1) quantifies the error structure of the data, i.e., the differences between the data and the model predictions resulting from measurement error, spatial and temporal averaging or correlations, and imperfect model representation. If “S” independent measurements are considered, for example following sequential concentration measurements returned from sensors or from concurrent measurements sampled in several locations, the likelihood of observing all of the measurements is the product of all of the individual likelihoods:
p O | Yk
S
pO j 1
j
| Yk
(3)
A. Gadgil et al.
268
For unbiased measurements with a normally distributed error, the likelihood of observing a sensor measurement, Os, given a model prediction, Ys,k, is given as: p(Os | Ys , k )
1
VH
§ 1 ªO Y º2 · s, k ¸ exp¨ « s ¨ 2 ¬ V H »¼ ¸ 2S © ¹
(4)
where Os is the concentration measured by a sensor in a room at t = ts; Ys,k is the airborne concentration predicted from the kth Monte Carlo realization that corresponds to Os; and VH2 is the error variance of the measurements. The error variance, VH2, describes not only the error in the sensor instruments, but also the error associated with comparing model predictions with sensor measurements having different spatial and temporal averaging. Alternative likelihood function can be used, as appropriate, without affecting the overall approach. This second stage of the approach is mathematically simple and can be executed very quickly, much quicker than the rate at which new data are likely to arrive from sensors.
3. First Illustrative Application This first illustration is to locate and characterize a hypothetical pollutant release in a five-room building, selected from the full description and results presented in Sohn et al. (2002). Uncertainties in source location, duration, and amount, and in some building characteristics, were estimated and updated using synthetic data.
3.1. Building description The study building is a single story building comprising three rooms, a common area (CA), and a bathroom (Figure 1). Each of the partitioned areas are treated as well-mixed zones. The zones connect to the outside via windows and doors, and interconnect via internal doors. The building does not have an HVAC system. The status of one of the CA zone windows and the door between the CA zone and Room 3 is “unknown” to the simulations (e.g., owing to failed position sensors at these locations). These are denoted in Figure 1 with question marks. All other windows and doors are closed. Wind blows at a steady 3 m/s on the exterior wall shared by the CA zone and Room 1. The temperatures of the rooms are indicated in Figure 1 and the outside temperature is unknown.
Rapid Data Assimilation in the Indoor Environment: Theory and Examples from Real-Time
269
Fig. 1 Plan of the five-room building. The arrows represent windows or doors; question marks indicate unknown open or closed status. Windows in the bathroom and Room 2, and interior door for Room 1 are open. All other windows and doors are closed. The wind blows at a steady 3 m/s
3.2. Airflow and pollutant transport simulation We selected the COMIS model (Feustel, 1999) to predict indoor airflow and pollutant transport. COMIS predicts the steady-state flows of air and the dynamic transport of pollutants by representing the building as a collection of well-mixed zones. Air flows between zones via cracks, doors, and windows (and also fans and ductwork, although we did not use these features here). Though we selected a multi-zone modeling approach for this application, our data interpretation algorithm may be used with any suitable indoor airflow and pollutant transport model. As part of the pre-event calculation, we generated a library of airflow and pollutant transport simulations by sampling the uncertainty distributions describing the source and building characteristics. Five thousand air flow and pollutant simulations, each of them equally likely, were generated using Latin Hypercube sampling techniques (Iman and Conover, 1980). Means and variances for several sample sizes were tested to ensure that five thousand simulations adequately sampled the problem solution space.
3.3. Description of synthetic data We generated synthetic data to represent measurements that might stream in from air monitoring sensors placed in the building. The synthetic data are based on an airflow and pollutant transport simulation that represents a possible pollutant release event. This simulation was excluded from the library of five thousand simulations describing the prior pollutant concentration predictions. We added measurement error to the model simulation using a two-part error structure: (1) a normally distributed error associated with a standard deviation proportional to the true value, i.e., a fixed coefficient of variation, and (2) a normally distributed random error that is independent of the magnitude of the measurement. We are not limited to this structure for errors. Many statistical
270
A. Gadgil et al.
methods for handling more complex error structures are available and can be used in place of Eq. (4) (see e.g., Morgan and Henrion, 1990; Sohn et al., 2000). We generated high, medium, and low quality synthetic data with progressively larger magnitudes of error components in the error structure. If adding the error generated a negative value for the pollutant concentration, we set the simulated measurement to zero. Details are described in Sohn et al. (2002), and not given here for brevity. As expected, the high quality data shows a more consistent pollutant concentration time series than is described by the low quality data.
3.4. Data interpretation Along with the three levels of data quality (Section 3.3), we also evaluated two different plans for data collection. In the first, concurrent sampling, we obtain sensor measurements from all five zones simultaneously at 5-minute intervals. In the second, sequential sampling, we obtain sensor measurements sequentially, one zone at a time, at 5-minute intervals. Though we present the results for the various quality of data and sampling schemes, it is important to emphasize that the results merely illustrate the types of data interpretation and “what-if” analyses that may be conducted using our interpretation algorithm. The results do not represent the success or failure of the interpretation approach. Figure 2 shows the estimation of the source location for the three qualities of data.
Fig. 2 Locating the source using (a) high, (b) medium, and (c) low quality measurements. Concurrent sampling draws a measurement from each zone every 5 minutes. Sequential sampling draws a measurement from one room at a time every 5 minutes in the order: (1) common area at t = 5 min, (2) Room 1 at t = 10 min, (3) Room 2 at t = 15 min, (4) bathroom at t = 20 min, (5) Room 3 at t = 25 min, and (6) common area at t = 30 min. The probabilities at t = 0 are before data interpretation
Rapid Data Assimilation in the Indoor Environment: Theory and Examples from Real-Time
271
With medium or high quality data, the interpretation correctly identified the source location at the first measurement event (t = 5 min.), when five measurements were obtained. With low quality data, the identification of the source location was slower, requiring more measurements to overcome the error in the data. More details are given in Sohn et al. (2002). Again, the medium and high quality data permit dramatic uncertainty reductions at t = 5 min., in all cases converging to the correct answers. The low quality data however required more data, and thus more time. With the sequential sampling plan, we gather data at a much slower pace – one, not five, data points every 5 minutes. It might represent a situation where a single (expensive) sensor is multiplexed to several sampling tubes. The rooms were sampled in the order (1) CA zone at t = 5 min, (2) Room 1 at t = 10 min, (3) Room 2 at t = 15 min, (4) bathroom at t = 20 min, (5) Room 3 at t = 25 min, and (6) CA zone at t = 30 min. See Figure 2. Given that the sequential sampling plan collects data five times slower than the concurrent sampling plan, the medium and high quality data did not locate the source until all of the rooms were sampled once (t = 25 min.), though reasonably good estimates were generated as early on as t = 10 min. The low quality data, however, did not locate the source even after 30 min.
4. Second Illustrative Application 4.1. Problem description Now we focus on trigger- or alarm-type sensors, rather than continuous-output devices. We base our case study on data from one of twelve tracer-gas experiments conducted at a building at the Dugway Proving Ground, Utah (Sextro et al., 1999). The paper from which this these illustrations are taken (Sreedharan et al., 2006), examines how well various sensor systems, each system consisting of sensors with different sensor characteristics (threshold level, response time, and accuracy), reconstruct the release event. These examples in Sreedharan et al. (2006) demonstrate the importance of a systems perspective in selecting sensors with desirable sensor characteristics. However, for lack of space, only one illustrative example will be shown here.
4.2. Approach We consider the following problem. A contaminant is released over a short duration somewhere in a building (this may include its indoor air intake vents). A network of threshold or alarm-type sensors operates to detect the contaminant. We seek to understand how sensor characteristics such as threshold level and response time affect the ability of a sensor interpretation algorithm to quickly detect and characterize the contaminant release.
272
A. Gadgil et al.
For the purposes of this paper, we assumed that each zone is equipped with one sensor, and each sensor has a single threshold, meaning the output is either “below threshold” or “above threshold.” We define several possible threshold levels that are at or above the minimum detection limit of the sensor. Three important parameters characterize sensor performance in this study: threshold level; response time (also sometimes called integration time); and accuracy. A sufficiently large release will trigger one or more sensors. A Bayes Monte Carlo algorithm is then initiated to determine key information about the release event, including the location of the release and the contaminant mass discharged. This information can help guide emergency response and post-event remediation.
4.3. Building characterization, data collection and model generation The study focuses on one unit in a multi-unit building located at the Dugway Proving Grounds, Utah. The unit consists of 660 m3 of interior volume and approximately 280 m2 of floor area on three levels. A mechanical air-handling unit (AHU) supplies air to the first and second floors, and its return is on the first floor. The AHU is a 100% recirculating unit (i.e., there is no deliberate outside air intake). A library of 5,000 simulated contaminant releases was generated using a COMIS model of this building, which was based on detailed measurements of the building flow paths. For the library, we sampled from statistical distributions of a set of key input parameters, as described earlier. Tracer gas dispersion data for examining the interpretation of signals streaming from triggered sensors was obtained from tracer gas experiments conducted in the same building, and described in Sextro et al. (1999). We generated hypothetical threshold sensor data by interpreting the tracer data from the experiment (Sextro et al., 1999) as if they were concentrations to which surface acoustic wave (SAW) sensors were exposed. SAW sensors are piezoelectric devices, often configured to provide alarms based on whether the incoming concentrations are above or below a predefined trigger or threshold level. False positive and false negative alarms may occur, according to the performance characteristics of each sensor, or the inability of the sensor to distinguish between the contaminant of interest and interfering chemicals that also may be present in the air. We generated hypothetical alarm data for sensors with different performance characteristics, based on discussions with several developers and users of SAW sensors. Three sensor attributes were varied: threshold level, response time, and error. The threshold levels were chosen relative to the measured concentrations during the first 120 minutes of the release event. The lowest selected threshold would cause 98% of the data to trigger the alarm, while the highest threshold would trigger an alarm for only 1% of the data. However, for presentation purposes, we normalize both threshold levels and concentration data by the concentration that
Rapid Data Assimilation in the Indoor Environment: Theory and Examples from Real-Time
273
would be found in the release zone if the entire release amount instantaneously mixed throughout that zone. That is, thresholds and concentrations are reported in terms of the theoretical maximum peak concentration that could be measured in the system under the perfectly well-mixed assumption. With this normalization, the lowest threshold level was 0.02% of the maximum peak, and the highest was 16%. Sensor response times ranged from 20 to 180 seconds. In the simulations, concentrations are averaged over the response time, and then compared to the appropriate threshold level. Note that averaging over the response time corresponds to an assumption that the SAW desorption cycle is brief relative to its adsorption cycle. In our simulations we ignored the duration of the desorption cycle (i.e., each sensor started integrating the next cycle of data as soon as it reported an alarm or no-alarm condition).
Fig. 3 Sample illustrating conversion of tracer gas concentration to threshold data: (a) concentration data; (b) threshold data without simulated error added; (c) threshold data with simulated error added
Simulations were run using data with and without synthetically added error. For simulations with added error, we generated sensor signals according to the following assumptions: (1) if the actual concentrations are within 25% of the sensor threshold level, the signal will be false 50% of the time; and (2) for concentrations outside of this range, the signal will be false either 10% or 30% of the time, depending on the assumed sensor error. In this implementation of Bayes’ rule, the likelihood function is based on the probability used to generate the false positives and negatives. For example, for data generated with a 30% error, the likelihood is 0.3 when the modeled concentration is
274
A. Gadgil et al.
more than 25% above the threshold level and the sensor has not signaled on; conversely, the likelihood is 0.7 if the sensor has signaled on. For the simulations using data without synthetically added error, we assume 5% error for all data. We did not assign 100% confidence to this data because of inherent uncertainty and variability. In practice the designer of the sensor system should have reliable information on the sensor’s actual rate of false positives and false negatives. Figure 3 illustrates the conversion of measured concentration data to simulated threshold data. Figure 3a shows normalized time-averaged concentration data, with the threshold level indicated. Figure 3b shows the threshold data that would result from a threshold sensor with no error and instantaneous response. Here, “1” signals that the concentration exceeds the threshold. Figure 3c shows the threshold data, corrupted with false negatives or positives. Because the false readings are generated stochastically, different realizations of the data in Figure 3c would exhibit different patterns of output signals. For the case-study examples, we systematically varied the sensor characteristics, as described above. In cases where error was specifically investigated, for each sensor attribute, we generated 50 sets of error-added threshold data, analogous to those displayed in Figure 3c, for each sensor in the system. Each combination of threshold level, response time, and error produced a data stream against which to apply the Bayes Monte Carlo algorithm. The algorithm was used to determine the release location and release magnitude; the time of release was assumed known.
4.4. Results for triggered sensors To demonstrate data interpretation using threshold sensor data, we limit this paper to only showing the ability of the sensor system to estimate the release location. Interested readers are referred to (Sreedharan et al., 2006) for more results and discussion. The information content in threshold sensor data is significantly less than that in direct concentration measurements. Nevertheless, the sensor system can successfully reconstruct the source, at least in some circumstances. We demonstrate this with an example in which the concentration data have been converted to threshold data using a threshold level of 2.3%, a sensor response time of 20 seconds, and without added error. We judge the sensor system performance by its ability to reduce the uncertainty in the estimates of the release location, mass, and duration parameters, and by the time required to do so. Figure 4 depicts the time required to identify the release location (Room 1.2a). At time zero, every zone is assumed to be equally likely as the release location. As sensor data arrive, the Bayes algorithm adjusts these probabilities, locating the release location with greater than 90% confidence within 1 minute. If rapid response hinges on locating a source very quickly, this example suggests that threshold sensors under this network configuration and data quality may be acceptable for real-time monitoring.
Rapid Data Assimilation in the Indoor Environment: Theory and Examples from Real-Time
275
Fig. 4 Probability of source being in location indicated, as estimated with the Bayes Monte-Carlo algorithm using threshold data with response time of 20 seconds, threshold level of 2.3%, and without added error. The actual release location is Room 1.2a. Time is referenced to the instantaneous release event
5. Conclusion Real-time environmental monitoring systems have the potential to help protect building occupants in the event of high-risk pollutant releases. Selection of sensor characteristics is best performed from a systems perspective. Here, we have demonstrated – albeit for a limited set of circumstances – that a network of continuous or single-level threshold sensors can be used to determine the location and magnitude of the release within a Bayes Monte Carlo framework. More importantly, treating the network as a system may lead to better choices for sensor characteristics like response time and error, than might be the case when considering sensors individually. With more complex buildings, system characterization is technically more challenging, and also more expensive. Hybrid methods that augment prior knowledge with sensor system data that monitors building operations to learn about airflows and contaminant transport may improve overall system performance. Such advances would not only be beneficial for designing indoor monitoring systems, but may potentially be extended to help diagnose and interpret data from largescale contaminant releases to the ambient atmosphere and to other environmental media. Such approaches also hold the promise of facilitating improvements in building performance with respect to energy use, thermal comfort, and indoor air quality. Acknowledgments This work was partly supported by the Office of Chemical Biological Countermeasures, of the Science and Technology Directorate of the Department of Homeland Security, and performed under U.S. Department of Energy Contract No. DE-AC03-76SF00098.
276
A. Gadgil et al.
References Brand KP, Small MJ (1995) Updating uncertainty in an integrated risk assessment: conceptual framework and methods, Risk Anal., 15(6), 719–731. Dakins ME, Toll JE, Small MJ, Brand KP (1996) Risk-based environmental remediation: Bayesian Monte Carlo analysis and the expected value of sample information, Risk Anal., 16(1), 67–79. Feustel HE (1999) COMIS – an international multizone air-flow and contaminant transport model, Energy and Buildings, 30, 3–18. Grewal MS, Andrews AP (2001) Kalman Filtering: Theory and Practice, Second Edition. Wiley, New York. Iman RL, Conover WJ (1980) Small sample sensitivity analysis techniques for computer models, with an application to risk assessment, Commun. Stat. -Theor. Meth., A9(17), 1749–1842. Morgan MG, Henrion M (1990) Uncertainty, A Guide to Dealing with Uncertainty in Quantitative Risk and Policy Analysis, Cambridge University Press, New York. Pinsky PF, Lorber MN (1998) A model to evaluate past exposure to 2,3,7,8,TCDD, J. Exp. Anal. Environ. Epidemiol., 8(2), 187–206. Sextro RG, Daisey JM, Feustel HE, Dickerhoff DJ, Jump C (1999) Comparison of modeled and measured tracer gas concentrations in a multizone building. Proceedings of the 8th International Conference on Indoor Air Quality and Climate – Indoor Air 99, Vol 1, pp. 696–701. Indoor Air 99, Edinburgh. Sohn MD, Small MJ, Pantazidou M (2000) Reducing uncertainty in site characterization using Bayes Monte Carlo methods; J. Environ. Eng., 126(10), 893–902. Sohn MD, Reynolds P, Singh N, Gadgil AJ (2002) Rapidly locating and characterizing pollutant releases in buildings. J.Air Waste Manage. Assoc., 52, 1422–1432. Sreedharan P, Sohn MD, Gadgil AJ, Nazaroff WW (2006) Systems approach to evaluating sensor characteristics for real-time monitoring of high-risk indoor contaminant releases. Atmos. Environ., 40, 3490–3502. Wolfson LJ, Kadane JB, Small MJ (1996) Bayesian environmental policy decisions: two case studies, Ecol. Appl., 6(4), 1056–1066.
Discussion B. Denby: Is it also possible for the user to provide in real time other parameters to the assimilation, e.g. which doors are closed, to limit the model ensemble? A. Gadgil: Most certainly, yes. We have tested this in our simulations and find it possible to update other parameters (such as door-positions) in real-time. As with other parameters, the updated values are
Rapid Data Assimilation in the Indoor Environment: Theory and Examples from Real-Time
277
probabilistic. So, one may obtain an evaluation that a particular door is likely closed 70%, and open 30%. A. Venkatram: Does the success of the technique depend on having substantial differences in the outcomes of events? Does the success depend on the uncertainty in the predictive model used to construct the library of possible outcomes? A. Gadgil: (1) If the outcomes for different values of a parameter are substantially the same (within the noise of the observation instruments), this technique will not narrow down that parameter value. This is of course an advantage, since the answers emerging from this technique are probabilistic, and a wide range of parameter values are indicated (rather than zooming down to a single predicted value based on some small noise in the data). (2) The success does depend on the quality of the predictive model used to construct the library. If the model is absolutely no good, then the technique too is pretty useless based on the no-good library of outcome predictions. However, if the model is imperfect but acceptably good (and that applies to almost all models of real world complex phenomena), then the technique is surprisingly useful. B. Rajkovic: The difference in making decision comes only from the difference in the quality of the data (i.e. H, M or L) A. Gadgil: Not quite. The difference in converging to a decision comes from a combination of quality of data (high, medium or low), and a number of data points within the duration when the model predictions are significantly different from one another. For example, if we have low quality data, and collect data points at a low rate, then by the time we accumulate plentiful datapoints, the transients caused by the event have subsided, and most of the data points are indistinguishable from one scenario to the next. This very point is explored in our recent paper cited below, (but not addressed owing to space limitations in my keynote address). “Influence of indoor transport and mixing time scales on the performance of sensor systems for characterizing contaminant releases,” P. Sreedharan, Sohn MD, Nazaroff WW, and Gadgil AJ (in press) Atmospheric Environment.
4.13 A Statistical Approach for the Spatial Representativeness of Air Quality Monitoring Stations and the Relevance for Model Validation Stijn Janssen, Felix Deutsch, Gerwin Dumont, Frans Fierens and Clemens Mensink
Abstract A methodology is presented for the assessment of the spatial representtativeness of air pollution monitoring data. The methodology relies on a statistical approach that links air quality expectation values with land use characteristics. The relevance of this issue for model validation is addressed and the technique is illustrated for the validation of BelEUROS model results.
Keywords Model validation, spatial representativeness
1. Introduction In a small country such as Belgium, air quality levels are sampled by a rather dense network of monitoring stations. More than 70 measuring stations are continuously collecting pollutant concentrations (O3, NO2, PM10,) on a half hourly basis. A great deal of these measuring stations is located in urban, suburban or industrial areas since it is there that air pollution levels are higher and adverse health effects due to air pollution are more important. When these measurements are to be used for air quality assessment throughout the whole region, an essential point that needs to be addressed is the spatial representativeness of the urban and industrial monitoring sites. This certainly holds for those urban or industrial sites that are used for model validation. For those locations it is questionable if point measurements can be directly compared to volume-averaged modelled concentrations. In this paper a technique is presented which addresses this spatial representativeness of monitoring locations. Subsequently, the relevance of this issue for model validations is examined making use of simulations obtained with the BelEUROSmodel. The BelEUROS model is an Eulerian chemistry transport model (Deutsch et al., 2007) that is used as a policy supporting tool in Belgium.
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
452
A Statistical Approach for the Spatial Representativeness of Air Quality Monitoring Stations 453
2. Methodology The methodology presented here for the assessment of spatial representativeness of a monitoring station relies on a general relation between mean air pollution concentration levels at the station and a parameter that determines the land use characteristics in the surroundings of the station. Before the relation as such is discussed, the definition of the land use indicator will be briefly expounded. The land use indicator, hereafter called the E-parameter, is derived from a combination of the CORINE Land Cover data set and a road network with traffic volumes. For a given area (~10 km²) the CORINE Land Cover pixels are determined and classified according to the 44 CORINE classes. The resulting classification histogram can be seen as a spectrum that represents a fingerprint of the land use characteristics in that particular area. As pointed out above, it is the aim to define a single value land use indicator that represents the characteristics of the local behaviour of the air pollution phenomenon. Therefore, the CORINE class distribution is transformed into a land use indicator Eaccording to the relation:
E
ª log «1 « «¬
¦a n ¦n i
CORINE Class i
i
CORINE Class i
i
ª º » b log «1 « » «¬ »¼
¦V l
j j
j
A
º » » »¼
(1)
In this formula, the index i runs over all CORINE classes. nCORINE Class i is the number of pixels of class i inside the specified area and ai is the pollution related coefficient for the CORINE class i. Since traffic is an important source of air pollution and is not very well represented in the CORINE data set, an additional term is added to the definition of the E-parameter. Here Vj is the traffic volume (#cars/time unit) for the line segment j with length lj and j runs over all line segments of the traffic data base inside the study area A. Note that the normalisations in each term ensure a flexible application in different study areas (e.g. a grid cell, a station buffer, …). The parameter b is a weight coefficient that defines the relative importance of the traffic volumes compared to the contribution of the CORINE data. The pollution coefficient ai represents the impact of a particular CORINE class on the air pollution levels. Contributions from different urban, industrial, traffic and natural classes are expected to be important. To reduce the number of free parameters in the methodology, similar classes are joint and only 11 super-classes are retained covering the general land use types such as industrial or commercial units or agricultural areas. For those 11 super-classes, the set of pollution coefficients ai and the traffic parameter b are optimized (see further) for each pollutant individually (e.g. NO2, PM10,…). In order to pin down the relation between the mean air pollutant concentrations and the land use characteristics, the E-parameter is determined inside a 2 km buffer around each monitoring site. After all, it is only at those locations that such a relation can be established since measurement data has to be at hand. When the expectation values are plotted versus the E-parameter for all available monitoring sites, a scatter plot as in Figure 1 is obtained. From this figure, a clear trend is revealed.
S. Janssen et al.
454
Rural stations (low E) have low mean NO2 concentrations, in urban or industrial sites (high E) increased NO2 levels are observed. This relation between land use and air pollution, hereafter called the “trend” function, can be approached by a polynomial fit. The functional form can then be used as an estimator for the pollutant concentration as a function of the land use indicator E Similar “trend” relations are obtained for the pollutants O3 and PM10 and they constitute the core of this methodology. The quality of the fit of the trend function (as RMS error) is used as criteria for the optimisation of the coefficients ai and b in the definition of E in Eq. (1). As a consequence the quality of the trend function is improved by finding an optimal parameterisation of the land use indicator E 90 80 70
3
¢NO2² [Pg/m ]
60 50 40 30 20 10 0 0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
E
Fig. 1 NO2 expectation values at the Belgian monitoring locations as a function of the Eparameter. Rural stations can be found at the left (low E-valuesҏҗ, urban or industrial sites at the right (high E-values). The trend function as a polynomial fit is printed in blue
Up to now, the E-values were determined for the monitoring sites (2 km buffer) only. However, the E-values can also be calculated on a regular grid (here 3 × 3 km²) for the entire Belgian territory. An example of such a E-map is given in Figure 2. In this map, the urbanized and industrialized areas in Belgium are clearly pronounced due to their high E-values. Rural areas like the Ardenne region in the south of Belgium show much lower E-values.
Legend 0.00 - 0.09 0.10 - 0.22 0.23 - 0.34
Fig. 2 E-map for the pollutant NO2 on a 3 × 3 km² regular grid. NO2 monitoring stations are indicated by the red dots
0.35 - 0.48 0.49 - 0.61 0.62 - 0.78 0.79 - 0.93 0.94 - 1.13 1.14 - 1.42 1.43 - 1.91
A Statistical Approach for the Spatial Representativeness of Air Quality Monitoring Stations 455
3. Methodology
3.1. Spatial representativeness of monitoring data In order to assess the spatial representativeness of the monitoring locations, the variability of the E-values in the vicinity of the site is examined. Therefore, all grid cells of the E-map in Figure 2 which fall inside a 7.5 km buffer of the station locations are collected and the variability of the E-values within this buffer is determined. The results for the NO2 stations are presented in Figure 3. For each station in this plot, the mean and standard deviation of the E-distribution within the 7.5 km buffer is printed (blue dots with error bars) in combination with the E-value of the corresponding monitoring site (green diamonds). As additional information, the minimum and maximum E-value in the buffer are also plotted (red triangles). The stations in Figure 3 are grouped according to the classification made by the network managers. This figure contains all relevant information for the assessment of the spatial representativeness of a monitoring site. First of all, it is interesting to explore the variability of the E-values in the buffer. This is expressed by the length of the error bars (one V). If this value is large, a significant variability in the land use characteristics is noticed within a 7.5 km radius. As a consequence it can be expected that the air pollution concentrations in the same area are subject to a similar variability which clearly reduces the spatial representtativeness of the monitoring site. Further, it can be examined how well the mean E-value in the buffer (blue dot) approximates the actual E-value of the monitoring site (green diamond). For a number of stations, both quantities differ to a large extent. In such cases, there is additional evidence for reduced spatial representativeness of the sampling site. After all, if the E-value of the site is far apart from the average land use characteristics in its surrounding area, its sampling values will by no means be representative for this same region. The same variability of the E-parameter can also be explored within buffers with a different radius. This idea is illustrated in Figure 4 where for four different NO2 stations (a rural, an urban background, an urban and an industrial station) the same (statistical) quantities are given for three different buffer sizes. From this plot, it can be deduced how land use characteristics vary in the vicinity of the monitoring site. For the rural and the urban background station, the buffer averages only slightly differ with increasing buffer size and they all correspond rather well with the station value (green diamond). For the urban and the industrial stations, discrepancies become larger and increase with larger buffer radii. This effect can be appreciated as a more limited spatial representativeness of those latter monitoring stations, compared to the rural and urban background ones. Similar plots can be obtained for the ozone and PM10 monitoring sites. Since the E-parameter represents a unique parameterisation for each pollutant, it has to be stressed that the spatial representativeness of a monitoring site highly depends on the type of pollutant. Although not presented here, this issue can also be studied within the presented framework.
S. Janssen et al.
456 RURAL
URB BACK
2.5
URBAN
INDUS
TRAFF
2.5
2.5
2.5
2.5
2
2
2
2
1.5
1.5
1.5
1.5
1.5
1
1
1
1
1
0.5
0.5
0.5
0.5
0.5
0
0
0
0
0
Avg E (Std dev) Min E 2
Max E
5
10 Station No
15
20
5 Station No
10
5 10 Station No
15
-0.5 0
5
10
15 Station No
20
25
41B003 41N043 41R002 41WOL1
-0.5 0
40HB23 40LD01 40LD02 40OB01 40WZ01 42M802 42R010 42R815 42R822 42R830 42R831 42R832 42R891 42R892 42R893 42R894 42R897 43R222 43R240 44M702 44M705 44R721 44R731 44R740 44R750 45R512
-0.5 0
41AND3 41B004 41B005 41B006 41B011 41MEU1 41R001 42R020 42R801 43N070 43R201 43R223 44R701 45R501 45R502
-0.5 0
40ML01 40SZ01 40SZ02 41R012 42N045 42R811 42R821 42R841 44R710
40HR01 42N015 42N016 42N035 42N040 42N046 42N054 43N063 43N066 43N073 43N085 43N093 43N100 43N121 43N132 44N012 44N029 44N050 44N052
E (NO2)
Station E
-0.5 0 5 Station No
Fig. 3 Variability of the NO2 E-parameter in the vicinity (7.5 km radius) of the Belgian monitoring sites. The average values ±1 standard deviation are given as blue dots with error bars. The maximum and minimum E-values in the buffer are given as the red triangles. The E-values of the monitoring sites are printed as green diamonds. Stations are grouped according to the classification given by IRCEL: rural, urban background, urban, industrial and traffic 44R710 (urban background)
45R502 (urban)
42R891 (industrial)
2
2
1.8
1.8
1.8
1.8
1.6
1.6
1.6
1.6
1.4
1.4
1.4
1.4
1.2
1.2
1.2
1.2
1
1
E
2
E
2
E
E(NO2)
43N093 (rural)
1
1
0.8
0.8
0.8
0.8
0.6
0.6
0.6
0.6
0.4
0.4
0.4
0.4
0.2
0.2
0.2
0.2
0
5 7.5 10 Buffer radius [km]
0
5 7.5 10 Buffer radius [km]
0
5 7.5 10 Buffer radius [km]
0
5 7.5 10 Buffer radius [km]
Fig. 4 Variation of the NO2 E-parameter as a function of the buffer size around four Belgian monitoring sites (43N093 – rural, 44R710 – urban background, 45R502 – urban and 42R891 – industrial). Symbols are as in Figure 3 and determined inside a 5, 7.5 and 10 km radius buffer around the monitoring site
A Statistical Approach for the Spatial Representativeness of Air Quality Monitoring Stations 457
3.2. Model validation The methodology developed so far can also be applied to improve model validations. As a matter of fact, a great deal of the air quality sampling sites are located in (sub-) urbanised or industrialized regions. As indicated before, it is questionable whether the sampling values collected at those locations are representative for the volume averaged concentrations calculated by the model in the grid cells. In this paragraph, the methodology will be illustrated with NO2 and PM10 results obtained by the BelEUROS model. The BelEUROS model is an Eulerian chemistry transport model that yields concentration fields on a 15 × 15 km² grid. In order to improve the comparison of model results with measurement values, the relation between air pollution and land use established in Section 2 is applied in a downscaling procedure. Based on this relation, inside each 15 × 15 km² parent model grid cell a mass conserving distribution of the modelled pollutant concentration is performed into 25 daughter 3 × 3 km² cells. The distribution is carried out according to the variability of the E-parameters of the parent BelEUROS grid cells. The result of this downscaling procedure of annual mean NO2 concentrations is presented in Figure 5. By examining the two maps, it can be recognized that mass conservation of the BelEUROS results is fulfilled. When the high resolution map (3 × 3 km² cells) is aggregated again to the low resolution BelEUROS grid (15 × 15 km²), the original map is obtained again. On the other hand it is clear that especially in urbanized regions, much more detail is discovered in the 3 × 3 km² air pollution map which makes a comparison with measurement data more appropriate. A first confirmation of the methodology is obtained by comparing the downscaled high resolution map with an interpolated map of measurement values. In Figure 6, a map of measured NO2 concentrations (annual average for 2002) is given as it is obtained by the interpolation model RIO. Note that here an extended version of the model presented by Hooyberghs et al. (2006) is used. By comparing both maps it is clear that the same patterns around the urban areas are revealed. The improvement of the model validation by using the downscaled results can also be demonstrated numerically by comparing the measurements once with the original low resolution outcome of the BelEUROS model and once with the downscaled high resolution outcome. Both validation results are summarized in Figure 7 where stations are grouped again according to the classification made by the network managers. For most of the rural and urban background stations no much difference is observed between the original and the downscaled model results. For the other station types with a lower spatial representativeness (see Figure 3), a clear improvement is observed. For those stations, the downscaled results are positively biased closer to the measured data. This can be explained by the fact that most of those stations are located in an urbanised area (higher E-value), which is accounted for by the downscaling based on the appropriate land use parameter but not by the original much broader grid cell result. This conclusion is also confirmed in the statistical analysis given in Table 1. The three quality indicators presented in this table show a general improvement in the model validation when the downscaled high resolution model results are used.
S. Janssen et al.
458
The same table also contains the results for PM10 for which a similar approach is adopted. Although the rather large negative bias in the PM10 model results could not be removed, the downscaling procedure has a net positive effect in the model validation skills. In general, the overall improvement of the model validation by taking into account spatial representativeness in the downscaling procedure can be quantified as in the order of 20%.
Legend NO2 [µg/m³]
Legend NO2 [µg/m³] <6
<6
7 - 10
7 - 10
11 - 14
11 - 14
15 - 18
15 - 18
19 - 22
19 - 22
23 - 26
23 - 26
27 - 30
27 - 30
31 - 34
31 - 34
35 - 38
35 - 38
> 38
> 38
Fig. 5 Map of the 2002 annual mean NO2 concentration in Belgium as calculated by the BelEUROS model. The map on the left shows the raw model results in the 15 × 15 km² grid. The map on the right presents the downscaled results on a 3 × 3 km² grid obtained by using the ȕvariability within the parent grid cells. Monitoring station locations are indicated by the black dots
Legend NO2 [µg/m³] <6 7 - 10 11 - 14 15 - 18 19 - 22
Fig. 6 Map of the 2002 annual mean NO2 concentration in Belgium obtained by interpolation of the measurements. The RIO model is used as the spatial interpolation tool
23 - 26 27 - 30 31 - 34 35 - 38 > 38
Table 1 Validation statistics for the annual mean NO2 and PM10 concentrations. Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and bias are obtained by comparing model results with measurements in all monitoring sites. Statistics are given for the original low resolution (15 × 15 km²) BelEUROS results and the high resolution (3 × 3 km²) downscaled results. Pollutant
NO2
Resolution
PM10
High
Low
High
RMSE (µg/m³)
8.11
9.57
8.75
Low 12.01
MAE (µg/m³)
5.88
7.02
7.31
10.37
BIAS (µg/m³)
0.72
–5.07
–6.91
–10.39
A Statistical Approach for the Spatial Representativeness of Air Quality Monitoring Stations 459 URBAN
INDUS
TRAFF
70
70
70
70 60
50
50
40
40
40
40
40
30
30
30
30
30
20
20
20
20
20
10
10
10
10
10
0 0
5 stations
10
0 0
5 stations
0
0
5
10 stations
0 15
0
40LD01 40OB01 40WZ01 42R010 42R822 42R830 42R832 42R891 42R892 42R893 42R894 42R897 43R240 44M702 44R721 44R731 44R740 44R750 45R512
60
50
41B004 41B005 41B006 41B011 41MEU1 41R001 42R020 42R801 43N070 43R201 43R223 44R701 45R501 45R502
60
50
40ML01 40SZ01 40SZ02 41R012 42R821
60
50
40HR01 42N015 42N016 43N066 43N073 43N085 43N100 43N121 43N132 44N012 44N029 44N050
¢NO2²
60
URB BACK
HighRes LowRes Measurement
5
10 stations
15
0 20
41B003 41N043 41R002 41WOL1
RURAL 70
0
5 stations
Fig. 7 Annual mean NO2 concentrations (2002). Measured values are printed as green squares, low resolution BelEUROS model results as blue stars and high resolution downscaled model results obtained by applying ȕ-variability, as red dots. The introduction of ȕ-variability in the downscaling procedure may improve model validation results specially against urban (background), industrial and traffic monitoring sites
4. Conclusions In this paper a methodology is presented for the assessment of spatial representativeness of air quality monitoring data. The methodology relies on general relations between air pollution expectation values and land use characteristics. These characteristics are captured by a single parameter, calculated from the CORINE Land Cover set complemented with traffic data and optimized per pollutant. The technique is applied here for the NO2 measuring stations but is applicable for other pollutants as well. The methodology developed for the assessment of spatial representativeness can also be applied to improve model validations. The technique was illustrated by making use of NO2 and PM10 concentration fields of the BelEUROS model. It was shown that model validation can be improved by downscaling model results using the same single parameter which captures the land use characteristics. Overall, the positive effect is estimated to be of the order of 20%.
References Deutsch F, Janssen L, Vankerkom J, Lefebre F, Mensink C, Fierens F, Dumont G, Roekens E (2007) Modelling changes of aerosol compositions over Belgium and Europe, Int. J. Environ. Pollut., 32, 162–173. Hooyberghs J, Mensink C, Dumont G, Fierens F (2006) Spatial interpolation of ambient ozone concentrations from sparse monitoring points in Belgium, J. Environ. Monit., 8, 1129–1135.
460
S. Janssen et al.
Discussion A. Venkatram: Your high correlations between the beta parameter (which depends primarily on land-use in the vicinity of the monitoring station) and concentrations of NO2, O3, PM10 suggests that you do not need an air quality model to estimate concentrations of these species. Could you provide some physical rationale for these extraordinary results? S. Janssen: It is important to stress that the high correlations between the beta parameter and the pollutant concentration only hold for annual (or long term) averages. In this way, the variable meteorological influence on the local air quality levels is filtered out and the correlations express a relation between average air pollutant concentrations and sources of primary emission and precursors of secondary components. J. Pleim: Does the chemical transport model have any sub grid information? S. Janssen: The belEUROS model used as a CTM in this study contains as sub grid information a detailed land use model for the parameterisation of dry deposition processes and some (biogenic) emission sources. R. Mathur: S. Janssen:
Is there a way to estimate the actual spatial representativeness scale (in terms of distance of area) from the parameters E? Based on the relation between the beta parameter and the average concentration levels, it is possible to estimate the actual spatial representativeness of a monitoring station. Given a certain acceptance interval for the average concentration level (e.g. 20%), the trend function gives a corresponding acceptance interval for the beta values. This interval can then be used to identify the area of representativeness (with corresponding beta values) in the surroundings of the station. Note that within this approach the area of representativeness is not necessarily circular but can have an erratic shape.
4.3 Air Quality Ensemble Forecast Coupling ARPEGE and CHIMERE over Western Europe Ana C. Carvalho, Laurent Menut, Robert Vautard and Jean Nicolau
Abstract The quality enhancement of the results encountered on numerical weather prediction ensemble runs has encouraged the air quality modellers’ community to test the same methodology to foresee air pollutants concentrations in the atmosphere. In air quality forecast it is important to know in advance if the event exceedences of a certain threshold value will happen in order to implement mitigation measures concerning air pollutant emission. The ensemble approach allows giving this information within a probability range. Within this work both perturbation on the circulation model and the chemical transport model will be implemented. The ensemble system is composed by the numerical weather prediction model ARPEGE, the meteorological model MM5 and the chemical transport model CHIMERE. Meteorological perturbations will be addressed firstly by a set of ten ensemble members derived by the ARPEGE model, plus a control run, which will force MM5 simulations. Since the concept of air pollution ensemble forecast is not the same than the one for meteorology, we propose here an original approach for the chemistry-transport model perturbations based on previously done CHIMERE sensitivities studies: focus will be made on plausible emissions scenarios and different daily emissions profiles, boundary layer evolution, vertical mixing and photolysis rates. The ensemble based simulations will cover July and August 2006, each includes the heat wave period that influenced the weather and air quality conditions of central Europe. Keywords Chemical, ensemble, forecast, ozone
1. Introduction The chaotic nature of the atmosphere arises in the numerical resolution of the equations that present different solutions for slightly different values of the initial conditions. Based on the definition of the chaotic dynamical system given by Lorenz (1963), the ensemble forecast concept was initially used to estimate the dependence of weather forecasts uncertainty to the initial boundary conditions errors (Hamill et al., 2000) by the ECMWF on Europe, and by the NCEP on the United States of America. Nowadays numerical weather prediction based on operational ensembles C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
367
368
A. Carvalho et al.
are available in many meteorological institutions because the mean of ensemble results have shown that this methodology provides better accuracy of weather forecast than the application of any single model (Delle Monache et al., 2006 and references within). Due to the encouraging results encountered on ensemble predictions results from NWP, Dabberd and Miller (2000) suggest also that air quality applications will benefit from probabilistic simulations, since they permit to estimate the likelihood of a certain event and its associated probability instead of a single value produced by a deterministic model. Galmarini et al. (2004) discusses the relevance of this type of results to the decision-making process showing that ensemble analysis and its variability reduces the risk of unreliability on the ensemble results and enriches the decision making process, since the available information is improved.
2. Air quality Modelling System The objective of this study is to incorporate the ensemble chemistry forecast into the operational mode giving this way a value for the pollutants concentrations and also the probability of its occurrence. Hence, the chosen air quality system is based on the actual one that is in used, in operational mode, in the PREV’AIR system (URL 1) since 2003. It includes the MM5 modelling system (Dudhia, 1993) and the CHIMERE chemical transport model (Vautard et al., 2001). The EMEP (URL 2) anthropogenic emissions of NOx, SO2, CO and non-methane volatile organic compounds (NMVOC) are calculated and interpolated to the CHIMERE grid. Concerning the meteorological input of this air quality system, global data are taking form the ARPEGE model used in its ensemble framework called PEARP. The ensemble comprises 11 forecast runs, a not perturbed control run, and 10 runs were the initial conditions are perturbed. The model runs with the same horizontal and vertical resolution in both modes, operational and ensemble forecast. The PEARP system is launched once a day for three days and results are considered as initial and boundary conditions for the MM5 model. The MM5 physical parameterisations used here are those described in (Chiriaco et al., 2005). The CHIMERE model settings are strictly those described in (Vautard et al., 2006). Each day, CHIMERE starts at 18H00 UTC (at D0), runs up to 24H00 at D + 2. The domain under study covers the Western Europe, with 54 km resolution, approximately, and 8 vertical levels from surface to 500 hPa. The simulation period ranges from 19 July to 3 August in order to cover the whole highly polluted period (23 July–3 August) with a reasonable chemical spinup time.
Air Quality Ensemble Forecast Coupling ARPEGE and CHIMERE over Western Europe
369
3. Methodology An ensemble design to be applied to air quality applications has more probabilities to succeed if one or more type of uncertainties sources are included, and hence if different parts of the numerical system are disturbed. In the present work, previous studies with the adjoint version of the CHIMERE model where taking into account, in order to get an insight of the factors to which the CHIMERE model was demonstrating to be more sensitive (Menut, 2003). Concerning the sensitivity of the modelled ozone maximum (afternoon), CHIMERE shown to be mainly sensitive to morning solvent and traffic emissions as well as to ozone boundary conditions and some reactions rates (oxidation of NO by O3 and by OH, the equilibrium reactions established between the acetaldehyde and the NO2, the oxidation of an aromatic group by the hydroxyl radical). A focus was also made on the NO2 photolysis rate. Concerning meteorological parameters temperature, wind speed and vertical diffusion coefficients are the ones that strongly affect the mode final results on ozone, Ox concentration in the afternoon and NO2 during the morning. Based on this prior knowledge 24 ensemble members were constructed. In this experiment the 11 meteorological ensemble members calculated by MM5, and driven by the PEARP system. Thirteen other ensemble members are derived by emissions, chemical or chemical related considerations. These perturbations were based on the meteorological PEARP-MM5 control run. Two different years for the EMEP emission inventory – 2002 and 2003 were chosen and the remainder perturbations were performed based on the 2003 emissions inventory. Vertical emissions distributions were taken into account considering different disaggregating factors between two and three levels. Since the actual validated operational runs with two vertical levels for annual emissions all the remainder disturbances were performed attending this consideration. Namely, a lag of one hour (±) on the hourly profiles for all the activity sectors; VOC emissions from solvent use perturbed (±40%, based on uncertainties emissions estimation taken from Theloke and Friderich (2000) and the IPCC (2006)), traffic NOx emissions were considered to vary between ±20% (Kühlwein and Friedrich, 2000). Random hourly perturbations are applied to the gas chemical species at the boundaries and to the attenuation coefficient that is calculated into the CHIMERE meteorological diagnostic preprocessor. The traffic behaviour at Mondays and Fridays during the morning was also included in one chemical ensemble member, in this case multiplying factors without considering emitted mass conservation from the inventory were applied between 5H00 and 8H00 UTC. Figure 1 summarises all the perturbations considered into the ensemble air quality system. Weather conditions and observed ozone concentrations during the simulation period Between the 23 of July and the 3 of August Europe as experienced several days of high concentrations of maximum ozone values daily values. These high values started to appear over Italy, and eastern part of Italy; then the high maximum ozone plume is produced over the central and western part of Europe, and finally it
A. Carvalho et al.
370
disappears from southern European countries around the 3 of August. It coincides with the end of the heat wave of 2006 that has affected several countries. In order to capture these air pollution events over the domain of simulation, the air quality systems was “speened-up” for a period of five days, the period of simulations started with the meteorological forecast 19th of July, at 18H00 UTC ending the forecast that starts at the 3rd of august.
Fig. 1 Disturbances introduced into the ensemble design
Results The BIAS, the root mean square error (RMS) and the Pearson correlation coefficient (COR) statistical parameters were applied in model validation. In order to have a robust measure of the central value of the simulated concentration distribution at each locations, model results were evaluated comparing the median of the ensemble results at each station. The ozone stations database comprises several European countries: France, United Kingdom, Belgium, Suisse, The Netherlands, Germany the Check Republic and Spain. Unfortunately, during the period under evaluation the information of most of the countries is not present and only France and Germany are represented in this study. Tables 1–3 summarise the distribution of the RMS, COR and BIAS calculated. The number of stations may vary according to the existence or not of measured data. The skill of the model was evaluated considering the type of the station, namely, rural, peri-urban and urban. The air quality model systems tend to present an error magnitude between (Table 1) 30 and 40 µg m-3 for all type of stations. The second more frequent class of error is between 40 and 50 µg m-3 at D1 and D2 for urban and peri-urban site locations. Globally, it is on D2 at rural sites that the system presents small errors on the ozone forecast magnitude.
Air Quality Ensemble Forecast Coupling ARPEGE and CHIMERE over Western Europe
371
The BIAS score is helpful in order to understand of the model over or under predicts the ozone concentrations. Regarding Table 3, results on the deviation between model results and observations indicates that the model under-predicts the measured ozone concentrations. Model errors spans –60 and +40 µg m-3. Due to the model system horizontal grid resolution, rural stations present errors of 20% for 84% for forecast day D1. Over most of the peri-urban and also over urban sites the model under predicts the measured value between –30 and –20 concentration units. But, it is in the urban sites that the air quality system tends to over predict ozone concentrations. Differences between D1 and D2 do not seem very significant for this parameter. The model system is able to better reproduce maximum and minimum values of the measured ozone tome series for peri-urban stations sites (Table 2). More then 86% of the station have correlation coefficients between 0.7 and 0.8. Results for rural and urban present correlation coefficients results are preferable encountered between correlation coefficient values between 0.6 and 0.8. The degradation of the forecast for D2 is more visible for peri-urban sites were the percentage of locations with correlation coefficients between 0.7–0.8 increases. Table 1 RMS distributions of skill score for D1 and D2 forecast days (µg m-3) calculated for Rural, peri urban (PURB) and urban (URB) stations, where: % – percentages and # – number of stations. RURAL
Classe % D1
PURB #
D2
D1
% D2
D1
URB #
%
D2
D1
D2
D1
# D2
D1
D2
10 d RMS < 20
4.9
4.9
4
4
4.4
5.1
3
3
–
20 d RMS < 30
20.7
25.6
17
21
22.1
22.0
15
13
15.3
15.3
9
9
30 d RMS < 40
48.8
43.9
40
36
33.8
37.3
23
22
47.5
44.1
28
26
40 d RMS < 50
22.0
22.0
18
18
27.9
27.1
19
16
25.4
30.5
15
18
50 d RMS < 60
3.7
2.4
3
2
11.8
8.5
8
5
8.5
10.2
5
6
60 d RMS < 70
1.2
1
3.4
–
2
The score calculated by Honoré et al. (JGR, submitted) for the PREV’AIR system, as a mean of all stations between 2003 and 2006 summer seasons for forecast D0 are: BIAS 12.3, RMS 28.2 and COR 0.67. In the present study the system was evaluated under a situation were high ozone concentration were measured during 2006. This explains the higher values for the RMS and the under prediction of model results indicated by the BIAS. In general, correlation coefficients are of the same order for all type of stations. Regarding the spatial distribution of the coefficient correlations for forecast day D1, it can be observed that the model is unable to reproduce maximum and minimum values of the time series mostly over the North and Central-eastern part of Germany (see Figure 2). The Auvergne region and the Mediterranean Pyrenees are the locations over France were the correlation coefficients show also results with values lesser than 0.5.
A. Carvalho et al.
372
Table 2 COR distributions of skill score for D1 and D2 forecast days (µg m-3) calculated for Rural, peri urban (PURB) and urban (URB) stations, where: % – percentages and # – number of stations. RURAL
Classe % D1 0.1 d COR < 0.2
PURB #
D2
D1
1.2
URB
%
D2
D1
#
D2
D1
%
D2
D1
#
D2
D1
D2
1
0.2 d COR < 0.3 1.2
1
–
1.7
1
0.3 d COR < 0.4 1.2
1.2
1
1
–
4.4
3
1.7
1.7
1
1
0.4 d COR < 0.5 1.2
3.7
1
3
2.9
2.9
2
2
8.5
1.7
5
1
0.5 d COR < 0.6 12.2
11.0
10
9
4.4
2.9
3
2
5.1
16.9
3
10
0.6 d COR < 0.7 22.0
23.2
18
19
5.9
8.8
4
6
30.5
35.6
18
21
0.7 d COR < 0.8 42.7
45.1
35
37
48.5
61.8
33
42
45.8
33.9
27
20
0.8 d COR < 0.9 19.5
14.6
16
12
38.2
19.1
26
13
8.5
8.5
5
5
0.9 d COR < 1.0 –
–
–
–
–
Table 3 BIAS distributions of skill score for D1 and D2 forecast days (Pg.m-3) calculated for Rural, peri-urban (PURB) and urban (URB) stations, where: % – percentages and # – number of stations. RURAL
Classe % D1
D2
D1
1.2
–60 d BIAS < –50
PURB #
% D2
D1
D2
URB #
D1
% D2
1
D2
1.7
D1
D2
1
–50 d BIAS < –40
1.2
2.9
1.5
2
1
5.1
3.4
3
2
–40 d BIAS < –30
3.7
3.7
3
3
20.6
18.2
14
12
13.6
16.9
8
10
–30 d BIAS < –20
11
9.8
9
8
27.9
33.3
19
22
20.3
20.3
12
12
–20 d BIAS < –10
32.9
37.8
27
31
27.9
27.3
19
18
18.6
16.9
11
10
–10 d BIAS < 0
35.4
31.7
29
26
13.2
16.7
9
11
16.9
22.0
10
13
0 d BIAS < 20
15.8
15.9
13
13
4.4
3.0
3
2
23.7
14
12
20 d BIAS < 40
1
D1
#
2.9
2
The score calculated by Honoré et al. (JGR, submitted) for the PREV’AIR system, as a mean of all stations between 2003 and 2006 summer seasons for forecast D0 are: BIAS 12.3, RMS 28.2 and COR 0.67. In the present study the system was evaluated under a situation were high ozone concentration were measured during 2006. This explains the higher values for the RMS and the under prediction of model results indicated by the BIAS. In general, correlation coefficients are of the same order for all type of stations. Regarding the spatial distribution of the coefficient correlations for forecast day D1, it can be observed that the model is unable to reproduce maximum and minimum values of the time series mostly over the North and Central-eastern part
Air Quality Ensemble Forecast Coupling ARPEGE and CHIMERE over Western Europe
373
Fig. 2 Pearson’s correlation coefficient for rural stations for forecast D1
of Germany (see Figure 2). The Auvergne region and the Mediterranean Pyrenees are the locations over France were the correlation coefficients show also results with values lesser than 0.5.
4. Conclusions A meteorological chemical air quality model ensemble system was planned and evaluated during the heat wave of 2006 over Western Europe. This system was designed based on ensemble meteorological runs and also on the previous knowledge form chemical adjoint results on most sensitivity parameters to the model. The median of the air quality system ensemble results were compared to the observations. For all the calculated skill scores classes, the degradation of the forecast skill between D1 and D2 is not very significant. In general, air quality model results are under predicted, more frequent RMS error values are between 30 and 40 Pg m-3. Pearson correlation coefficients have best fits over peri-urban sites, indicating that the model is able reproduce better the maximum and minimum of the results distributions at these sites. Concerning spatial distribution of the COR at forecast D1, the variability introduced by the ensemble members is not sufficient to reproduce time series in the northern, near the coast, and central-eastern part of Germany as well as over the French mountain regions of Auvergne and Mediterranean Alpes. These results may be justified by the coarse model resolution.
374
A. Carvalho et al.
References Chiriaco M, Vautard R, Chepfer H, Haeffelin M, Wanherdrick Y, Morille Y, Protat A, Dudhia J (2005) The ability of MM5 to simulate Ice clouds: systematic comparison between simulated and measured fluxes and lidar/radar profiles at SIRTA atmospheric observatory. Month. Wea. Rev., 134, 897–918. Dabberdt WF, Miller E (2000) Uncertainty, ensembles and air quality dispersion modeling: applications and challenges, Atmos. Environ., 34, 4667–4673. Delle Monache L, Deng X, Zhou Y, Stull R (2006) Ozone ensemble forecasts: 1. A new ensemble design, J. Geophys. Res., 111, D05307, doi:10.1029/ 2005JD006310. Dudhia J (1993) A nonhydrostatic version of the Penn State – NCAR mesoscale model: validation tests and simulation of an Atlantic cyclone and cold front. Month. Wea. Rev., 121, 1493–1513. Galmarini S et al. (2004) Ensemble dispersion forecasting – Part I: concept, approach and indicators, Atmos. Environ., 38, 4607–4617. Hamill TM, Mullen SL, Snyder C, Toth Z, Baumhefner DP (2000) Ensemble Forecasting in the Short to Medium Range: Report from a Workshop, Bull. Am. Meteorol. Soc., 81, 2653–2664. Honoré C et al. Predictability of regional air quality in Europe: the assessment of three years of operational forecasts and analyses over France, J. Geophys. Res., under revision. IPCC (2006) IPCC Guidelines for National Greenhouse Gas Inventories (http://www.ipcc-nggip.iges.or.jp/public/2006gl/pdf/3_Volume3/V3_5_Ch5_No n_Energy_Products.pdf) Kühlwein J, Friedrich R (2000) Uncertainties of modelling emissions from road transport. Atmos. Environ. 34, 4603–4610. Lorenz EN (1963) Deterministic non-periodic flow. J. Atmos. Sci., 20, 130–141. Menut L (2003) Adjoint modeling for atmospheric pollution process sensitivity at regional scale, J. Geophys. Res., 108(D17), 8562, doi:10.1029/2002JD002549. Theloke J, Friedrich R (2002) NMVOC Emissions from Solvent Use in Germany 2000. Annual Report 2001 of the EUROTRAC subproject Generation and evaluation of emission data – GENEMIS. Munich 2002. Vautard R, Beekmann M, Roux J, Gombert D (2001) Validation of a deterministic forecasting system for the ozone concentrations over the Paris area. Atmos. Environ., 35, 2449–2461. Vautard R et al. (2006). Is regional air quality model diversity representative of uncertainty for ozone simulation? Geophys. Res. Lett., 33, L24818, doi:10.1029/2006GL027610. URL 1: http://prevair.ineris.fr/en/introduction.php URL 2: http://www.emep.int
Air Quality Ensemble Forecast Coupling ARPEGE and CHIMERE over Western Europe
375
Discussion S.T. Rao: Are you implementing ensemble modeling in operational forecasting of air quality? A. Carvalho: Although that is the ultimate objective for the moment we are on a test phase concerning the best sources of uncertainty to be introduced in the ensemble model system. S.T. Rao: How do you communicate probability of exceeding a certain threshold concentration to the general public in your move towards probabilistic modeling? A. Carvalho: We believe we need to think about that a little bit more. In particular we need to learn about with the meteorological experience gathered until today. Over some regions it could be very important to make an alert with a low probabilistic forecast and over other regions a high probabilistic forecast may introduce unnecessary measures. A. Venkatram: Is there any empirical evidence to indicate that the median of the ensemble predictions is “better” than the control estimate? Clearly there is no formal proof for this result. A. Carvalho: There are some studied stations were the median of the ensemble prediction is not better than the control estimate. Nonetheless, the results have shown that this is not true for all station neither for all the meteorological conditions that drives the ozone formation. Also in this point we will need to quantify the performance of the model for control and the ensemble average estimates. B. Rajkovic: In ensemble approach having meteorological and chemistry models is it possible to see which of the two has more uncertainties i.e. which creates larger spread in the final forecast? A. Carvalho: For a blocking meteorological situation over Western Europe (the case of June 26t, 2006, presented here), the analysed surface ozone fields have shown that the meteorological perturbations induced the larger spread, but the high ozone values over the domain are only attained with the disturbances introduced into the CHIMERE model.
4.7 Application of a Regional Atmospheric Emission Inventory to Ozone and PM Modelling over the French North Region: The summer 2006 Heat Wave Case Study E. Terrenoire and V. Fèvre-Nollet
Abstract This paper reports on the first application of a Nord-Pas-de-Calais regional emission inventory to the CHIMERE-MM5 modelling system. Using statistical indexes, we compare computed data to hourly observations and daily maximum of ozone and PM concentration measured with the monitoring station network data period in order to evaluate the performance of the modelling system for the selected area. This comparison is presented for a coastal location (Dunkerque: 51.03° N/2.37° E) and an inland location (Lille: 50.63° N/3.07° E) during the June July summer 2006 period. The model gives an excellent reproduction of the daily ozone cycle both for inland and coastal region but underestimates (20 µg/m3 on average) the O3 hourly maxima over 180 µg/m3. The model underestimates the PM10 measured concentration and has difficulties to reproduce hourly maxima over 60 µg/m3 for both inland and coastal stations. The system has difficulties to reproduce O3 concentration when air masses flows are continental. The poor representation of monitoring coastal station for particles makes the system validation more delicate for this area. We recommend using high resolution (3 km) and grid nudging for the production meteorological data for the coastal location due to specific local dynamical conditions.
Keywords Atmospheric regional modelling, costal area, inland area, model evaluation, O3, PM10, regional emission inventory
1. Introduction Photochemical models are largely used for the study of air pollution. It is a key tool for the decision maker in the development of strategies to control and reduce both photochemical and particular pollution. Examples of Chemistry Transport Models (CTMs) are the Community Multiscale Air Quality Model (CMAQ) (Byun et al., 1998), Meso-NH (Tulet et al., 2003). They are generally coupled to meteorological models such as the Fifth-Generation Pennsylvania State University/National Center C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
404
Application of a Regional Atmospheric Emission Inventory to Ozone and PM Modelling
405
for Atmospheric Research (NCAR) Mesoscale Model (MM5) (Grell et al., 1994) which provide 4D meteorological data. Meteorological data can also be calculated online with models such as the Meso-NH model. Recent evaluations of those models for their abilities to reproduce and predict the hourly photoxydant and particular tropospheric pollution have been performed over different cities: using MM5CMAQ (San Jose et al., 2002; Sokhi et al., 2006; Vautard et al., 2003). We propose in this study to evaluate the capacity of the MM5-CHIMERE modelling system to reproduce the O3 and PM10 ground level concentration over the north region of France during the 2006 summer June–July period with a zooming over the Lille (50.63° N–3.07° E) and Dunkerque (51.03° N–2.37° E) areas being respectively an inland and a coastal location.
2. Tools
2.1. The CHIMERE model CHIMERE runs over a range of spatial scales from the continental (1,000 km) to the meso-scale scale (100 km) with resolutions varying from 1 to 100 km. It requires meteorological data, boundary conditions, land-use information and biogenic and anthropic emissions. The model offers the option to use different gas phase mechanisms. For this study we use the MELCHIOR2 mechanism describing 44 species and 116 reactions (Derognat, 1998) following the concept of chemical operator. Due to the high PM potential emission of our studied area (Figure 1) the aerosol option was used (Bessagnet et al., 2004). Anthropogenic ground emissions are taken from the 2003 EMEP database (http:/w / ww.emep.int). Those data have a 0.5 x 0.5 °resolution. For the Nord-Pas de Calais region we used the 2004 updated regional emission inventory data (Martinet, 2004). Resolution is 1 km2. The benefit of such emission inventories to model outputs results has been shown by Terrenoire et al. (2007). For more details on the CHIMERE description, please refer to the official CHIMERE web site: http://euler.lmd.polytechnique.fr/chimere
2.2. Meteorological input data The PSU/NCAR mesoscale model, known as MM5, is a non-hydrostatic, terrain following sigma-coordinate designed to simulate or predict mesoscale atmospheric circulation. More information on MM5 is given at http://www.ncar.ucar.edu. The MM5 model version used in this work is the 3.7.3 version. It includes classical parameterizations such as Planetary Boundary Layer (PBL), radiation, microphysics and convection. We used the FNL reanalysis data sets (every 6 hours, 1 × 1q resolution) downloaded from National Center for Environmental Prediction (NCEP)
406
E. Terrenoire, V. Fèvre-Nollet
to provide initial and boundary conditions for MM5. The model is designed with three nested domains (Figure 1) and 23 vertical layers stretching from surface to 100 hPa. The 25 × 35 horizontal mother grid is centred over the city of Dunkerque (51.03° N–2.37° E). The 9 km resolution inner domain is used to provide 4D meteorological data to CHIMERE.
3. Methodology
3.1. Domain of simulation The Nord Pas de Calais region is part of a North West European area which runs from South-East England to the Rhine. It has 4 million inhabitants, one of the highest density in Europe. A dense road infrastructure and a high industrial activity have yielded an alarming decrease of the air quality in this area over the last 20 years. For this case study, we present the results using domain 2 (Figure 1) with the lower left and upper right grid box centre at (1.0° E/49.5° N) and (4.5° E/51.5° N) respectively. For information the modelled data are extracted from the first layer which is about 40 m deep.
3.2. Statistical indexes used We evaluate the MM5-CHIMERE modelling system by quantifying the agreement between predicted and observed values using the following statistical indexes: the correlation coefficient (CC), the Normalized Mean Square Error (NMSE), the Index of Agreement (IA) and the fractional Bias (FB). CC, NMSE and IA provide a measure of the correlation of the predicted and measured times series of concentrations. IA varies from 0 to 1 which represents the perfect agreement between the observed and calculated values. FB represents a measure of the agreement between the mean predicted and observed concentrations. Perfect agreement between the two sets of data is reached when NMSE = FB = 0.
4. Results and Discussion
4.1. Meteorological analysis of the modelled period High temperatures have been recorded by Météo-France over the June–July 2006 period in France. A heat wave has affected the north of France during the last three
Application of a Regional Atmospheric Emission Inventory to Ozone and PM Modelling
407
weeks of July 2006. Monthly mean anomalies were between 4°C and 5°C for July and make this month the second hottest one since 1950 after August 2003. The records of the Lesquin suburban station (near Lille) show three main periods with maximum daily temperature above 25°C. The Dunkerque coastal station shows similar tendencies with three main periods where maximum temperatures are above 25°C Drops in temperature (e.g.: 14 , 25 June and 05, 20 July) are generally linked, for both locations, with precipitation due to thunderstorms triggered by cold fronts penetrating inland. It is worth noting that July minimum temperatures were always above 15°C. Computed backward trajectory showed during North-Easterly flow period that air masses arriving over North of France were coming from central Europe which implies that those air masses were possibly already charged with pollutants (e.g., O3).
Fig. 1 Domain architecture for the MM5-CHIMERE system used for this application. MM5 Domain 1: 25 × 35 × 23 grid cells and 81 km resolution. MM5 domain 2: 28 × 37 × 23 grids cells and 27 km resolution. MM5 domain 3: 40 × 55 × 23 grid cells and 9 km resolution. Chimere domain 1: 55 × 40 × 8 grid cells and 10 km resolution. Chimere domain 2: 71 × 41 × 8 grids cells and 5 km resolution
4.2. Evaluation of the MM5-CHIMERE modelling system 4.2.1. Comparison of Hourly Measured and Modelled O3 Concentration for an Inland and Coastal Station
Comparison of the calculated and observed hourly maxima ozone values’s time series averaged over eight inland stations around Lille shows that the model reproduced the daily O3 cycle well, including minima. The model also catches well the decrease of O3 concentrations due to the diminution of photochemical activity and wet deposition when cold fronts are advected over the north of France. There
E. Terrenoire, V. Fèvre-Nollet
408
200
200
180
180
160
160
C a lc u la t e d ( µ g /m 3 )
Calculated data (µg/m3)
are three main periods where the model is underestimating the maxima. They correspond to high temperature periods describe. They are linked to continental North-East or Easterly flow air. The fact that those air masses are already charged with pollution (NOx, O3) could explain the underestimation of the model output. Similar results are found for the coastal location. Figure 2 shows scatter plots of observed versus modelled maximum O3 concentration for an inland and coastal station. Overall, 80% of the values lie within the factor of the two dashed line. However, over 180 µg/m3 for inland station and 160 µg/m3 for coastal station, the model is underestimating the measured values by about 50 and 80 µg/m3 respectively.
140 120 100 80 60
140 120 100 80 60
40
40
20
20 0
0 0
20
40
60
80
100
120
140
Measured data (µg/m3)
160
180
20
0
20
40
60
80 100 120 140 160 180 200 Measured (µg/m3)
Fig. 2 Scatter plots of observed versus daily modelled maxima O3 concentration for a suburban inland station (Salomé 50.53° N/2.84° E on the left) and a suburban coastal station (Petite-Synthes 51.01° N/2.33° E on the right)
4.2.2. Comparison of Hourly Measured and Modelled PM10 Concentration for an Inland and Coastal Station
Comparison of the calculated and observed hourly maxima PM10 values’s time series averaged over six inland stations around Lille shows that the model reproduces the mean tendency of the PM10 measured concentration but underestimate it systematically. Difficulties concerning the reproduction of PM10 peaks can be explain by the fact that PM10 emission can be very localised (iron and steel factories) and that the model is averaging concentrations over 5 km grid boxes. Similar results are found for the coastal location. The measured maxima PM10 concentrations are more frequent and higher than for the inland location due to the dense industrialization of the area. Most of the peaks are reproduced but underestimated except for two of them. Figure 3 shows scatter plots of observed versus modelled maximum PM10 concentration for an inland and coastal station. The under-estimation of PM10 measured concentration is evident. For inland location the model is able to reproduce PM10 concentration bellow 50 µg/m3 but underestimated higher
100
100
90
90
80
80
C alcu lated d ata (µg /m 3 )
Calculated data (µg/m 3)
Application of a Regional Atmospheric Emission Inventory to Ozone and PM Modelling
70 60 50 40 30 20
409
70 60 50 40 30 20 10
10
0
0 0
10
20
30 40 50 60 70 Measured data (µg/m 3)
80
90
100
0
10
20
30 40 50 60 70 Measured data (µg/m 3)
80
90
100
Fig. 3 Scatter plots of observed versus daily modelled maxima PM10 concentration for an urban inland station (Marcq 50.67° N/3.08° E on the left) and a traffic coastal station (Calais 50.95° N/1.87° E on the right)
measured values while for coastal location the ability for the model to predict PM10 concentration is very variable in time.
4.3. Statistical evaluation results Statistical indexes are great tools to evaluate the model performance for specific location. However, it cannot be the only indicator of the model’s success. The model averaged concentration over a 5 km grid cell in our case. Some of the stations are too close to a local source pollutant which can explain the difficulties that have been observed for the model to predicted observed values. This is particularly true for primary pollutant and particles. Results of the statistical comparison are shown in Table 1 for O3 and 2 for PM10. Coastal locations are shown in bold and CC based on maximum values are shown in bracket. The correlation coefficients are satisfactory for the inland location (0.55 on average) and very good for maxima value (0.75 on average). For coastal stations the CC is very variable (0.31–0.53). Suburban stations are more representative for O3 value for both inland and coastal location while urban stations are for PM10 concentration. The IA between modelled and measured data is about 0.7 for O3 values and 0.5 for PM10 values. The FB is low and often negative (on average –0.1 for O3 and –0.5 for PM10) indicating the under-estimation model tendency which is stronger for PM10. NMSE is quite low (0.25) for O3 and indicates the good performance of the model while they are higher for PM10 (1.10).
E. Terrenoire, V. Fèvre-Nollet
410
Table 1 Statistical indexes for O3 during the June–July period. Station Name Lesquin Baisieux Halluin Salomé Graveline Outreau Sangatte Petite-Synthes Armentière Tourcoing Marcq Lomme Calais
Type Suburban Suburban Suburban Suburban Suburban Suburban Suburban Suburban Urban Urban Urban Urban Traffic
CC 0.57 (0.78) 0.55 (0.68) 0.56 (0.82) 0.65 (0.78) 0.45 (0.07) 0.31 (0.72) 0.38 (0.22) 0.53 (0.45) 0.56 (0.80) 0.51 (0.77) 0.57 (0.74) 0.55 (0.69) 0.42 (0.19)
NMSE 0.26 0.24 0.27 0.24 0.42 0.17 0.19 0.37 0.26 0.34 0.26 0.26 0.28
IA 0.72 0.74 0.72 0.74 0.63 0.55 0.62 0.76 0.70 0.63 0.70 0.71 0.71
FB –0.10 –0.04 –0.10 –0.07 0.02 –0.01 0.19 –0.15 –0.16 –0.25 –0.14 –0.12 –0.02
Table 2 Statistical indexes for PM10 during the June–July period. Station Name Salomé Graveline Outreau Sangatte Petite-Synthes Lille-Faidherbe Lakanal Tourcoing Marcq Lomme St Pol Nord Grande-Synthe Calais Boulogne centre Dunkerque Fort-Mardyck Mardyck
Type Suburban Suburban Suburban Suburban Suburban Urban Urban Urban Urban Urban Urban Urban Traffic Traffic Centre traffic Industrial Industrial
CC 0.16 (0.40) 0.25 (0.36) 0.17 (0.25) 0.18 (0.40) 0.11 (0.04) 0.30 (0.57) 0.24 (0.53) 0.21 (0.54) 0.28 (0.41) 0.22 (0.39) 0.16 (0.13) 0.05 (0.33) 0.16 (0.16) 0.14 (0.39) 0.19 (0.22) 0.07 (0.10) 0.16 (0.24)
NMSE 0.90 0.85 1.18 0.66 1.03 1.70 0.58 0.77 0.74 1.00 0.88 1.33 0.71 1.58 0.78 1.29 1.21
IA 0.53 0.46 0.58 0.71 0.46 0.52 0.61 0.48 0.57 0.44 0.26 0.23 0.66 0.46 0.57 0.40 0.38
FB –0.44 –0.37 –0.54 –0.31 –0.39 –0.66 –0.34 –0.41 –0.41 –0.51 –0.43 –0.43 –0.24 –0.64 –0.37 –0.55 –0.42
5. Summary Conclusion The ability of the MM5-CHIMERE modelling system to predict air pollution concentration over an inland and coastal location of Nord-Pas de Calais region of France has been evaluated. We used the regional emission inventory for the NordPas de Calais for anthropogenic emissions. The modelling system reproduces well the O3 cycle for both inland and coastal region. It has difficulties in predicting high maxima when air masses flow are continental but it has correctly reproduced the influence of cold weather fronts on O3 and PM10 pollution. It systematically underpredicts PM10 concentration for both locations. This is partially due to the lack of representation of some monitoring stations especially on the coastal area. Further work should include testing and improving PBL scheme specific to local dynamical and thermal coastal weather conditions (sea-breezes, low PBL height). Moreover, update of the emission inventory as well as the use of the nesting seems primordial.
Application of a Regional Atmospheric Emission Inventory to Ozone and PM Modelling
411
Finally, we encourage comparing the performance of the MM5-CHIMERE system over the North-Pas de Calais region with a Lagrangian Particle Dispersion model. Acknowledgments The authors would like to acknowledge the Laboratoire de Meteorologie Dynamique de Palaiseau and the CHIMERE user list who provided useful comments on the use of CHIMERE. We are also grateful toward the Pennsylvania State University and the University Corporation for Atmospheric Research for providing the MM5 code. Finally, thanks to Atmo Nord-Pas de Calais for providing the hourly monitored O3 and PM10 data concentration.
References Bessagnet B, Hodzic A, Vautard R, Beekmann M, Rouil L, Rosset R (2004) Aerosol modeling with CHIMERE-first evaluation at continental scale. Atmospheric Environment 38, 2803–2817. Byun D, Young J, Gipson J, Godowitch J, Binkowski F, Roselle S, Benjey B, Pleim J, Ching J, Novak J, Coats C, Odman T, Hanna A, Alapaty K, Mathur R, McHenry J, Sankar U, Fine S, Xiu A, Jang C (1998) Description of the Models3 community multiscale air quality (CMAQ) model. In: Proceedings of the American Meteorological Society 78th Annual Meeting. January 11–16, Phoenix, AZ. Derognat C (1998) Elaboration d’un code chimique simplifié applicable à l’étude de la pollution photooxydante en milieu urbain et rural. Rapport de stage de master, Université Parie 6, France. Grell GA, Dudhia J, Stauffer J (1994) A description of the Fifth-generation Penn State/NCAR Mesoscale Model (MM5). NCAR Tech Note TN-398, pp.122 Martinet Y (2004) Conception, validation et exploitation d’un cadastre des émissions de polluants atmosphériques sur la région Nord-Pas de Calais. Thèse de doctorat en chimie, Université de Lille 1, France. San Jose R, Perez JL, Blanco JF, Barquín R, González RM (2002) An operational version of MM5–CMAQ modelling system over Madrid City. Forth Symposium on the Urban Environment, May 20–24 2002. American Meteorology Society. Sokhi RS, San Jose R, Kitwiroon N, Fragkou E, Perez JL, Middleton DR (2006) Prediction of ozone levels in London using the MM5-CMAQ modelling system. Environmental Modelling and Software. 21, 566–576. Terrenoire E, Nollet V, Deconinck A (2007) PM Modelling in the French north area using the CHIMERE model: case study. JIQA, Lille, France, 2007. Tulet P, Crassier V, Solmon F, Guedalia D, Rosset R (2003) Description of the mesoscale non-hydrostatic chemistry model and application to a transboundary pollution episode between northern France and southern England. Journal of Geophysical Research, 108 (D1), 4021 Vautard R, Beekmann M, Roux J, Gombert D (2001) Validation of a deterministic forecasting system for the ozone concentrations over the Paris area. Atmospheric Environment, 35, 2449–2461.
4.5 Application of Advanced Particulate Matter Source Apportionment Techniques in the Northern Italy Basin Marco Bedogni, Simone Casadei, Guido Pirovano, Giovanni Sghirlanzoni and Andrea Zanoni
Abstract This work describes the results of the CAMx modelling system application aimed at reconstructing the particulate matter concentration over the Northern Italy basin and analysing the role played by the main emission sources on the pollution levels. Simulations have been performed on yearly basis in the frame of two different modelling exercises. The present analysis has been performed taking into account an overlapping period, covering February 2004. As a first step, CAMx results have been compared to a considerable data set of observed concentrations of NO2 and PM10. Model has been able to correctly reproduce the daily evolution and spatial variability of NO2, while some underestimations of PM10 concentrations have been highlighted. Then the Particulate Source Apportionment tool (PSAT) has been applied, in order to discriminate the contribution of several key emission sectors such as transport, domestic heating and power plants, in relation to three emission areas. Road transport appears to be the most responsible sector of PM10 concentration, especially in the area between Milan and Turin. The urban area of Milan is characterised by high PM10 concentration with an important contribution by road transport and domestic heating. The analysis has also pointed out the increasing relevance of the emissions coming from areas outside the city during critical pollution episodes, when severe stagnant conditions emphasise the role of secondary particulate matter. Finally, the comparison between the two PSAT applications has put in evidence, as a whole, quite coherent distributions of the relative contribution to PM10 concentration, thus proving the robustness of the source apportionment results. The main differences concerned domestic heating contribution, clearly related to the discrepancies in the emission inventories. Keywords Chemical transport models, PM10, Po valley, Source apportionment
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
385
386
M. Bedogni et al.
1. Introduction Transport and dispersion modelling is a useful tool to develop emission reduction scenarios in order to support management systems. The Northern Italy basin is often subject to high concentrations of particulate matter, mainly during winter time, when severe stagnant conditions lead to frequent exceedances of the air quality standards. The definition of effective emission reduction policies for PM is a challenging task due to the relevant non–linearities that influence the interactions among the different sources. The CAMx (ENVIRON, 2006) chemical and transport model (CTM) has been applied over the Northern Italy basin in order to evaluate the concentration of photochemical pollutants, such as NO2 and O3, and particulate matter (PM10) over a whole year in the framework of two different modelling exercises. PSAT algorithm, implemented into CAMx by Environ (ENVIRON, 2005), was then used to perform a source apportionment analysis over an overlapping period in order to discriminate the contribution of different emission sectors and areas to the modelled PM concentrations in several receptors. PSAT is embedded in CAMx code thus allowing to reduce the effect of non-linearities, arising in more traditional approach such as the zero-out modelling (i.e. removing a subset of sources). In Sections 2 and 3 the CAMx configuration and the evaluation of the model performances are described while PSAT application is illustrated in Section 4.
2. Description of the Simulations The CAMx model has been applied over two different domains (Figure 1) and periods. The APAT-CTNACE (APAT, 2007) domain covers the whole Po valley with an extension of 640 × 410 km2 subdivided according to a grid system with a 10 km step size. The vertical domain has been subdivided into 13 vertical levels, the first layer being 70 m thick. The CITYDELTA (Cuvelier et al., 2007) domain covers a smaller area, centred on Milan and having an extension of 300 × 300 km2 with a grid step size of 5 km. The vertical domain is subdivided in 11 vertical levels with the first layer of 30 m. As the two simulations cover different years, the analysis of the results has been focused on a shorter overlapping period, February 2004. The two simulations have been performed feeding the CAMx model with different meteorological and emission data sets. Table 1 provides a brief outline of the two simulation features. Over the CITYDELTA computational domain, the two emission inventories provided quite coherent estimations of the main pollutants (Sghirlanzoni and Zanoni, 2007), apart from PM10, where CITYDELTA emissions were almost double than CTN-ACE, due to a discrepancy in the estimation of the domestic heating contribution.
Application of Advanced Particulate Matter Source Apportionment Techniques
387
Fig. 1 APAT (solid line) and CityDelta (dashed line) exercise simulation domains. The domains are divided into three areas for PSAT analysis: Milan critical zone (black), the leftover area of Lombardy region (grey) and the rest of the domain (white) Table 1 Comparison of the computational domains and input data sets. CTN-ACE
CITYDELTA
Computational grid
640 × 410 km2 (grid step size = 10 km) First layer = 70 m
300 × 300 km2 (grid step size = 5 km) First layer = 30 m
Emissions
Italian local emission inventories gridded at 5 km resolution
Meteorology
LAMI model at 7 km horizontal resolution First layer = 70 m
Boundary conditions
CHIMERE continental fields Version 200402C – without dust
EMEP 2004 at 50 km Gridded at 5 km on the basis of Italian local emission inventories MM5 model at 5 km horizontal resolution First layer = 30 m CHIMERE continental fields Version including dust (Vautard et al., 2005)
The results of the simulations have been compared with a set of air quality stations all placed inside the Lombardy region, in order to limit the influence of the boundary conditions in the CITYDELTA exercise.
3. Results Figure 2 shows the comparison of the monthly mean concentration of PM10 computed in the two exercises. The simulations show a similar spatial distribution of the PM10 concentration over the common domain, although the absolute values are very different. Both simulations show the maximum concentration near the Milan urban area, with values higher than 50 (µg m-3) for the CITYDELTA simulation and always lower than 35 (µg m-3) for the CTNACE run. Obtained results have been compared to available observations of both NO2 and PM10. As an
M. Bedogni et al.
388
example, Figure 3 shows the comparison of the computed and observed daily mean concentrations. Concerning NO2, CAMx has put in evidence a good agreement in both simulations, with some overestimations that took place for the CTNACE exercise. Differently, CAMx reproduced quite well the PM10 concentrations in the CITYDELTA simulation, while for the CTNACE run the computed concentrations were clearly underestimated (up to 40%). Discrepancies between observed and computed concentrations in CTNACE run are probably related to an underestimation of the primary PM emissions and an overestimation of the dispersion strength over the whole domain. Dilution induces a widespread underestimation of primary and secondary PM10. It is worth noting that the chemical composition analysis (not shown) has demonstrated that the underestimation is mainly due to SOA (Secondary Organic Aerosols) and winter sulphate mass determination (Sghirlanzoni and Zanoni, 2007). Such a result puts in evidence that besides the meteorological forcing, there could be a lack in the reconstruction of the heterogeneous phase processes.
5150
5100000
TRENTO
BRESCIA
VERONA
VENEZIA
VARESE
5000000 TORINO
(UTM - km)
UTM
TRENTO
TRIESTE MILANO
PIACENZA
MODENA BOLOGNA RAVENNA
GENOVA
4900000
SONDRIO
5100
AOSTA
BERGAMO
5050
BRESCIA VERONA
MILANO
NOVARA
5000 TORINO
PIACENZA ALESSANDRIA PARMA
4950
FIRENZE
MODENA
4800000 300000
400000
500000
600000 UTM
700000
800000
900000
[µg/m3]
GENOVA
4900 400
450
500
60 56 52 48 44 40 36 32 28 24 20 16 12 8 4 0
550
600
650
(UTM - km)
Fig. 2 Monthly mean concentration of PM10 computed by CAMx over the CTNACE (left) and CITYDELTA (right) domains
4. Application of the Particulate Source Apportionment Technology (PSAT) PSAT algorithm has been applied to the winter month of February 2004 in order to discriminate the contribution of several key emission sectors such as road transport, domestic heating, and power plants, over three emission areas: Milan critical zone, the leftover area of Lombardy region and the remaining part of the domain (Figure 1). The temporal evolution of particulate matter concentration has then been tracked over the whole domain and, with further details, in the Milan urban area.
Application of Advanced Particulate Matter Source Apportionment Techniques 100
200
90
180
80
160
70
Computed (P g.m-3)
Computed (ppb)
389
60 50 40 30
140 120 100 80 60
20
40
10
20
0
0 0
10
20
30
40
50
60
Observed (ppb)
70
80
90
100
0
20
40
60
80
100 120 140 160 180 200
Observed (P g.m-3)
Fig. 3 Comparison of computed and observed daily mean concentration of NO2 (left) and PM10 (right) for the CTNACE (circles) and CITYDELTA (black diamonds) simulations
Firstly, PSAT has been used to calculate the contributions of the different emission areas and groups to the PM10 concentration over the domain. Figure 4 shows the contribution of domestic heating and road transport to the monthly average concentration of PM10 (µg m-3) respectively. Significant contributions from domestic heating are centred around the major cities such as Milan, Turin and Venice with mean values between 5 and 7 (µg m-3). The contribution of road transport is relevant over the whole domain with concentrations ranging from 5 (µg m-3) in the Po valley up to 11 (µg m-3) near the urban area of Milan. The receptor analysis has then allowed to evaluate the weight of each emission source on the monthly average concentration of PM10 in the urban area of Milan. Figure 5a shows the contributions of four emission groups belonging to the 3 different areas and the boundary conditions (BC) to the total PM10 concentration. Although the two simulations provided very different absolute PM10 concentrations, the source apportionment analysis has put in evidence quite similar distributions of the relative contributions among the different sources. The estimated PM10 monthly mean concentration in Milan for the two simulations is 34.6 (µg m-3) and 56.7 (µg m-3) respectively. Milan critical zone contributes only for the 32% to this concentration for CTNACE run, but such a contribution increases up to 47% for CITYDELTA, due to the influence of the domestic heating. The emission sectors analysis shows that road transport from Lombardy plays the main role, accounting for a fraction ranging from 20% to 25% of PM10 concentrations for the two simulations. More precisely, the critical zone (MIL) contributes to about the 15%, while the leftover area of the region (LOM) accounts for a fraction ranging from 5% to 11%. The emissions of sector Others, located outside Milan critical zone, have an impact of about 20%, mainly related to ammonia emissions from the agricultural sector. It is clearly visible the low contribution of the power plants sector (lower than 4%). The main difference between the two source apportionment analyses is related to the domestic heating sector of the critical zone, that represents the highest contribution for the CITYDELTA run. There, APAT-CTNACE estimates a contribution of 12% to the PM10 concentration, while in CITYDELTA it grows up to 23%. As shown in
M. Bedogni et al.
390
Figure 5b and c the discrepancy entirely relies on the primary fraction of the PM10 concentration and is clearly related to the above mentioned differences in PM10 domestic heating emissions. As expected, the primary PM10 is mostly related to the Milan area emissions (Figure 5b), while the relative contributions to the secondary PM10 are much more smoothed (Figure 5c), confirming that particulate matter pollution is a basin scale problem. The only relevant discrepancy in the contributions to the secondary PM10 concerns the sector Others of the Lombardy region. Further analysis (not shown) has put in evidence that the difference is due to the sulphate concentration that, as previously discussed, was higher for the CITYDELTA run.
TRENTO
TRENTO AOSTA
AOSTA
TRIESTE BRESCIA
MILANO TORINO
TRIESTE
TORINO
PIACENZA
MODENA BOLOGNA RAVENNA
GENOVA
FIRENZE
FIRENZE
PM10 [µg/m3]
PM10 [µg/m3]
1
2
3
4
5
6
7
8
VENEZIA
VERONA
PIACENZA
MODENA BOLOGNA RAVENNA
GENOVA
BRESCIA
MILANO
VENEZIA
VERONA
9 10 11
1
2
3
4
5
6
7
8
9 10 11
Fig. 4 Contribution of domestic heating (left) and road transport (right) to the monthly mean concentration of PM10 (µg m-3) over the domain for the CTNACE simulation 40%
Road Transport
Domestic Heating
Others
Power Plants
25%
2% 0%0% 2% 2% 1%
3% 3%
Power Plants
LEFT
MIL
40%
BC
35% % contribution to [SPM10]
35%
30% 25% 20% 15% 10% 5% LEFT
BC+IC
MIL
LOM
LEFT
MIL
LOM
LEFT
MIL
LOM
LEFT
MIL
0% LOM
% contribution to [PPM10]
LOM
b)
LEFT
Road Tr.
MIL
LOM
Domestic H.
LEFT
Others
BC+IC
a)
Power Plants BC
30% 25% 20% 15% 10% 5% 0% LEFT
LOM
BC+IC
Others
MIL
LEFT
Domestic H.
LEFT
MIL
Road Tr.
LOM
LOM
MIL
MIL
3%
12%
LOM
5%
5%
6% 6%
13%
13% 7% 8%
LEFT
8%
9% 6%
MIL
10%
15%
12%
LOM
11%
LEFT
15%
23% 15% 15%
MIL
20%
0%
40%
BC
30%
LOM
% contribution to [PM10]
35%
c)
Fig. 5 Contribution (%) of emission areas and groups to the PM10 monthly mean concentration in Milan: comparison between APAT-CTNACE (grey) and CYTYDELTA (black) simulations for February 2004
Application of Advanced Particulate Matter Source Apportionment Techniques
391
Source apportionment analysis has also been performed by taking into account critical episodes only. To this aim, days with a computed daily mean concentration over the air quality limit of 50 (µg m-3) have been selected for both simulations. These days are characterised by an estimated average PM10 concentration of 63.7 (µg m-3) and 74.3 (µg m-3) respectively. During exceedances days (Figure 6), for both simulations, road transport contribution increases for each emission area as well as the influence of the areas outside Milan raises for sector Others (OT2 and OT3) and power plants (PP2). On the opposite, the contribution of critical zone domestic heating (DH1) decreases by 2%, and the role of boundary conditions (BC) for CTNACE drops to 8%, indicating a meteorological stagnation condition. It is clearly visible how, during high concentration episodes, the contribution of the areas outside Milan critical zone increases as well as the impact of sectors characterised by relevant emissions of PM10 precursors (NOx, SOx, VOC and NH3). Such a result highlights that stagnant conditions favour the formation and accumulation of secondary particulate matter, emphasising the interactions that take place over the whole domain more than at the local scale. 3
Milan PM10 = 63.7 µg/m PP1 PP2 PP3 BC 0% 3% 2% 8%
Milan PM10 = 74.3 µg/m3
RT1 15%
OT3 8%
BC 12%
RT2 12%
OT2 18%
RT3 9% OT1 DH3 DH2 6% 3% 6%
DH1 10%
PP1 PP2 PP3 0% 2% 1%
RT1 14% RT2 5%
OT3 9%
RT3 4%
OT2 13% DH1 22%
OT1 9% DH3 3%
DH2 6%
Fig. 6 Contribution (%) of emission sectors (Power Plants, Road Transport, Domestic Heating, Other sectors) divided by emission areas (1-Milan critical zone, 2-rest of Lombardy region, 3-rest of the domain) to the PM10 mean concentration during critical episodes for the CTNACE (left) and CITYDELTA (right) simulations. BC = Boundary Conditions
5. Conclusions PSAT algorithm, implemented into CAMx chemical and transport model, has been applied on the Northern Italy basin to study the contribution of different emission groups and areas to the modelled PM concentrations. The same investigation has been carried out over two computational domains with different input data sets. The source apportionment analysis gave similar results in terms of relative contribution, though the absolute values of the modelled total PM concentrations were very different. PSAT analysis testified that Milan emissions account for 30–40% of the
392
M. Bedogni et al.
PM10 concentration inside the city at most, thus confirming that particulate matter pollution is a basin scale problem. Sector analysis has also highlighted that there are not any prevailing sectors, but both road transport, domestic heating and agriculture play a critical role. Significant dissimilarities in the relative contribution to PM10 concentration are noticeably related to evident discrepancies in the emission inventories. Finally, the analysis highlighted that stagnant conditions favour the formation and accumulation of secondary particulate matter, emphasising the interactions that take place over the whole domain more than at the local scale. In conclusion, the obtained results proved that PSAT is a coherent algorithm able to give useful indications on the relationships between emissions and air concentrations, tracking the fate of primary PM10 as well as secondary PM components. For these reasons PSAT could be a very functional tool not only from a scientific point of view, to better understand the correctness of the model simulations, but also to support air quality management policies devoted to the reduction of the primary and secondary pollution levels. Acknowledgments CESI Ricerca contribution has been financed by the Research Fund for the Italian Electrical System under the Contract Agreement between CESI RICERCA and the Ministry of Economic Development – D.G. for Energy and Mining Resources stipulated on June 21, 2007 in compliance with the Decree n.73 of June 18, 2007. Agency contribution has been sustained by the Milan Council.
References APAT (2007) Rapporto Tecnico Sull’Applicazione di Modellistica al Bacino Padano Adriatico, Technical Report, APAT-CTNACE. Cuvelier C, Thunis P, Vautard R, Amman M, et al. (2007) CityDelta: a model intercomparison study to explore the impact of emission reductions in European cities in 2010, Atmospheric Environment, 41, 189–207. ENVIRON (2005) Development of the CAMx Particulate Source Apportionment Technology (PSAT), Final Report, Environ Int. Corp. ENVIRON (2006) CAMx (Comprehensive Air Quality Model with extensions) User’s Guide Version 4.31, Internal Report, Environ Int. Corp. Sghirlanzoni GA, Zanoni A (2007) Il particolato fine nel bacino padano: analisi modellistica del ruolo delle fonti., Master thesis, Politecnico di Milano, Italy. Vautard R, Bessagnet B, Chin M, Menut L (2005) On the contribution of natural aeolian sources to particulate matter concentrations in Europe: testing hypotheses with a modelling approach. Atmospheric Environment, 39, 3291– 3303.
Application of Advanced Particulate Matter Source Apportionment Techniques
393
Discussion T. Watkins: Have you considered the use of observation based receptor modelling approaches as a complementary source apportionment approach? (For example, positive matrix factoration) G. Pirovano: Not yet, because we have not a specific experience in applying receptor models and also because we should need specific field data, not available at the moment. M. Astitha: 1 – What aerosol size distribution approach did you use in CAMx? Fixed size approach? 2 – Can you comment on the good correlation you showed for nitrate aerosol compared to the observations? 3 – Did your emission inventory include high emission sources for Italy? G. Pirovano: 1) Yes, we used the fixed size approach that is based on two static size bins: Fine (0–2.5 Pm) and Coarse (2.5–10 Pm). The Fixed size approach is less accurate than the moving size approach also implemented in CAMx, but it is much less time consuming and, so far, is mandatory if you are applying the PSAT probing tool. 2) The slide showed a qualitative comparison between computed and observed aerosol components, because the available observations concerned a different time period. Nevertheless, such a comparison put in evidence that the model correctly reproduced the statistical distribution of the aerosol fraction of both Nitrate and Ammonium during a winter period. 3) Yes, emissions from SNAP sectors 1, 3 and 9 were considered as high sources according to a height distribution profile similar to the height profile implemented in the EMEP model.
4.12 Comparison of Six Widely-Used Dense Gas Dispersion Models for Three Actual Railcar Accidents Steven Hanna, Seshu Dharmavaram, John Zhang, Ian Sykes, Henk Witlox, Shah Khajehnajafi and Kay Koslan
Abstract The simulations are compared of six widely-used dense gas dispersion models of downwind chlorine gas concentrations following three railcar accidents. The six models are ALOHA, HGSYSTEM, SLAB, SCIPUFF, PHAST and TRACE. The three railcar accidents, where as much as 60 t of chlorine were released, are Festus, MO (release from a ruptured 1 in. line while offloading), and Macdona, TX, and Graniteville, SC (release from a large hole due to an accident). Input data were obtained from public sources. Source emissions rates were refined based on source modeling with PHAST and TRACE and derivations using fundamental thermodynamic equations. When using the same source emissions rates, the models’ simulations of 10-minute averaged cloud centerline concentration, at downwind distances ranging from 0.1 to 10 km, agree with each other within plus and minus a factor of two most of the time. For a very large release (Graniteville), the 10-minute averaged 2000 ppm, 400 ppm, and 20 ppm contours are predicted to extend downwind about 1.3, 3.1, and 14 km, respectively, from the source. There is also agreement among the models simulations of the plume widths and heights to the 2000, 400, and 20 ppm contours. A major conclusion of the study is that estimation of the source or release term is important for reliable results.
Keywords Chlorine railcar accidents, dense gas models, dispersion model comparisons
1. Introduction and Approach The objective of the current study is to test several of the more widely-used hazardous gas dispersion models using three recent railcar release scenarios. Publiclyavailable source observations and meteorological data for three recent rail car accidents have been collected: August 14, 2002, Festus, Missouri; June 29, 2004, Macdona, Texas; and January 6, 2005, Graniteville, South Carolina. These recent accidents span several types of accidental releases – ranging from a release during loading/unloading operations in a fixed facility (hose rupture at Festus) to worstC. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
443
444
S. Hanna et al.
case accidents involving large releases from a tank railcar during transportation (Macdona and Graniteville). In all cases the chlorine was stored at ambient temperature as a pressurized liquefied gas in a rail car with capacity of about 60–90 t. The release durations ranged from about 1 minute to three hours. In an earlier study completed about 15 years ago, Hanna et al. (1993) evaluated 15 dense gas dispersion models (including five of the six that are the subject of the current paper) with observations from eight field experiments. There were about six of the 15 models that gave similar reliable performance (mean biases less than about 50% and relative scatter less than a factor of two or three) for the eight research-grade field experiments. There are numerous models available for application to releases of dense gases (Hanna et al., 1996). The project team identified six of the more widely-used models that include dense gas algorithms: TRACE, PHAST, CAMEO/ALOHA, HGSYSTEM, SLAB, and SCIPUFF. TRACE and PHAST are proprietary models. The other models are in the public domain and are available at no charge. The question for this study is whether there is agreement among these models for simulating dispersion of dense gases for “typical” chlorine railcar release scenarios. The purpose of the study is not to reproduce exactly the accidents and the concentration distributions. The input data that are used are taken entirely from publicly-available documents such as reports by the National Transportation Safety Board (NTSB), which investigates each accident.
2. Descriptions of Three Railcar Accidents Festus, Missouri, August 14, 2002, 9:20 am CDT. The chlorine was being offloaded from a railcar parked at a chemical facility. A 1 in. hose ruptured, with a ragged aperture. Photos show a visible chlorine gas cloud of depth about 1 m and width about 20 or 30 m around the railcar. The location was 1785 Highway 61, Festus, MO (35 mi south of St. Louis, at N lat 38° 10’ 45”, W lon 90° 23’ 15”). 48,000 lb of liquid chlorine were released over three hours, for an average release rate of 2.02 kg/s. Macdona, Texas, June 28, 2004, 5:03 am CDT. A collision of two trains occurred along a rail line where the main line met a side track in a rural area. The total release (including the initial large two phase cloud and the subsequent smaller vapor release) lasted about seven hours, ending when wooden plugs were hammered into the puncture, which was about 2” by 11” on the lower part of the end of the car. However, most of the mass was released in the flashing plume during the first 3 minutes. Witnesses reported a visible cloud in the neighborhood of the accident shortly after the release. The location was near 9221 Nelson Rd., Macdona, TX (N lat 29° 19’ 40”, W lon 98° 40’ 20”). About 120,000 lb of liquid chlorine were released. Graniteville, SC, January 6, 2005, 2:45 am EST. The train collided with a parked train in an area consisting of a mixture of woods, fields, industrial buildings, and residences. There was a hole with diameter about 4” or 5” in the side of the car.
Comparison of Six Widely-Used Dense Gas Dispersion Models for Three Actual Railcar
445
Witnesses reported a visible cloud near the accident. Initially, the dense cloud spread in all directions near the release point. At larger distances from the release point, the cloud followed the general airflow towards the northeast. The location was near the intersection of US Route 1 and State Route 191, Graniteville, SC (N lat 33° 34’ 00”, W lon 81° 48’ 30”). The NTSB report stated that about 120,000 lb of chlorine were released. Most was released as a two-phase jet in the first minute.
3. Description of Models The SLAB model (Ermak, 1990) was developed primarily to address effects of dense clouds from evaporating pools. The model can handle horizontal jets. It is freely distributed by the Environmental Protection Agency (EPA). The science in SLAB is regarded as excellent by peer reviewers. The HGSYSTEM model (Witlox and McFarlane, 1994) is a hazard-assessment software package developed at Shell. HEGADAS is the area source dispersion module in HGSYSTEM (Witlox, 1994a, b). HGSYSTEM also includes a model, HEGABOX, for describing instantaneous releases. HGSYSTEM is distributed by the EPA, by the American Petroleum Institute (API), and by Shell. The ALOHA model is a linked source emission and dispersion model for hazardous chemical releases to the atmosphere. It is part of the CAMEO software developed and distributed by the National Oceanic and Atmospheric Administration’s (NOAA’s) HAZMAT Office (NOAA and EPA, 1992). CAMEO/ALOHA is in use by most US fire departments and emergency responders. The SCIPUFF model was originally developed for application to stack plumes. It was greatly enhanced in the 1990s and 2000s under Department of Defense (DOD) support. SCIPUFF is the transport and dispersion module in the HPAC modeling system, which is now in use throughout all DOD agencies (Sykes et al., 2004). It can handle dense gases (Sykes et al., 1999). The TRACE model is a widely-used proprietary hazardous gas model that has been available for about 20 years (SAFER Systems, 1996). It contains linked source and dispersion models. The PHAST model is another widely-used proprietary modeling system in the same category as TRACE. It has linked source emissions modules and transport and dispersion models and is distributed by DNV (Witlox and Holt, 1999). All of the models have been extensively evaluated with field data (e.g., Hanna et al., 1993) and shown to have good performance.
4. Approach Detailed information (e.g., source emissions conditions, meteorology, local topography and specific release location (UTM or Lat-Long)) was gathered for each of the accidental releases. This information came primarily from reports by the NTSB
446
S. Hanna et al.
and other government agencies. Using the emission and source release information and other guidance, the modeling team ran the models (SLAB, CAMEO/ALOHA, HGSYSTEM, SCIPUFF, TRACE, and PHAST). TRACE and SAFER were run in two modes: Phase 1, where the model calculated its own source emission or release rate and subsequent dispersion; and Phase 2, where the model was run with the optimum source release terms as determined as an amalgamation of TRACE and PHAST calculations, plus independent derivations using the thermodynamic equations. The current paper includes only the Phase 2 results where the models were run with a common optimum source term. The modelers were asked to provide the following outputs for each accident: x Maximum near-ground 10-minute averaged chlorine concentration, C, in ppm, at several specified downwind distances for each site. x Maximum downwind distances, widths, and heights to the 2000, 400, and 20 ppm concentration contours for each site. These concentrations were chosen based on health effects standards in the literature. x Width and height of the 2000, 400, and 20 ppm concentration contours at the specified downwind distances. x Contour plots of cloud movement (a routine output of some models).
5. Input Data for Models
5.1. Source emissions inputs A single best estimate of the source emission rate for each accident was determined using the PHAST and TRACE models, basic derivations, and information in the official accident reports. A general assumption was that the released chlorine would not form a significant liquid pool, and that most of the unflashed liquid would remain airborne as an aerosol with small drop sizes. For the Festus accident, the actual discharge rate was much lower than that expected for a 1” hose, due to the fact that an excess flow valve partially closed, causing an effective hole diameter of about 1/8 in. The resulting continuous discharge lasted for three hours. For the Macdona and Graniteville accidents, the release consisted of an initial rapid release of the pressurized liquid as a two-phase jet (where the liquid exists as small drops carried by the gas jet), followed by a longer period of a much smaller gas phase release. For the publicly-available models, a decision had to be made about where the plume should be initialized – near the hole as the plume exits the tank and completes its expansion or flashing, or at the end of the momentum jet where the plume’s velocity drops to the ambient velocity and all aerosol has evaporated (usually within a few tens of meters of the hole). The latter method was used in most cases. The former method was used for HGSYSTEM for Festus and Macdona.
Comparison of Six Widely-Used Dense Gas Dispersion Models for Three Actual Railcar
447
For Macdona and Graniteville, it was assumed that, after the initial large twophase release was concluded, there was a much smaller vapor release that extended for several hours. The modelers simulate the initial two-phase release and the subsequent vapor release. However, because the concentrations were much less for the vapor releases, those results are not given in this paper.
5.2. Meteorological data inputs The meteorological data for the three releases were obtained from the National Climate Data Center (NCDC). The data are from the “official” National Weather Service sites that are closest to the accident sites. Because the largest source emission rates occurred in the first hour at Macdona and Graniteville, and the Festus release lasted only three hours, the models have been run assuming that the meteorological conditions for the first hour apply for the modeling period. The specific assumptions are listed below: Festus – wind speed = 2.6 m/s, wind direction = 310°, T = 20.0°C, T(dew point) = 19.4°C, weather = overcast with drizzle, stability = neutral. Macdona – wind speed = 2.6 m/s, wind direction = 150°, T = 24.4°C, T(dew point) = 23.3°C, weather = overcast, stability = neutral. Graniteville – wind speed = 2.1 m/s, wind direction = 190°, T = 10.0°C, T(dew point) = 10.0°C, weather = clear but with fog/haze, stability = slightly stable.
6. Results As an example of the results, Figure 1, for Graniteville, contains plots of the variation with downwind distance, x, of the maximum 10-minute averaged concentration on the plume centerline for the six models. Good agreement among the models can be seen. The model simulations generally follow a straight line on the log-log plot, consistent with the power law, C2/C1 = (x2/x1)-p, with a value of p between about 1.5 and 2.0. Tables were prepared summarizing the model comparisons for the three accidents. In general, the six models’ simulations of cloud centerline concentration agree within plus and minus a factor of two. For a large release (Graniteville), the 2,000 ppm, 400 ppm, and 20 ppm contours are predicted to extend downwind about 1.3, 3.1, and 14 km, respectively, from the source. The maximum widths for these same contours have a median of 625, 900, and 1625 m. The maximum heights have a median of 16, 26, and 97 m. Thus the ratio of the maximum width to the maximum height is in the range of about 15–40, indicating a very flat and shallow cloud that is likely caused by the dense gas effect.
S. Hanna et al.
448 1,000,000
Cl2 Concentration (ppmv)
100,000
10,000
TRACE PHAST SCIPUFF SLAB HGSYSTEM ALOHA
1,000
100
10 0.1
1
10
x (km)
Fig. 1 Plot of maximum 10 minutes average concentration (ppm of chlorine) on plume centerline versus downwind distance, x, for the six models for Graniteville
For all accidents, the fraction of model predictions of plume heights within a factor of two of the median is nearly always more than 0.8. The fraction of model predictions of plume widths within a factor of two of the median is slightly less than for the heights. As mentioned earlier, since standard samplers were unavailable to observe concentrations at the sites during the period of the initial release, there is no way of knowing which model is “best”. The mass emission rate and its duration evolved over the course of this project, due to the availability of new NTSB reports and other public documents, and due to ongoing analyses of the thermodynamic equations. It is noted that the Macdona and Graniteville releases are now thought to have occurred over much shorter times than stated in the NTSB reports. Most of the Macdona release is believed to have occurred in the first 3 minutes, while most of the Graniteville release is believed to have occurred in the first ½ minute. Although much effort was put into setting up inputs (source emissions conditions and meteorological conditions) for the scenarios, there is still much uncertainty in the mass emission rates and thermodynamic conditions. Consequently, the results in the final model comparisons of concentrations and plume dimensions should be considered conditional. It is concluded that these six widely-used models closely agree in their estimates of downwind dispersion when given the same source emission terms.
Comparison of Six Widely-Used Dense Gas Dispersion Models for Three Actual Railcar
449
7. Further Comments on Importance of Source Term The agreement in the dispersion calculations among the six models is dependent on accurate estimation of the source term. For a hole in a pressurized chlorine railcar, even the models that have source emissions modules have disagreements on the fraction of the unflashed liquid that enters a pool under the railcar, the variation of the release as the location of the hole on the railcar shifts position, the liquid aerosol drop sizes, and many other factors. Also, it is very important to point out that knowledge of the hole size is critical. In order to characterize the hole properly as well as match the scenario, one must review photographs, NTSB reports, and other documents, and discussions with personnel on-site at the time of the accident. It is expected that, for hole diameters greater than about 10 cm, most of the mass in the tank would be released within the first few minutes, before emergency personnel arrived on the scene. Also, because emergency responders are involved in assessing safety and keeping persons away from the site, it is unlikely that a model will be run by them during the initial period. Acknowledgments This study has been supported by the Research Foundation for Health and Environmental Effects (RFHEE), and has involved collaboration with the Chlorine Institute and the Chlorine Chemistry Council.
References Ermak DL (1990) Users Manual for SLAB: An Atmospheric Dispersion Model for Denser-than-Air Releases. UCRL-MA-105607, Lawrence Livermore National Laboratory, Livermore, CA. Hanna SR, Chang JC, Strimaitis DG (1993) Hazardous gas model evaluation with field observations, Atmos. Environ. 27A, 2265–2285. Hanna SR, Drivas PJ, Chang JC (1996) Guidelines for Use of Vapor Cloud Dispersion Models, Second Edition. Published by AIChE/CCPS, 345 East 47th St., New York, NY 10017, 285 pp. + diskette. NOAA/HMRAD and EPA/CEPPO (1992) ALOHA Users Manual and Theoretical Description. Reports available from NOAA/HMRAD, 7600 Sand Point Way NE, Seattle, WA 98115 and on CAMEO/ALOHA web site. SAFER Systems (1996) Description of Modeling Algorithms, TRACE Version 8.0. Looseleaf notebook available from SAFER Systems, 4165 E. Thousand Oaks Blvd., Suite 350, Westlake Village, CA 91362. Sykes RI, Cerasoli CP, Henn DS (1999) The representation of dynamic flow effects in a Lagrangian puff dispersion model, J. Hazard. Mater. A:64, 223–247. Sykes RI, Parker SF, Henn DS (2004) SCIPUFF Version 2.1 Technical Documentation, Titan Corporation, P.O. Box 2229, Princeton, NJ, 292 pp. Witlox HWM (1994a) The HEGADAS model for ground-level heavy-gas dispersion –I Steady-state model, Atmos. Environ. 28, 2917–2932.
450
S. Hanna et al.
Witlox HWM (1994b) The HEGADAS model for ground-level heavy-gas dispersion –II Time-dependent model, Atmos. Environ. 28, 2933–2946. Witlox HWM, Holt A (1999) A unified model for jet, heavy and passive dispersion including droplet rainout and re-evaporation”, International Conference and Workshop on Modelling the Consequences of Accidental Releases of Hazardous Materials, CCPS/AIChE, San Francisco, California, September 28–October 1, 315–344. Witlox HWM, McFarlane K (1994) Interfacing dispersion models in the HGSYSTEM hazard-assessment package, Atmos. Environ. 28, 2947–2962.
Discussion U. Pechinger: The speaker suggested that the models may overestimate concentrations at distances of ±1 km from the source at the Graniteville accident. Question: Could the survival of people be also due to the indoor/outdoor effect, or the difference in plume transport direction, or the effects of obstacles? S. Hanna: The six dense gas dispersion models agree fairly closely in their estimates of chlorine concentrations. The problem is that, if the publicshed health standards are used to estimate casualties based on the models’ predicted chlorine concentrations, there are many more casualties expected than were observed at distances beyond a few hundred meters from the accident. It is believed that the dense gas models are not significantly biased, and the observed absence of casualties at distances beyond 200 m could be due to several possible causes, including (a) neglect of removal due to photolysis, chemical reactions, and dry deposition; (b) extreme conservatism of the health standard (usually an uncertainty factor of at least an order of magnitude is applied), and (c) the fact that most people were indoors with the windows closed. The latter effect is mentioned in the question (i.e., indoor/outdoor effect). I do not think that the uncertainties in plume transport direction would be important, because people were living and working in all directions from the accident. The effects of obstacles may be important in the near field, but their influence would be less at distances beyond 100 or 200 m. Ashok Gadgil: The speaker comments on the fact that many fewer fatalities occurred in the real accidents compared to the predictions from the dispersion models linked with the health effect model. I believe that a large part of the discrepancy can be explained by the reduction in dose to indoor occupants, because of the highly reactive chlorine being removed by adsorption and reaction on the indoor surfaces. We have illustrated this effect quantitatively in our recent public-
Comparison of Six Widely-Used Dense Gas Dispersion Models for Three Actual Railcar
451
cations, in press, in Atmos. Envir.: Chan, Nazaroff, Price and Gadgil, “Effectiveness of Urban Shelter-in-place II: Residential Districts”. So direct comparison of predicted and actual fatalities is probably a poor metric unless this effect is taken into account. S. Hanna: Most of the response to question 1 also applies to this question. We agree that one possible cause is the reduction in dose due to the fact that people were indoors, as explained in question 2. Furthermore, as Dr. Gadgil points out, some chlorine will be absorbed and reacted on the indoor surfaces, although I do not believe that this is a major contributor. I have read the publications that Dr. Gadgil mentions, and feel that his practical approaches to estimating indoor concentrations should be incorporated in model systems.
4.11 Comprehensive Surface-Based Performance Evaluation of a Size- and Composition-Resolved Regional Particulate-Matter Model for a One-Year Simulation M.D. Moran1, Q. Zheng1, M. Samaali2, J. Narayan1, R. Pavlovic2, S. Cousineau2, V.S. Bouchet2, M. Sassi2, P.A. Makar1, W. Gong1, S. Gong1, C. Stroud1 and A. Duhamel2
Abstract A comprehensive performance evaluation has been carried out for the first annual simulation made with AURAMS, a size- and composition-resolved, off-line, regional particulate-matter (PM) modelling system. The year simulated was 2002, the modelling domain covered most of North America, and the horizontal grid size was 42 km. The large evaluation data set consisted of filter-based and continuous surface air-chemistry measurements made by five Canadian and U.S. networks and precipitation-chemistry measurements made by seven Canadian and U.S. networks. Completeness criteria were used to exclude stations with incomplete records, and units conversions were performed to maximize uniformity and comparability. Quantities used in the performance evaluation included annual air concentrations of SO2, NO2, O3, HNO3, PM2.5, PM10, PM2.5-SO4, PM2.5-NO3, PM2.5-NH4, PM2.5-CM, PM2.5-EC, and PM2.5-TOM, and annual concentrations in precipitation of SO4=, NO3-, and NH4+. The extensive evaluation has allowed inferences about factors contributing to some model weaknesses.
Keywords Annual simulation, AURAMS, model evaluation, particulate matter
1. Introduction This paper presents some results from a comprehensive performance evaluation of the first extended (one-year) simulation made with a regional air-quality (AQ) modelling system called AURAMS (A Unified Regional Air-quality Modelling
1 2
Environment Canada, 4905 Dufferin Street, Toronto, ON, M3H 5T4, Canada Environment Canada, 2121, Route Transcanadienne, Montreal, QC, H9P 1J3, Canada
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
434
Comprehensive Surface-Based Performance Evaluation of a Size and Com
435
System). More extensive quarterly and annual results may be found in Moran et al. (2007). AURAMS performance has been evaluated previously for short-term episodic simulations (e.g., Bouchet et al., 2003; Makar et al., 2004; McKeen et al., 2005, 2007), but the evaluation of a long-term model simulation can provide additional valuable information. For example, systematic model errors are easier to detect and identify when longer simulation periods are considered, and model performance across different seasons can also be compared, which provides insights into relative model performance for distinctly different meteorological conditions. Results from a comparable annual evaluation of the U.S. EPA CMAQ model have been described by Eder and Yu (2006).
2. Model Description and Setup AURAMS consists of three main components: (a) a prognostic meteorological model, GEM; (b) an emissions processing system, SMOKE; and (c) an off-line regional chemical transport model, the AURAMS CTM. The Global Environmental Multiscale (GEM) meteorological model is an integrated weather forecasting and data assimilation system that was designed to meet Canada’s operational needs for both short- and medium-range weather forecasts (Côté et al., 1998). For the 2002 simulation, GEM version 3.2.0 with physics version 4.2 was run on the variableresolution North American regional horizontal grid. The grid consisted of a 353 × 415 horizontal global grid on a rotated latitude-longitude map projection. The horizontal grid spacing was approximately 24 km (0.22°) on the 270 × 353 uniform regional “core” grid and the 28 vertical hybrid-coordinate levels reached from the Earth’s surface to 10 hPa. The hourly gridded anthropogenic emissions files used for the simulation were prepared from the 2000 Canadian, 2001 U.S., and 1999 Mexican national emissions inventories with version 2.2 of the SMOKE emissions processing system (see http://www.smoke-model.org/index.cfm). Emissions of 20 model species were considered. Hourly biogenic emission fields were predicted in the CTM based on BEIS v3.09. Note that it was discovered after the annual model run that the NO emission fields that had been used were too high by a factor of 46/30 due to an error during the emissions processing. The multi-pollutant, regional AURAMS CTM was developed as a tool to study the formation of ozone, PM, and acid deposition in a single “unified” framework. The PM size distribution is currently represented using 12 size bins ranging from 0.01 to 41 ȝm in diameter and nine chemical components: sulphate (SO4), nitrate (NO3), ammonium (NH4), elemental carbon (EC), primary organic matter (POM), secondary organic matter (SOM), crustal material (CM), sea salt, and particle-bound water. PM is assumed to be internally mixed. Process representations include emissions from surface and elevated sources, transport, vertical diffusion, gas-phase, aqueous-phase, and inorganic heterogeneous chemistry, SOM formation, dry and wet deposition, and particle nucleation, condensation, coagulation, and activation (e.g., Gong et al., 2006).
436
M. Moran et al.
The AURAMS CTM v1.3.1 was run on a 150 × 106 uniform, continental-scale North American horizontal grid for the 2002 simulation. The horizontal grid spacing was 42 km, and 28 terrain-following vertical levels reached from the Earth’s surface to 29 km. A time step of 900 s was used, and AURAMS-predicted fields were output once an hour. Both GEM and the AURAMS CTM were run for the 13-month period from 1 December 2001 to 31 December 2002. GEM was run in 30-hour segments starting 24 hours apart from analyzed fields, where the first six hours of each segment were treated as a “spin-up” period and were discarded. The remaining 24 hours of successive simulations were then “stitched” together to create a complete set of hourly meteorological fields for input to the AURAMS CTM. The first month of the simulation (i.e., December 2001) was treated as a spin-up period and was not used in the evaluation.
3. Measurement Data Gas-phase air-chemistry measurements for North America for 2002 were obtainned from five networks, two in Canada (CAPMoN, NAPS) and three in the U.S. (AQS, CASTNet, STN). PM air-chemistry measurements were obtained from eight networks and subnetworks, three in Canada (CAPMoN, NAPS-continuous, NAPSfilter) and five in the U.S. (AQS-continuous, AQS-filter, AQS-STN, CASTNet, and IMPROVE). Note that these are heterogeneous data sets: different networks have different goals and objectives, choose different types of sampling locations, use different sampling techniques and protocols, and measure different species (e.g., Eder and Yu, 2006). For example, some networks report hourly values, some report daily averages every third day, and one reports weekly averages. For this reason, AURAMS predictions for 2002 have been evaluated against individual network measurements as well as multi-network measurements (see Moran et al., 2007). Figure 1a–g show the locations of stations that measured air concentrations of a number of gas- and particle-phase species in 2002. Precipitation-chemistry measurements for North America for 2002 were obtainned from eight networks and subnetworks, six in Canada (BCPCSN, CAPMoN, NBPMN, NSPSN, PQMPA, REPQ) and two in the U.S. (NADP-AIRMoN, NADPNTN). There is greater uniformity in sampling protocols and standard operating procedures across the precipitation-chemistry networks, but some networks do daily sampling, some do weekly sampling, and some do both. Figure 1h shows the locations of stations in 2002 for these eight networks. Note that only PM total organic matter (TOM) is measured, and for AURAMS TOM is the sum of POM and SOM. Unit conversions were performed so that all model air-concentration predictions and network measurements could be compared in ppbV for gas-phase species and in ȝg m-3 STP (0°C) for PM species. And completeness criteria were imposed: (a) for air-chemistry data at least 75% of 2002 samples at a station had to be valid, and (b) for precipitation-chemistry data a station must have operated for at least 90% of 2002 and the percentage total precipitation of valid samples had to be at least 70%.
Comprehensive Surface-Based Performance Evaluation of a Size and Com
437
a
b
c
d
e
f
g
h
Fig. 1 Locations of stations in 2002 measuring (a) SO2, (b) NO2, (c) O3, (d) HNO3, (e) PM2.5 mass, (f) PM2.5-SO4, (g) PM2.5-NH4, air concentrations and (h) SO4 concentration in precipitation
4. Results 4.1. Annual concentrations in air Figure 2 shows a comparison of AURAMS annual mean concentration predictions with measurements for two primary gas-phase species (SO2, NO2) and two secondary gas-phase species (O3, HNO3). Measurements from different networks are marked with different symbols, and the calculated statistics (correlation coefficient R, root mean square error, normalized mean bias, and normalized mean error) follow the definitions given by Eder and Yu (2006). The number of data points varies by
M. Moran et al.
438
species due both to the number of stations measuring the species and the number of complete measurements. For 2002 there were 463 valid annual SO2 values but only 92 valid HNO3 values. The better agreement for NO2 than for SO2 likely reflects the tendency for SO2 to be emitted from fewer but larger sources. Figure 2c reveals some stratification between annual O3 values in Canada vs the U.S., and AURAMS can be seen in Figure 2d to overpredict HNO3 concentrations, consistent with the ~50% overprediction error for NO emissions. Model performance was significantly better for PM2.5 than for PM10, but model predictions were biased low for both (Figure 3). Figure 4 shows corresponding scatterplots for six chemical components of PM2.5. Model performance varies greatly by chemical component, reflecting the complex nature of PM2.5, which has both primary and secondary sources with different atmospheric pathways. For the inorganic ions, correlation coefficients for PM2.5 SO4, NO3, and NH4 range from 0.75 to 0.90. Similar results were obtained for the CAPMoN and CASTNet networks that had no PM size cut (see Moran et al., 2007). The two carbonaceous components EC and TOM, on the other hand, had R values of 0.61 and 0.50 but also strong negative biases. And for PM2.5 CM the model showed little skill.
4.2. Annual concentrations in precipitation Figure 5 examines model skill in predicting wet removal. Prediction of annual precipitation by the GEM model is very good, with an R value of 0.85 and a small positive bias. Model predictions of SO4, NO3, and NH4 concentrations in precipitation are also good, with R values of 0.83, 0.72, and 0.80, respectively. Like SO4
c
d
Fig. 2 Scatterplots of measured vs predicted 2002 annual mean surface air concentrations (ppbV) of (a) SO2, (b) NO2, (c) O3, and (d) HNO3. The 2:1, 1:1, 1:2, and linear best-fit lines are also plotted
Comprehensive Surface-Based Performance Evaluation of a Size and Com
439
and NH4 annual air concentrations, SO4= and NH4+ annual concentrations in precipitation are biased low. Annual NO3- concentration in precipitation, on the other hand, is biased high, consistent with the positive bias for HNO3 in air (Figure 2d). Model performance for SO4, NO3, and NH4 wet deposition (not shown) was comparable to that for concentration in precipitation (see Moran et al., 2007).
a
b
Fig. 3 Same as Figure 2 but for (a) PM2.5 and (b) PM10 mass [ȝg m-3 STP (0°C)]
a
b
c
e
d
f
Fig. 4 Same as Figure 3 but for six PM2.5 chemical components: (a) SO4; (b) NO3; (c) NH4; (d) CM; (e) EC; and (f) TOM
M. Moran et al.
440
a
c
b
d
Fig. 5 Scatterplots of measured vs predicted 2002 annual (a) precipitation (mm) and (b) SO4=, (c) NO3-, and (d) NH4+ concentrations in precipitation (mg L-1)
5. Discussion The results of this evaluation have provided a number of insights into current model performance and model weaknesses. Overall model prediction for sulphur and oxidized and reduced nitrogen species in both air and precipitation was quite good, with the exception of those overpredictions resulting from too high NO emissions. The marked underprediction of the PM carbonaceous components has suggested improvements to the current speciation of PM emissions and to the treatment of secondary organic aerosol formation, especially for biogenic precursors. The underprediction of O3 and PM2.5 SO4 concentrations among others has pointed to weaknesses in the zero-gradient chemical lateral boundary conditions that were used for this annual simulation (Samaali et al., 2006). See Moran et al. (2007) for a more detailed discussion and analysis. Acknowledgments Almost all of the AQ data sets that have been used in this study were obtained from the National Atmospheric Chemistry (NAtChem) data base, an AQ data “clearinghouse” operated by Environment Canada (see www.mscsmc.ec.gc.ca/natchem/index_e.html). Thanks are due as well to the individual AQ networks that made their data available to NAtChem, including AQS, CAPMoN, CASTNet, IMPROVE, NADP, NAPS, and STN.
Comprehensive Surface-Based Performance Evaluation of a Size and Com
441
References Bouchet VS, Moran MD, Crevier L-P, Dastoor AP, Gong S, Gong W, Makar PA, Menard S, Pabla B Zhang L (2003) “Wintertime and summertime evaluation of the regional PM air quality model AURAMS: Proc. 26th NATO/CCMS ITM, May 26–29, Istanbul, Turkey. Côté J, Desmarais J-G, Gravel S, Méthot A, Patoine A, Roch M, Staniforth A, (1998) Mon. Wea. Rev., 126, 1373–1395. Eder, B and Yu S, 2006, Atmos. Environ., 40, 4811–4824. Gong W, Dastoor AP, Bouchet VS, Gong S, Makar PA, Moran MD, Pabla B, Ménard S, Crevier L-P, Cousineau S, Venkatesh S (2006) Atmos. Res., 82, 248–275. Makar PA, Bouchet VS, Gong W, Moran MD, Gong S, Dastoor AP, Hayden K, Boudries H, Brook J, Strawbridge K, Anlauf K, Li S-M (2004) “AURAMS/ Pacific2001 Measurement Intensive comparison”. Proc. 27th NATO/CCMS ITM, October 25–29, Banff, Alberta, Canada. McKeen S, Wilczak J, Grell G, Djalalova I, Peckham S, Hsie E-Y, Gong W, Bouchet V, Ménard S, Moffet R, McHenry J, McQueen J, Tang Y, Carmichael GR, Pagowski M, Chan A, Dye T (2005) J. Geophys. Res., 110, D21307, doi:10.1029/2005JD005858, 16 pp. McKeen S, Chung SH, Wilczak J, Grell G, Djalalova I, Peckham S, Gong W, Bouchet V, Moffet R, Tang Y, Carmichael GR, Mathur R, Yu S (2007) J. Geophys. Res., 112, D10S20, doi:10.1029/2006JD007608, 20 pp. Moran MD, Zheng Q, Samaali M (2007) Long-term multi-species performance evaluation of AURAMS for first 2002 annual run. EC internal report, Toronto, Ontario (in preparation). (Available from first author: email [email protected]) Samaali M, Pavlovic R, Moran MD, Cousineau S, Bouchet VS, Gong W, Makar PA, Zhang J (2006) Influence of the type of chemical lateral boundary condition on regional chemical transport model forecasts. Poster A51C-0097, 2006 AGU Fall Meeting, December 11–15, San Francisco. Seigneur C, Moran MD (2004) Using models to estimate particle concentration. Chapter 8 in Particulate Matter Science for Policy Makers: A NARSTO Assessment, P. McMurry, M. Shepherd, and J. Vickery, Editors, February, 42 pp. Cambridge University Press, Cambridge.
442
M. Moran et al.
Discussion M. Schaap: You showed a large overestimation of dust in populated areas. Could you comment on the sources involved and the main uncertainties associated with it? W. Gong: Dust or crustal material in this version of the model comes from anthropogenic primary emissions only. There is a considerable uncertainty in the inventory estimates of dust emissions, particularly for fugitive dust emissions, to which emissions from paved and unpaved roads, construction, and agricultural activities make the largest contributions. The spatial allocation factor fields used to disaggregate these emissions, which are reported on a jurisdictional basis, to the model grid also tend to emphasize urban areas (e.g., kilometres of road, number of dwellings). D. Steyn: Earlier performance evaluation (e.g. as presented at 26th ITM) showed markedly poor performance. What has caused this improvement in model performance? W. Gong: There have been a number of upgrades in the model both in terms of model science and input emissions data (better and more up-todate inventories) that contributed to the improved performance. It needs to be pointed out, though, that this evaluation is for a long time period (one year). Air quality models tend to perform better for longer than for shorter time scales due to averaging of short-term fluctuations.
4.15 Development of a New Canadian Operational Air Quality Forecast Model D. Talbot1, M.D. Moran2, V. Bouchet1, L.-P. Crevier1, S. Ménard1, A. Kallaur2 and the GEM-MACH Team
Abstract Development and implementation of the next generation of Canada’s operational air quality (AQ) forecast model is underway at Environment Canada (EC). The goal of this project is the replacement in 2008 of the current operational off-line AQ forecast model, CHRONOS, by GEM-MACH, an on-line chemical transport model. To construct GEM-MACH, chemistry modules have been implemented directly inside GEM, EC’s operational multi-scale meteorological forecast model. This new on-line AQ forecast model will be able to exploit EC’s massively parallel supercomputer via the parallelism options already implemented in GEM. Physical and chemical processes related to AQ are solved on GEM’s “native” grid, thus avoiding the spatial and temporal interpolations of the meteo-rological fields that are required by CHRONOS. CHRONOS : Canadian Hemispherical Regional Ozone and NOx System GEM : Global Environmental Multiscale model MACH : Modelling Air quality and CHemistry
Keywords CHRONOS, forecast model, model evaluation 1. Introduction The prediction of air quality (AQ) is a complicated problem since it requires the time evolution of a large number of coupled meteorological and chemical fields to be modelled. The traditional approach to modelling AQ has been to separate the chemical transport model (CTM) from the meteorological model (MM). In this “off-line” approach, the meteorological “driver” model is run first, and then predicted meteorological fields such as wind, temperature, turbulence intensity, cloud liquid water, and precipitation are input to the CTM and used to model the emission, transport and diffusion, chemical transformation, and dry and wet removal of a set of trace chemical species. 1 2
Environment Canada, 2121, Route Transcanadienne, Montreal, QC, H9P 1J3, Canada Environment Canada, 4905 Dufferin Street, Toronto, ON, M3H 5T4, Canada
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
470
Development of a New Canadian Operational Air Quality Forecast Model
471
The off-line approach has several advantages. One computational advantage is that in retrospective studies, where a series of CTM runs may be required for the same period, as might be the case for a set of emission-change scenarios, sourceapportionment runs, or process sensitivity runs, the MM need only be run once and then its output files can be used for multiple CTM runs. A second advantage is that it permits a “division of labour”, in that different people or even different groups are able to develop, maintain, and run the MM and the CTM separately, thereby distributing the work load and allowing individuals to develop deeper expertise with one model as opposed to broader expertise across two models. The off-line approach, however, also suffers from a number of disadvantages: x It does not fully reflect the complex, two-way coupling between meteorological and chemical fields in the real atmosphere, where chemical fields can influence meteorological processes (e.g., Jacobson, 1998). In the off-line approach, the modelled coupling is one-way and only the influence of meteorological fields on chemical processes can be considered. x Meteorological fields are provided to the CTM at fixed time intervals, in effect imposing time filtering and removing information about higherfrequency atmospheric processes (e.g., Grell et al., 2004). x If the MM and CTM grids are not identical, then horizontal and/or vertical interpolation will be required, which will then introduce physical inconsistencies and errors (e.g., Byun, 1999). x Running the MM first means that all of the meteorological fields needed by the CTM must be saved on disk and then input by the CTM. Particularly for high-resolution simulations with small time steps or very long simulations, the data volume that must be saved can become very large. As well, for forecasting applications, the meteorological and chemical fields both must be predicted only once, so that no time is saved by calculating them sequentially in an off-line CTM rather than simultaneously in an on-line CTM. Given the above considerations, it is no surprise that on-line AQ models have now started to appear. Three early examples are the GATOR-GCMM model (Jacobson, 2001, 2006), the MM5/Chem model (Grell et al., 2000, 2004), and the MC2-AQ model (Plummer, 1999; Plummer et al., 2001; Kaminski et al., 2002; Yang et al., 2003). Chemistry modules have also been introduced recently into operational numerical weather prediction models in Canada (GEM model: Dastoor and Larocque, 2004; O’Neill et al., 2006) and the USA (WRF model: Grell et al., 2005). Environment Canada (EC) has employed an off-line regional AQ modelling system to issue national two-day public ozone forecasts since summer 2001 and PM2.5 forecasts since summer 2003. The MM used by EC is the GEM model (Côté et al., 1998a, b; Yeh et al., 2002) and the CTM is the CHRONOS model (Pudykiewicz et al., 1997; Sirois et al., 1999; McKeen et al., 2005, 2007; Tarasick et al., 2007). Development of the next generation of Canada’s operational AQ forecast model is now underway at EC. The goal of this project is to replace CHRONOS in 2008 with GEM-MACH, a new on-line chemical transport model whose host MM is GEM. The rest of this paper will provide an overview of the GEM-MACH design and the steps required to implement it.
D. Talbot et al.
472
2. Planned GEM-MACH Configuration GEM was a logical choice to be the host model for GEM-MACH given that (a) it has already been used successfully in EC’s current operational AQ forecast system, and (b) it has strong development and maintenance support due to its role as EC’s operational regional and global weather forecast model. In addition, though, GEM offers two more major advantages. The first advantage is that the GEM code has already been extensively parallelized for optimal performance on EC’s massively parallel supercomputers using both coarse parallelism tools (OpenMP) and fine parallelism tools (Message Passing Interface [MPI]), whereas CHRONOS has only been coarsely parallelized and so is limited to the number of CPUs on a single shared-memory “node” (was 8, recently upgraded to 16). This means that, like GEM, a single GEM-MACH simulation will be able to be run on hundreds of CPUs. The second advantage is that GEM supports three different grid configurations: (a) a global uniform grid; (b) a global variable grid; and (c) a nested regional limited-area grid. Although the initial implementation of GEM-MACH will be in a regional configuration, it could also be run in a global configuration, which means that it could serve as both a regional CTM and as a global CTM in the same way that GEM now serves EC as both a regional MM and a global MM. GEM15 is the regional configuration of GEM currently used by EC to make short-range forecasts (up to two days). As described by Mailhot et al. (2006), GEM15 employs a variable-resolution global grid with a portion covering North America at a uniform resolution of 15 km. GEM15 starts with a relatively short cutoff time and completes a forecast after only 35 minutes since it runs on 400 CPUs. Following the GEM15 integration, it is expected that GEM-MACH will be launched from the same meteorological analysis in a LAM (Limited Area Model) configuration with meteorological boundary conditions given by the GEM15 forecast. The tentative GEM-MACH grid is a sub-area of the uniform portion of the GEM15 with co-located grid points (Figure 1). For reference, the current CHRONOS grid is also shown in Figure 1. Since GEM-MACH will be so tightly embedded in GEM15, it will share most of the GEM15 grid configuration and integration characteristics. The following table (Table 1) shows how with the proposed configuration, GEM-MACH will be a higher-resolution model in both space and time relative to the current operational CHRONOS model. Table 1 Comparison of GEM-MACH and CHrONOS grid configuration and integration characteristics. GEM-MACH
CHRONOS
Rotated lat-long map projection ǻx = 15 km, 465 × 400 grid points 58 eta levels, top at 10 hPa (~30 km) GEM15 topography ǻt = 450 s, 48-hour integration
Secant polar-stereographic projection ǻx = 21 km, 350 × 250 grid points 24 Gal-Chen levels, top at ~6 km Topography at 21 km resolution ǻt = 3,600 s, 48-hour integration
Development of a New Canadian Operational Air Quality Forecast Model
473
As well, physical and chemical processes related to AQ will be solved on GEM’s “native” grid, with no need to interpolate in space or in time as is now done in CHRONOS. Furthermore, the AQ component of the model is fully consistent with the meteorological component since both use the same semi-Lagrangian advection scheme and the same physics schemes where appropriate. Although it has been assumed that chemistry will not impact on meteorology at the present time, this new on-line configuration does open the door for an eventual two-way feedback between chemistry and meteorology.
Fig. 1 Tentative GEM-MACH grid (465 × 400, dx = 15 km) in dark gray underneath the smaller operational CHRONOS grid (350 × 250, dx = 21 km) in a lighter gray
3. Emissions in GEM-MACH Work was required to develop an emissions interface in GEM to include two different types of emissions (gridded surface emissions and major point sources) and to propagate the information to the chemistry module. Prior to this project, GEM had no need to input a set of time-dependent files and distribute the values to the domain-decomposed grid “tiles” during an integration, but the GEM developers now view the new emissions interface more generally as a “forcing” interface that can also be used to input other time-dependent gridded and point-format files to GEM for other purposes.
3.1. Anthropogenic emissions The emissions processing system SMOKE is run off-line to produce a “library” of anthropogenic emission files on the GEM-MACH grid (CEMPD, 2007). Emission fields are prepared for each month for each day of the week for each hour of the day. They are assumed to be constant for the eight model time steps subdividing each hour. Major point source emissions correspond to non-gridded point locations that are assigned to the nearest model grid cell. The newest emission inventories that are available will be used in the operational version of GEM-MACH. In the meantime, the prototype model uses the 2000 Canadian, 2001 U.S., and 1999 Mexican national emission inventories.
474
D. Talbot et al.
3.2. Biogenic emissions The initial version of GEM-MACH is based on BEIS v3.09 with updated emissions factors from v3.13 (Vukovich and Pierce, 2002; Schwede et al., 2006). The biogenics vegetation database consists of the standard U.S. EPA BELD3 dataset modified over Canada with data from the 2001 Canadian National Forest Inventory and 2001 Canadian Agricultural Census (Pierce et al., 2000; CFS, 2007; Statistics Canada, 2002). Over regions not covered by the BELD3 data, 1-km USGS land characteristics data are used to generate the biogenic emissions (USGS, 2001). To save computational time, the vegetation data are pre-processed to generate reference emissions on the model grid for four species (NO, isoprene, monoterpenes, and other VOCs) as well as for the leaf area index. At the moment, two such datasets are constructed, one for summer and one for winter conditions. At run time the reference emissions are read by the model and modulated by the season, temperature, and photosynthetically active radiation to calculate the final biogenic emissions at each time step and grid point.
4. Chemistry in GEM-MACH For the initial development phase, the inclusion of chemistry in GEM has been viewed as an implementation task in which existing chemistry process modules from other EC CTMs are incorporated into the GEM chemistry library after any modifications required for the GEM environment (e.g., GEM vertical coordinate system). The main source of these process modules is the AURAMS (A Unified Regional Air-quality Modelling System) CTM, an off-line, size-resolved, chemically-characterized regional PM model (e.g., Gong et al., 2006; Tarasick et al., 2007), but process modules from CHRONOS (Pudykiewicz et al., 1997), from GRAHM (Dastoor and Larocque, 2004), and from GEM-AQ (O’Neill et al., 2006) have also been considered. Both GRAHM and GEM AQ are GEM-based and have served as prototypes for GEM-MACH. The initial set of process representations that are being implemented in order to achieve equivalence with CHRONOS are (a) plume rise for major point sources, (b) transport and vertical diffusion of chemical species, (c) gas-phase chemistry, (d) secondary organic aerosol (SOA) formation, (e) dry deposition of gases, (f) dry deposition and sedimentation of particles, (g) wet removal of gases, (h) wet removal of particles, (i) inorganic heterogeneous chemistry, (j) cloud modulation of photolysis rates, and (k) aqueous-phase chemistry. Descriptions of some of these process representations can be found in Gong et al. (2006), Jiang (2003), Makar et al. (2003a, b), Stockwell and Lurmann (1989), and Zhang et al. (2001, 2002). Additional process representations, including those related to aerosol kinetics, natural PM emissions, and subgrid-scale convective transport, will be added in later versions.
Development of a New Canadian Operational Air Quality Forecast Model
475
In order to be useful in the operational setting, a 48-hour forecast of GEMMACH must be completed in less than two hours, and the envisioned configuration of the model is quite ambitious. Strategies to reduce the duration of integration are being developed, and some decisions may have to be made that could include such steps as the simplification of some process representations or confining chemistry calculations to the lower part of the atmospheric column.
5. Summary The development of GEM-MACH at EC is expected to lead to a new generation of AQ forecast model and will open new doors for the future. This on-line AQ model will take advantage of the MPI capabilities of GEM, making it possible for a more computational demanding model to be implemented operationally. Not only is the domain of integration enlarged, temporal and spatial resolution increased, and spatial interpolation avoided, but more complex chemical process representations will also be included. A new chemical library along with a mechanism to read the diverse pollutant emissions have been developed to work within the current GEM host model. Acknowledgments The numerous contributions to this effort of the other members of the EC GEM-MACH team, including D. Davignon, S. Gaudreault, S. Gong, W. Gong, S. Gravel, H. Landry, P.A. Makar, B. Pabla, M. Sassi, and C. Stroud, are gratefully acknowledged, as are contributions to the development of the GEM chemistry interface by S. Chabrillat of the Belgium Institute for Space Aeronomy in Brussels, Belgium.
References Byun DW (1999) Dynamically consistent formulations in meteorological and air quality models for multiscale atmospheric studies. Part II: Mass conservation issues. J. Atmos. Sci., 56, 3789–3807. CEMPD (2006) http://www.smoke-model.org/index.cfm (visited 6 June 2007). CFS (2007) http://www.pfc.forestry.ca/monitoring/inventory/index_e.html (visited 6 June 2007). Côté J, Desmarais J-G, Gravel S, Méthot A Patoine A, Roch M, Staniforth A (1998a) The operational CMC/MRB Global Environmental Multiscale (GEM) model. Part 1: Design considerations and formulation, Mon. Wea. Rev., 126, 1373–1395. Côté J, Desmarais J-G, Gravel S, Méthot A, Patoine A, Roch M, Staniforth A, (1998b) The operational CMC-MRB Global Environment Multiscale (GEM) model. Part II: Results, Mon. Wea. Rev., 126, 1397–1418.
476
D. Talbot et al.
Gong W, Dastoor AP, Bouchet VS, Gong S, Makar PA, Moran MD, Pabla B, Ménard S, Crevier L-P, Cousineau S, Venkatesh S (2006) Cloud processing of gases and aerosols in a regional air quality model (AURAMS), Atmos. Res., 82, 248–275. Grell GA, Emeis S, Stockwell WR, Schoenemeyer T, Forkel R, Michalakes J, Knoche R, Seidl W (2000) Application of a multiscale, coupled MM5/chemistry model to the complex terrain of the VOTALP valley campaign, Atmos. Environ., 34, 1435–1453. Grell GA, Knoche R, Peckham SE, McKeen SA (2004) Online versus offline air quality modeling on cloud-resolving scales, Geophys. Res. Lett., 31, L16117 1–4. Grell GA, Peckham SE, Schmitz R, McKeen SA, Frost G, Skamarock WC, Eder B, (2005) Fully coupled “online” chemistry within the WRF model, Atmos. Environ., 39, 6957–6975. Jacobson MZ (1998) Studying the effects of aerosols on vertical photolysis rate coefficient and temperature profiles over an urban airshed, J. Geophys. Res., 103, 10593–10604.Jacobson MZ (2001) GATOR-GCMM: a global through urban scale air pollution and weather forecast model. 1. Model design and treatment of subgrid soil, vegetation, roads, rooftops, water, sea ice, and snow, J. Geophys. Res., 106, 5385–5402. Jacobson MZ (2006) Comment on “fully coupled ‘online’ chemistry within the WRF model” by Grell et al., 2005, Atmospheric Environment 39, 6957–6975, Atmos. Environ., 40, 4646–4648. Jiang W (2003) Instantaneous secondary organic aerosol yields and their comparison with overall aerosol yields for aromatic and biogenic hydrocarbons, Atmos. Environ., 37, 5439–5444. Kaminski JW, Plummer DA, Neary L, McConnell JC, Struzewska J, Lobocki L (2002) First application of MC2-AQ to multiscale air quality modelling over Europe, Phys. Chem. Earth, 27, 1517–1524. Makar PA, Moran MD, Scholtz MT, Taylor A (2003a) Speciation of volatile organic compound emissions for regional air quality modelling of particulate matter and ozone. J. Geophys. Res., 108 (D2), 4041, doi:10.1029/2001JD000797, 51 pp. Makar PA, Bouchet VS, Nenes A (2003b) Inorganic chemistry calculations using HETV ҟ a vectorized solver for the SO42-ҟNO3-ҟNH4+ system based on the ISORROPIA algorithms, Atmos. Environ., 37, 2279–2294. Mailhot J, Bélair S, Lefaivre L, Bilodeau B, Desgagné M, Girard C, Glazer A, Leduc A-M, Méthot A, Patoine A, Plante A, Rahill A, Robinson T, Talbot D, Tremblay A, Vaillancourt P, Zadra A, Qaddouri A (2006) The 15-km version of the Canadian regional forecast system, Atmos.-Ocean, 44, 133–149. McKeen S, Wilczak J, Grell G, Djalalova I, Peckham S, Hsie E-Y, Gong W, Bouchet V, Ménard S, Moffet R, McHenry J, McQueen J, Tang Y, Carmichael GR, Pagowski M, Chan A, Dye T (2005) Assessment of an ensemble of seven real-time ozone forecasts over Eastern North America during the summer of 2004, J. Geophys. Res., 110, D21307, doi:10.1029/2005JD005858, 16 pp.
Development of a New Canadian Operational Air Quality Forecast Model
477
McKeen S, Chung SH, Wilczak J, Grell G, Djalalova I, Peckham S, Gong W, Bouchet V, Moffet R, Tang Y, Carmichael GR, Mathur R, Yu S (2007) Evaluation of several real-time PM2.5 forecast models using data collected during the ICARTT/NEAQS 2004 field study, J. Geophys. Res., 112, D10S20, doi:10.1029/2006JD007608, 20 pp. O’Neill NT, Campanelli M, Lupu A, Thulasiraman S, Reid JS, Aubé M, Neary L, Kaminski JW, McConnell JC (2006) Evaluation of the GEM-AQ air quality model during the Québec smoke event of 2002: analysis of extensive and intensive optical disparities, Atmos. Environ., 40, 3737–3749. Pierce T, Kinnee EJ, Geron CD (2000) Development of a 1-km vegetation database for modelling biogenic fluxes of hydrocarbons and nitric oxide, 6th Int’l Conf. on Air-Sea Exchange of Gases and Particles, July, Edinburgh. Plummer DA (1999) On-line chemistry in a mesoscale model: assessment of the Toronto emission inventory and lake-breeze effects on air quality, Ph.D. thesis, York University, Toronto, Canada, 281 pp. Plummer DA, McConnell JC, Neary L, Kaminski J, Benoit R, Drummond J, Narayan J, Young V, Hastie DR (2001) Assessment of emissions data for the Toronto region using aircraft-based measurements and an air quality model, Atmos. Environ., 35, 6453–6463. Pudykiewicz JA, Kallaur A, Smolarkiewicz PK (1997) Semi-Lagrangian modelling of tropospheric ozone, Tellus, 49B, 231–248. Schwede D, Pouliot G, Pierce T (2006) Changes to the Biogenic Emissions Inventory System Version 3 (BEIS3). 5th Annual CMAS Conference, October 16–18, Chapel Hill, North Carolina, Community Modeling & Analysis System Center, University of North Carolina at Chapel Hill, North Carolina (Available from http://www.cmascenter.org/conference/2005/abstracts/2_7.pdf). Sirois A, Pudykiewicz JA, Kallaur A (1999) A comparison between simulated and observed ozone mixing ratios in eastern North America, J. Geophys. Res., 104, 21397–21423. Statistics Canada (2002, 2001) Census of Agriculture, Ottawa, Canada (see http:// www.statcan.ca/english/freepub/95F0301XIE/index.htm) Stockwell WR, Lurmann FW (1989) Intercomparison of the ADOM and RADM Gas-Phase Chemical Mechanisms, Electrical Power Research Institute Topical Report, EPRI, 3412 Hillview Avenue, Palo Alto, CA, 254 pp. Tarasick DW, Moran MD, Thompson AM, Carey-Smith T, Rochon Y, Bouchet VS, Gong W, Makar PA, Stroud C, Ménard S, Crevier L-P, Cousineau S, Pudykiewicz JA, Kallaur A, Moffet R, Ménard R, Robichaud A, Cooper OR, Oltmans SJ, Witte JC, Forbes G, Johnson BJ, Merrill J, Moody JL, Morris G, Newchurch MJ, Schmidlin FJ, Joseph E (2007) Comparison of Canadian air quality forecast models with tropospheric ozone profile measurements above mid-latitude North America during the IONS/ICARTT campaign: evidence for stratospheric input, J. Geophys. Res., 112, D10S20, doi:10.1029/2006JD007782. USGS (2001) Global land cover characteristics data base, Earth Resources Observation and Science Centre, U.S. Geological Survey, Sioux Falls, South Dakota (Available from http://edcsns17.cr.usgs.gov/glcc/).
D. Talbot et al.
478
Vukovich J, Pierce T (2002) The implementation of BEIS3 within the SMOKE modeling framework. 11th Emission Inventory Symp., April 16–18, Atlanta, Georgia, 7 pp. (Available from http://www.epa.gov/ttn/chief/conference/ei11/ modeling/vukovich.pdf). Zhang L, Moran MD, Brook JR (2001) A comparison of models to estimate in-canopy photosynthetically active radiation and their influence on canopy stomatal resistance. Atmos. Environ., 35, 4463–4470. Zhang L, Moran MD, Makar PA, Brook JR, Gong S (2002) Modelling gaseous dry deposition in AURAMS – A Unified Regional Air-quality Modelling System, Atmos. Environ., 36, 537–560.
Discussion E. Terrenoire: Any performances of GEM-MACH in respect to the CRONOS model? D. Talbot: No, too early for results. P. Davidson: When you update the GEM-MACH with new weather analysis, will you use GEM-MACH analysed weather or weather from GEM15 operational? Will there be a discontinuity with warm-start chemistry? D. Talbot: 1. GEM-MACH will start with the same weather analysis as GEM15. 2. The chemistry of the new run will be taken from the previous run. There will be a small inconsistency with the new meteorology analysis but this is what we have best to start the chemistry.
4.14 Estimation of the Modelling Uncertainty Related with Stochastic Processes Oxana Tchepel, Alexandra Monteiro and Carlos Borrego
Abstract In the present work a methodology for quantification of modelling uncertainty using decomposed measured data is proposed. The original measured data are decomposed to deterministic and short-term components before the statistical evaluation of the modelling results is performed against the measurements. Using Fourier analysis, the spectral density was obtained for different types of air quality monitoring stations. Next, short-term fluctuations were subtracted from the original data using an iterative moving average filter and taking into account the contribution of higher frequencies determined from the spectral analysis. The methodology was used to estimate uncertainties of the results obtained with CHIMERE model for Portugal. The modelling outputs for one year are compared with the measurement data from different types of air quality stations after the subtraction of short-term variations. The comparison shows a better agreement after the application of the decomposed time series methodology.
Keywords Air pollution modelling, spectrum analysis, time-series decomposition
1. Introduction Quantification of uncertainties in predicted pollutant concentrations is one of the challenging issues in air pollution modelling. Although there is no common set of statistical parameters to access the uncertainty, the defined methodologies are usually based on a comparison of modelling results with observation data (Borrego et al., 2007). However, the observation data have uncertainties too. These uncertainties are due to errors related with instrumental techniques (systematic error due to imprecise calibration and random errors associated with imperfections in measurement techniques), but also include sampling uncertainty associated with inherent variability of the data and representativeness of the samples. Therefore, and taking into account different spatial representativeness of the data obtained at a monitoring point and that from model calculations (performed for a grid cell), a discrepancy between the measured and predicted values is inevitable. It is related, primarily, to the short-term fluctuations (stochastic processes) of the pollutant concentration. The physics of the pollutant dispersion contributes to large natural variability in concentration and to inherent uncertainty of the model predictions (Weil, 1992). C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
461
O. Tchepel et al.
462
Besides that, the emissions should also be considered as a stochastic process that can be measured precisely, in principle, but not in practice when a large number of sources with complex emission patterns are present. In the scope of this work, a methodology for quantification of modelling uncertainty related with stochastic processes, using decomposed measured data, is proposed. To achieve the defined objective: (i) a spectrum analysis of the air quality measurement data was performed, and (ii) the stochastic short-term variation is removed from the original data prior to the model validation. Finally, an example of application of the methodology is presented for one-year of CHIMERE model predictions.
2. Data Processing Methodology 2.1. Original data Air quality monitoring data for 2004 were analysed for different stations located in Lisbon and Porto regions (Table 1). The proximity to the emission sources and, therefore, different temporal and spatial representativeness expected for the station was a selection criterion. The time series consist of one-year pollutant concentration values with 1-hour resolution for NO, NO2 and O3. Only data with completeness above 80% were considered in the study. To transform the data set to normality and to stabilise the variance, the data were log-transformed (y = ln (concentration)) before the analysis. Table 1 Characteristics of the air quality monitoring stations considered for the analysis.
Name Location Type
Benfica Lisbon Urban traffic
Chelas Lisbon Urban background
Chamusca Lisbon Region Regional background
Boavista Porto Urban traffic
Vila N.Telha Porto/Maia Suburban background
2.2. Spectrum analysis The spectrum analysis allows to characterise the time series in frequency domain and it is complementary to the autocorrelation function defined in the time domain. Using the spectrum analysis, a contribution of different frequencies to the variance can be determined and, thus, to distinguish different phenomena presented in the time series.
Estimation of the Modelling Uncertainty Related with Stochastic Processes
463
A time series Xt of length N is presented as a linear combination of finite number of harmonic functions with frequencies {fj} and random amplitudes {Aj} and {Bj}:
Xt
> N / 2@
ȝ
¦ j 1
ª¬ A j cos( 2ʌ f j t) B j sin( 2ʌ f j t)º¼ ,
(1)
where µ is a constant term, (N/2) is the greatest integer less than or equal to N/2. For one-year air quality measurements with hourly resolution N is equal to 8,760 (365*24) and the highest frequency is f =(N/2)/N = ½ = 0.5, which correspond to a wave period (period = 1/frequency) of 2 hours. Therefore, the measurement data with 1-hour resolution do not allow analysing waves with period less then 2 hours. The fast Fourier transform (FFT) algorithm was used to study the variance in the frequency-domain. The periodogram values (Pj) at frequency fj were calculated as:
Pj
A
2 j
B 2j u N / 2
(2)
The parameter Pj can be interpreted in terms of variance and allow to identify which frequencies give most important contribution to the variability.
2.3. Data filtering Depending on the analysis objective and the properties of the phenomena under study, it could be important to remove some frequencies from the series. For this purpose the data filtering techniques can be used. The Kolmogorov-Zurbenko (KZ) filter commonly applied to the air quality data (Rao et al., 1997; Hogrefe et al., 2006) is used to decompose the time series into deterministic and stochastic components. The KZ filter is a low-pass filter that removes higher frequency variations from the data. The KZ(m,k) filter of the original time series x is computed as a simple moving average of m points applied k times (number of iterations) The filter is designed taking into account the desired separation frequency wc (Rao et al., 1997). The application of the KZ filter allows to decompose the original time series C(t) on baseline (CB) (deterministic) and short-term (CS) (stochastic) components in time t (Rao et al., 1997): C(t)=CB(t)+CS(t). Thus, the output of the filtering process corresponds to the baseline component and the short-term component which is defined as a difference between the original and the filtered data.
3. Photochemical Model Application The air quality was modelled with the CHIMERE chemistry-transport model (Schmidt et al., 2001), forced by the MM5 meteorological fields (Figure 1), and carried out for 2004 year, regarding gaseous and particulate pollutants (Monteiro et al., 2007).
O. Tchepel et al.
464 meteorological model MM5
boundary conditions
emissions
climatological model GOCART
anthropogenic biogenic
CHIMERE gaseous chemistry
aerosols module
44 species, 116 reactions
Fig. 1 The MM5-CHIMERE air quality modelling system
transport
gas deposition/aerosols
horizontal (PPM) vertical diffusion
dry wet
pollutants concentration
pollutants deposition
gases/aerosols
gases/aerosols
In the present application, the model is run at a regional scale over a domain covering all mainland Portugal, with a 10 × 10 km2 grid size with a vertical resolution of eight vertical layers of various thickness extending from ground to 500 hPa. Boundary conditions are provided by a prior large-scale simulation, covering Western Europe with a 50 × 50 km2 resolution. Boundary conditions for the regional simulation are taken from the monthly means of the MOZART and GOCART models, as in Hodzic et al. (2005). Regarding anthropogenic emissions, the most updated annual emission inventory (2003 year) was used, disaggregated at the municipality level (Monteiro et al., 2007). Simpson et al. (1999) methodology was adopted to calculate biogenic emissions with the CHIMERE model. Time disaggregation was obtained by application of monthly, weekly and hourly profiles from the University of Stuttgart (Monteiro et al., 2007).
4. Results and Discussion Figure 2 presents the result of the statistical analysis of the one-year log-transformed observation and modelling time series considered in the analysis. A comparison of the modelling data with the measurements shows underestimation of the NO concentration predicted by the model (median value) and wider spread. It can be seen that the model predicts the same distribution for the urban background station (Chelas) and for the urban traffic station (Benfica), while the measurements reveal significant difference between these two observation points located in proximity. This tendency is also observed for NO2 time series. In general, for NO2 and O3 the model outputs are in good agreement with the observations in terms of median, although the spread of the data in the interquartile range is different.
Estimation of the Modelling Uncertainty Related with Stochastic Processes
465
These log-transformed data were used for the spectrum analysis after removing the linear trend and the subtracting the annual average log concentration. An example of the results of the spectrum analysis for NO2 is presented in Figure 3 for several locations. The plot represents the relative contributions of different frequencies to the variance. Measurement data
Modelling results Modelling results
6
6
4
4
2
2
0
Chamusca_O3
Boavista_O3
Vila N.T._O3
Chamusca_NO2
Chelas_NO2
99% 1% 75% 25% Median
Vila N.T._NO2
Vila N.T._O3
Chamusca_O3
Boavista_O3
Chamusca_NO2
Chelas_NO2
Vila N.T._NO2
-8
Benfica_NO2
-6
-8 Chelas_NO
-4
-6
Benfica_NO
-4
Benfica_NO2
-2
Chelas_NO
-2
Benfica_NO
0
Chamusca_NO
ln(Concentration)
8
Chamusca_NO
ln(Concentration)
Measurement data
8
99% 1% 75% 25% Median
Fig. 2 Box plot of the measurement and CHIMERE modelling data. Concentration in Pg m-3
The variance spectrum of the measured time series reveals that higher frequencies (small period, T = 1/f) have important contribution in areas with direct influence of traffic, while for the rural station seasonal variations are most important. Nevertheless, all stations present clearly identified peaks for the period of T = 12 hours and T = 24 hours (f § 0.08 h-1 and f § 0.04 h-1). For the Benfica urban traffic station, the contribution of the wave with the 12 hours period to the total variance is even higher than day-to-night variations represented in the spectrum by the wave with T = 24 hours. This pattern is clearly related with fluctuations in the traffic emissions. The analysis of the CHIMERE variance spectrum shows that seasonal variations are underestimated in the modelled timeseries. Moreover, one-week variance peak is missing in the model spectrum due to incorrection on working day/weekend emissions disaggregation. However, the 12and 24-hours peaks are well identified and in some cases are even overrated. Based on the spectrum analysis the separation frequency for the KZ filter was defined with the purpose to remove all fluctuations with the period less than 12 hours from the original data. The filter parameters m = 3 and k = 3 were selected, providing the separation frequency wc = 0.0905 h-1. The KZ3,3 filter was applied to the log-transformed measurement and predicted hourly concentration data, thus allowing the deterministic component and short-term noise to be separated. The short-term component is obtained as the filter residual. It is characterised by zero mean and covariance with the baseline of about 2% showing a good effectiveness of the timescales separation for NO2 and O3. The covariance for NO is about 8%. The distribution parameters of the short-term components for different pollutants are presented in Figure 4. The data reveal very good agreement between Ozone modelling data and observations at rural background station (Chamusca) but significant spread for NO and NO2. For the urban stations, both traffic (Benfica)
466
O. Tchepel et al.
and background (Chelas), the model is underestimating NO and overestimating NO2 short-term variability. For O3 short-term component at urban station (Boavista), the model predictions are characterized by larger spread in the 98percentile but narrower in the interquartile range.
Fig. 3 Variance spectrum for the log-transformed hourly NO2 measurements and CHIMERE predictions for urban traffic station (Benfica), urban background station (Chelas), sub-urban background station (Vila N Telha) and regional background station (Chamusca)
Estimation of the Modelling Uncertainty Related with Stochastic Processes Modelling results Modelling results
Measurement data Measurement data
1.5
1.5
1.0
1.0
0.5
0.5
0.0
0.0 ln (Concentration)
-0.5
-1.0
-1.5
-0.5
-1.0
-1.5
-2.0
Chamusca_O3
Boavista_O3
Vila N.T._O3
Chamusca_NO2
Chelas_NO2
Vila_N.T._NO2
Benfica_NO2
Chamusca_NO
Chamusca_O3
Boavista_O3
Vila N.T._NO2
C hamusca_NO2
Chelas_NO2
Vila N.T._NO2
Benfica_NO2
Chelas_NO
Benfica_NO
Chamusca_NO
-2.5 99% 1% 75% 25% Median
Chelas_NO
-2.0
-2.5
Benfica_NO
ln (Concentration)
467
99% 1% 75% 25% Median
Fig. 4 Box plot of the log-transformed stochastic variations
After the filtering, the data were back-transformed to the original concentration units and Root Mean Squared Error (RMSE) and Pearson Correlation coefficient were calculated in order to analyze the relationship between the observations and the model predictions. Table 2 summarizes the statistics and confirms a better agreement between the observed and predicted values after the short-term components being removed. Thus, the analysis of the deterministic components shows a decrease in RMSE and improves the correlation between the two time series in comparison with the same parameters calculated for the original data without filtering. Table 2 Model validation using original and filtered hourly concentration data. Parameter Pollutant Station Benfica Chelas Vila N.T. Chamusca Boavista
NO Orig. 134.8 25.2 20.6 0.9 –
RMSE (Pg m-3) NO2 Det. Orig. Det. 115.6 48.9 45.9 18.1 21.5 18.9 15.7 22.3 20.6 0.7 5.3 4.6 – – –
Orig. – – 30.8 21.4 32.1
O3
Det. – – 27.9 19.8 30.9
Pearson correlation coefficient NO NO2 O3 Orig. Det. Orig. Det. Orig. 0.21 0.25 0.55 0.58 – 0.11 0.15 0.39 0.40 – 0.10 0.10 0.43 0.43 0.40 0.42 0.46 0.35 0.40 0.60 – – – – 0.49
Det. – – 0.43 0.65 0.54
Orig. – original data; Det. – deterministic component after the filtering
5. Conclusions The proposed approach contributes to better understanding of the model prediction uncertainty related with short-term variations of the pollutants concentration. The analysis of the measurement data in frequency domain shows that the contribution of short- and long-term components to the total variability is different at different locations and accentuates the important role of the intra-day (<24 h) fluctuations at urban stations with direct influence of traffic emissions, while the seasonal variations are most important at rural station. Moreover, it reveals that short period cycles of less than 12 hours resemble random fluctuations.
468
O. Tchepel et al.
The model validation methodology based on the pre-filtered data allows to separate the estimation of the model error in predicting the baseline concentration from the error in predicting the stochastic variations related not only to the model physics and application issues, but also to representativeness of the measurements. The application of this approach to validation of decomposed one-year hourly time series obtained by CHIMERE model confirms the improvement of the model performance after removing the short-term component. Acknowledgments The methodology presented in this work was developed within AIR4EU project (CE SSPI-CT-2003-503596). The authors wish to thank the Portuguese Environmental Agency for financing and giving access to measured data. Thanks are extended to the Portuguese Ministério da Ciência e Ensino Superior, for the PhD grant of A. Monteiro (SFRH/BD/10922/02). The authors are also grateful to the Network of Excellence ACCENT (GOCE-CT-2004-505337).
References Borrego C, Monteiro A, Ferreira J Miranda, AI, Costa AM, Carvalho AC, Lopes M (2007) Modelling uncertainty estimation procedures for air quality assessment, Int. Environ. J. doi:10.1016/j.envint.2007.12.005. Hodzic A, Vautard R, Bessagnet B, Lattuatic M, Moreto F (2005) Long-term urban aerosol simulation versus PM observations. Atmos. Environ. 39, 5851–5864. Hogrefe C, Porter PS, Gego E, Gilliland A, Gilliam R, Swall J, Irwin J, Rao ST (2006) Temporal features in observed and simulated meteorology and air quality over the Eastern United States, Atmos. Environ. 40, 2041–5055. Monteiro A, Borrego C, Miranda AI, Gois V, Torres P, Perez AT (2007) Can air quality modelling improve emission inventories? In 6th Int. Conference of Urban Air Quality 2007 (UAQ), 27–30 March 2007, Limassol, Cyprus. Published in CD-Rom. Rao ST, Zurbenko IG, Neagu R, Porter PS, Ku JY, Henry RF (1997) Space and time scales in ambient ozone data, Bull. Amer. Meteor. Soc. 78(10), 2153–2166. Schmidt H, Derognat C, Vautard R, Beekmann M (2001) A comparison of simulated and observed ozone mixing ratios for the summer of 1998 in Western Europe, Atmos. Environ. 35, 2449–2461. Simpson D, Winiwarter W, Börjesson G, Cinderby S, Ferreiro A, Guenther A, Hewitt CN, Janson R, Khalil MAK, Owen S, Pierce TE, Puxbaum H, Shearer M, Skiba U, Steinbrecher R, Tarrasón L, Öquist MG (1999) Inventorying emissions from nature in Europe, J. Geophys. Res. 104, 8113–8152. Weil JC, Sykes RI, Venkatram A (1992) Evaluating air-quality models: review and outlook, J. Appl. Meteorol. 31, 1121–1145.
Estimation of the Modelling Uncertainty Related with Stochastic Processes
469
Discussion M. Sofiev: During the talk you have mentioned that in some cases model shortterm fluctuations appear to be stronger than in the observed data. What could be the cause for this (indeed, unexpected) behaviour? O. Tchepel: The observed behaviour has a complex pattern and it is different for the different pollutants and station types. It is important to stress that the short-term component represents fluctuations with wave period greater than 2 hours and less than 12 hours. The cause for this behaviour can be the emission inputs that result in overestimation of NO fluctuations by the model at rural background station and underestimation at urban traffic station. However, more studies are needed to establish the cause-effect relation.
4.8 Evaluating Regional-Scale Air Quality Models Alice B. Gilliland, James M. Godowitch, Christian Hogrefe, and S.T. Rao
Abstract Numerical air quality models are being used to understand the complex interplay among emission loading, meteorology, and atmospheric chemistry leading to the formation and accumulation of pollutants in the atmosphere. A model evaluation framework is presented here that considers several types of approaches, referred to here as the operational evaluation, diagnostic evaluation, dynamic evaluation, and probabilistic evaluation. The operational evaluation is used to reveal the overall performance of the model, and diagnostic evaluation approaches are then used to identify what processes and/or inputs significantly influence the predictted concentrations and whether they are simulated correctly. Dynamic evaluation entails assessing a model’s ability to reproduce observed changes in pollutant concentrations stemming from changes in weather and emissions. Probabilistic evaluation approaches will provide the confidence that can be placed on model results for air quality management or forecasting applications. Here, we present example results from several different model evaluation studies that consider questions related to the operational, diagnostic, and dynamic evaluation of a model, and discuss their complementary goals toward model improvements and characterization of model performance.
Keywords Air quality modelling, evaluation methods, regional scale
1. Introduction Photochemical air quality models are being used to simulate ozone (O3), particulate matter d2.5 Pg m-3 (PM2.5), and other pollutants across regional domains. Performance evaluations play a critical role in both regulatory and research applications of the models. For example, air quality model simulations must be evaluated against observational data prior to using the model to make decisions about emission control strategies. In research, improvements to process-level model algorithms or inputs are in part judged based on whether these changes improved model performance. In model applications that have either or both regulatory and research purposes, models can further be used to infer relationships between atmospheric pollutant concentrations and relevant processes, meteorology, and emissions. Given the influence that model evaluation results can have on regulatory decisions and scientific C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
412
Evaluating Regional-Scale Air Quality Models
413
conclusions about air pollution, it is critical that model evaluation studies are comprehensive and characterize model performance in insightful ways that not only reveal how well model predicted pollutant levels compare to observed data, but also increase confidence in the inputs (e.g., meteorology and emissions) and the modelled processes. Here, a model evaluation framework is presented that organizes evaluation approaches to represent how they differ and complement one another, and a few examples are discussed.
2. Proposed Air Quality Model Evaluation Framework In Figure 1, we present a framework for model evaluation approaches, which is based on the purpose and specific questions being asked as part of an analysis.
Fig. 1 A suggested framework for organizing and identifying the purpose and questions addressed in various evaluation analyses
As the first step in model evaluation, model predictions are compared to observed data and statistical metrics are computed, which is referred to here as “operational evaluation.” Typically, most of the observational data is focused on the endpoint pollutants that are monitored for air quality, such as O3 or PM2.5 and component species of PM2.5. However, the ability of a model to predict the endpoint pollutant of interest does not address whether the predicted concentrations result from correct or incorrect processes, which is commonly referred to as diagnostic evaluation. For secondary pollutant species that are not directly emitted, diagnostic evaluation
414
A.B. Gilliland et al.
methods are critical for insuring confidence in a model as a tool and for identifying model improvements. Figure 1 also includes a new evaluation approach referred to as “dynamic evaluation” that focuses on the model predicted change in air quality concentrations in response to either emission or meteorological changes. This requires historical case studies where known emission changes or meteorological changes occurred that could be confidently estimated, and dynamic evaluation also requires that these changes had an observed impact on air quality. Operational, diagnostic, and dynamic evaluation approaches complement one another by not only characterizing how well the model captured the air quality levels at that time, but how well the model captures the role and contributions of individual inputs and processes and the air quality responses to changes in these factors. For the remainder of this discussion, examples will be shown of how these three approaches in concert capture a more comprehensive evaluation of model performance for specific model applications and support the priority of further model improvement. A fourth aspect of model evaluation in Figure 1, referred to as probabilistic evaluation, attempts to capture the level of confidence in model results for regulatory or forecasting applications, and a classic example would be ensemble modelling for meteorology forecasting. With computer efficiencies improving exponentially, methods such as ensemble modeling that introduce a range of uncertainties into air quality model predictions become increasingly realistic for decision-making or forecasting. This topic of model evaluation is only included here in a very limited extent, but additional research and advancements are needed to develop more innovative and creative approaches that consider the confidence in air quality models for various applications (see Gégo et al., 2003). The following examples illustrate how these evaluation approaches can help provide increased confidence that model performance is well characterized and suitable for air quality regulatory and forecast application. Example results are shown using the Community Multiscale Air Quality (CMAQ) model version 4.5 (Byun and Schere, 2006) For the purpose of illustration, only scatterplot illustrations are shown, but it is of course critically important to examine the full range of spatial and temporal scales.
3. Operational and Diagnostic Evaluation Methodologies: Complementary Roles Previous studies have provided operational model evaluation results for O3 for both retrospective and forecasting cases (e.g., Eder et al., 2006; Tesche et al., 2006). While the results on average show quite good performance in most studies, the results are often based on more than 500 observational sites and extremely large subcontinental regions. An example of typical operational evaluation results for O3 are shown in Figure 2, where results from a summer 2002 CMAQ model simulation are compared against observational data. If one looks only at the scatterplot and statistical metrics, it gives the impression that the model performance is very good.
Evaluating Regional-Scale Air Quality Models
415
Fig. 2 Example scatterplot for daily 8-hour maximum O3 from Summer 2005 comparing observations from the Air Quality System (AQS) network and the Community Multiscale Air Quality (CMAQ) model along with mean bias (MB), Normalized Mean Bias and Error (NMB and NME), and root mean square error (RMSE) from the same daily maximum 8-hour O3 concentrations
However, further analysis of the operational model evaluation results for O3 elucidates that model performance for O3 is not equally good across all conditions. For example, Appel et al. (2007) compared model performance at different ranges of O3 levels as well as evaluation under different synoptic meteorological regimes and demonstrated that the model’s underpredictions are associated with high pressure, stagnant conditions typical of high O3 events in the U.S and overpredictions are associated with frontal passages. Hogrefe et al. (2001) have also shown that the model predictions of O3 are challenged most for the high-frequency variations that occur below the diurnal time scales. These types of evaluation results are needed to identify specific conditions associated with meteorological forcing that need further diagnostic evaluation for model improvements. Modeling PM2.5 introduces many additional challenges since it is comprised of a number of aerosol chemical species such as sulphate, nitrate, ammonium, organic and elemental carbonaceous materials and because the emission inputs are largely uncertain for many agricultural and diffuse sources. Continued research is needed to refine the modelled representation of the chemical transformation processes as well as the influences of emissions and meteorology. Operational evaluations of PM2.5 components such as sulphate aerosol concentrations compare reasonably on the seasonal time scale compared to other aerosol species such as nitrate and carbonaceous aerosols where scientific advancements and model improvements are needed (e.g., Morris et al., 2006). For model improvement of nitrate, as an example, diagnostic evaluations are needed to identify the factors that contribute to model deficiencies. Bhave et al. (2006) provide a summary of recent diagnostic work to understand and improve nitrate predictions related to chemical transformation processes, specifically the heterogeneous N2O5 pathway for HNO3 production. Gilliland et al. (2003, 2006) and Pinder et al. (2006) demonstrate how critical NH3 emissions as well as the heterogeneous N2O5 pathway can be to nitrate aerosol predictions. Here, an example is shown of additional diagnostic evaluation work that is ongoing to look more carefully at the role of meteorological forcing to wintertime nitrate predictions. Figure 3 illustrates that meteorological model inputs can have a substantial
416
A.B. Gilliland et al.
impact on model’s predictions of total nitrate, and demonstrates the need for improving the estimated removal via wet and dry deposition.
Fig. 3 Predictions versus observations of total nitrate from the Clean Air Status and Trends Network (CASTNet) from January 2002 using two separate sets of meteorological model inputs. “Meteorology version 1” simulation used a non-graupel microphysics scheme and had a large surface temperature cold bias that affect wet and dry deposition. “Meteorology version 2” used the same meteorological model (MM5) but used a microphysics scheme with graupel and had improved surface temperatures
4. Dynamic Evaluations: Challenges and Relevance The previous examples provide illustrations of how operational and diagnostic evaluation studies can provide initial characterization of model performance issues and direction for model improvement. More uncommon are dynamic evaluation studies that explicitly focus on the model-predicted pollutant responses stemming from changes in emissions or meteorology. Gilliland et al. (2008) provide the most direct example of a dynamic evaluation study, where air quality model simulations were evaluated before and after major reduction in the NOx emissions. The U.S. Environmental Protection Agency’s NOx SIP Call required substantial reductions in NOx emissions from power plants in the Eastern U.S. during summer O3 seasons beginning in June 2004. Gégo et al. (2007) and USEPA (2006) offer examples of how observed O3 levels have decreased noticeably after the NOx SIP Call was implemented. Since air quality models are used to estimate how air quality will change due to various emission control strategies, the NOx SIP Call is an excellent opportunity to evaluate a model’s ability to simulate the response of O3 to known and quantifiable O3 changes. Figure 4 provides an example from this study where changes in O3 are compared from before (summer 2002) and after (summers 2004 and 2005) the NOx emission reductions occurred. Meteorological differences were much greater between 2002 and 2004 than 2002 and 2005, and, hence, larger O3 decreases in 2004 were also due to the cooler/wetter conditions in 2004. Figure 4 also illustrates model underestimation of O3 decreases as compared to observations, which could be due to either the underestimation of NOx emission reductions or a dampened chemical response in the model to those emission changes, or other factors. Analysis methods such as the e-folding distances (Godowitch et al., 2007; Gilliland et al., 2008) have been used
Evaluating Regional-Scale Air Quality Models
417
to show that NOx emissions in these simulations are not impacting O3 levels as far downwind as observations suggest, which could be a factor here. Dynamic evaluation approaches introduce several new challenges. First, retrospective case studies are needed that offer observed changes in air quality that can be closely related to known changes in emissions or meteorology. The NOx SIP call has offered a very strong case study to test model responses via dynamic evaluation, but next steps must include further diagnostic evaluation to identify what chemical, physical, or emission estimation uncertainties are contributing to the current model results. Findings from additional analysis of this case study can ultimately lead to model improvements that are directly relevant to the way air quality models are used for regulatory decisions.
Fig. 4 Summer (2004–2002) and (2005–2002) comparison of the average of upper 95th% of maximum daily 8-hour average O3 values at the Air Quality System (AQS) network sites in the Eastern U.S. Results are shown using both the CMAQ CB4 and SAPRC99 chemical mechanisms. See Gilliland et al. (2007) for further description
5. Summary The topic of this paper, evaluation of regional air quality models, is indeed challenging and broad. The intention here is to present a perspective about how many different studies all contribute to a multi-faceted area of research referred to as regional photochemical air quality model evaluation. It can be challenging to characterize model performance for a number of air pollutants via operational methods, but we encourage analyzing model results in ways that characterize model performance across a range of scales and dis-aggregation. Diagnostic evaluation perspectives are needed to extend operational results to the next stage of identifying processes or model inputs that have an influential role on model predictions and how they compare to observations. The model’s sensitivity to meteorological and emission uncertainties should also be addressed within a diagnostic evaluation context, as well as the more traditional diagnostic studies such as chemical indicators
418
A.B. Gilliland et al.
that consider the chemical state within the model simulation. As a next challenge to traditional evaluation studies, we introduce dynamic evaluation to stress-test the model’s ability to reproduce known changes in air quality “forcings” such as meteorological and emission changes that can directly impact the way that air quality models are used in regulatory decision making. Acknowledgments The research presented here was performed under the Memorandum of Understanding between the U.S. Environmental Protection Agency (EPA) and the U.S. Department of Commerce’s National Oceanic and Atmospheric Administration (NOAA) and under agreement number DW13921548. This work constitutes a contribution to the NOAA Air Quality Program. Although it has been reviewed by EPA and NOAA and approved for publication, it does not necessarily reflect their policies or views.
References Appel KW, Gilliland AB, Sarwar G, Gilliam R (2007) Evaluation of the Community Multiscale Air Quality (CMAQ) model version 4.5: Sensitivities impactting model performance; Part I – ozone, Atmos. Environ., 41, 9603–9615. Bhave et al. (2006) 6th Annual CMAS Conference, October 1–3, 2007, Chapel Hill, NC, http://www.cmascenter.org/conference/2006/ppt/session1/bhave.ppt Byun D, Schere KL (2006) Review of the governing equations, computational algorithms, and other components of the Models-3 Community Multiscale Air Quality (CMAQ) modeling system. Appl. Mech. Rev., 59, 51–77. Eder B, Kang D, Mathur R, Yu S, Schere K (2006) An operational evaluation of the Eta–CMAQ air quality forecast model, Atmos. Environ., 40, 4894–4905. Gégo et al. Probabilistic assessment of regional scale ozone pollution in the eastern United States (2003) In Air Pollution in Regional Scale. Proceedings of the NATO Advanced Research Workshop, Kallithea, Halkidiki, Greece, June 13–15, 2003. NATO Science Series: IV. Earth and Environmental Sciences. D. Melas, and D. Syrakov (Eds.). Kluwer, Dordrecht, 87–96. Gégo E, Porter PS, Gilliland A, Rao ST (2007) Observation-based assessment of the impact of nitrogen oxides emissions reductions on ozone air quality over the eastern United States, J. Appl. Met. Climatol., 46, 994–1008. Gilliland AB, Hogrefe C, Pinder RW, Godowitch JM, Rao ST (2008) Dynamic evaluation of regional air quality models: assessing changes in O3 stemming from changes in emissions and meteorology, Atmos. Environ. doi:10.1016/ j.atmosenv.2008.02.018. Gilliland AB, Appel KW, Pinder R, Roselle SJ, Dennis RL (2006) Atmospheric environment, seasonal NH 3 emissions for an annual 2001 CMAQ simulation: inverse model estimation and evaluation, Atmos. Environ, 40, 4986–4998.
Evaluating Regional-Scale Air Quality Models
419
Gilliland AB, Dennis RL, Roselle SJ, Pierce TE (2003) Seasonal NH3 emission estimates for the Eastern Unites States using ammonium wet concentrations and an inverse modeling method, J. Geophys. Res.-Atmos., 108, 10.1029/ 2002JD003063. Godowitch JM, Hogrefe C, Rao ST (2007) Influence of point source NOx emission reductions on modeled processes governing ozone concentrations and chemical/ transport indicators, in review with J. Geophys. Res.-Atmos. Hogrefe C, Rao ST, Kasibhatla P, Hao W, Sistla G, Mathur R, McHenry J (2001) Evaluating the performance of regional-scale photochemical modeling systems: Part II - O 3 predictions, Atmos. Environ., 35, 4175–4188. Morris RE, Koo B, Guenther A, Yarwood G, McNally D, Tesche TW, Tonnesen G, Boylan J, Brewer P (2006) Model sensitivity evaluation for organic carbon using two multi-pollutant air quality models that simulate regional haze in the southeastern United States, Atmos. Environ., 40, 4960–4972. Pinder RW, Adams PJ, Pandis SN, Gilliland AB (2006) Temporally resolved ammonia emission inventories: Current estimates, evaluation tools, and measurement needs, J. Geophys. Res.-Atmos., 111, doi:10.1029/2005JD006603 Tesche TW, Morris R, Tonnesen G, McNally D, Boylan J, Brewer P (2006) CMAQ/CAMx annual 2002 performance evaluation over the eastern US, Atmos. Environ., 40, 4906–4919. USEPA (2006) NOx Budget Trading Program, EPA-430-R-07-009. http://www. epa.gov/airtmarkets
4.6 Has the Performance of Regional-Scale Photochemical Modelling Systems Changed over the Past Decade? C. Hogrefe, J.-Y. Ku, G. Sistla, A. Gilliland, J.S. Irwin, P.S. Porter, E. Gégo, P. Kasibhatla, and S.T. Rao
Abstract This study analyzes summertime ozone concentrations that have been simulated by various regional-scale photochemical modelling systems over the Eastern U.S. as part of more than ten independent studies. Results indicate that there has been a reduction of root mean square errors (RMSE) and an improvement in the ability to capture ozone fluctuations stemming from synoptic-scale meteorological forcings between the earliest seasonal modelling simulations and more recent studies. However, even the more recent model simulations exhibit RMSE values of about 15 ppb and there is no evidence that differences in RMSE between these recent simulations are attributable to systematic improvements in modelling capability. Moreover, it was determined that certain aspects of model performance have not changed over the past decade. One such aspect is that the RMSE of simulated time series can be reduced by applying temporal averaging kernels of up to seven days while the benefit of longer averaging windows appears to vary from year to year. In addition, it is found that spatial patterns simulated by these modelling systems typically have lower correlations and higher centered RMSE than temporal patterns. Analogous to the errors in the simulated time series, these errors in the spatial patterns can be reduced through the application of spatial averaging kernels.
Keywords Regional-scale air quality modelling, model evaluation, model intercomparison
1. Introduction In the United States, grid-based photochemical modelling systems consisting of separate modules for estimating emissions, meteorology, and air quality have been used for several decades to simulate ozone concentrations, most often in the context of assessing the effectiveness of emission control strategies. While early model applications were limited to a few ozone episodic days, there are increasingly more seasonal, annual and multi-year simulations over the past decade, and the scope of applications has broadened to include air quality forecasting and assessments of the C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
394
Has the Performance of Regional-Scale Photochemical Modelling Systems
395
impacts of climate change on air quality. This study analyzes summertime ozone concentrations that have been simulated as part of more than ten independent studies utilizing such modelling systems. While these studies were not coordinated to form harmonized modelling or a controlled ensemble, each of them represents best modelling practices, reflecting the state of science at the time the simulations were performed. The object of this analysis is to assess how our ability to simulate regional-scale ozone concentrations and their variability has changed over the past decade. To this end, we have attempted to quantify the ability of the various simulations to capture temporal and spatial patterns and to characterize model performance on different temporal and spatial scales. No attempt was made to perform diagnostic evaluations for determining the underlying reasons for differences in model behaviour. Moreover, the focus of the analysis is on the comparison between the observed and simulated spatial and temporal patterns rather than on differences in absolute values and biases. Section 2 contains a brief overview of the observations and modelling simulations analyzed in this study. Results are presented in Section 3, and Section 4 discusses the implications of our analysis for various modelling applications.
2. Description of Observations and Model Simulations The observed daily maximum 8-hour ozone concentrations for the period 1993– 2005 were determined from hourly ozone observations at surface monitors from the U.S. EPA’s AQS database. In order to be included in the analysis, monitors had to be (a) located within the analysis domain spanning the land area common to all modelling simulations listed in the next section, and (b) have at least 50% nonmissing days during June–August between 1993 and 2005. The application of these screening criteria resulted in the selection of 248 monitor locations.
396
C. Hogrefe et al.
An overview of all modelling simulations analyzed in this study is provided in Table 1. In the following sections, the various modelling simulations will be referred to by the abbreviations listed in the first column. All simulations are listed roughly in chronological order based on when they were performed as part of various studies. In particular, the simulations for the summer of 1995 by M1–M3 were performed significantly earlier (in the late 1990s) compared to all other simulations analyzed here. Also included in Table 1 is a reference to the publications which provide more details about the individual simulations. For the comparison with observations, model values were extracted for the grid cells in which the 248 monitors described above were located.
3. Results and Discussion 3.1. Model evaluation of temporal and spatial patterns Model performance for all simulations was summarized through the use of root mean square errors (RMSE) and the match between observed and simulated temporal and spatial patterns. Pattern matching, in turn, was quantified through the use of correlation coefficients, the ratio of simulated to observed standard deviations, and the centered pattern RMSE. The centered pattern RMSE is calculated after the means of observed and simulated ozone concentrations are subtracted from each observed and simulated data point (Taylor, 2001). As shown in Taylor (2001), these correlation coefficients, ratio of standard deviations, and centered pattern RMSE d between observations and model predictions is an array of data points sampled through time and/or space. In this analysis, we focus on the daily maximum 8-hour ozone concentrations and determine the models’ ability to capture both the temporal patterns (time series, results are presented in Figure 1a, b) and spatial patterns (maps of concentrations, results are presented in Figure 2a, b). Figure 1a displays the so-called Taylor diagram (Taylor, 2001) in which the match between the observed and simulated spatially-averaged time series of June–August daily maximum 8-hour ozone at 248 monitors for each modelling simulation is indicated by the position of a single letter on a polar diagram. In this polar coordinate system, the position of the letter for a given modelling simulation on the diagram indicates (a) the correlation between observed and simulated time series as angle counterclockwise from 90º, (b) the ratio of simulated to observed standard deviation of the time series as radial distance from the origin with a radius of 1 corresponding to am exact match between observed and simulated standard deviations, and (c) the centered pattern RMSE as distance from the reference point indicated by a black dot (Taylor, 2001). In terms of correlation coefficients, most simulations with the exception of the simulation of the summer of 1995 by models M1–M3 exhibit values greater than 0.8, ranging as high as 0.95 for the simulation of the summer of 2002 by both models M6 and M10. In terms of standard deviations, almost all simulations underestimate the observed standard deviations, typically by about 20%,
Has the Performance of Regional-Scale Photochemical Modelling Systems
397
but by as much as 40% for the simulation of the summer of 2002 by model M9. The centered pattern RMSE, i.e. the distance from the reference point indicated by a black dot, is largest for the simulations of the summer of 1995 by models M1–M3 and for the summers of 1993, 1994 and 1996 by model M8. Most other modelling simulations show similar performance to each other, with correlations between 0.85 and 0.95 and an underestimation of observed standard deviations by about 20%. While Figure 1a compared the behaviour of observed and simulated time series of daily maximum 8-hour ozone concentrations that were spatially averaged on each day, Figure 1b shows box plots of the total RMSE of the simulated time series of daily maximum 8-hour ozone concentrations, grouped by simulation. The distributions depicted by each box/whisker represent the RMSE of time series calculated separately at each of the 248 monitor locations. It is evident that the simulations of the summer of 1995 by models M1–M3 have a larger RMSE compared to all other simulations. In particular, they show a larger RMSE compared to the simulation of the same summer by model M8 which was performed almost a decade after these earlier simulation, indicating an improvement in model performance as measured by RMSE over this time span. Furthermore, the variations in model performance for more recent simulation periods (2001 or later) do not appear to be attributable to any systematic improvement in modelling capabilities. This is evidenced by the
Fig. 1 (a) Taylor diagram of spatially averaged time series, (b) boxplot of RMSE of June–August time series at 248 monitors
Fig. 2 (a) Taylor diagram of summertime averaged spatial patterns, (b) boxplot of RMSE of spatial patterns across 248 monitor locations on 92 summer days
398
C. Hogrefe et al.
substantial within-group variations of “F” – “H” and “Q” – “T”, two sets of simulations that were each performed as part of a single study. These within-group variations as large or larger than the variations between all of the simulations for more recent time periods. While Figure 1a, b illustrate the ability of the various simulations to capture temporal patterns (i.e. time series), Figure 2a, b present the corresponding results when spatial patterns are evaluated. For Figure 2a, the Taylor diagram was constructed for the spatial maps of observed and simulated summer-average daily maximum 8-hour concentrations for each simulation. Several features are noteworthy. First, the correlations between observed and simulated spatial patterns of summertimeaverage ozone concentrations are markedly lower than those between the observed and simulated spatially-averaged time series (Figure 1a). Second, the centered pattern RMSE (as indicated by the distance from the reference point) is similar for all simulations with the exception of the simulation of the summer of 1995 by model M3. Figure 2b shows box plots of the total RMSE of the simulated spatial patterns of daily maximum 8-hour ozone concentrations, grouped by simulation. Similar to the box plot for the time series in Figure 1b, the simulations of the summer of 1995 by models M1–M3 show a larger RMSE compared to other simulations. In particular, they show a larger RMSE compared to the simulation of the same summer by model M8, which was performed almost a decade after these earlier simulation, indicating an improvement in the performance of these modelling systems as measured by RMSE over this time span. The results presented in this section present evidence that model performance can vary based on the meteorological and emission conditions simulated and can exhibit trends. It is beyond the scope of this study to unequivocally ascribe the changes in model performance to improvements in any particular component of the modelling system (emissions, meteorological modelling, or photochemical modelling). However, variability in meteorology certainly affects model performance; therefore, the following section focuses on investigating the modelling systems’ ability to capture the effect of synoptic-scale variations.
3.2. Model performance for synoptic regimes Because synoptic-scale meteorological conditions exert a significant influence on the ground-level ozone concentrations, it is of interest to evaluate the modelling systems’ response to these forcings. To this end, we characterized meteorological conditions through a map-typing procedure applied to gridded fields of mean sea level pressure (MSLP). The Kirchhofer correlation-based map typing procedure was used to determine the 10 most frequent MSLP patterns from all summer days between 1995–2005 (the gridded MSLP fields used in this analysis were not available prior to 1995), and each day was then assigned to the MSLP pattern best representing it. Details of this procedure are described in Hegarty et al. (2007). Next, both observed and simulated average ozone concentrations were computed for each pattern and at each station, and the all-pattern observed or simulated average was
Has the Performance of Regional-Scale Photochemical Modelling Systems
399
subtracted to determine the observed and simulated anomaly for each pattern and each station. Finally, at each station, the correlation between the observed and simulated anomalies across the 10 MSLP patterns was computed, and the results for the medians, 10th and 90th percentiles across all sites are shown in Table 2. In general, median correlations exceed 0.6, indicating that the modelling systems typically catch a substantial portion of the meteorologically-induced ozone variability on the synoptic-scale. In addition, the simulations of the summer of 1995 by models M1– M3 stand out as having the lowest correlations. In particular, the correlations are lower than for the simulation of the same summer by model M8. This suggests that the ability of photochemical modelling systems to capture the phase of synopticscale ozone build-up and removal events has improved over the past decade. On the other hand, simulations for more recent time periods exhibit large interannual variability in model performance but no systematic change. Table 2 Correlation coefficients between the observed and simulated anomalies across the ten MSLP patterns for all model simulations. No results are shown for M8-1993 and M8-1994 because the gridded MSLP fields used in the synoptic typing analysis were not available for these time periods. M6- M6- M6M8- M8- M8- M8- M8M7 M9 M10 M11 M12 M13 2001 2002 2003 1995 1996 1997 1998 1999 10th % 0.24 0.22 0.22 0.67 0.49 0.66 0.76 0.6 0.64 0.43 0.44 0.67 0.41 0.51 0.77 0.77 0.55 0.59 0.72 Median 0.67 0.68 0.64 0.86 0.81 0.87 0.89 0.79 0.85 0.78 0.73 0.86 0.75 0.85 0.89 0.89 0.8 0.84 0.87 90th % 0.88 0.87 0.88 0.94 0.94 0.94 0.95 0.94 0.94 0.92 0.88 0.95 0.92 0.95 0.95 0.95 0.94 0.94 0.95 M1 M2 M3 M4 M5
3.3. Evaluation of model performance of different temporal and spatial scales Hogrefe et al. (2001) showed that regional-scale modelling systems typically perform better in capturing signals on time scales longer than one day. To investigate this issue further, we constructed time series of running average one-day, three-day, …, 31-day time series for both observations and model predictions and computed the standardized centered RMSE of these time series for each averaging period and model simulation. Results are presented as box plots in Figure 3a. The standardized centered RMSE in this plot was normalized by the observed standard deviation to account for reduced variability when averaging kernels are applied. The box/ whiskers represent results across the 21 model simulations, for each model simulation, the median time series across all 248 monitors was chosen. The median standardized centered pattern RMSE generally decreases for averaging lengths up to 15 days but then shows an increase for greater averaging lengths. For individual models, the centered pattern RMSE begins to increase even beyond averaging lengths of seven days. We also investigated the effects of spatial averaging on model performance. To this end, a fine 1 × 1 km grid was overlayed on the analysis domain, and each observation and corresponding model value was assigned to the closest grid cell. Moving-average kernels of 1 × 1, 41 × 41, …, 601 × 601 grid cells were then
400
C. Hogrefe et al.
applied to these gridded fields, and the standardized centered RMSE was computed for the spatial patterns obtained by each averaging kernel. Only the averaged values at the original monitor locations were considered. Results are shown in Figure 3b. The box/whiskers represent results across the 21 model simulations. It can be seen that spatial averaging decreases the standardized centered pattern RMSE for all model simulations analyzed here for all averaging distances, i.e. all modelling systems are better able to capture the large-scale concentration patterns than localized features in the observed maps.
3.4. Implications The results presented in the previous sections have considerable implications for applications of regional-scale photochemical modelling systems. Despite a reduction of RMSE and an improvement in the ability to capture ozone fluctuations stemming from synoptic-scale meteorological variability between the earliest seasonal modelling simulations and more recent studies, RMSE of modelled ozone time series still show values of 15 ppb. While this error can be reduced by applying temporal averaging kernels of up to seven days, the benefit of longer averaging windows appears to vary from year to year. For forecasting applications in which temporal averaging is not feasible and where the focus is on predicting single-day peak concentrations, this implies that bias-correction approaches such as the Kalman filter are needed to improve the accuracy of the model-based forecast. Second, spatial patterns simulated by these modelling systems typically have lower correlations and larger centered RMSE than temporal patterns. For studies seeking to utilize model-predicted concentration maps for applications such as health impact assessments, these points to the need for developing and applying statistical techniques aimed at combining information from both observations and model simulations to best represent spatial variability.
Fig. 3 (a) Boxplots of standardized centered RMSE for simulated time series as function of temporal averaging window length across 21 model simulations. (b) As in (a) but for spatial patterns and spatial averaging windows
Has the Performance of Regional-Scale Photochemical Modelling Systems
401
Acknowledgments and disclaimer Christian Hogrefe gratefully acknowledges partial support for this work through a research fellowship from the Oak Ridge Institute for Science and Education (ORISE). Part of the work also was performed with support from the New York State Department of Environmental Conservation and under the Memorandum of Understanding between the U.S. Environmental Protection Agency and the U.S. Department of Commerce’s National Oceanic and Atmospheric Administration and under agreement number DW13921548. The results presented in this paper do not necessarily reflect the policies or views of the supporting agencies.
References Appel KW, Gilliland AB, Sarwar G, Gilliam RC (2007) Evaluation of the Community Multiscale Air Quality model version 4.5: Uncertainties and sensitivities impacting model performance; Part I – ozone, Atmos. Environ., Vol. 41, doi:10.1016/j.atmosenv.2007.08.044, pp. 9603–9615. Eder B, Yu S (2006) A performance evaluation of the 2004 release of Models-3 CMAQ, Atmos. Environ., 40, 4811–4824. Gilliland AB, Hogrefe C, Godowitch JL, Rao ST (2007) Dynamic evaluation of regional air quality models: assessing changes in O3 stemming from emissions and meteorology, Atmos. Environ., doi:10.1016/j.atmosenv.2008.02.018. Hegarty J, Mao H, Talbot R (2007) Synoptic Controls on Summertime Surface Ozone in the Northeastern U.S., J. Geophys. Res., 112, D14306, doi:10.1029/ 2006JD008170 Hogrefe C, Rao ST, Kasibhatla P, Hao W, Sistla G, Mathur R, McHenry J (2001) Evaluating the performance of regional-scale photochemical modeling systems: Part II – ozone predictions. Atmos. Environ, 35, 4175–4188. Hogrefe CW, Hao K Civerolo, J-Y Ku, Gaza RS, Sedefian L, Sistla G, Schere K, Gilliland A, Mathur R (2007a) Daily photochemical simulations of ozone and fine particulates over New York State: findings and challenges, J. Appl. Meteor, 46, 961–979. Hogrefe C, Lynn B, Knowlton K, Goldberg R, Rosenzweig C, Kinney PL (2007b) Long-term regional air quality simulations in support of health impact analyses, preprints, NATO 28th ITM, Aveiro, Portugal, September. 25–29, 2007. Kasibhatla P, Chameides WL (2000) Seasonal modeling of regional ozone pollution in the eastern United States. Geophys. Res. Lett., 27, 1415–1418. Nolte CG, Gilliland AB, Hogrefe C (2007) Linking global to regional models to assess future climate impacts on air quality in the United States: 1. surface ozone concentrations. J. Geophys. Res., in press. Ozone Transport Commission (2007) Draft Modeling Technical Support Document. http://www.otcair.org/projects_details.asp?FID=101&fview=modeling# Taylor KE (2001) Summarizing multiple aspects of model performance in a single diagram, J. Geophys. Res., 106, 7183–7192.
402
C. Hogrefe et al.
Discussion S.T. Rao: Do you expect that there is a limit to model improvement, and do you think that we have reached that now for ozone? C. Hogrefe: Yes, I do expect that there is a limit to model improvement. At least in an empirical sense, the results presented in this paper suggest that this limit has been reached for ozone. While there may be the potential for better model predictions of ozone through the use of higher-resolution modelling or updated chemical mechanisms, my expectation is that the resulting improvement in model performance would be incremental at best. P. Builtjes: With regard to Peter Builtjes comment about model performance would have been better if high resolution modelling (say 4 km) were performed, you should refer to the study by C. Mass in the Bulletin of American Meteorological Society, which analysed the performance of a meteorological model with two different grid cell size, 12 and 4 km, for a summer season. These results revealed the lack of superiority of the 4 km modelling over the 12 km modelling for the meteorological variables he had analyzed. Reviews of these results, there is no assurance that air quality models would perform better with higher resolution. Of course, there may be case studies where people showed better performance, but these are not longterm simulations. With episodic type modelling (two to three days simulations), there is not enough data to properly evaluate model predictions and errors. C. Hogrefe: I agree with this comment. Conceptually, higher-resolution modelling may lead to improved performance in some areas of strong gradients in terrain or emission densities, but this hypothesis can only be tested with longer-term simulations and monitoring networks that are denser than the routine meteorological and air quality networks. Moreover, the analysis you are referring to suggests that there is no systematic improvement in the performance of meteorological models run at 4 km vs 12 km resolution, implying that the potential benefits of higher resolution air quality modelling may also be limited and sporadic Y.P. Kim: Since UAM does not contain aerosol module, the ozone levels might be different from the models (e.g. CMAQ) with the aerosol module.
Has the Performance of Regional-Scale Photochemical Modelling Systems
403
C. Hogrefe: The effect of ozone-aerosol interactions on simulated ozone concentrations typically is a few ppb or less during summertime conditions. Therefore, I do not expect this effect to a major contributor to the differences in model performance between UAM-V and CMAQ seen in this study.
4.10 Modelling Evaluation of PM10 Exposure in Northern Italy in the Framework of CityDeltaIII Project C. Carnevale, G. Finzi, E. Pisoni, and M. Volta
Abstract This work presents an application of the multiphase model TCAM to evaluate the impact of three different emission control strategies in northern Italy. The domain, including the whole of Lombardia and part of confining regions is often affected by severe PM10 levels, far from the European standard laws. This fact is due to high industrial and residential sites, to a close road net and to frequently stagnating meteorological conditions; for these reasons, the area is a very important benchmark for modelling simulations. The impact evaluation has been performed in terms of both yearly mean value and 50 µg/m3 threshold exceedance days in nine points of the domain, chosen to be representative of the chemical and meteorological regimes of the area under study. The results show that even if the three emission reduction scenarios improve air quality all over the domain, in particular in the area with higher emission density, the PM10 levels remain far from the 2020 European standards.
1. Introduction Multiphase models, simulating the physical-chemical processes involving seconddary pollutants in the troposphere, are key tools to evaluate the effectiveness of emission control strategies. In this paper, the chemical and transport model TCAM (Carnevale et al., 2008) is applied. TCAM is a module of GAMES (Gas Aerosol Modelling Evaluation System) integrated modelling system (Volta and Finzi, 2006), including the emission model POEM-PM (Carnevale et al., 2006), the CALMET meteorological model (Scire et al., 1990), a pre-processor providing the initial and boundary conditions required by the model and the System Evaluation Tool (SET). The modelling system has been validated over northern Italy in the frame of CityDelta project (Cuvelier et al., 2007). The model has been used to asses the effectiveness of three different emission control strategies at 2020, in the frame of CityDelta project (Cuvelier et al., 2007). The first emission scenario is related to the emission reduction expected assuming the European current legislation (CLE). The second one (CLEC) is equal to the first one, but the PM2.5 emissions are set to zero inside the Milan metropolitan area. Finally, the third scenario is computed starting from the CLEC one, and applying the best available technologies for nitrogen oxides reductions (Most feasible reduction, MFR). C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
426
Modelling Evaluation of PM10 Exposure in Northern Italy in the Framework of CityDeltaIII
427
2. The TCAM Model TCAM (Transport and Chemical Aerosol Model) is a multiphase three-dimensional Eulerian model, working in a terrain-following co-ordinate system (Carnevale et al., 2008). The model formalizes the physical and chemical phenomena involved in the formation of secondary air pollution solving, for each time step, the following mass-balance equation:
wCi wt
Ti Ri Di Si
(1)
where Ci (µg m–3) is the concentration of i species, Ti (µg m–3 s–1) is the transport/ diffusion term, Ri (µg m–3 s–1) is the multiphase term, Di (µg m–3 s–1) includes the wet and dry deposition and Si is the emission term. TCAM implements a split operator technique (Marchuk, 1975) allowing to separately treat the horizontal transport, the vertical phenomena (including transport-diffusion, emissions and deposition) and the chemistry. The advection scheme implemented in TCAM is based on a finite differences scheme and it solves horizontal transport of both gas and aerosol species. The module describes the convective and the turbulent transport (Seinfeld and Pandis, 1998) solving the PDE transport equation using chapeau functions (Pepper et al., 1979) and the non linear Forester filter (Forester, 1977). The dry deposition is described by a resistance based approach, with different deposition velocity for each pollutant. Wet deposition (for both gas and aerosol species) is described by a scavenging approach (Seinfeld and Pandis, 1998), with scavenging coefficient defined distinctly for gas and aerosol species. For gases, two components are calculated: (1) the uptake of ambient gas concentration into falling precipitation, which can occur within and below clouds, and (2) the collection by precipitation of cloud droplets containing dissolved gas species. For particles, separate in-cloud and below-cloud scavenging coefficients are determined. Within clouds, all aerosol mass is assumed to exist in cloud droplets (all particles are activated as condensation nuclei), so scavenging is due to the efficient collection of cloud droplets by precipitation. Below clouds, dry particles are scavenged by falling precipitation with efficiency depending on particle size. TCAM allows the simulation of gas chemistry using different chemical mechanisms, based both on the lumped structure (Carbon Bond 90 [Gery et al., 1989]) and on the lumped molecule (SAPRC90 [Carter, 1990], SAPRC97 [Carter et al., 1997], SAPRC99 [Carter, 2000]) approach. In order to describe the mass transfer between gas and aerosol phase an extended version of SAPRC97 mechanism, the COCOH97 (Wexler and Seinfeld, 1991), is implemented in the model. The chemical kinetic system is solved by means of the Implicit-Explicit Hybrid (IEH) solver (Chock et al., 1994), that splits the species in fast and slow ones, according to their reaction velocity. The system of fast species is solved by means of the implicit Livermore Solver for Ordinary Differential Equations (LSODE) (Hindmarsh, 1975) implementing an Adams predictor-corrector method in the non-stiff case (Wille, 1994), and the Backward Differentiation Formula method in the stiff case (Wille, 1994). The slow species system is solved by the Adams-Bashfort method. The aerosol module implemented in TCAM describes the
C. Carnevale et al.
428
most relevant aerosol processes: the condensation, the evaporation (Seinfeld and Pandis, 1998), the nucleation of H2SO4 (Jaecker-Voirol et al., 1989) and the aqueous oxidation of SO2 (Seinfeld and Pandis, 1998). The aerosol module describes the particles by means of a fixed-moving approach (Wexler et al., 1994); a generic particle is represented with an internal core containing the non volatile material, like elemental carbon, crustal and dust. The core dimension of each size class is established at the beginning of the simulation and is held constant during the simulation. The volatile material is supposed to reside in the outer shell of the particle whose dimension is evaluated by the module at each time step on the basis of the total mass and of the total number of suspended particles. The aerosol module describes the dynamics of 21 chemical compounds: twelve inorganic species (H2O, SO4=, NH4+, Cl–, NO3–, Na+, H+, SO2(aq), H2O2(aq), O3(aq), elemental carbon and other), and nine organics, namely a generic primary organic species and eight classes of secondary organic species. Each chemical species is split in n (namely n = 10) size bins. The estimation of equilibrium pressures of the condensing inorganic species is computed by means of the SCAPE2 thermodynamic module (Kim et al., 1993), while the Condensible Organic Compounds included in COCOH97 mechanism are considered as fully condensed due to their very low volatility. Water is assumed to be always in equilibrium between the gas and the aerosol phases.
3. Simulation Setup The simulation domain has a dimension of 300 × 300 km2 (Figure 1). It includes the Lombardia region as well as portions of Piemonte, Liguria, Veneto and EmiliaRomagna. The site is characterized by complex terrain, by high industrial and urban emissions and by a close road net. The domain has been horizontally divided into 5 × 5 km2 grid cells and vertically in 11 varying levels ranging from 20 to 3,900 m above ground level. As reference case, the 2004 year has been simulated. The input data are provided to the model by meteorological, emission and boundary condition GAMES pre-processors, starting from data shared by JRC-IES in the frame of CityDelta-CAFE exercise (Cuvelier et al., 2007).
5150
SONDRIO
NE
3000
TRENTO
5100
2600
NW
VARESE
2400
N
2200
BERGAMO Vimercate Osio
5050
MagentaMILANO NOVARA Limito Verziere Juvara C
2000 1800
BRESCIA Rezzato
E
VERONA
1600 1400
W
1200 1000
TORINO
5000
PIACENZA
800 600
ALESSANDRIA
400
PARMA
Fig. 1 Simulation domain with orography (m. a.g.l.)
2800
200
SE
4950
GENOVA
SW
MODENA
S
4900
400
450
500
550
600
6 50
0
Modelling Evaluation of PM10 Exposure in Northern Italy in the Framework of CityDeltaIII
429
3.1. Emission fields The emission fields have been estimated by means of POEM-PM model (Carnevale et al., 2006) processing two inventories: the Lombardia Region inventory, with a 5 × 5 km2 resolution, and the EMEP (European Monitoring and Evaluation Programme) one (Vestreng et al., 2004), following a resolution of 50 × 50 km2. The inventories include yearly emission of NOx, VOC, CO, NH3, SOx, PM10 and PM2.5 for each CORINAIR sector. Temporal modulation is performed using monthly, weekly and hourly profiles provided by EMEP (Vestreng et al., 2004). Speciation profiles for organic compounds are defined mapping UK classes (227 species) into SAROAD ones (Carnevale et al., 2006). Chemical and size profiles of emitted PM have been performed using EMEP profiles (Vestreng et al., 2004), provided by JRC. The simulations concern the base case (2004) and three different emission scenario at 2020 (Table 1): (1) the CLE scenario, computed applying to the emission the current legislation up to 2020, (2) the CLE + City (CLEC) scenario, which considers the CLE scenario with PM2.5 set to 0 in the Milan metropolitan area (see dot line in Figure 1), and (3) the MFR (Most Feasible Reduction) scenario, in which the emissions are computed starting from the CLEC scenario and supposing that for nitrogen oxides the best emission reduction technology is applied. Table 1 highlights the heavy emission reduction estimated using the current legislation scenario (close to 50% with the exception of ammonia). The CLEC scenario is equal to CLE one with the exception of PM2.5 (and consequently of PM10), which shows a reduction of about the 80% (compared to the 60% of the CLE emission) with respect to the base case, due to the switch off of the Milan metropolitan area emissions, while the most feasible reduction scenario implies only an extra reduction of NOx with respect the CLEC one. Table 1 Base case total emissions (kt/year) and emission scenario reductions (in comparison to base case). NOx
VOC
PM2.5
PM10
SO2
NH3
208
258
30
38
63
80
54
58
57
53
54
6
2020 CLEC (%)
54
58
77
68
54
6
2020 MFR (%)
74
58
77
68
54
6
Base Case (kt/year) 2020 CLE (%)
4. Base Case Results Figure 2 presents the mean PM10 concentration and the 50 Pg/m3 exceedance days computed for the base case. The highest levels are reached in the Po Valley and in the south-east area of the domain where high NH3 emissions favour the formation of secondary inorganic aerosol.
430
C. Carnevale et al.
5150
5150
SONDRIO
NE
TRENTO
5100
65 60
NW
N
VARESE
45
BRESCIA
MagentaMILANO NOVARA Limito Verziere Juvara C
Rezzato
E
VERONA
BERGAMO
40
25
PIACENZA
20
PARMA SE
4950
GENOVA
SW
Rezzato
5
140
E
VERONA
W
120 100 80
TORINO
5000
PIACENZA
60
ALESSANDRIA
40
PARMA
10
MODENA
160
BRESCIA
MagentaMILANO NOVARA Limito Verziere Juvara C
15
ALESSANDRIA
220
180
N
Vimercate Os io
5050
30
5000
VARESE
35
W
TRENTO
200
NW
50
Vimercate O sio
NE
5100
55
BERGAMO 5050
SONDRIO
70
20
SE
4950
MODENA
0
GENOVA
SW
S
0
S
4900
4900
400
450
500
550
600
400
650
450
500
550
600
650
Fig. 2 PM10 mean concentration (µg/m3, left) and 50 µg/m3 exceedance days (right)
The validation of the aerosol phase simulation results has been performed comparing computed and observed 2004 daily mean concentration in a set of stations selected to be representative of the different emission and meteorological conditions over the domain (Figure 1). Validation results (Table 2) highlight that the model is able to represents the mean value of the period for the entire year, with values of normalized mean error (NME) lower than 0.25, with the exception of Rezzato station and Verziere, where the measured concentration are influenced by local emission phenomena that are very difficult to reproduce with a resolution of 5 × 5 km2. The values of correlation coefficient (CORR) are comparable to performances presented in literature for different models (Cuvelier et al., 2007).
Table 2 Performance indexes computed for PM10 concentration series.
Mean OBS (Pg/m3) Mean TCAM (Pg/m3) NME CORR
Osio
Vimercate
Rezzato
Juvara
Limito
Magenta
Verziere
42.79
42.31
49.11
50.98
43.19
49.29
51.83
44.94
34.08
37.32
60.19
48.62
40.16
69.43
–0.05
0.19
0.24
–0.18
–0.12
0.18
–0.33
0.46
0.36
0.55
0.6
0.49
0.47
0.64
5. Scenario Analysis The evaluation of the impact of the three different scenarios has been performed with respect to the yearly mean values and the 50 Pg/m3 exceedance days in nine selected point (NW, N, NE, W, C, E, SW, S, SE) representative of the different meteorological and chemical regimes in the domain (Figure 1). In the base case, both the indicator present values out of the current air quality standard limits in the
Modelling Evaluation of PM10 Exposure in Northern Italy in the Framework of CityDeltaIII
431
center east of the domain (C, E, SE). The impact on mean values (Table 3) of the three emission scenarios is very similar, in particular in the areas where base case concentrations are lower than 30 ҏPg/m3. The differences are remarkable in the Milan metropolitan area (C point) where MFR and CLEC scenarios show reductions consistently higher than CLE. The impact of the three scenarios on the number of exceedance days (Table 4) is noticeable in all the points of the domain. The differences between the three scenarios in the center point are higher. In fact in this point, the MFR shows a reduction of 192 days with respect to the base case, while the CLE shows a reduction of 97 days. In this point, the local impact of the CLEC scenario could be highlighted, with 40 exceedance days less than the CLE one. For each scenario, the number of exceedance days in the higher concentration areas exceed (C point) or are very close (E, SE points) to the 2020 air quality standard of seven days per year. It is important to note that for both the indicators the impact of the CLE and CLEC scenarios outside the Milan metropolitan area is the same, suggesting that local emission reduction has effect only close to the emission ablation area. Table 3 Scenario impact for mean concentration (µg/m3).
Base Case CLE-Base Case CLEC-Base Case MFR-Base Case
NW
N
NE
W
C
E
SW
S
SE
8.9 –4.8
17.1 –7.5
5.5 –2.1
24.3 –14.1
67.5 –18.2
36.3 –17.1
20.2 –9.1
12.2 –4.5
35.3 –14.1
–4.8
–7.3
–2.0
–14.1
–24.1
–17.0
–9.0
–4.5
–14.3
–4.6
–9.5
–2.6
–13.1
–37.1
–18.4
–11.3
–6.1
–19.2
Table 4 Scenario impact for number of exceedance days. Base Case CLE-Base Case CLEC-Base Case MFR-Base Case
NW 1 –1
N 15 –14
NE 0 0
W 31 –30
C 210 –97
E 87 –80
SW 18 –16
S 2 –1
SE 85 –72
–1
–14
0
–30
–138
–80
–16
–1
–72
–1
–15
0
–30
–192
–76
–18
–2
–78
6. Conclusion The work presents an application of the Transport and Chemical Aerosol Model (TCAM) over a northern Italy domain. The validation phase, performed in the frame of CityDelta project, shows that the model is able to correctly reproduce the measured PM10 daily mean concentration series, in terms of both mean values and correlation coefficient. The scenario analysis has been performed with respect to three different emission scenarios over the selected domain at 2020. The results show that all the scenarios have high impact on the simulated air quality indexes. However, the value of the exceedance days index are not going to respect the 2020 air quality standards in the more industrialized area of the domain.
432
C. Carnevale et al.
Acknowledgments The authors are grateful to Dr. Marco Bedogni (Agenzia Mobilità e Ambiente, Italy) and Dr. Guido Pirovano (CESI, Italy) for their valuable cooperation in the frame of CityDelta project. The work has been partially supported by MIUR (Italian Ministry of University and Research) and by AgipPetroli.
References Carnevale C, Decanini E,Volta M (2008) Design and validation of a multiphase 3D model to simulate tropospheric pollution, Science of Total Environment 390(1), 166–176. Carnevale C, Gabusi V, Volta M (2006) POEM-PM: an emission modelling for secondary pollution control scenarios, Environmental Modelling and Software 21, 320–329. Carter WPL (1990) A detailed mechanism for the gas-phase atmospheric reactions of organic compounds, Atmospheric Environment, 24A, 481–518. Carter WPL, Luo D, Malkina IL (1997) Environmental chamber studies for development of an updated photochemical mechanism for VOC reactivity assessment, Technical report, California Air Resources Board, Sacramento (CA). Carter WPL (2000) Documentation of the SAPRC-99 Chemical Mechanism for VOC Reactivity Assessment, Contract 92-329,95-308, California Air Resources Board. Chock DP, Winkler SL, Sun P (1994) A Comparison of Stiff Chemistry Solvers for Air Quality Modeling, Air & Waste Management Association 87th Annual Meeting. Cuvelier C et al. (2007) Citydelta: a model intercomparison study to explore the impact of emission reductions in European cities in 2010, Atmospheric Environment 41(1), 189–207. Forester CK (1977) Higher order monotonic convection difference schemes, Journal of Computational Physics 23, 1–22. Gery MW, Whitten GZ, Killus JP (1989) A photochemical kinetics mechanism for urban and regional-scale computering modeling, Journal of Geophysical Research 94, 2925–2956. Hindmarsh AC (1975) LSODE and LSODEI, Two New Initial Value Ordinary Differential Equation Solvers, ACM-SIGNUM Newsletter 15(4), 10–11. Jaecker-Voirol A, Mirabel P (1989) Heteromolecular nucleation in the sulfuric acid-water system, Atmospheric Environment 23, 2053–2057. Kim YP, Seinfeld JH, Saxena P (1993) Atmospheric gas aerosol equilibrium I: thermodynamic model, Aerosol Science and Technology 19, 157–187. Marchuk GI (1975) Methods of Numerical Mathematics, Springer, New York. Pepper DW, Kern CD, Long PE (1979) Modelling the dispersion of atmospheric pollution using cubic splines and chapeau functions, Atmospheric Environment 13, 223–237.
Modelling Evaluation of PM10 Exposure in Northern Italy in the Framework of CityDeltaIII
433
Scire JS, Insley EM, Yamartino RJ (1990) Model formulation and users guide for the CALMET meteorological model. Technical Report, California Air Resources Board, Sacramento (CA). Seinfeld JH, Pandis SN (1998) Atmospheric Chemistry and Physics, Wiley, New York. Vestreng V, Adams M, Goodwin J (2004), Inventory review 2004. emission data reported to CRLTAP and under the NEC Directive, Technical report, EMEP/ EEA Joint Review Report. Volta M, Finzi G (2006) GAMES, a comprehensive Gas Aerosol Modelling Evaluation System, Environmental Modelling and Software 21, 578–594. Wexler AS, Seinfeld JH (1991) Second-Generation Inorganic Aerosol Model, Atmospheric Environment 25A, 2731–2748. Wexler AS, Lurmann FW, Seinfeld JH (1994) Modelling urban and regional aerosols-I, model development. Atmospheric Environment 28(3), 531–546. Wille DR (1994) New Stepsize Estimators for Linear Multistep Methods, Numerical Analysis Report 247, inst-MCCM.
4.9 Ozone Modeling over Italy: A Sensitivity Analysis to Precursors Using BOLCHEM Air Quality Model Alberto Maurizi, Mihaela Mircea, Massimo D’Isidoro, Lina Vitali, Fabio Monforti, Gabriele Zanini, and Francesco Tampieri
Abstract The sensitivity of ozone to the reduction of NOx and VOC over Italy has been investigated with the air quality model BOLCHEM, which includes two photochemical mechanisms: SAPRC-90 and CB-IV. The study has been carried out for some case studies during the years 1999 and 2003. The results show the relative importance of precursors in reducing the ozone levels and allow identifying regions of Italy where local emissions reduction strategies are less effective. This study also shows the effect of the errors in isoprene inventories on ozone concentrations.
Keywords Air Quality, BOLCHEM, ozone, photochemistry
1. Introduction Tropospheric ozone pollution is a wide spread air quality problem since its potential impact on human health and environment. In the troposphere, ozone is formed in sunlight by a series of complex chemical reactions that involve nitrogen oxide (NOx) and volatile organic compounds (VOC). Therefore, the amount of ozone in the air depends on not only the amounts of precursors, NOx and VOC, but also on weather conditions such as actinic flux, temperature, pressure, wind speed, humidity. In recent years, a number of studies have investigated the split into VOCsensitive and NOx-sensitive regimes in various areas of the world (Junier et al., 2005; Gabusi and Volta, 2005; Kang et al., 2004; Baertsch-Ritter et al., 2004). Nevertheless, because ozone is the product of many and non-linear atmospheric processes, the ozone response to the reduction of the precursor emissions are still unknown. In this study, we show the sensitivity of ozone concentration to the reduction of NOx and VOC emissions and to the increase of isoprene emissions over the very complex topography of Italy. The air quality model BOLCHEM comprises the limited area meteorological model BOLAM (Buzzi et al., 2003), an algorithm for transport, dispersion and chemical transformations of pollutants. BOLCHEM was C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
420
Ozone Modeling over Italy: A Sensitivity Analysis to Precursors Using BOLCHEM
421
successful validated over Italy (D’Isidoro et al., 2005; Mircea et al., 2007). The simulations were performed for two summer periods: 5–8 August 1999 and 9–12 August 2003. High temperatures and high actinic fluxes (lack of clouds) favourable for photochemical reactions characterized both periods. For the former period, the effects of precursors reduction on ozone was studied with both SAPRC90 (Carter, 1990) and CB-IV (Gery et al., 1989) photochemical mechanisms.
2. Ozone Sensitivity to Reductions of VOC and NOx Figure 1 shows the differences in ozone concentrations simulated for the 5–8 August 1999 with both SAPRC90 and CB-IV. The simulations were performed with a 20 km horizontal resolution grid and 33 sigma vertical levels. The lower level was approximately at 20 m above the surface. The simulated periods start on first day at 00UTC and end last day at 00UTC. The meteorological fields were supplied by ECMWF analyses and lateral boundary conditions were updated every 6 hours. The weather fields were re-initialised every 48 hours in order to avoid an excessive error growth in the meteorological forecast. The chemical fields were driven by hourly surface emissions and 3 hourly lateral boundary conditions after the initialization. Emissions, initial and boundary conditions were obtained from the MINNI model (Zanini et al., 2004). The differences in ozone concentrations plotted in Figure 1 represent the differences between the ozone produced with 65% VOC and the ozone produced with 65% NOx. A region has a NOx chemical regime (NOxsensitive area) if the reduction of NOx emissions is more effective in reducing the ozone concentration. In contrast, if the reduction of VOC emissions can reduce the ozone concentrations, the region is in a VOC chemical regime (VOC-sensitive area). Thus, the positive values in Figure 1 indicate the NOx-sensitive areas while the negative values correspond to the VOC-sensitive areas. According to this definition, the Italian peninsula results almost entirely NOx-sensitive for the two simulated periods. However, there are few but well-defined VOC-sensitive regions, generally located over great urban/industrial areas and harbours such as Genoa, Milan, Rome, Naples, Taranto and southern Sicily. These VOC-sensitive areas are stronger and larger when CB-IV photochemical mechanism is used.
422
A. Maurizi et al.
Fig. 1 Differences in ozone concentrations (ppb) between the ozone concentrations calculated with the emissions of VOC reduced and with the emissions of NOx reduced for 6 August 1999 at 12 UTC (CB-IV-left and SAPRC90-right)
Figure 2 shows the differences in ozone concentrations simulated for the period 9–12 August 2003 with SAPRC90. The simulations for the year 2003 were accomplished using the 1999 emission database since no other updated inventory is available. The comparison of Figure 2 with Figure 1 (right) clearly shows the sensitivity of the chemical regimes to weather conditions, since the simulations have used the same emission inventory and same chemical mechanism. The comparison shows that in August 2003, the intensity of NOx regimes had diminished around Venice, Naples and Messina from 10–35 ppb to 5–10 ppb, while the extension of NOx area with 5–10 ppb intensity had increase.
Fig. 2 Differences in ozone concentrations (ppb) between the ozone concentrations calculated with the emissions of VOC reduced and with the emissions of NOx reduced using SAPRC90 for 11 August 2003
Ozone Modeling over Italy: A Sensitivity Analysis to Precursors Using BOLCHEM
423
Overall, the present simulations show that the distribution and intensity of the chemical regimes are equally controlled by chemical mechanisms and weather conditions (Figures 1 and 2). Isoprene emission inventory is one of the most affected by errors since isoprene has both biogenic and anthropogenic sources. Figure 3 shows the differences in ozone concentrations for 6 August 1999 at 6 and 12 UTC calculated with both photochemical mechanisms for an increase in isoprene concentrations of 300% (Gabusi and Volta, 2005). It can be seen that, even in conditions of low photochemical activity (6 UTC), the increase of isoprene leads to substantial increase in the concentration of ozone. The ozone concentrations predicted at 6 UTC with SAPRC90 are higher than those predicted by CB-IV while CB-IV predicted higher ozone concentrations at 12 UTC. This behaviour shows that the increase of ozone concentrations depends strongly on photochemical mechanism.
Fig. 3 Differences in Ozone concentrations (ppb) calculated with CB-IV (upper panels) and SAPRC90 (lower panels) at 6 and 12 UTC for 6 August 1999
424
A. Maurizi et al.
3. Conclusions The main results of this study are: 1. In some cases, the differences in the predicted ozone concentrations due to the photochemical mechanisms can be comparable to those obtained by reducing the emissions of NOx or VOC. 2. In general, the distribution of VOC- or NOx-limited areas does not show a marked dependence on the photochemical mechanism, although in few limited areas some differences are evident. For instance, in the same meteorological and environmental conditions, a region can occasionally result VOC- or NOx sensitive according with the photochemical mechanism used. 3. The local reduction of VOC was efficient for Milano and Venice areas. In the other regions, significant increase in ozone concentration was observed by reducing locally both the NOx and VOC emissions. 4. The increase of isoprene leads to substantial increase in the concentration of ozone at some locations (up to 25%), therefore, uncertainties in isoprene emissions can bias the air quality design. Acknowledgments Excellence ACCENT and the project GEMS (Global and regional Earth-system Monitoring using Satellite and in-situ data), and by the Italian Ministry of Environment through the Program Italy-USA Cooperation on Science and Technology of Climate Change.
References Baertsch-Ritter N, Keller J, Dommen J, Prevot ASH (2004) Effects of various meteorological conditions and spatial emission resolutions on the ozone concentration and ROG/NOx limitation in the Milan area, Atmos. Chem. Phys., 4, 423–438. Buzzi A, D’Isidoro M, Davolio S (2003) A case-study of an orographic cyclone south of the Alps during MAP SOP, Quart. J. Roy. Met. Soc., 129, 1795–1818. Carter WPL (1990) A detailed mechanism for the gas-phase atmospheric reactions of organic compounds, Atmos. Environ., 24A, 481–518. Gery MW, Witten GZ, Killus JP, Dodge MC (1989) A photochemical kinetics mechanism for urban and regional scale computer modeling. J. Geophys. Res., 94 (D10), 12925–12956. D’Isidoro M, Fuzzi S, Maurizi A, Monforti F, Mircea M, Tampieri F, Zanini G, Villani MG (2005) Development and Preliminary Results of a Limited Area Atmosphere-Chemistry Model: BOLCHEM, First ACCENT Symposium, Urbino 12–16, September, 2005. Gabusi V, Volta M (2005) Seasonal modelling assessment of ozone sensitivity to precursors over northern Italy, Atmos. Environ., 39, 2795–2804. Junier M, Kirchner F, Clappier A, van den Bergh H (2005) The chemical mechanism generation programme CHEMATA-Part2: comparison of four chemical
Ozone Modeling over Italy: A Sensitivity Analysis to Precursors Using BOLCHEM
425
mechanisms for mesoscale calculation of atmospheric pollution, Atmos. Environ., 39, 1161–1171. Kang D, Aneja VP, Mathur R, Ray JD (2004) Observed and modelled VOC chemistry under high VOC/NOx conditions in the Southeast United States national parks, Atmos. Environ., 38, 4969–4974. Mircea M, D’Isidoro, Massimo, Maurizi, Alberto, Vitali L, Monforti F, Zanini G, Tampieri, Francesco (2008) A comprehensive performance evaluation of the air quality model BOLCHEM to reproduce the ozone concentrations over Italy, Atmos. Environ., 42, 1169–1185. Zanini G, Monforti F, Ornelli P, Vialetto G, Brusasca G, Calori G, Finardi G, Silibello C (2004) The MINNI project, 9th Conference on Harmonization within Atmospheric Dispersion Modelling for Regulatory Purposes, 1–4 June 2004, Garmisch-Partenkirchen, Germany.
Discussion P. Builtjes: Is there a difference in model performance between using CB-IV and SAPRC90? M. Mircea: Yes, it is. However, the model results generally fulfil the EPA’s criteria with both photochemical mechanisms.
4.2 Saharan Dust over the Eastern Mediterranean: Model Sensitivity Pavel Kishcha, Slobodan Nickovic, Eliezer Ganor, Levana Kordova and Pinhas Alpert
Abstract 3D-dust distributions, daily predicted by the Tel Aviv University Weather Research Center, are used by the Ministry of Environmental Protection to distinguish between natural and anthropogenic aerosols. Two different dust prediction systems were used to perform a sensitivity experiment and to compare the accuracy of the models. The first one was the original DREAM model with four dust particle size classes (0.7, 6.1, 18.0 and 38.0 Pm); only the first two classes have a radius less than 10 Pm and can be used in comparisons with PM10 measurements. The second model used was the modified DREAM model (DREAM-8) with a more detailed set of dust particle size classes with a radius of between 0.1 and ~7 Pm (0.15, 0.25, 0.45, 0.78, 1.3, 2.2, 3.8, and 7.1 Pm). The sensitivity experiment was carried out by using a quantitative comparison between PM10 measurements of surface dust concentrations and those predicted by models, for the high dust activity season in Israel, from February to June 2006. Quantitative comparisons showed that the use of eight particle size classes including four less than 1 Pm in dust modelling, instead of only two classes, improves dust forecasts. Keywords Forecast, PM10 measurements, Saharan dust
1. Introduction 3D-dust distributions of the Saharan dust transport over the Mediterranean region are daily predicted by the Tel Aviv University Weather Research Center and displayed on the web-site (http://wind.tau.ac.il/dust8/dust.html). The Israeli Ministry of Environmental Protection uses the forecasts to distinguish between natural and anthropogenic aerosols. As found by Kishcha et al. (2007), the use of four particle size bins in the dust modelling instead of only one size bins improves dust forecasts. This study was aimed at evaluating dust distributions predicted by modified DREAM model with a more detailed set of eight dust particle size classes, in order to better understand the model’s capabilities for providing reliable dust forecasts. Provided by the Israeli Ministry of Environment Protection, the PM10 measurements for three different sites in Israel (Beer-Sheva (31.3o N, 34.8o E), TelAviv University (32.1o N, 34.8o E) and Carmiel (32.9o N, 35.3o E) were used for C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
358
Saharan Dust over the Eastern Mediterranean: Model Sensitivity
359
model evaluation. The period under investigation was that of maximum dust activity in the Eastern Mediterranean, from February to May 2006.
2. Dust Prediction Systems Two different dust prediction systems were used to perform a sensitivity experiment and to compare accuracy of the models. The first one is the original DREAM model, and the second one is the modified DREAM model (DREAM-8) with a more detailed set of dust particle size classes, as described below. The fourparticle-size original DREAM model incorporates the state-of-the-art parameterizations of all the major phases of atmospheric dust life such as production, diffusion, advection and removal (Nickovic et al., 2001). Dust productive areas in the model are specified using the US Geological Survey (USGS) data of 30 seconds resolution on land cover. For each soil texture class, the fractions of clay and silt are estimated with four particle size radii of 0.7, 6.1, 18.0 and 38 Pm respectively. In DREAM, the dust cycle is described by a set of K-independent Euler-type concentration equations allowing no inter-particle interactions, where K = 4 indicates the number of particle size class. The area covered by the model is 20o W to 45o E and 15o N to 50o N. The model has a horizontal resolution of 0.3q and 24 vertical levels between the surface and ~15 km. One can see, however, that only the first two size classes in the original DREAM model have their radii less than 10 Pm, which could be used in comparisons with PM10 measurements. These fine particles could be transported far away from their sources in Sahara by means of intensive cyclones, which are responsible for dust transport over the Mediterranean. The other two coarse particle classes are deposited near their sources because of their weight. In order to improve 3D-dust forecasting, the modified version of DREAM model, which is called hereafter DREAM-8, with a more detailed set of eight dust particle size classes between 0.1 and ~7 Pm, has been recently developed (Nickovic, 2005; Pérez et al., 2006). All eight dust particle size classes (0.15, 0.25, 0.45, 0.78, 1.3, 2.2, 3.8, and 7.1 Pm) contribute to dust concentrations on comparing with PM10 measurements.
3. PM10 Measurements The 24-hour dust forecasts were compared with PM10 measurements, taken at the following three different sites in Israel: Beer-Sheva (31.3o N, 34.8o E,), Tel-Aviv University (32.1o N, 34.8o E,) and Carmiel (32.9o N, 35.3o E,), which are located in the southern, central and northern parts of Israel respectively. PM10 surface aerosol concentrations for the high dust activity season in Israel, from February to June 2006, were used for comparisons with model data. The dust models, used in this study, produce forecast of Saharan dust, which is transported from the Sahara into
360
P. Kishcha et al.
the Mediterranean. No local air pollution is included in the models. On the other hand, PM10 measurements include all kids of atmospheric aerosols both natural and anthropogenic. To compare correctly PM10 vs model data, we have to distinguish between mineral dust, transported from the Sahara, and local PM10 aerosols. Without conducting some chemical analysis this is not a simple task. To get around this problem we used the following approach. When aerosol concentrations retrieved from PM10 measurements were highly correlated with each other at different sites, this was the evidence that we deal with aerosols from remote sources, i.e. from the Sahara desert. As estimated, for example, for days with PM10 > 100 Pg/m3, the correlation is higher than 0.5 between any two PM10sites. On the contrary, low correlation between PM10 measurements taken at different sites is the evidence of local aerosols. In particular, for days when PM10 < 100 Pg/m3, the correlation is about 0.2. Therefore, for those days without significant dust, i.e. when PM10 < 100 Pg/m3, we could characterize an average level of local aerosols by estimating their mean and standard deviation. The threshold of local PM10 aerosols was determined as their mean plus standard deviation. As estimated, the threshold in Tel-Aviv (59.7 ȝg/m 3 ) is higher than that in Beer-Sheva (56.6 ȝg/m 3) or in Carmiel (48.7 ȝg/m3 ). Only PM10 measurements, which exceeded the threshold of local aerosols, have been compared with model data.
4. Quantitative Comparisons Between Model and PM10 Data As an illustration, Figure 1a shows a comparison between 24-hour model-predicted surface dust concentrations and PM10 measurements taken at Tel Aviv during March 2006. It is clearly seen that for dusty days, when Saharan dust is transported over Israel, dust aerosols dominate PM10 concentrations. With respect to model capabilities to produce dust forecasts, we see that the old DREAM model frequently underestimates PM10 data. DREAM-8 produces more accurate forecasts than the old DREAM. The correspondence between model data and PM10 measurements higher than the threshold of local aerosols over Tel-Aviv, Israel, was evaluated by means of scatter-plots (Figure 1b). The bisector curve, shown in the scatter-plot, indicates ideally accurate forecasts: the points on or close to the bisector represent the best correspondence between the model-simulated data and the PM10 measurements.
Saharan Dust over the Eastern Mediterranean: Model Sensitivity
361
Fig. 1 a – the comparison between PM10 and model data at Tel Aviv during March 2006. The horizontal line corresponds to the level of local aerosols. b – The scatter-plot between the common logarithm of PM10 and model data at Tel Aviv obtained during the period from February to June 2006. The dashed lines show the root-mean-square interval of the original DREAM deviations from the bisector, while the thin lines show the root-mean-square interval of DREAM-8
The root-mean-square interval of deviations of points from the bisector can be used in order to characterize the range of forecast accuracy. One can see that the root-mean-square interval for DREAM-8 is half as wide as that for the original DREAM (Figure 1b). This conclusion is supported by the results of the correlation analysis. A high correlation 0.54 between PM10 and DREAM-8 data was obtained,
362
P. Kishcha et al.
in contrast to that of 0.27 for the old DREAM model. This suggests the advantage of DREAM-8 compared to the original DREAM.
5. A Typical Route for Dust Transport to Israel As described by Alpert and Ziv (1989), a typical route for dust transport into Israel in spring is from the Sahara desert, through Egypt, into the Eastern Mediterranean. Sometimes, typical dust transport into the Eastern Mediterranean ends with an anticlockwise movement of dust around the Eastern Mediterranean, based on DREAM-8 predictions of dust transport and wind distributions between 13 and 14 March 2006. The dust transport over Gulf of Suez, Sinai, Israel to Cyprus and Turkey was associated with a low-pressure system centred over Greece. The SeaWIFS satellite picture corroborates model simulations by displaying dust around the Eastern Mediterranean cost on March 13 (not shown). Note that Ganor (1991) described a similar event of counter-clockwise dust transport around the Eastern Mediterranean, observed on April 29, 1987. For the dust event on 13–14 March 2006, the comparison between PM10 measurements of surface dust concentration, taken in Tel Aviv with 24-hour model-predicted data, shows that both models produced dust forecasts of quite good accuracy.
6. Unusual Clockwise Dust Transport We noticed that, sometimes, Saharan dust could be transported to the Eastern Mediterranean in an unusual long-distance clockwise movement, from the western part of the Sahara desert through Southern Europe towards Alps and then to the Eastern Mediterranean, based on DREAM-8 predictions. This dust transport started on April 5, 2006 (Figure 2a). The synoptic situation was characterized by the specific structure consisting of an extensive high, dominating the whole of North Africa and the western Mediterranean, and an intensive low over the eastern Atlantic, near the Iberian Peninsula. The low significantly affected the western part of the Sahara desert, as revealed by the wind distribution at 3,000 m altitude. A strong airflow with speeds higher than 20 m/s moved from the Western Sahara to the Western Mediterranean. Such a synoptic situation produced favourable conditions for the development of a heavy dust storm accompanied by an intensive dust intrusion into southern Europe. On the following day, April 6, as the low shifted eastward from the Atlantic across Europe, dust was transported through the central Mediterranean, Italy, Greece, into the Black Sea (Figure 2b). On April 7, dust movement shifted eastward in such a way that dust reached Cyprus and, eventually, the Eastern Mediterranean (Figure 2c).
Saharan Dust over the Eastern Mediterranean: Model Sensitivity
363
a)
b)
c)
d)
e) DREAM-8
250 200
DREAM orig.
150
150
100
100
50
50
0
0
C oncentration, ug/m ^3
200
PM10
07
/0 4 07 /200 /0 4/ 6 0 0 8 2 0 0 2 :0 0 /0 6 4 1 08 /2 00 4:00 /0 6 4/ 0 0 9 2 0 0 2 :0 0 6 /0 1 4 09 /200 4:00 /0 6 4/ 0 1 0 2 0 0 2 :0 0 /0 6 4 1 10 /2 00 4:00 /0 6 4/ 2 0 0 2 :0 06 0 14 :0 0
C oncentration, ug/m^3
250
Fig. 2 a–d are dust loading predicted by DREAM-8 on 5–8 April 2006 respectively, e – is the comparison between PM10 and 24-hour model-predicted data at Tel-Aviv
On April 8, however, another intensive low developed over the Eastern Mediterranean near Turkey, which caused strong wind over Lybia. As the low shifted eastward between 8 and 10 April, the typical counter-clockwise dust transport was observed over the Eastern Mediterranean and Israel (Figure 2d). The aforementioned dust event was unusual for the season in question – the spring. In
364
P. Kishcha et al.
fact, it was a combination of clockwise long-distance transport through southern Europe supplemented in the end, on April 7, 2006, with counter-clockwise transport through Egypt. This situation suggests that both fine dust particles from the Western Sahara and coarse dust particles from the Eastern Sahara could have been presented in the atmosphere. The comparison between PM10 measurements with model-simulated data clearly shows the advantage of DREAM-8 compared to the original DREAM (Figure 2e). In particular, the maximum PM10 concentration is about 240 Pg/m3, while the one of DREAM-8 is about 100 Pg/m3. We also see that the original DREAM model hardly displays dust. This suggests that the use of eight dust-particle-size classes including four less than 1 Pm (DREAM-8) instead of two classes (the original DREAM model) could be significant for long-distance dust transport predictions.
7. Conclusions The quantitative comparison between the PM10 measurements and those, predicted by the DREAM models, showed that the models are capable of giving mainly acceptable forecasts. For DREAM-8, the correlation between model data and PM10 data was found to be higher than that for the original DREAM, indicating that the use of eight-particle size bins in the dust forecasting instead of only two size bins improves dust forecasts. The advantage of DREAM-8 compared to the original DREAM is particularly significant for long-distance dust transport predictions. During spring 2006, Saharan dust was transported into the Eastern Mediterranean not only in the typical route (from the Sahara desert through Egypt into Israel), but also in an unusual clockwise movement from the western part of the Sahara desert through Southern Europe into Israel. Acknowledgments This study was supported by the Israeli Ministry of Environment’s grant, by the GLOWA-Jordan River BMBF-MOST project and also by the BMBF-MOST grant #1946 on climate change. The authors gratefully acknowledge Boris Starobinets for helpful comments and discussion.
References Alpert P, Ziv B (1989) The Sharav cyclone: observations and some theoretical considerations. J. Geophys. Res. 94, 18495–18514. Ganor E (1991) The composition of clay minerals transported to Israel as indicator of Saharan dust emission. Atmos. Environ., 25A, 12, 2657–2664. Kishcha P, Alpert P, Shtivelman A, Krichak S, Joseph JH, Kallos G, Katsafados P, Spyrou C, Gobbi GP, Barnaba F, Nickovic S, Perez C, Baldasano JM (2007) Forecast errors in dust vertical distributions over Rome (Italy): multiple particle size representation and cloud contributions, J. Geophys. Res., 112, D15205, doi:10.1029/2006JD007427.
Saharan Dust over the Eastern Mediterranean: Model Sensitivity
365
Nickovic S (2005) Distribution of dust mass over particle sizes: Impacts on atmospheric optics, paper presented at Fourth ADEC Workshop: Aeolian Dust Experiment on Climate Impact, Ministry of the Environ., Nagasaki, Japan, 2005. Nickovic S, Kallos G, Papadopoulos A, Kakaliagou O (2001) A model for prediction of desert dust cycle in the atmosphere, J. Geophys. Res., 106, 18113– 18129. Pérez C, Nickovic S, Baldasano JM, Sicard M, Rocadenbosch F, Cachorro VE (2006) A long Saharan dust event over the Western Mediterranean: lidar, sun photometer observations and regional dust modeling. Journal of Geophysical Research, 111, D15214, doi:10.1029/2005JD006579.
Discussion D. Syrakov: What is the threshold in your first figure? What does it mean? Is there concentrations less then it and how have you determined it? P. Kishcha: The dust models, used in this study, produce forecast of Saharan dust, which is transported from the Sahara into the Mediterranean. No local air pollution is included in the models. On the other hand, PM10 measurements include all kids of atmospheric aerosols both natural and anthropogenic. To compare correctly PM10 vs model data, we have to distinguish between mineral dust, transported from the Sahara, and local PM10 aerosols. Without conducting some chemical analysis this is not a simple task. To get around this problem we used the following approach: When aerosol concentrations retrieved from PM10 measurements were highly correlated with each other at different sites, this was the evidence that we deal with aerosols from remote sources, i.e. from the Sahara desert. As estimated, for example, for days with PM10 > 100 Pg/m3, the correlation is higher than 0.5 between any two PM10-sites. On the contrary, low correlation between PM10 measurements at different sites is the evidence of local aerosols. In particular, for days when PM10 < 100 Pg/m3, the correlation is about 0.2. Therefore, for those days without significant dust, i.e. when PM10 < 100 Pg/m3, we could characterize the average level of local aerosols by estimating their mean and standard deviation. The threshold of local PM10 aerosols was determined as the mean plus standard deviation. Only PM10 measurements, which exceeded the threshold of local aerosols, have been compared with model data.
P. Kishcha et al.
366
S. Hanna: In your statistical evaluations (e.g. correlation coefficients) of the observed and modeled concentrations, you applied a threshold to the observed concentrations but not to the modeled concentrations. Shouldn’t you also apply the same threshold to the modeled concentrations? This is, the statistical calculations should use only those data pairs where both observed and modeled concentrations exceed the threshold. P. Kishcha: Two models were evaluated by using PM10 measurements: DREAM-8 and the old DREAM. It was obtained that the old DREAM model underestimated dust concentrations, while DREAM-8 produced more accurate forecast. We could not apply the same threshold to model-simulated data as we did to PM10 measurements: otherwise we could remove low model values produced by the old DREAM model, and thereby artificially improve its forecasts. G. Kallos: In order to use eight size classes in dust particles you need to have an accurate source area classification. Do you have such data available or you use just what was in the model version developed at the University of Athens several years ago (SKIRON project)? P. Kishcha: In DREAM-8 with eight particle size bins we use the same dust sources as in the original DREAM model with four particle size bins (and also in SKIRON). Ref.: Nickovic et al., JGR, 18, 113– 118,129, 2001. Y.P. Kim: What was the dry deposition algorithm in the model? P. Kishcha: The dry deposition of dust particles was parameterized according to the scheme of Georgi JGR, 9794-–9806, 1986.
4.1 The Effect of Heterogeneous Reactions on Model Performance for Nitrous Acid Golam Sarwar, Robin L. Dennis and Bernhard Vogel
Abstract Recent studies suggest that emissions, heterogeneous reactions, and surface photolysis of adsorbed nitric acid may produce additional nitrous acid in the atmosphere. The effects of these sources on nitrous acid formation are evaluated using the Community Multiscale Air Quality modeling system. Predicted nitrous acid with and without these sources are compared with observed data from northeast Philadelphia. The incorporation of these sources greatly improves the model performance for nitrous acid. It also increases the average hydroxyl radical and ozone by 10% and 1.7 ppbv, respectively. Keywords Emissions, heterogeneous reaction, nitrous acid, surface photolysis reaction
1. Introduction The importance of nitrous acid (HONO) chemistry in producing hydroxyl (OH) and hydroperoxy (HO2) radicals is well established. Alicke et al. (2002, 2003) suggested that the photolysis of HONO may provide as much as 34% of daily integrated OH levels. Zhou et al. (2002) reported that the photolysis of HONO may produce up to 24% of the total daily radical production. Using concurrently measured HONO and OH at the Meteorological Observatory Hohenpeissenberg in summer 2002 and 2004, Acker et al. (2006) suggested that the photolysis of HONO produced 42% of the integrated photolytic HOx (OH + HO2) formation. While the effect of HONO on OH is established, the chemical reactions producing HONO are not well understood. Most air quality models, including the Community Multiscale Air Quality (CMAQ) modeling system, employ only homogeneous chemical reactions for HONO. Recent studies suggest that emissions, heterogeneous reactions, and surface photolysis of adsorbed nitric acid may produce additional HONO in the atmosphere (Vogel et al., 2003; Zhou et al., 2003). In this study, the effects of these sources on HONO are evaluated using the CMAQ modeling system.
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science +Bus iness Media B.V . 2008
349
G. Sarwar et al.
350
2. Methodology 2.1. Model description Model simulations were performed using the CMAQ modeling system (version 4.6) (Binkowski and Roselle, 2003; Byun and Schere, 2006). The CMAQ chemical transport model was configured to use the mass continuity scheme to describe advection processes, the Asymmetric Convective Model version 2 (ACM2) (Pleim, 2007) to describe vertical diffusion processes, the multiscale method to describe horizontal diffusion processes, an adaptation of the ACM algorithm for convective cloud mixing, and the Carbon Bond (CB05) mechanism to describe the gas-phase chemical mechanism (Yarwood et al., 2005; Sarwar et al., 2007). Aqueous chemistry, aerosol processes, and dry and wet deposition were also included. The meteorological driver for the CMAQ modeling system was the PSU/NCAR MM5 system version 3.5 (Grell et al., 1994). Predefined clean air vertical profiles for initial and boundary conditions provided in the CMAQ modeling system were used. Model simulations were performed for July 2001; the model was spun up for seven days to minimize the effect of initial conditions on predictions.
2.2. HONO chemistry The CB05 mechanism contains five homogeneous reactions related to HONO (Yarwood et al., 2005). These reactions and their rate constants are shown in Eqs. (1)–(5) (NO = nitric oxide, NO2= nitrogen dioxide, H2O = water vapor, k = expression for rate constant, first order rate constants are in units of second-1, second order rate constants are in units of cm3 molecule-1 second-1, third order rate constants are in units of cm6 molecule-2 sec-1, T = temperature in Kelvin, M = the total pressure in molecules/cm3, photolysis rate for Eq. (3) is at 400 N (typical summer noon). While this study used the CB05 mechanism, the chemical reactions used in other widely used atmospheric chemical mechanisms are also similar. For example, the CB-IV mechanism contains the same five chemical reactions for HONO (Gery et al., 1989). The Statewide Air Pollution Research Center (SAPRC99) mechanism contains four chemical reactions for HONO (Eqs. (2)–(4) and an additional photolytic reaction) (Carter, 2000). NO + NO2 + H2O ĺ 2.0*HONO; k = 5.0 × 10-40 NO + OH ĺ HONO;
k={ko[M]/(1+ko[M]/k)}Fz Z={(1/N)+log10[ko[M]/k]2}-1 ko = 7.0 × 10-31 (T/300)-2.6 k = 3.6 × 10-11 (T/300)-0.1
(1)
The Effect of Heterogeneous Reactions on Model Performance for Nitrous Acid
F = 0.6 and N = 1.0 HONO + hv ĺ NO + OH; photolysis rate = 0.002 sec-1 OH + HONO ĺ NO2 + H2O; k = 1.8 × 10-11 e(-390/T) HONO + HONO ĺ NO + NO2 + H2O; k = 1.0 × 10-20
351
(2) (3) (4) (5)
As shown below, CMAQ predictions with observed data indicate incorporating only the homogeneous reactions results in HONO predictions that are seriously deficient. HONO emissions are currently not included in the CMAQ modeling system. Several investigators have reported HONO emissions from motor vehicles (Winer and Biermann, 1994; Kirchstetter et al., 1996; Kurtenbach et al., 2001). Kirchstetter et al. (1996) measured HONO and NOx emissions from on-road vehicles at Caldecott Tunnel in San Francisco, California and reported a value of 2.9 × 10-3 for the HONO/NOx emissions ratio. Kurtenbach et al. (2001) conducted measurements in the Wuppertal Kiesbergtunnel and reported a value of 8 × 10-3 for the same ratio. Winer and Biermann (1994) also reported a value of 8 × 10-3 for the same ratio. For this study, HONO emissions were estimated using a value of 8 × 10-3 for the HONO/NOx emissions ratio for on-road and off-road vehicles. Several heterogeneous reactions have been suggested to produce HONO formation in the atmosphere (Aumont et al., 2003). However, most of these reactions appear to be not significant for HONO production in the atmosphere. The heterogeneous reaction involving NO2 and H2O has been shown to be important for HONO production in the atmosphere (Vogel et al., 2003): 2NO2 + H2O ĺ HONO + HNO3
(6)
Laboratory studies suggest that reaction (6) is first order in NO2 (Finlayson-Pitts and Pitts, 2000) and can occur on aerosol and ground surfaces. This heterogeneous reaction is implemented into the CMAQ modeling system with a rate constant of 3.0 × 10-3 × S/V m-1 following Kurtenbach et al. (2001) (S/V is the ratio of surface area to volume of air). The reaction can occur on aerosol as well as ground surfaces. Aerosol surface areas are generally smaller than ground surface areas; thus aerosol surface areas are less effective in producing HONO in the atmosphere. Ground surface areas provided by leaves can be estimated using the Leaf Area Index (LAI). Jones (2006) suggests that the use of LAI underestimates leaf surface areas by at least a factor of two since it only accounts for areas on one side of the leaves. Thus, the S/V ratios for leaves were estimated as follows: S/V = 2*LAI / surface layer height
(7)
Buildings and other structures can also enhance ground surface areas in urban environments. However, estimates of these areas are not readily available. In the absence of such information, an ad hoc approach was used to estimate the ground surface areas for buildings and other structures in urban environments. Svensson
G. Sarwar et al.
352
et al. (1987) studied the kinetics of the reaction involving NO2 and H2O and suggested a value of 0.2 m-1 for typical urban environments. They, however, also indicated that some building materials may provide an order of magnitude greater surface area than their simple projected surface areas due to porosity and roughness. For this study, the S/V ratio for buildings and other structures at the grid-cell with the highest urban environment was assigned a value of 0.3 m-1. The S/V ratios for buildings and other structures for other urban environments were linearly scaled to this S/V ratio by assuming that the values are proportional to the percent urban area in any grid-cell. Using this procedure, the total S/V ratio for the grid-cell containing northeast Philadelphia was estimated to be 0.28 m-1, which is almost 5 times lower than the value of 1.3 m-1 used by Cai et al. (2007) for New York. HONO produced via the heterogeneous reaction on ground surfaces was released into the first layer of the model. Several recent studies also suggest the possibility of the production of HONO via surface photolysis (Zhou et al., 2002, 2003; Vogel et al., 2003; Acker et al., 2006). In their study, Vogel et al. (2003) used a hypothetical species to add a source for HONO production in the surface layer of the model via photolysis since the species that may undergo photolysis to produce HONO was not known. Zhou et al. (2003) recently conducted laboratory experiments and suggested that adsorbed nitric acid (HNO3) on surfaces can undergo photolysis to produce HONO and NO2. A photolysis reaction producing HONO and NO2 was added to the CMAQ modeling system (Eq. (8)) by assuming that the adsorbed amount is equal to the HNO3 deposited via dry deposition since the last precipitation (O3P = ground state oxygen atom). HNO3(ads) ĺ 0.5 HONO + 0.5 O3P + 0.5 NO2 + 0.5 OH
(8)
Zhou et al. (2003) reported a photolysis rate of 1.3 × 10-3 minute-1 at noontime tropical condition for the surface photolysis of adsorbed HNO3, which is 24 times greater than the gaseous HNO3 photolysis rate in the CMAQ modeling system. The surface photolysis rate of adsorbed HNO3 was scaled to the photolysis rate of gaseous HNO3 used in the CMAQ modeling system. HONO and NO2 produced via the surface photolytic reaction was released into the first layer of the model only.
2.3. Observed data Model predictions are compared with measurements from the Northeast Oxidant and Particle Study conducted at northeast Philadelphia (http://lidar1.ee.psu.edu) in July 2001. Continuously measured data from the study were converted into hourly averaged data which were used to calculate an average diurnal profile. Average values for night and day were calculated using the average diurnal profile.
The Effect of Heterogeneous Reactions on Model Performance for Nitrous Acid
353
3. Results Four different model simulations were performed as shown in Table 1. Table 1 Summary of cases investigated. Case A B C D
Emissions Included NOx, SO2, CO, VOC, NH3, aerosol NOx, SO2, CO, VOC, NH3, aerosol, HONO NOx, SO2, CO, VOC, NH3, aerosol, HONO NOx, SO2, CO, VOC, NH3, aerosol, HONO
Chemical mechanism used CB05 CB05 CB05 + reaction 6 CB05 + reaction 6 and 8
The average predicted and observed HONO at northeast Philadelphia are presented in Table 2. Observed HONO mixing ratio was 50% greater at night than that during the day. Predicted HONO mixing ratio for the case A was only 0.01 ppbv at night compared to an observed value of 1.26 ppbv. Contrary to the observed data, predicted HONO mixing ratio was greater during the day by at least a factor of 4 over the mixing ratio at night. Predicted HONO mixing ratio was significantly lower than the observed data both at night and during the day. Table 2 Summary results.
Case A B C D
Night Obs. HONO (ppbv) 1.26 1.26 1.26 1.26
Pred. HONO (ppbv) 0.01 0.11 0.95 0.96
Obs./ pred. (ratio) 126 11 1.3 1.3
Day Obs. HONO (ppbv) 0.85 0.85 0.85 0.85
Pred. HONO (ppbv) 0.04 0.07 0.25 0.60
Obs./ Pred. (ratio) 21 12 3.4 1.4
When HONO emissions were added to the model (case “B”), predicted HONO reached to 0.11 ppbv at night and was slightly higher at night than during the day. Predicted HONO mixing ratio was still lower than the observed data by a large margin both at night and during the day. When the heterogeneous reaction was also added to the model (case “C”), predicted HONO mixing ratio further improved at night and reached to within 30% of the observed data. However, predicted HONO mixing ratio was still lower than the observed data during the day by a factor of 3.4. When the surface photolysis of adsorbed HNO3 was added to the model (case “D”), predicted HONO mixing ratio improved to 0.60 ppbv during the day due to the increased HONO production and reached to within 40% of the observed data. The observed ratio of HONO/HNO3 was 2.9 at night. For the case A, the ratio was only 0.006 and improved to 0.6 for the case D. During the day, the observed ratio was 0.73 compared to a value of only 0.02 for the case A. It improved to 0.2 for the case D. The relative contribution of these production pathways to predicted HONO for the case D is dominated by two pathways. The reaction #6 contributed 56% to the predicted HONO and was the largest contributor. The reaction #8 was the second
354
G. Sarwar et al.
largest contributor and contributed 32% to the predicted HONO. The HONO emissions and the homogeneous reactions (Eqs. (1)–(5)) contributed only 8% and 4% to the predicted HONO, respectively. HONO undergoes photolysis in the atmosphere to produce OH radicals. The additional HONO enhanced the average OH by 10% in the case D compared to that of the case A. The inclusion of the additional HONO sources increased the average O3 by 1.7 ppbv. The increased O3 is a contribution of additional VOC oxidation via enhanced OH as well as the photolysis of additional NO2 generated from the surface photolysis of adsorbed HNO3 (reaction #8).
4. Summary The results of this study suggest that heterogeneous reaction and surface photolysis of adsorbed HNO3 are important sources of HONO in the atmosphere. Most air quality models, however, do not currently account for these sources; thus, models tend to under-predict HONO by a large margin. The incorporation of these sources improves HONO predictions. The improved HONO predictions can increase OH as well as O3. Surface areas of leaves can be estimated using the LAI. Better estimates of the surface areas of buildings and other structures in urban environments are needed to improve HONO predictions via the heterogeneous reaction. However, such information is not readily available; efforts should be directed in determining such values. The surface photolysis of adsorbed HNO3 producing HONO and NO2 during the day is an emerging topic; many related issues are still unknown. However, it appears to provide the missing HONO source during the day without which model predictions remain under-predicted compared to observed data. Thus, this reaction should be further explored in laboratory as well as field studies before it can be confidently used in air quality models. Disclaimer The research presented here was performed under the Memorandum of Understanding between the U.S. Environmental Protection Agency (EPA) and U.S. Department of Commerce’s National Oceanic and Atmospheric Administration (NOAA) and under agreement number DW13921548. This work constitutes a contribution to the NOAA Air Quality Program. Although it has been reviewed by EPA and NOAA and approved for publication, it does not necessarily reflect their policies or views.
References Acker K, Möller D, Wieprecht W, Meixner FX, Bohn B, Gilge S, Plass-Dülmer C, Berresheim H (2006) Strong daytime production of OH from HNO2 at a rural mountain site, Geophys. Res. Lett., 33, L02809, doi:10.1029/2005GL024643.
The Effect of Heterogeneous Reactions on Model Performance for Nitrous Acid
355
Alicke B, Platt U, Stutz J (2002) Impact of nitrous acid photolysis on the total hydroxyl radical budget during the limitation of oxidant production/Pianura Padana Produzione di Ozono study in Milan, J. Geophys. Res., 107, doi:10.1029/2000JD000075. Alicke B, Geyer A, Hofzumahaus A, Holland F, Konrad S, Pätz HW, Schäfer J, Stutz J, Volz-Thomas A, Platt U (2003) OH formation by HONO photolysis during the BERLIOZ experiment, J. Geophys. Res., 108, 8247, doi:10.1029/ 2001JD000579. Aumont B, Chervier F, Laval S (2003) Contribution of HONO sources to the NOx/HOx/O3 chemistry in the polluted boundary layer, Atmos. Environ., 37, 487–498. Binkowski FS, Roselle SJ (2003) Community Multiscale Air Quality (CMAQ) model aerosol component, I: Model description, J. Geophys. Res., 108(D6): 4183, doi:10.1029/2001JD001409. Byun D, Schere KL (2006) Review of the governing equations, computational algorithms, and other components of the Models-3 Community Multiscale Air Quality (CMAQ) modeling system, Appl. Mech. Rev., 59, 51–77. Cai C, Hogrefe C, Schwab JJ, Katsafados P, Kallos G, Ren X, Brune WH, Zhou X, He Y, Demerjian KL (2007) Performance evaluation of an air quality forecast modeling system for a summer and winter season – Part II: HONO formation processes and their implications for HOx budgets, J. Geophys. Res., in review. Carter, WPL (2000) Implementation of the SAPRC-99 chemical mechanism into the Models-3 Framework, report to the United States Environmental Protection Agency. Available at http://www.cert.ucr.edu/~carter/absts.htm#s99mod3 Finlayson-Pitts BJ, Pitts JN Jr (2000) Chemistry of the Upper Lower Atmosphere, Theory, Experiments and Applications, Academic, San Diego, CA. Gery MW, Whitten GZ, Killus JP, Dodge MC (1989) A photochemical kinetics mechanism for urban and regional scale computer modeling. J. Geophys. Res., 94(D10), 12925–12956. Grell G, Dudhia J, Stauffer D (1994) A description of the fifth-generation Penn State/NCAR Mesoscale model (MM5), NCAR Tech. Note NCAR/TN-398+STR. Jones MR (2006) Ammonia deposition to semi-natural vegetation, PhD dissertation, University of Dundee, Scotland. Kirchstetter TW, Littlejohn D (1996) Measurements of nitrous acid in motor vehicle exhaust, Environ. Sci. Technol., 30, 2843–2849. Kurtenbach R, Becker KH, Gomes JAG, Kleffmann J, Lorzer JC, Spittler M, Wiesen P, Ackermann R, Geyer A, Platt U (2001) Investigations of emissions and heterogeneous formation of HONO in a road traffic tunnel, Atmos. Environ., 35, 3385–3394. Pleim JE (2007) A combined local and nonlocal closure model for the atmospheric boundary layer. part I: model description and testing, J. Appl. Meteor. Clim., 46, 1383–1395. Sarwar G, Luecken V, Yarwood G, Whitten G, Carter WPL (2007) Impact of an updated carbon bond mechanism on predictions from the Community Multiscale Air Quality (CMAQ) modeling system: preliminary assessment, J. Appl. Meteor. Clim., accepted.
356
G. Sarwar et al.
Svensson R, Ljungstrom E, Lindqvist O (1987) Kinetics of the reaction between nitrogen dioxide and water vapour, Atmos. Environ., 21, 1529–1539. Vogel B, Vogel H, Kleffmann J, Kurtenbach R (2003) Measured and simulated vertical profiles of nitrous acid – Part II, model simulations and indications for a photolytic source, Atmos. Environ., 37, 2957–2966. Winer AM, Biermann HW (1994) Long pathlength differential optical absorption spectroscopy (DOAS) measurements of gaseous HONO, NO2 and HCHO in the California South Coast Air Basin. Res. Chem. Intermed., 20, 423–445. Yarwood G, Rao S, Yocke M, Whitten G (2005) Updates to the Carbon Bond Chemical Mechanism: CB05, Final Report to the US EPA, RT-0400675, Available at http://www.camx.com/publ/pdfs/CB05_Final_Report_120805.pdf Zhou X, Civerolo K, Dai H, Huang G, Schwab JJ, Demerjian KL (2002) Summertime nitrous acid chemistry in the atmospheric boundary layer at a rural site in New York State, J. Geophys. Res., 107(D21), 4590, doi:10.1029/ 2001JD001539. Zhou X, Gao H, He Y, Huang G, Bertman SB, Civerolo K, Schwab J (2003) Nitric acid photolysis on surfaces in low-NOx environments: significant atmospheric implications, Geophys. Res. Lett., 30(23), 2217, doi:10.1029/2003GL018620.
Discussion A. Venkatram: How do you adsorb HNO3 in the model? How do you justify assuming that all the day deposited HNO3 is available for photolysis? Does the inclusion of HNO3 photolysis affects HNO3 in the atmosphere? G. Sarwar:
W. Gong: G. Sarwar:
In the model, it is assumed that HNO3 that is removed from the atmosphere via dry deposition gets adsorbed on surfaces. Adsorbed HNO3 can then undergo photolysis to produce HONO during the day. A fraction of the deposited HNO3, indeed, may not be available all day for photolysis. However, the detailed processes that form HONO in the atmosphere are still unknown. In this study, we explored the effects of heterogeneous reaction, homogeneous reactions, emissions, and surface photolysis of HNO3 on HONO. Since we used all available adsorbed HNO3 for photolysis, it does represent the upper limit of HONO formed via the surface photolysis process. The inclusion of surface photolysis of adsorbed HNO3 did not significantly affect HNO3 in the atmosphere. Did the changes involving HONO make any difference in model predictions of O3?
The Effect of Heterogeneous Reactions on Model Performance for Nitrous Acid
357
The changes involving HONO made some differences in ozone predictions. Predicted ozone increased by up to 2.4 ppbv. The diurnally averaged ozone increased by 1.4 ppbv. The increases started during the morning hours and continued throughout the day. However, the increases in predicted ozone were not adequate to explain the observed early morning rise. M. Brauer: How was HONO measured in the field campaign? Depending on the measurement approach there may be artifacts from other nitrogen species. G. Sarwar: Ambient HONO were measured using ion chromatography technique that has a detection limit of about 0.11 pptv and accuracy of about 5%.
4.4 Uncertainty in Air Quality Decision Making Bernard Fisher
Abstract This paper considers the decisions that should be made arising from a prediction using a model, taking into account the uncertainty associated with the prediction. Sometimes taking account of uncertainty leads to better decisions than just taking a decision on the basis of a single, central value. This is illustrated by examples taken from air quality models. Regulatory models need to be simple, leading to effective decision making, but their use implies accepting greater uncertainty. The paper describes an approach to this dilemma. Keywords Decision analysis, loss function, regret, uncertainty, ozone, nitrogen dioxide, limit value
1. Introduction to Decision Theory There has been considerable interest recently in the uncertainty associated with predictions made by air quality models (see Borrego et al., 2006, 2007). However less attention is paid as to how this leads to a decision. A decision should always be related to an action. Let the parameter ș describe the state of nature, with prior probability p(ș), which describes the uncertainty, and let d denote the action or decision, so that L(d,ș) is the regret or loss. Then the average expected risk or loss is
L (d )
L(d ,T ) p (T ) ¦ T
(1)
The preferred action (Morgan and Henrion, 1990) is the action d = dmin which minimises L(d ) and the risk, or loss, for this action is equal to Lmin
Lmin
min d
L(d ,T ) p(T ) ¦ L(d ¦ T T
min ,T ) p (T )
(2)
In air pollution one goes to great effort to evaluate the concentration ș and its probability distribution, which depends on the uncertainty in the parameters within the model. Often Monte Carlo simulation may be used. However on its own the Monte Carlo analysis does not lead to a decision (see discussion in Fisher and Willows, 2007). Sometimes measured values are used to calibrate the model, in an
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
376
Uncertainty in Air Quality Decision Making
377
attempt to reduce the uncertainty, but not eliminating it. What is generally neglected is the problem of deciding on the form of L(d,ș). A decision can be made by ignoring the uncertainty in the value of ș. Let us assume that if one ignores the uncertainty, one takes a best estimate of ș, say the mean T . The decision ignoring uncertainty diu would then be based on taking the minimum of L(d ,T ) . The extra loss incurred by ignoring uncertainty is ( L(d ¦ T
iu ,T )
L(d min ,T )) p (T )
(3)
which from the definition of dmin is always positive or equal to zero. In Eqs. (1) and (2) it is assumed that there are a finite number of discrete states ș may occupy. ș could vary continuously over a range, in which case the summation would become an integral. The range over which ș varies could also be replaced by a number of discrete intervals. Later we will consider a simple example of the variation of ș. However the decision itself is always discrete. There may be alternatives, but one is always looking for the preferred option. The development has parallels with hypothesis testing, when the decision-maker in that case has to decide about acceptable errors. In the present case acceptable error is described by the form of L(d,ș).
2. Air Pollution Example: Limit Value Exceedences We shall consider a simple air pollution example, considering whether an air quality limit value, denoted by the concentration șc, is exceeded. There are two possible decisions d0 corresponding to a belief that șc is ‘not exceeded’ and decision d1 that the limit value is ‘exceeded’. There is a loss associated with taking decision d0 no exceedence, when there really is an exceedence, associated with health effects, given by
L ( d 0 ,T )
a0 (T T c ) when ș> șc and zero otherwise.
(4)
Similarly there is a loss associated with decision d1 exceedence, if there really is no exceedence, associated with unnecessary control measures, given by
L(d1 ,T )
a1 (T c T ) when ș< șc and zero otherwise.
(5)
If one ignores uncertainty and the model tells us that ș= T , taken to be greater than șc, then the decision is d1 (exceedence). It is assumed for convenience that the uncertainty is described by a rectangular probability distribution around the mean T , so that p (T )
1 , where T - w/2 d ș d T + w/2 and zero otherwise, w
(6)
B. Fisher
378
where w is the range of uncertainty. A loss occurs, when ignoring uncertainty, if the true concentration is less than the limit value, though the predicted concentration is above. In the above idealised case the loss would be L (d1 )
Tc
³T
w / 2
L(d1 ,T ) p (T )dT
a1 w (T T c ) 2 2 2
(7)
If one took account of uncertainty, the expected loss, for decision d1, would be as above, but for decision d0, would be
L (d0 )
T w/ 2
³T L(d ,T ) p(T )dT 0
c
a0 w (T T c ) 2 2 2
(8)
If L (d1 ) d L (d 0 ) then the preferred decision taking account of uncertainty would be d1, the same as the decision ignoring uncertainty. However if L (d1 ) > L (d 0 ) the preferred decision would be d0. Thus the expected value of including una w certainty is either 0, when L (d1 ) < L (d 0 ) , or L (d1 ) - L (d 0 ) = 1 (T T c ) 2 2 2 a0 w 2 (T T c ) , when L (d1 ) > L (d 0 ) . 2 2 If a0~a1 and T > șc then would generally expect that d1 was the preferred decision, since L (d1 ) - L (d 0 ) v - (T T c ) w <0 and there is no advantage in taking into account uncertainty. However if a0<>a1 one would take decision d1 and this is probably the situation envisaged when implementing European Directive limit values.
3. Types of Environmental Standard The form of L(d,ș) is critical to making decisions, but has received little attention in the air pollution literature. Setting an environmental standard sets a discrete boundary on what is really the fuzzy set of acceptable concentrations. The fuzziness has been acknowledged by recognising that standards fall into two categories (Mark Crane, personal communication). Type A standards are designed to be precautionary so that if they are met, this is anticipated not to result in any adverse environmental effects. Exceedence of this standard does not necessarily lead to adverse environmental consequences because the standard tends to be designed to be conservative and precautionary. An example of this kind is the concept of critical loads. The second type of value, Type B, is a standard that, if exceeded, is anticipated to result in adverse environmental effects. Examples of Type B values might be the standards used to take action after pollution incidents, or standards where failure
Uncertainty in Air Quality Decision Making
379
leads directly to a regulatory response. Examples include the limit values in EU Directives. When a particular standard is set, a judgement is made, at least implicitly, regarding the level of precaution incorporated and determines where the standard lies in the spectrum from Type A to Type B. The range from higher than Type A to lower than Type B values spans the probability of unacceptable environmental consequences. The form of L(d,ș) should describe the degree of precaution associated with the standard. Classification of standards into Type A and B categories is equivalent to allowing for uncertainty. A Type A standard may be used as a screening tool (where uncertainty is high, and a precautionary approach to the uncertainty has been taken), whereas a Type B standard is a legally mandatory pass/fail threshold (for which there is little doubt). The rest of this paper considers how these results may be used in actual model applications. The Environment Agency has supported the use of simple statistical models for assessing airborne concentrations and deposition of sulphur oxides, nitrogen oxides and particulates. These will be considered from the viewpoint of regulation and decision making.
4. Application to Regulatory Models The models have been tested and calibrated against measurements although the number of monitoring sites is limited. Some examples are as follows. For acid deposition, D, associated with sulphur and nitrogen deposition, the range of uncertainty is generally such that ½w ~ 0.3D. However the environmental standard is the critical load which varies spatially. A more direct application is to air quality for which limit values have been set which apply everywhere. In their comparison of a simple statistical model of NO2 with measurements, Abbott et al. (2007) obtained the following comparisons shown in Figures 1 and 2. 60 National Network Verification Sites y =x y = x - 30% y = x + 30%
Modelled NO2 (ugm-3)
50
40
30
20
10
0 0
10
20
30
40
50
60
Measured NO2 (ugm-3)
Fig. 1 Verification of background annual mean NO2 model 2002
B. Fisher
380
If appears that choosing ½w ~ 0.3C, where C is the best prediction of the concentration, is a reasonable assumption for NO2. Figures 3 and 4 show similar comparison plots of measured and modelled PM10 concentrations at background and roadside monitoring sites. The range of uncertainty, ½w ~ 0.5C, is somewhat larger. Similarly the 0.3 uncertainty range is quoted for EMEP modelling of particle concentrations (Tarrason et al., 2005), but with the largest uncertainty in the particle fraction involving elemental and organic carbon, which makes up 30–50% of the total mass. From Section 2 we found that one can ignore uncertainty in the a w a w decision, if 1 (T T c ) 2 - 0 (T T c ) 2 < 0. If it is positive then one gets a 2 2 2 2 better decision by including uncertainty. Assuming ½w = 0.3 T , T ~ șc (1 + H) where H << 1 (the prediction is close to the limit), it is only worth considering (a a 0 ) 0.6H uncertainty value when a1 (H 0.3) 2 - a 0 (H 0.3) 2 >0 or 1 . ! 2 (a 0 a1 ) (H 0.09) 90 80
National Network Verification Sites y=x y = x - 30% y = x + 30%
Modelled NO2 (ugm-3)
70 60 50 40 30 20 10 0 0
10
20
30
40
50
60
70
80
90
Measured NO2 (ugm-3)
Fig. 2 Verification of roadside annual mean NO2 model 2002 30
Modelled annual mean PM10 (ugm-3, TEOM)
National Network Verification Sites y=x
25
y = x - 50% y = x + 50%
20
15
10
5
0 0
5
10
15
20
25
30
Measured annual mean PM10 (ugm-3, TEOM)
Fig. 3 Verification of background annual mean PM10 model 2002
Uncertainty in Air Quality Decision Making
381
40 National Network Verification Sites
Modelled annual mean (ugm-3, TEOM)
35
y=x y = x - 50%
30
y = x + 50%
25 20 15 10 5 0 0
5
10
15
20
25
30
35
40
Measured annual mean (ugm-3, TEOM)
Fig. 4 Verification of roadside annual mean PM10 model 2002
The term on the right is usually less than 1 and only equals 1 when H = 0.3. One may assume that near the limit value of 40 ȝg m-3 for PM10 as an annual average, there is an optimum, so that a1 ~ a0, and uncertainty can be neglected in decisions concerning exceedence. However one might expect a1 > a0 near the stricter indicative limit value of 20 ȝg m-3 and uncertainty should be included in an assessment of exceedence. For NO2 the health benefits of emission reductions set by the magnitude of a0 are less easy to evaluate. Generally one should expect that for stricter limit values, more attention should be paid to the uncertainty in the decision.
5. Regulatory vs Research Models An issue that arises for a regulator is when to use a simplified, regulatory model to make decisions, or when to invest in a more complex model, or in improved monitoring. One might decide to choose a simpler model with greater uncertainty, because it is more practical to apply. The decision about choosing one model in preference to another (a research model rather than a regulatory model) does not affect the loss function L(d,ș) associated with taking the wrong decision, but rather the form of the probability p(ș), through the range of uncertainty w. (The consequence of improving models or making more measurements is to reduce w.) The influence of w can be broadly described as follows. When the uncertainty is small (w<<1) T – șc -w/2 is more likely to be positive when T – șc >0, L (d1 ) = 0 and d1 is preferred and uncertainty can be ignored. One can be sure when the limit value has been exceeded. When T -w/2< șc < T +w/2 uncertainty affects the decision. If the uncertainty is large (w>>1), the expected value of considering uncertainty L (d1 ) – L (d 0 ) will be (roughly) proportional to (a1 a0 ) w 2 > 0 and uncertainty cannot be ignored. Generally the loss from making the wrong decision L (d ) increases, when the range of uncertainty w increases, except if uncertainty can be ignored. The effect of
B. Fisher
382
using a simplified, regulatory model is to increase w. If uncertainty can be ignored in both the research and the regulatory model, then there is no advantage in using the research model for decision making. It is important to consider both the difference between the best estimate and the limit value, T – șc, and the range of uncertainty w in model inter-comparisons. Two recent regulatory examples are briefly mentioned here, both involving the extraction of a simplified model from a complex system. Bounding curves can be used to describe NO2 in industrial plumes. NO2 plume bounding curves have been compared with chemical transport models with full chemistry, and real and idealised meteorology (Yu et al., 2007). At night, the upper bounding curve is given by [ NO2 ] [ NOx ]
[O3 ] ([ NOx ] [O3 ] / 4)
(9)
The ozone background (O3) used for the bounding curve is selected on the basis of results from the CMAQ model. In the day, the lower bounding curve is given by [ NO2 ] [ NOx ]
[O3 ] ([ NOx ] [O3 ])
(10)
Scatter plots based on CMAQ and NAME III model runs and bounding curves (see Yu et al., 2007 for details) for two representative episodes are generally well below the 1-hour mean NO2 limit value of 200 ȝg m-3, and so a regulatory model based on bounding curves could be applied and uncertainty ignored. In the other example the excess downwind ozone concentration averaged over a 24-hour period following the time when an air parcel passes over an industrial VOC source, describing the VOC chemical reactivity, was considered. Ozone formation downwind of an industrial VOC source, based on the chemical reactivity and estimates from the CMAQ model, were compared in a sensitivity experiment (see Yu et al., 2007, for details) carried out for a June 2001 episode over the UK. In this experiment an imaginary VOC source was added to the base case emissions and located in the Thames Estuary area. The imaginary VOC source, containing only ethene, was assumed to have a constant emission strength of 10 t per hour. The differences in the daily maximum 1-hour ozone mixing ratios and the 24-hour averaged O3 mixing ratios from the base case for each day of the simulation period were calculated and compared to estimates based on the chemical reactivity (Derwent and Nelson, 2002). A maximum 24-hour average ozone excess of 30 ȝg m-3 is attained on the 24 June 2001 using CMAQ, which is substantially lower than the 175 ȝg m-3 estimate based on the chemical reactivity. The two estimates straddle the ozone target value of 120 ȝg m-3. Hence in this case further investigation is needed before deciding between the two methods. Caution is also needed because the two calculations are not strictly directly comparable, apart from the rather uncharacteristically high emission rate.
Uncertainty in Air Quality Decision Making
383
6. Conclusions Decisions depending on the exceedence of air quality limit values should take into account the loss, if a wrong decision is made, and the decision can in some circumstances depend on the uncertainty. In other conditions ignoring uncertainty would lead to the same decision as including uncertainty. Thus there are situations when considering uncertainty is important and others when it is not. It depends on how close the best prediction is to the limit value, and this can have implications when applying regulatory models. Acknowledgments The views expressed in this paper are those of the author and are not necessarily those of the Environment Agency.
References Abbott J, Stedman V, Vincent V (2007) Annual audits of the contribution to pollutant concentrations from processes regulated by the Environment Agency: method, development SC030172/SR2, application of method SC030172/SR3. http://publications.environment agency.gov.uk/pdf/SCHO0307BMDS-e-e.pdf http://publications.environment agency.gov.uk/pdf/SCHO0307BMDP-e-e.pdf Borrego C, Miranda AI, Costa AM, Monteiro A, Ferreira J, Martins H, Tchepel O, Carvalho AC (2006) Uncertainties of models and monitoring, Air4EU project report M.2 http://www.air4eu.nl/reports_products.html Borrego C, Monteiro A, Costa AM, Miranda AI, Builtjes P, Kerschbaumer A, Lutz M (2007) Estimation of modelling uncertainty for air quality assessment: the AIR4EU Berlin case. Proceedings of the 6th International Conference on Urban Air Quality, Limassol, Cyprus, 27–29 March 2007. Derwent RG, Nelson N (2002) Development of a reactivity index for the control of the emissions of organic compounds, UK Environment Agency R&D Technical Report P4-105 RC8309. Fisher B, Willows R (2007) Uncertainty in air pollution models used for regulatory and risk assessment purposes. International Technical Meeting on Air Pollution Modelling and its Applications, Leipzig. Developments in Environmental Science, Volume 6, C. Borrego and E. Renner (Editors) pp. 392–401, Elsevier. Morgan MG, Henrion M (1990) A Guide to Dealing with Uncertainty in Quantitative Risk and Policy Analysis, Cambridge University Press, Cambridge, ISBN 0-521-36542-2. Tarrason L et al. (2005) Transboundary Acidification, Eutrophication and Ground Level Ozone in Europe. EMEP Status Report, Norwegian Meteorological Institute, EMEP Report 1/2005. Yu Y, Sokhi RS, Middleton DR (2007) Estimating contributions of Agencyregulated sources to secondary pollutants using CMAQ and NAME III models. Environment Agency Science Report: SC030171.
384
B. Fisher
Discussion D. Steyn: Do you think policy makers are equipped to behave rationally when presented with a complex forecast and uncertainties estimates? B. Fisher: It is unreasonable to expect the policy maker, or lawyer, to be able to make rational judgements when presented with complex forecasts with uncertainty estimates. The modeller or expert should present a summary of the evidence favouring options, using quantitative decision analysis techniques as appropriate, with the full working available for review if required. A. Venkatram: The modeler can provide information on model uncertainty. However, there is little guidance on constructing a realistic loss function. Can you provide an example? B. Fisher: At the ITM we tend to concentrate on obtaining improved estimates of the probability distribution of the model forecast. Therefore consideration of the loss function has only been hinted at in the paper. The loss function involves wider issues relating to the evidence supporting, say, an air quality limit value. It could be a rectangular function of concentration penalising any exceedence of the limit value. This is one idealised case. Given the uncertainty in the limit value one would expect the loss function to be smooth around the limit value and for it to increase with increasing concentrations exceeding the limit value. The slope of the loss function would depend on the estimated economic or health damage caused by concentrations exceeding the limit value.
5.2 Diagnostic Analysis of the Three-Dimensional Sulfur Distributions over the Eastern United States Using the CMAQ Model and Measurements from the ICARTT Field Experiment Rohit Mathur, Shawn Roselle, George Pouliot and Golam Sarwar
Abstract Previous comparisons of air quality modeling results from various forecast models with aircraft measurements of sulfate aerosol collected during the ICARTT field experiment indicated that models that included detailed treatment of gas- and aqueous-phase atmospheric sulfate formation, tended to overestimate airborne SO42- levels. To understand the three-dimensional distributions and fate of atmospheric SO42- and to diagnose the possible reasons for these over-predictions, we perform detailed analysis of modeled SO42- budgets over the eastern U.S. during the summer of 2004 using an instrumented version of the Community Multiscale Air Quality (CMAQ), namely the sulfur-tracking model. Two sets of three-dimensional model calculations are performed using different gas-phase chemical mechanisms: (1) the widely used CBM4 mechanism, and (2) the SAPRC mechanism.
Keywords Aerosols, chemical mechanisms, CMAQ, ICARTT, sulfate
1. Introduction The regional and global distribution of atmospheric sulfur compounds is of interest because of their important impacts on the environment and the climate. A large fraction of tropospheric sulfur oxides originate from SO2 which is emitted into the atmosphere as a result of anthropogenic combustion activities. The formation of atmospheric H2SO4 through rapid oxidation of emitted SO2 via both gas- and aqueous-phase pathways in the atmosphere has been widely studied as deposition of these compounds has led to acidification of lakes and forests. H2SO4 in the atmosphere can nucleate or condense on existing particles to produce SO42- aerosol which constitutes a relatively large fraction of the total ambient fine particulate matter (or PM2.5; particles with diameter less than 2.5 ȝm). Sulfate aerosols can further affect the climate by backscattering solar radiation and by changing the albedo of clouds (Charlson et al., 1992). Consequently, the accurate C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
496
Diagnostic Analysis of the Three-Dimensional Sulfur Distributions
497
characterization of the three-dimension distributions of tropospheric SO42- is of interest. In the eastern U.S., SO42- constitutes a large fraction of the airborne fine particulate matter. While measurements of surface-level SO42- and PM2.5 concentrations are available from a variety of surface networks, similar aloft measurements are only available during infrequent intensive field studies. During July and August 2004, airborne measurements of a variety of trace species were made from extensively instrumented aircrafts deployed as part of the International Consortium for Atmospheric Research on Transport and Transformation (ICARTT) field study (Fehsenfeld et al., 2006; Singh et al., 2006) and provide a unique opportunity to examine the performance of existing atmospheric chemistry-transport models in representing the processes that shape the three-dimensional distribution of airborne pollutants. Comparison of air quality modeling results from various forecast models with aircraft measurements of sulfate aerosol collected during the ICARTT field experiment indicated that models with detailed treatment of gas- and aqueous-phase atmospheric sulfate formation, tended to over-predict airborne SO42- levels (McKeen et al., 2007; Yu et al., 2008). To understand the three-dimensional distributions and fate of atmospheric SO42- and to diagnose the possible reasons for these over-predictions, in this study we perform detailed analysis of SO42- budgets over the eastern U.S. during the summer of 2004 simulated using the Community Multiscale Air Quality (CMAQ) and through comparisons of these model results with surface and aloft measurements.
2. The Modeling System The Community Multiscale Air Quality (CMAQ) model (Byun and Schere, 2006) driven with meteorological fields from the Eta model (Black, 1994) is used to examine the three-dimensional atmospheric chemical conditions during July– August 2004. Details on the linkage between the Eta and CMAQ models can be found in Otte et al. (2005). To be consistent with previous analysis, the input emissions data were constructed in a manner similar to that used in forecast mode. The emission inventories used in the model calculations discussed here were constructed to represent the 2004 period. NOx emissions from point sources were projected to 2004 (relative to a 2001 base inventory) using estimates derived from the annual energy outlook by the Department of Energy (http: //www.eia.doe.gov/oiaf/ aeo/index.html). Mobile emissions were estimated using the least-squares regression approximations to the MOBILE6 model following the approach of Pouliot and Pierce (2003). Area source emissions were based on the 2001 National Emissions Inventory, version 3 (http://www.epa.gov/ttn/chief), while BEIS3.12 (Pierce et al., 2002) was used to estimate the biogenic emission. CMAQ simulations were performed for the July 11–August 18, 2004 period. The aerosol module used in CMAQ is described in Binkowski and Roselle (2003) with updates described in Bhave et al. (2004). The aerosol distribution is modeled
498
R. Mathur et al.
as a superposition of three lognormal modes that correspond nominally to the Aitken (diameter (Dp) < 0.1 ȝm), accumulation (0.1 < Dp < 2.5 ȝm), and coarse (Dp > 2.5 ȝm) modes. The model results for PM2.5 concentrations are obtained by summing species concentrations over the first two modes. The horizontal model domain was discretized using grid cell sizes of 12 km. Twenty-two layers of variable thickness set on a sigma-type coordinate were used to resolve the vertical extent from the surface to 100 hPa. Daily 24-hour duration model simulations were conducted using the meteorological output from the 12 UTC Eta cycle.
3. Results and Discussion The Eta-CMAQ system was deployed during the summer of 2004 to provide developmental fine particulate matter forecasts over the eastern United States (Mathur et al., 2005). That configuration of the modeling system used the CBM4 chemical mechanism (Gery et al., 1989). Comparisons of modeled surface-level daily average PM2.5 compositional characteristics with corresponding measurements from the Speciated Trends Network (STN), indicated a slight high bias in predicted surface-level SO42-. Comparisons of predicted SO42- levels aloft with measurements from the NOAA-WP3 and NASA DC-8 aircraft, however indicated though the model captured the general characteristics of SO42- vertical distribution, it exhibited a systematic and often significant high bias aloft (see Figure 1).
Fig. 1 Comparison of average modeled SO42- vertical profiles with corresponding measurements from the (a) NOAA-WP3 and (b) NASA DC-8 aircrafts. The average profiles are constructed using modeled-observed pairs over all flights during July 15–August, 2004
Since the in-cloud aqueous phase oxidation of S(IV) to S(VI) constitutes a major fraction of the SO42- production, and since the CBM4 mechanism is known to be biased high in its predictions of H2O2 (which is the primary termination pathway for HO2 radicals in the mechanism), it can be hypothesized that in-cloud SO2 oxidation by H2O2 in this model formulation contributes to the noted SO42- overprediction. To further examine the modeled SO42- budgets, an instrumented version
Diagnostic Analysis of the Three-Dimensional Sulfur Distributions
499
of CMAQ was used to analyze sulfate production pathways. This model version, referred to as the CMAQ sulfur-tracking model, tracks sulfate production from gas-phase and aqueous-phase chemical reactions, as well as contributions from emissions and initial and boundary conditions. Five aqueous-phase reactions are individually tracked, including S(IV) oxidation by hydrogen peroxide (H2O2), ozone (O3), methyl-hydrogen peroxide (MHP), peroxyacetic acid (PAA), and catalysis by iron (Fe) and manganese (Mn). Contributions from each pathway are tracked in separate modeled species and are advected, diffused, processed through clouds, and depo-sited (both wet and dry). Figure 2c presents a breakdown of the various modeled SO42- production pathways along the NOAA-WP3 flight track on August 6, 2004 and illustrates that along this particular flight path, with the CBM4 mechanism approximately half of the simulated SO42- was produced through the aqueous oxidation pathways, amongst which the in-cloud H2O2 oxidation was the dominant contributor. Comparisons of predicted H2O2 concentrations with measurements from the NASA DC-8 (not shown) further confirmed the high bias in predicted H2O2 concentrations suggesting that H2O2 biases inherent in the CBM4 mechanism could be magnifying the role of modeled in-cloud SO2 oxidation.
Fig. 2 (a) Flight path of the NOAA WP3 on August 6, 2004; (b) Comparison of modeled and observed SO42- along flight path; Relative contribution of various pathways to modeled SO42- with the (c) CBM4 mechanism and (d) SAPRC mechanism
Additional model calculations with the more detailed and contemporary SAPRC mechanism were performed to further examine and quantify the uncertainties in the relative contributions of gaseous and aqueous oxidation pathways to SO42- formation. Comparisons of SO42- predictions with the CBM4 and SAPRC mechanisms for the August 6, 2004 WP3 flight are shown in Figure 2b; the contributions of the various pathways to the simulated SO42- using the two mechanisms for this case are shown in Figure 2c and d. Interestingly, model configurations with either mechanisms result in similar predictions of SO42- concentrations (Figure 2b), though there
500
R. Mathur et al.
are significant differences in the relative contributions of the gaseous and aqueous pathways (Figure. 2c and d). As expected, aqueous SO42- production from the H2O2 oxidation pathway was considerably reduced (by about half) in the SAPRC mechanism configuration due to improved prediction of ambient H2O2 (compared to NASA DC-8 measurements; not shown). However, the reductions in SO42- production from the aqueous pathway were compensated by a corresponding increase in contribution from the gas-phase OH oxidation pathway (Figure 2d). Figure 3 presents a comparison of modeled H2O2 and OH concentrations predicted using the CBM4 and SAPRC mechanisms along the same flight path. As expected, the H2O2 concentrations modeled with the SAPRC mechanism are significantly lower than those predicted using the CBM4 mechanism. However, the SAPRC OH concentrations are significantly larger than those modeled using the CBM4 mechanism. This combined with the availability of additional SO2 (from reduced aqueous-phase conversion) results in the noted increase in SO42- production from the gas-phase pathway in the SAPRC model configuration.
Fig. 3 Comparison of (a) H2O2 and (b) OH predictions using the CBM4 and SAPRC mechanisms along the August 6, 2004 NOAA WP3 flight path
The model’s ability to simulate the regionally averaged vertical profiles of sulfur species sampled by the NOAA WP3 aircraft campaigns during the study period is illustrated in Figure 4, which presents comparisons of the average composite vertical profiles for SO42- and the SO2/total-sulfur ratio. In constructing these profiles we averaged both the observed and the modeled data within each vertical model layer and over all the flights. These vertical profiles may thus be regarded as representing the mean conditions that occurred over the northeastern U.S. during the study period. As illustrated in the comparisons both the CBM4 and the SAPRC chemical mechanism model configurations over-predict the observed SO42- aloft.
Diagnostic Analysis of the Three-Dimensional Sulfur Distributions
501
Both configurations also significantly underestimate the aloft SO2/total sulfur ratio, suggesting that in both model configurations the S(IV) to S(VI) conversion occurs more efficiently than that suggested by the aircraft measurements. Detailed analysis of the SO42- production pathways in the model however indicate that even though the CBM4 and SAPRC configurations of the model yield similar levels of SO42aloft, the relative importance of the gas and aqueous production pathways is significantly different between the two chemical mechanisms and highlights the uncertainties in these mechanisms especially for aloft conditions.
Fig. 4 Comparisons modeled and observed regionally-averaged vertical profiles of SO42- and the SO2/total Sulfur ratio
In recent years regional air quality models are being applied to study and address increasingly complex multi-pollutant air pollution issues over time scales ranging from episodic to annual cycles. Majority of the chemical mechanisms currently in use in such models have been validated against smog chamber measurements designed to represent surface-level photochemistry. Our results indicate that the extrapolation of these mechanisms to represent chemistry associated with multi-day transport and free-tropospheric conditions need to be scrutinized in more detail. Acknowledgments The research presented here was performed under the Memorandum of Understanding between the U.S. Environmental Protection Agency (EPA) and the U.S. Department of Commerce’s National Oceanic and Atmospheric Administration (NOAA) and under agreement number DW13921548. This work constitutes a contribution to the NOAA Air Quality Program. Although it has been reviewed by EPA and NOAA and approved for publication, it does not necessarily reflect their policies or views.
502
R. Mathur et al.
References Bhave PV, Roselle SJ, Binkowski FS, Nolte CG, Yu SC, Gipson GL, Schere KL (2004) CMAQ aerosol module development Recent enhancements and future plans, Proc. of the 2004 Models-3/CMAQ Conference, October 18–20, 2004, Chapel Hill, NC, available at:http://www.cmascenter.org/conference/2004/ abstracts/Model%20Development/bhave_abstract.pdf Binkowski FS, Roselle SJ (2003) Models-3 Community Multi-scale Air Quality (CMAQ) model aerosol component:1. Model description, J. Geophys. Res., 108(D6), 4183, doi:10.1029/2001JD001409. Black T (1994) The new NMC mesoscale Eta Model: description and forecast examples. Weather Forecast., 9, 265–278. Byun DW, Schere KL (2006) Review of governing equations, computational algorithms, and other components of the models-3 Community Multiscale Air Quality (CMAQ) modeling system, Appl. Mech. Rev., 59, 51–77. Charlson et al. (1992) Climate forcing by anthropogenic aerosols, Science, 255, 423–430. Fehsenfeld FC et al. (2006) International Consortium for Atmospheric Research on Transport and Transformation (ICARTT): North America to Europe – Overview of the 2004 summer field study, J. Geophys. Res., 111, D23S01, doi:10.1029/ 2006JD007829. Gery MW, Whitten GZ, Killus JP, Dodge MC (1989) A photochemical kinetics mechanism for urban and regional scale computer modeling. J. Geophys. Res., 94, 12,925–12,956. Mathur R, Kang D, Yu S, Schere KL, Pleim J, Young J, Pouliot G, Otte T (2005) Particulate matter forecasts with the Eta-CMAQ modeling system: towards development of a real-time system and assessment of model performance, Proc. of the 2005 Models-3/CMAQ Conference, September 26–28, 2005, Chapel Hill, NC, Available at: http://www.cmascenter.org/conference/2005/ppt/1_11.pdf. McKeen et al. (2007) Evaluation of several PM2.5 forecast models using data collected during the ICARTT/NEAQS 2004 field study, J. Geophys. Res., 112, D10S20, doi:10.1029/2006JD007608. Otte TL et al. (2005) Linking the Eta Model with the Community Multiscale Air Quality (CMAQ) Modeling System to Build a National Air Quality Forecasting System, Weather Forecast., 20, 367–384. Pierce T, Geron C, Pouliot G, Kinnee E, Vukovich J (2002), Integration of the Biogenic Emission Inventory System (BEIS3) into the Community Multiscale Air Quality Modeling System, Preprints, 12th Joint Conf. on the Apps. of Air Pollut. Meteor. with the A&WMA, Amer. Meteor. Soc., Norfolk, VA, 20–24 May 2002, J85–J86. Pouliot G, Pierce T (2003) Emissions processing for an air quality forecasting model, 12th Intl. Conf. on Emission Inventories, San Diego, CA, April 28–May 1, 2003.
Diagnostic Analysis of the Three-Dimensional Sulfur Distributions
503
Singh HB, Brune WH, Crawford JH, Jacob DJ, Russell PB (2006) Overview of the summer 2004 Intercontinental Chemical Transport Experiment – North America (INTEX-A), J. Geophys. Res., 111, D24S01, doi:10.1029/2006JD007905. Yu S, Mathur R, Schere K, Kang D, Pleim J, Young J, Tong D, Pouliot G, McKeen SA, Rao ST (2008) Evaluation of real-time PM2.5 forecasts and process analysis for PM2.5 formation over the eastern U.S. using the Eta-CMAQ forecast model during the 2004 ICARTT study, J. Geophys. Res., 113, D06204, doi:10.1029/2007/ JD009226.
Discussion D. Steyn: You mention hourly versus weekly averaging time as a possible source of error. Is it possible that differences are due to move than the statistics of hourly versus longer averaging but rather reflect different chemical processes? R. Mathur: In the comparisons presented, the surface measurements represent a 24-hour average where as the aircraft measurements were instantaneous. Certainly, averaging of data over longer time periods removes the inherent variability and can result in better agreement between the model and the measured values. You are correct that the averaging process implicitly represents different processes, with larger averaging times representing cumulative effects of several processes. For instance, in this case, surface-level 24-hour average SO42- concentrations at a location represent the cumulative affects both gas- and aqueous phase sulphate formation and well as transport and turbulent mixing processes. Ideally, to have greatest confidence in model results it would be desirable that the model captures temporal variability at all resolvable scales. B. Fuher: Are you able to put this very interesting model validation in to context? For example have there been other studies, either routine monitoring or field experiments, which demonstrate similar over prediction of sulphate? R. Mathur: Other than the model inter-comparison of McKeen et al. (J. Geophys. Res., 112, D10S20, doi:10.1029/2006JD007608, 2007) and our previous analysis with ICARTT data (Yu et al., J. Geophys. Res., in press), we are not aware of any previous studies that have identified similar systematic over-predictions in sulphate or have attempted to diagnose the reasons for the noted over-estimation. It should be noted that the summer of 2004 (period of the ICARTT study) was characterized by unusually wet conditions with widespread cloudiness. Consequently, the noted overestimation of sulphate by the models could not only be related to representation of
504
R. Mathur et al.
chemical processing of S(IV) to S(VI) but also to representation of the clouds and their impacts on sulphate production. Our analyses in this study has focussed on diagnosing aspects related to chemical processing; additional work underway is investigating the possible effects related to representation of clouds and associated missing processes on ambient sulphate levels.
5.5 Formation of Secondary Inorganic Aerosols by High Ammonia Emissions Simulated by LM/MUSCAT Eberhard Renner and Ralf Wolke
Abstract Ammonia (NH3) is the most abundant gaseous base and responsible for neutralizing a large fraction of acidic gases promoting the formation of atmospheric particles. Therefore, the contribution of ammonia to the formation of secondary particles (PM2.5 and PM10) in a regional scale is examined. The aerosols result from SO2 and NOx via sulfuric and nitric acid formation in the gas- and the liquid-phase and following subsequent reactions with ammonia. A period in May and a period in August/September 2006 were simulated by a nested application of the model system LM-MUSCAT.
Keywords Ammonia, ammonium sulfate, ammonium nitrate, chemical transport
1. Introduction Atmospheric particulate matter (PM) in ambient air has been associated with human health effects (Dockery et al., 1993; Pope et al., 1995; Brunekreef, 1997; Hoek et al., 2002). Since particles with aerodynamic diameters smaller than 10 Pm (PM10) are able to pass the larynx, the European air quality standards prescribe a limit for PM10. Currently, the focus of environmental sciences and politics includes besides primary particulate emissions also the formation and growth processes of secondary particles (e.g., Andreani-Aksoyoglu et al., 2004). In order to observe the fate of high ammonia emissions especially from agriculture and livestock husbandry in terms of particulate matter, the formation of secondary inorganic particles (PM2.5 and PM10) in a regional scale is investigated. The dominant contribution to the particle mass is given by heterogeneous condensation of gaseous compounds on pre-existing aerosol particles. Especially the generation of sulfuric and nitric acids by the precursor species SO2 and NOx and their reactions with ammonia leading to ammonium sulfate and ammonium nitrate are relevant. The aerosol-forming, photolytic and gas-chemical reactions together with the transport and diffusion processes are integral part of the chemistry-transport model MUSCAT, which was on-line coupled to the Local Model (LM) of German Weather Service. A regime for long-term simulations was tested. The horizontal C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
522
Formation of Secondary Inorganic Aerosols by High Ammonia Emissions Simulated
523
resolution of the outer region (Europe) was chosen to be 15 km and of the inner nested region (Germany and bordering regions of central Europe) 7.5 km. The present study estimates, to which extent the gaseous and particulate emissions of ammonia, in a region of high ammonia emissions, especially from agriculture and livestock husbandry, contribute to the formation of secondary particulate matter. For this purpose, long-term real-weather simulations were performed for a spring and a late summer period.
2. Model Description 2.1. Meteorological-chemical model system LM-MUSCAT The three-dimensional non-hydrostatic model LM (Local Model) is the operational meteorological model of the German Weather Service (Doms and Schättler, 1999). The governing equations are formulated in terrain-following coordinates. Radiative processes as well as cloud microphysics and turbulence are treated in a parameterized manner. The model technique is capable of self-nesting and fourdimensional data assimilation, where the model GME (Global Model of the Earth) by German Weather Service serves as the global driver model. In that way, the horizontal grid spacing is reduced from 60 km (GME) down to 7.5 km for the innermost region. In both models 50 layers are used for the vertical discretization. The three-dimensional chemistry transport model MUSCAT (Multi-Scale Atmospheric Transport Model) performs the transformation of the gaseous and particulate species together with the transport processes advection, turbulent diffusion, dry/wet deposition, and sedimentation (Knoth and Wolke, 1998; Wolke and Knoth, 2000). Due to the online coupling with LM, these calculations are involved in the meteorological model, thus exploiting the current atmospheric conditions. Regardless, the implicit-explicit procedures of the time integration scheme of MUSCAT are widely independent of the meteorological model and allow for autonomous time steps and different horizontal grid resolutions in selected areas. For this purpose, the three-dimensional LM fields of wind, temperature, humidity, density, pressure, exchange coefficients, etc., are interpolated temporally and spatially without flux divergences. The chemical part of MUSCAT contains the gas phase mechanism RACM (Stockwell et al., 1997) considering 76 reactive gas species, 217 chemical and 22 photolytic reactions. Particle formation and appropriate interactions with the gas phase are included. The radiation activity and the biogenic emission are also calculated in MUSCAT, where the informations on cloud cover, temperature and other meteorological parameters are taken from LM. Anthropogenic emissions are accounted for by point, line, and area sources in different layers above the surface. The annual emission intensities are disaggregated in time and space.
E. Renner, R. Wolke
524
2.2. Aerosol model The aerosol model, used for these simulations, is mass-based and quiet similar to that of the EMEP Eulerian model (Simpson et al., 2003). The formation of ammonium sulfate and ammonium nitrate is calculated by an equilibrium approach. The dominant contribution to mass accretion is treated as heterogeneous condensation of gaseous compounds on pre-existing aerosols. Here, the particulate matter is formed by the reactions between ammonia and sulfuric or nitric acid, which are generated from the gas phase precursor species SO2 and NOx. In the following, the relevant sources and sinks of particulate matter are specified in detail. Finally, the different aerosol species are subsumed to derive PM10. 1. Primary particulate matter (PPM): PPM is that part of PM, which is directly emitted by the various point and area sources (industry, traffic, etc.). PPM10 is the sum of PPM2.5 (particles with aerodynamic diameters smaller than 2.5 ȝm) and PPM10–2.5 (particles with aerodynamic diameters between 2.5 and 10 ȝm). 2. Formation of ammonium sulfate: The precursor gas SO2 can be oxidized in the gas phase by OH radicals to form sulfuric acid: SO2 + OH + O2 + H2O ĺ ... ĺ H2SO4 + HO2
(1)
In addition to the gas phase reaction an oxidation pathway in clouds with a simple first order reaction constant Rk is considered. Rk is calculated as a function of relative humidity (%) and cloud cover (H): Rk = 8,3e-5(1 + 2H) (min–1) for RH < 90% Rk = 8,3e-5(1 + 2H)>1,0 + 0,1(RH–90.0@ (min–1) for RH t 90%. This parameterisation is described in Schaap et al. (2004). It enhances the oxidation rate under cool and humid conditions. If ammonia is available, the sulfuric acid reacts fast and irreversible with ammonia resulting in ammonium sulfate on the particle surface: H2SO4 + 2 NH3 ĺ {(NH4)2SO4}a } ĺ {(NH4)1.5SO4}a H2SO4 + NH3 ĺ {NH4HSO4}a
(2)
Because of the different stoichiometric weights in the products, a mean stoichiometric value of 1.5 is used for ammonium sulfate (Ackermann, 1997).
Formation of Secondary Inorganic Aerosols by High Ammonia Emissions Simulated
525
In the investigated case 5% of SO2-emisisions are emitted primarily as ‘free’ sulfate ions (Simpson et al., 2003). This is a further effective pathway of the formation of ammonium sulfate by the direct reaction of this sulfate with ammonia in the condensed phase. The reaction is similar to Eq. (2). Any “excess” of sulfuric acid or sulfate ions, which could not be neutralized by ammonia, is assumed to exist in the condensed phase and therefore to participate in the formation of particle mass, too. 3. Formation of ammonium nitrate: In general, the atmospheric ammonia is neutralized first of all by the sulfuric acid (see above). The remaining ammonia can then combine with nitric acid to form ammonium nitrate. The nitric acid is mainly yielded by the following reaction chains in the gas phase and partly in the liquid phase: Day:
NO2 + OH ĺ HNO3
Night: NO2 ĺ .. ( NO3 , N2O5 ) ... ĺ HNO3
(3) (4)
In contrast to the fast and irreversible formation of ammonium sulfate, ammonium nitrate is a semi-volatile compound which forms on the aerosol surface in equilibrium with its gaseous precursors nitric acid and ammonia (Mozurkewich, 1993; Nenes et al., 1998; Zhang et al., 2000): HNO3 + NH3 ļ {NH4NO3}a
(5)
This equilibrium is dependent on the ambient atmospheric temperature and humidity. It is shifting to the gas phase under dry and warm conditions. 4. Aerosol sinks: The dry deposition velocity is determined by means of a resistance approach accounting for the atmospheric turbulence state, the kinetic viscosity, and the gravitational settling in dependence on the particle size. An essential role in removing particulate matter from the atmosphere is played by wet deposition, which is distinguished between in-cloud and sub-cloud aerosol scavenging. These processes are parameterized in dependence on the scavenging and collection efficiency (Tsyro and Erdman, 2000). 5. Total PM10: The total particulate mass PM10 is composed of the various primary and secondary constituents described in the foregoing points: PM10 = PPM2.5 + PPM10–2.5 + {(NH4)1.5SO4}a + {SO42-}a + {NH4NO3}a
(6)
526
E. Renner, R. Wolke
3. Emissions The inventories of the anthropogenic emissions are based on the EMEP/CORINAIR data (see CORINAIR), improved by TNO (Builtjes et al., 2003), using the SNAP-codes (Selected Nomenclature of Air Pollution) for characterizing the different source types (e.g., energy transformation, industrial combustion, road transport, agriculture). The considered list of chemical species includes the main pollutants CO, NOx, SO2, NH3, PPM2.5, PPM10–2.5, methane, and non-methane volatile organic compounds (NMVOC). The temporal variation of the emissions is represented by time functions, which break down the annual totals into hourly values in dependence on the source category, and the month and day of emissions. The differenttiation of the NMVOC follows the VOC-split by Winiwarter and Züger (1996). The biogenic VOC and NO emissions are calculated within the chemistrytransport model MUSCAT for each time step and grid cell in dependence on the land-use type, temperature and radiation. Consequently, the meteorological situation has a direct influence on the emission amount. There are numerous volatile organic compounds, which are emitted from plants. In this study, only the emissions of isoprene and terpenes by forests are taken into account. According to the parameterization of Guenther et al. (1993), the emission rates are formulated as products of plant specific standard emission rates with a temperature and/or a light dependent function. Agriculturally used areas may be significant sources of NO caused by microbiological processes. The biogenic NO emissions are parameterized using empirical relationships, which depend on the land use type, season, and soil temperature. The functional dependence follows Williams et al. (1992) and Stohl et al. (1996).
4. Results In order to identify the influence of different meteorological situations on the formation of secondary particulate matter the simulated concentrations for two days in May are presented. The discussion is focussed exemplarily on the 7th of May in 2006, a sunny day with moderate winds from East and the 26th of May, a day with westerly winds. The figures show the concentration distributions of the relevant gaseous and particulate species only in the N1-region in the lowest modelling layer. In Figures 1 and 2 the concentrations of the gaseous precursors SO2, NO2 and NH3, the secondarily formed ammonium sulfate and ammonium nitrate and PM10 are displayed. Additionally the temperature and the relative humidity are presented. At the 7th of May the PM10 concentrations in Germany are about 10 ȝg/m3. The ammonium sulfate concentrations are about 5 ȝg/m3 whereas no ammonium nitrate was formatted at this time in this region. The reason for this is the circumstance that we have at this situation extreme dry and hot air in this region. So, the equilibrium between the gaseous precursors nitric acid and ammonia and ammonium nitrate as a semi-volatile compound, which is formed on the aerosol surface, is shifted to the gas phase. At the 26th of May we found a different situation. At the westerly winds we have in the northern part of Germany air masses, which are cool and wet. As a
Formation of Secondary Inorganic Aerosols by High Ammonia Emissions Simulated
527
result we see comparable ammonium sulfate concentrations of about 5 ȝg/m3. Contrary to the former situation we have additionally ammonium nitrate concentrations of about 10 ȝg/m3 at this time in this region. The issue of the simulations was finally to get the additional PM10 burden caused by the NH3 emissions in this region and the later formation of secondary ammonium sulfate and ammonium nitrate. One can summery that at situations with low PM10 concentrations 50–70% of the PM10 concentrations within the determined region can be secondary formed. By situations with higher and very high PM10 burden this fraction will decline more and more. Further studies have shown (not shown here), that the availability of SO2 is the limited factor of the formation of secondary aerosol mass.
Fig. 1 Concentrations, relative humidity and temperature at 7th of May
Fig. 2 Concentrations, relative humidity temperature at 26th of May
528
E. Renner, R. Wolke
References Ackermann I (1997) MADE: Entwicklung und Anwendung eines AerosolDynamikmodells für dreidimensionale Chemie-Transport-Simulationen in der Troposphäre. Mitteilungen aus dem Institut für Geophysik und Meteorologie der Universität Köln, Heft 115, 1–153. Andreani-Aksoyoglu S, Prévôt ASH, Baltensperger U, Keller J, Dommen J (2004) Modeling of formation and distribution of secondary aerosols in the Milan area (Italy). Journal of Geophysical Research 109 (D5), D05306, doi:10.1029/ 2003JD004231. Builtjes PJH, van Loon M, Schaap M, Teeuwisse S, Visschediijk AJH, Bloos JP, (2003) TNO-report R 2003/166. Brunekreef B (1997) Air pollution and life expectancy: is there a relation? Journal of Occupational and Environmental Medicine 54, 781–784. CORINAIR (Co-ordinated Information on the Environment in the European Community – AIR). Web address: http://reports.eea.eu.int/EMEPCORINAIR3/en Dockery DW, Pope CA, Xu X, Spengler JD, Ware JH, Fay ME, Ferris BG, Speizer FE (1993) An association between air pollution and mortality in six U.S. cities. New England Journal of Medicine 329, 1753–1759. Doms G, Schättler U (1999) The Nonhydrostatic Limited-Area Model LM (LokalModell) of DWD: Part I: Scientific Documentation (Version LM-F90 1.35). German Weather Service, Offenbach. Guenther AB, Zimmerman PR, Harley PC, Monson RK, Fall R (1993) Isoprene and monoterpene emission rate variability: model evaluations and sensitivity analyses. Journal of Geophysical Research 98 (D7), 12609–12617. Hoek B, Brunekreef G, Goldbohm S, Fischer P, van den Brandt PA (2002) Association between mortality and indicators of traffic-related air pollution in the Netherlands: a cohort study, The Lancet 360, 1203–1209. Knoth O, Wolke R (1998) An explicit-implicit numerical approach for atmospheric chemistry-transport modelling. Atmospheric Environment 32, 1785–1797. Mozurkewich M (1993) The dissociation constant of ammonium nitrate and its dependence on temperature, relative humidity and particle size. Atmospheric Environment 27 A, 261–270. Nenes A, Pilinis C, Pandis SN (1998) Isorropia: a new thermodynamic model for multiphase multicomponent inorganic aerosols. Aquatic Geochemistry 4, 123– 152. Pope CA, Thun MJ, Namboodiri MM, Dockery DW, Evans JS, Speizer FE, Heath CW (1995) Particulate air pollution as a predictor of mortality in a prospective study of U.S. adults. American Journal of Respiratory and Critical Care Medicine 151, 669–674. Schaap M, van Loon M, ten Brink HM, Dentener FJ, Builtjes PJH (2004) Secondary inorganic aerosol simulations for Europe with special attention to nitrate. Atmospheric Chemistry and Physics, 4, 857–874. Simpson D, Fagerli H, Jonson JE, Tsyro S, Wind P (2003) Transboundary Acidification, Eutrophication and Ground Level Ozone in Europe. PART I,
Formation of Secondary Inorganic Aerosols by High Ammonia Emissions Simulated
529
Unified EMEP Model Description. EMEP/MSC-W: EMEP Status Report 2003, ISSN 0806-4520. Stockwell WR, Kirchner F, Kuhn M, Seefeld S (1997) A new mechanism for regional atmospheric chemistry modeling. Journal of Geophysical Research 102 (D22), 25847–25879. Stohl A, Williams E, Wotawa G, Kromp-Kolb H (1996) A European inventory of soil nitric oxide emissions and the effect of these emissions on the photochemical formation of ozone. Atmospheric Environment 30, 3741–3755. Tsyro S, Erdman L (2000) Parameterization of aerosol deposition processes in EMEP MSC-E and MSC-W transport models. EMEP/MSC-E & MCS-W Note 7/00, Norwegian Meteorological Institute, Oslo. Williams EJ, Guenther A, Fehsenfeld FC (1992) An inventory of nitric oxide emissions from soils in the United States. Journal of Geophysical Research 97 (D7), 7511–7519. Winiwarter W, Züger J (1996) Pannonisches Ozon Projekt, Teilbericht Emissionen. Endbericht, Seibersdorf Report OEFZS-A-3817. Wolke R, Knoth O (2000) Implicit-explicit Runge-Kutta methods applied to atmospheric chemistry-transport modelling. Environmental Modelling and Software 15, 711–719. Zhang Y, Seigneur C, Seinfeld JH, Jacobsen M, Clegg SL, Binkowski FS (2000) A comparative review of inorganic aerosol thermodynamic equilibrium modules: differences, and their likely causes. Atmospheric Environment 34, 117–137.
5.3 Heterogeneous Chemical Processes and Their Role on Particulate Matter Formation in the Mediterranean Region Marina Astitha, George Kallos, Petros Katsafados and Elias Mavromatidis
Abstract The impact of particulate matter on air quality and the environment is an important subject for areas like the Greater Mediterranean Region, mostly due to the coexistence of major anthropogenic and natural sources. Such coexistence can create air quality conditions that exceed the imposed air quality limit values. Particulate matter formation and the factors enhancing or reducing such formation in the Mediterranean Region will be the primary focus of the work presented herein. Natural particulate matter appears mainly in the form of desert dust, sea salt and pollen among others and anthropogenic particulate matter appears as particulate sulfate and nitrate. The processes affecting the formation of new types of aerosols are based on the heterogeneous uptake of gases onto dust particles. New model development will be presented referring to the implementation of sea salt production and heterogeneous chemical processes leading to new aerosol formation in the photochemical model CAMx. Results from these simulations showed reasonable agreement with the available measurements. These results also revealed interesting effects of the coexistence of natural and anthropogenic particulate matter concerning the direct and indirect impacts on air quality and the environment.
Keywords Heterogeneous processes, modelling, Mediterranean, particulate matter
1. Introduction The direct and indirect impacts of aerosols on the environment and climate are the subject of scientific investigation during the last years and have gained the interest of the public since the IPCC 3rd Assessment Report (Penner et al., 2001). Focusing on the Mediterranean Region, the coexistence of natural and anthropogenic type of sources is a factor enhancing the air quality degradation without the capability to ensure the exact causes of such degradation. In this work, natural particulate matter appears in the form of desert dust and sea salt and anthropogenic particulate matter appears as particulate sulfate and nitrate. This categorization is due to the fact that Sahara is the desert responsible for many severe dust outbreaks that influence the C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
505
506
M. Astitha et al.
area and the Mediterranean Sea is considered an important source for sea salt and sulfur. The first step towards the identification of the feedback effects of excessive dust load on air quality was done by studying the direct shading effect of dust particles on photochemical processes. This was accomplished by implementing dust optical depth in the calculation of photolysis rates (Astitha et al., 2006) during severe dust transport episodes. The shading effect of dust particles caused a decrease in ozone and sulfate concentration at the surface, considered of minor importance (ozone and particulate sulfate decrease was maximum 2–3% of the initial concentration). The second step is the investigation of the processes affecting the formation of new types of aerosols based on the heterogeneous uptake of gases onto dust particles (Levin et al., 2005; Alastuey et al., 2005). These processes require extensive investigation since they are complex and uncertain and model development is needed to properly include the relevant physicochemical processes. For this purpose, advanced atmospheric and photochemical models are implemented with the aid of air pollutant measurements from stations in the region. The models used are the SKIRON/Eta atmospheric modeling system with the implementation of the dust module and the CAMx photochemical model. New model development will be presented referring to the implementation of sea salt production and heterogeneous chemical processes leading to new aerosol formation in the photochemical model.
2. Model Development A short description of the modeling systems used for performing simulations is provided in this section: The SKIRON/ETA is a modeling system developed at the University of Athens from the Atmospheric Modeling and Weather Forecasting Group (Kallos, 1997; Kallos et al., 2006; Nickovic et al., 2001). It has enhanced capabilities with the unique one to simulate the dust cycle (uptake, transport, deposition). The CAMx model (Environ, 2006) is an Eulerian photochemical model that allows for integrated assessment of air-pollution over many scales ranging from urban to super-regional (http://www.camx.com).
2.1. Particulate matter of natural origin Natural species like sea salt and desert dust are implemented inside the photochemical model, allowing the interaction between species of different origin (natural, anthropogenic). Desert dust surface fluxes were derived from SKIRON/Eta model in four size sections (centered diameters of 1.5, 12, 36, 76 ȝm). Only the first two sizes were implemented in the emissions processor, producing emission fields of crustal material to be imported in the photochemical model. Sea salt production was directly developed inside the air quality model. The production of sea salt was based on the work of Gong (2003), ǽhang et al. (2005)
Heterogeneous Chemical Processes and their Role on Particulate Matter Formation
507
for the open ocean function and the work of de Leeuw et al. (2000) for the surf zone function. The coding implementation was similar to that of CMAQ model (Shankar et al., 2005 and personal communication). Sea salt was speciated into sodium, chloride and sulfate aerosol using the factors fNA = 0.3856, fCl = 0.5389, fSO4 = 0.0755 respectively.
2.2. Heterogeneous uptake of gases on dust particles Air pollutants in the atmosphere can be categorized as natural or anthropogenic depending on their origin. Depending on their formation mechanism, they can also be categorized as of 1st generation (primary) and 2nd generation (secondary). In this work we apply a new terminology considering the 3rd generation species as the ones formed by the synthesis of the 1st and the 2nd generation. The schematic diagram in Figure 1 explains the synthesis described above. The synthesis of the 3rd generation pollutants is driven by the heterogeneous reactions occurring in the surface of a wet particle (i.e. desert dust for this work). A first-order kinetic relationship is used to represent the change in gases and aerosols concentration due to uptake on dust particles as shown in Eq. (1):
dC
g
k g ,r C
dt
g
(1)
where Cg is the gas concentration and kg,r (s-1) is the removal rate of species g to a particle surface with radius r. Following the formulation by Fuchs and Sutugin (1970) and Saylor (1997) the removal rate is calculated as shown in Eq. (2): k g ,r
4 S D g rN r 1 f (K n, J )K n
1.333 0.71K n1 4(1 J ) 1 K n1 3J
f (Kn , J )
(2)
where Kn is the Knudsen number, r the radius of the particle, Dg the gas-phase molecular diffusion coefficient in air, Nr is the particle number density and Ȗ is the uptake coefficient (different for each gas). The reactions chosen for this work are the following: dust
1.
SO
2.
NO 2 ( gas ) o DNO 3 ( aerosol )
3.
HNO3 ( gas ) o DNO3 b( aerosol )
2
( gas ) o DSO
4
( aerosol
)
Ȗ = 1.0*10-4
dust
Ȗ = 1.0*10-4
dust
Ȗ = 0.01
dust
4.
O3 ( gas ) o O2
Ȗ = 5.0*10-5
M. Astitha et al.
508
ANTHROPOGENIC EMISSIONS
Fig. 1 Schematic diagram of the synthesis between primary and secondary pollutants leading to new particle formation (3rd generation pollutants)
GASES
PARTICULATE MATTER
SO2, NO, NO2, VOC CO, NH3, CO2
PM2.5 PM10
2nd GENERATION POLLUTANTS SECONDARY POLLUTANTS O3, SO4, NO3, NH4
NATURAL EMISSIONS
AEROSOLS 1st GENERATION POLLUTANTS
3rd GENERATION POLLUTANTS
DUST SEA SALT (NaCl)
PARTICLES COVERED WITH GASES OR AEROSOLS
The reactions are considered irreversible concerning the uptake of each gas on aerosol surface and the dust particles are assumed to be of spherical shape.
3. Results and Discussion The new processes of sea salt production and heterogeneous uptake on dust particles are implemented in CAMx air quality model, providing the ability to study the 3rd generation aerosol formation. Two periods were chosen for simulation: August 1–20, 2001 and April 1–20, 2003. The main reason for choosing the month of April is that severe desert dust episodes occur in the Mediterranean Region during the transient seasons and photochemical processes are considerable, while August is a month where photochemical processes are at their peak along with the transport of desert dust in the region. The two long-term simulations revealed the ability of the model to describe the aforementioned processes as well as the shortcomings, uncertainties and future work followed by the new development. Meteorological fields and desert dust fluxes were derived from SKIRON/Eta modeling system for the periods mentioned above. Particulate matter is treated using the multi-sectional approach in CAMx model. Three size sections are selected, each having diameters in ranges: 0.03–0.1, 0.1–2.5, 2.5–10.0 ȝm. Results will be presented for the generation and the transport of the new aerosols formed on dust particles as described by chemical reaction 1 in the previous section. CRST_x is the dust particle, PSO4_x is the particulate sulfate of anthropogenic origin with the contribution from sea salt and DSO4_x is the particulate sulfate formed on dust particle via heterogeneous reaction 1, where x denotes the size section (1–3). Figure 2 presents the evaluation of the sea salt production, performed by comparing time series (3h average) of bulk sodium aerosol measurements from a coastal station (Finokalia station, ECPL chemistry, professor N. Mihalopoulos) and model output for the period 1–11 August 2001. The scatter plot diagram shown in Figure 3a refers to the same measuring station at Finokalia, for total sea salt concentration and the sum of the modeled sodium and chloride concentration. Figure 3b shows the comparison of model output with daily measurements from several stations in
Heterogeneous Chemical Processes and their Role on Particulate Matter Formation
509
Spain (X. Querol, 2007, personal communication) for coarse sodium aerosol. The model seems to slightly underpredict the sodium aerosol, and that can be attributed to the horizontal resolution (dx = 0.235q, dy = 0.18q) which is not adequate to resolve correctly the coastal zone. Another reason can be the fact that, the wind direction in the coastal zone is not taken into consideration in the production of sea salt, so the emissions and consequently the concentrations are rather smoothed as shown in Figure 2. SODIUM AEROSOL (ȝg/m 3) 8,000
ȝg/m 3
7,000 6,000 5,000 4,000 3,000 2,000 1,000
11/8/2001 6:00-9:00
11/8/2001 15:00-18:00
10/8/2001 21:00-0:00
10/8/2001 3:00-6:00
9/8/2001 18:00-21:00
10/8/2001 12:00-15:00
9/8/2001 0:00-3:00
9/8/2001 9:00-12:00
8/8/2001 6:00-9:00
7/8/2001 21:00-0:00
8/8/2001 15:20-18:00
7/8/2001 3:00-6:00
7/8/2001 12:25-15:00
6/8/2001 0:00-3:00
6/8/2001 9:00-12:00
6/8/2001 18:00-21:00
5/8/2001 6:00-9:00
5/8/2001 15:00-18:00
4/8/2001 3:00-6:00
4/8/2001 21:00-0:00
3/8/2001 18:00-21:00
4/8/2001 12:00-15:30
3/8/2001 3:00-6:00
3/8/2001 10:30-12:00
2/8/2001 9:00-12:00
2/8/2001 18:00-21:00
2/8/2001 00:00 - 3:00
1/8/2001 9:00 - 10:30
1/8/2001 15:00 - 18:00
1/8/2001 00:00 - 3:00
0,000
date (UTC) CAMx-NAtotal
MEASUREMENTS-bulk Na
Fig. 2 Comparison of sodium aerosol bulk measurements at Finokalia station (Crete, Greece) with CAMx model output for 1–11 August 2001 (3h average concentrations) a
b
MEASURED vs MODELED VALUES OF SODIUM CHLORIDE AEROSOL (ȝg/m3) AT FINOKALIA 2-11 AUGUST 2001
Sodium Aerosol(PM10) Obs vs CAMx for sites in SPAIN 1-13 August 2001 (Daily average)
3,00
7,00 MODEL (NACL=NA+CL)
y = 0,623x - 0,1676 R2 = 0,6034 r=0,77
2,50
3
8,00
MODEL OUTPUTS (ȝg/m )
9,00
6,00 5,00 4,00 3,00 2,00
y = 0,6185x + 0,5674
1,00
R = 0,7148 r=0,845
2,00 1,50 1,00 0,50
2
0,00 0,00
2,00
4,00
6,00
Measurem ents (NACL=1,8NA)
8,00
10,00
0,00 0,00
0,50
1,00
1,50
2,00
2,50
3,00
3
OBSERVATIONS (ȝg/m )
Fig. 3 (a) Comparison of sodium chloride measurements at Finokalia station (3h average, Crete, Greece) with CAMx model output, (b) Comparison of sodium aerosol (daily average PM10) from several stations in SPAIN for August 2001
During the second simulation for April 2003, the model performance for sea salt production was equally reliable. Figure 4a presents the comparison of modeled versus observed sodium and chloride aerosols (PM10) for the period 1–18 April 2003. The correlation coefficient reached high values coinciding with the August simulation. In Figure 4b the comparison of sulfate aerosol measurements from Finokalia during April 2003 with model output is quite good as well. The measurements were two to four days in duration and the modeled values were averaged for the same time period. In this figure modeled sulfate concentration is taken to be the sum of PSO4 and DSO4.
M. Astitha et al.
510 Modeled vs Observed Aerosol Concentration (PM10, ȝg/m3 ) Finokalia station, Crete 1-18 April 2003
Modeled vs Observed Aerosol Sulfate 3
Concentration (PM10, ȝg/m ) Finokalia station, Crete 1-18 April 2003
5,0 4,5
y = 0,8775x - 0,6155 R2 = 0,8076 r=0,898
4,0
6
y = 0,5798x + 0,0348 R2 = 0,7731 r=0,879
y = 0,8072x + 0,0618 R2 = 0,6099 r=0,78
5
3,0
Model Output
Model Output
3,5
2,5 2,0 1,5 1,0
4 3 2 1
0,5
0
0,0 0,0
0,5
1,0
1,5
2,0
2,5
3,0
3,5
4,0
4,5
0
5,0
1
2
CL-Aerosol
Na-Aerosol
Linear (Na-Aerosol)
3
4
5
6
Observations
Observations Linear (CL-Aerosol)
SO4-Aerosol
Linear (SO4-Aerosol)
Fig. 4 Comparison of modeled versus observed concentrations of (a) chloride and sodium, and (b) sulfate aerosol at Finokalia station for 1–18 April 2003
In general, the simulations showed rather small amounts of particulate sulfate produced from heterogeneous reactions on dust particles (DSO4) in cases where small amounts if dust and SO2 were available. Such result is evident in the time series of Figure 5a, where the modeled values for PSO4 and DSO4 are plotted for Erdemli, Turkey for 1–9 April 2003. This was not the case when there was significant amount of SO2 and available amounts of dust particles. In Figure 5b, DSO4 concentration (grey line) can be greater than PSO4 (dotted line), changing completely the general picture of total sulfate concentration. It should be noted that the discussed production of 3rd generation SO4 varies also with altitude, due to the transport of dust particles in higher vertical layers. The column mass load for the fifth day of the simulation for April 2003 is presented in Figure 6. The column mass load is plotted for each aerosol (g/m2) in the middle (0– 4 km) and the upper troposphere (4–8 km) showing the differences in all three pollutant loads. Significant role in these results plays the value of the gamma (Ȗ) uptake coefficient, which is considered 10-4 in this work (Zhang and Carmichael, 1999; Usher et al., 2002; Bauer and Koch, 2005; among others) and the weather conditions in the study area. Due to lack of data in the areas with high dust and SO2 concentrations, the comparison with the model results (for example Figure 4b), did not reveal the above discussed patterns in the daily average values. As part of an on-going work, the results presented herein await for more evaluating data. 2h Average CAMx Sulfate Concentration at Erdemli, Turkey Layer 1: 0-50m 1-9 April 2003
3,5E+01
2h average CAMx Sulfate Concentration at Saudi Arabia (near Red Sea) Layer 1: 0-50m 1-10 April 2003
7,00 3,0E+01
6,00 2,5E+01
ȝ g /m 3
ȝ g /m 3
5,00 4,00
2,0E+01
1,5E+01
3,00
1,0E+01
2,00 1,00
5,0E+00
0,00
0,0E+00
1/4/03 1/4/03 2/4/03 2/4/03 3/4/03 3/4/03 4/4/03 4/4/03 5/4/03 5/4/03 6/4/03 6/4/03 7/4/03 7/4/03 8/4/03 8/4/03 9/4/03 9/4/03 0:00- 12:00- 0:00- 12:00- 0:00- 12:00- 0:00- 12:00- 0:00- 12:00- 0:00- 12:00- 0:00- 12:00- 0:00- 12:00- 0:00- 12:002:00 14:00 2:00 14:00 2:00 14:00 2:00 14:00 2:00 14:00 2:00 14:00 2:00 14:00 2:00 14:00 2:00 14:00
PSO4total
DSO4total
PSO4+DSO4
1/4/03 1/4/03 2/4/03 2/4/03 3/4/03 3/4/03 4/4/03 5/4/03 5/4/03 6/4/03 6/4/03 7/4/03 8/4/03 8/4/03 9/4/03 9/4/03 10/4/03 10/4/03 0:00- 14:00- 4:00- 18:00- 8:00- 22:00- 12:00- 2:00- 16:00- 6:00- 20:00- 10:00- 0:00- 14:00- 4:00- 18:00- 8:00- 22:002:00 16:00 6:00 20:00 10:00 24:00 14:00 4:00 18:00 8:00 22:00 12:00 2:00 16:00 6:00 20:00 10:00 24:00
PSO4_2
DSO4_2
PSO4+DSO4
Date (UTC)
Fig. 5 Two hour average modeled sulfate concentration (PSO4 and DSO4) for two grid cells in the domain affected by desert dust transport. Left plot: Erdemli, Turkey. Right plot: Saudi Arabia
Heterogeneous Chemical Processes and their Role on Particulate Matter Formation
511
Fig. 6 Column integrated dust aerosol CRST (left), particulate sulfate PSO4 (middle) and particulate sulfate formed on dust DSO4 (right), for April 5, 2003, 12 UTC (g/m2). Size range: 0.1–2.5ȝm diameter. The upper plots are for the vertical column 0–4 km and the lower plots are for 4–8 km
4. Conclusions The preliminary results presented in this work emphasized on the sea salt particle production and the heterogeneous uptake of sulfur dioxide on dust particles using air quality modeling techniques. The generation of new aerosols on dust surfaces can be significant for both the middle and the upper troposphere, not because of the high amounts of produced species, but due to the different properties of such generation. This new 3rd generation aerosol can be higher than the sulfate produced from anthropogenic sources, depending on the weather and air quality conditions. Dust particles acting as reactive surfaces in a wet environment might lead to new climate modifiers for “desert dust-sensitive areas” like the Mediterranean Region. Overall, the new development presented in the previous sections, revealed the capability of an air quality model like CAMx to handle the complex processes of heterogeneous reactions on dust particles. Also revealed the uncertainty of such processes based on the assumptions during the implementation of the relevant coefficients and algorithms. This work is still ongoing, mostly the evaluation of the heterogeneous reactions and provides the capability for future improvement of the physicochemical processes treated by the model. Acknowledgments This work is realized in the framework of Act. 8.3 of the Operational Program Competitiveness (3rd Community Support Program) and is co-funded by 75% from the European Social Fund (EU), 25% from the Greek General Secretariat for Research and Technology (GSRT) under the project PENED 2003. The authors would like to acknowledge the contribution from Professor N. Mihalopoulos and X. Querol for providing air quality measurements.
512
M. Astitha et al.
References Alastuey et al. (2005) Characterisation of TSP and PM2.5 at Izana and Sta Cruz de Tenerife (Canary Islands, Spain) during a Saharan dust episode. Atmos. Environ., 39, 4715–4728. Astitha M, Kallos G, Katsafados P, Pytharoulis I, Mihalopoulos N (2006) “Radiative effects of natural PMs on photochemical processes in the Mediterranean Region”. 28th NATO/CCMS ITM, May 2006, Leipzig, Germany. Bauer SE, Koch D (2005) Impact of heterogeneous sulfate formation at mineral dust surfaces on aerosol loads and radiative forcing in the Goddard Institute for Space Studies general circulation model, J. Geophys. Res., 110(D17). de Leeuw, G., F.P. Neele, M. Hill, M.H. Smith, E. Vignati (2000) Production of sea spray aerosol in the surf zone, J. Geophys. Res., 105(D24), 29397–29409. Environ (2006) User’s Guide to the Comprehensive Air Quality Model with Extensions (CAMx). Version 4.31. ENVIRON Inter. Corp., Novato, CA. Fuchs NA, Sutugin AG (1970) Highly Dispersed Aerosols. Ann Arbor Science Publishers, Ann Arbor, MI. Gong SL (2003) A parameterization of sea-salt aerosol source function for suband super- micron particles. Glob. Biog. Cycles, 17(4), 1097. Kallos G (1997) The regional weather forecasting system SKIRON: an overview. Proceedings of the symposium on regional weather prediction on parallel computer environments, University of Athens, Greece, pp. 109–122. Kallos GA, Papadopoulos P, Katsafados, Nickovic S (2006) Trans-Atlantic Saharan dust transport: Model simulation and results. J. Geophys. Res., 111. Levin Z, Teller A, Ganor E, Yin Y (2005) On the interactions of mineral dust, sea-salt particles and clouds: A measurement and modelling study from the Mediterranean Israeli Dust Experiment campaign. J. Geophys. Res., 110, D09204. Nickovic S, Kallos G, Papadopoulos A, Kakaliagou O (2001) A model for prediction of desert dust cycle in the atmosphere. J. Geophys. Res., 106, D16. Penner JE et al. (2001) Aerosols, their direct and indirect effects, in Climate Change 2001: The Scientific Basis: Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change, edited by J.T. Houghton et al., pp. 291–348, Cambridge University Press, New York. Saylor RD (1997) An estimate of the potential significance of heterogeneous loss to aerosols as an additional sink for HO radicals in the troposphere. Atmos. Environ., 21, 3653–3658. Shankar U, Bhave PV, Vukovich JM, Roselle SJ (2005) Implementation and initial applications of sea salt aerosol emissions and chemistry algorithms in the CMAQ v4.5-AERO4 module. 4th Ann. CMAS Models-3 Users’ Conference. Usher CR, Al-Hosney H, Carlos-Cuellar S, Grassian VH (2002) A laboratory study of the heterogeneous uptake and oxidation of sulfur dioxide on mineral dust particles, J. Geophys. Res.-A, 107, 4713, doi:10.1029/2002JD002051. Zhang Y, Carmichael GR (1999) The role of mineral aerosol in tropospheric chemistry in East Asia-A model study, J. Appl. Met., 38, 353–366.
Heterogeneous Chemical Processes and their Role on Particulate Matter Formation
513
Zhang KM, Knipping EM, Wexler AS, Bhave PV, Tonnesen GS (2005) Size distribution of sea-salt emissions as a function of relative humidity. Atmos. Environ., 39, 3373–3379.
Discussion G. Sarwar: Please explain the reason for the decrease in ozone concentration. M. Astitha: The decrease in Ozone concentration is related to shading effects (effects of dust on photolysis rates) and heterogeneous processes (ozone with dust). M. Mircea: How many dust bins have your model? Did you study the impact of the number of size bins (of their size/on the gas uptake)? M. Astitha: At the present version of the dust flux model, we used four size bins but we used only the first two with particle size less than 10 ȝm. No we did not. Our new version of dust model includes eight particle sizes (up to PM10). I found your question interesting and I will look for such an impact in my future work. A. Venkatram:
M. Astitha:
Did you examine the sensitivity of your results on O3 and sulfate to the uptake coefficients for O3 and SO2? The inclusion of the heterogeneous conversion of SO2 to SO4 appears to have led to overestimation of SO4. Are you worried about these results? The uptake coefficients for O3 and sulphate on dust particles were chosen based on the literature. I haven’t done sensitivity experiments changing O3 and SO2 uptake coefficients but I intend to do this in the future. The amount of SO4 produced from the uptake of SO2 on dust is in general much smaller than the anthropogenic sulphate, except for some specific locations. We have seen the overestimation and we plan to carry out a series of sensitivity tests to see how the situation will be improved.
B. Fisher: How are the uptake coefficients determined? M. Astitha: The uptake coefficients (Ȗ values) of gases onto the particle surface are determined mostly through laboratory and field experiments.
5.7 Modelling Regional Aerosols: Impact of Cloud Processing on Gases and Particles over Eastern North America and in Its Outflow During ICARTT 2004 W. Gong1, J. Zhang1, M.D. Moran1, P.A. Makar1, S.L. Gong1, C. Stroud1, V.S. Bouchet2, S. Cousineau2, S. Ménard2, M. Samaali2, M. Sassi2, B. Pabla1, R. Leaitch1, A.M. Macdonald1, K. Anlauf1, K. Hayden1, D. Toom-Sauntry1, A. Leithead1 and J.W. Strapp3
Abstract A regional aerosol model, AURAMS (A Unified Regional Air-quality Modelling System), is used to simulate gases and aerosols over eastern North America for the ICARTT field campaign period during summer 2004. The model performance is evaluated against both ground-based and airborne observations during the field campaign. A model sensitivity study is used to assess the impact of cloud processing on the aerosol characteristics in the air masses over eastern North America and its outflow to the North Atlantic during the study period.
Keywords Air-quality modelling, cloud processing of gas and aerosol, regional aerosols
1. Introduction During the summer of 2004, several coordinated field campaigns were conducted over North America, the North Atlantic, and western Europe as part of the International Consortium for Atmospheric Research on Transport and Transformation (ICARTT). These field programs were designed to study the emission of aerosol and ozone precursors, their chemical transformations and removal during transport to and over the North Atlantic, and their impact downwind on the European continent (Fehsenfeld et al., 2006). 1
Environment Canada, 4905 Dufferin Street, Toronto, ON, M3H 5T4, Canada. Environment Canada, 2121, Voie de Service Nord No. 404 Route Transcanadienne, Montreal, QC, H3P 1J3, Canada. 3 Environment Canada, 4905 Dufferin Street, Toronto, ON, M3H 5T4, Canada. 2
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
539
540
W. Gong et al.
One of the campaigns, an aircraft study conducted by the Environment Canada scientists using the National Research Council of Canada (NRCC) Convair 580, focused on the chemical transformation of gases and aerosols by clouds. The summer of 2004 in eastern North America was characterized as being cooler and wetter than normal, with relatively more frequent southwesterly flow (Fehsenfeld et al., 2006). The observations at New England and Nova Scotia coastal sites and over the Gulf of Maine showed that the fine particulate matter (PM2.5) in the plumes from eastern North America was composed mainly of acidic sulphate and highly oxidized organics, an indication of significant atmospheric processing. An important question is how much of that PM2.5 was contributed by cloud processing. In this study we address the above question through the application of a regional aerosol model (AURAMS) to the ICARTT field campaign period over eastern North America. The model’s performance is examined against surface ozone and PM measurements from the AIRNow network, the speciated PM measurements from the IMPROVE network, and observations from the Convair 580 flights. The impact of cloud processing on aerosols over eastern North America and its outflow to the North Atlantic is then assessed with a model sensitivity study.
2. Simulation Setup AURAMS is a multi-pollutant, regional air-quality modelling system with size segregated and chemically speciated representation of aerosols. It has been described elsewhere (e.g., Gong et al., 2006; Tarasick et al., 2007). Simulation with AURAMS version 1.3.2 was carried out on an 85 × 105 polar-stereographic grid over eastern North America with a horizontal resolution of 42 km (true at 60q N). The simulation period is July 7–August 19, 2004. The first seven days are counted as model spin-up. An older version of AURAMS was run in real-time during the ICARTT field campaign to provide guidance for flight planning (Bouchet et al., 2004) and to participate in a multi-model inter-comparison and evaluation effort (McKeen et al., 2005, 2007). The real-time AURAMS run was also evaluated against ozonesonde data by Tarasick et al. (2007). Some of the important updates since the real-time run include changes in anthropogenic emission inventory, biogenic emission processing and vegetation database, and chemical lateral boundary conditions. For this simulation, the anthropogenic emissions files were prepared from the 2000 Canadian, 2001 U.S., and 1999 Mexican national emissions inventories with version 2.2 of the SMOKE emissions processing system (http://www.smokemodel.org/index.cfm). State-specific adjustments to the NOx emission from major point sources were also incorporated to account for the emission changes due to the NOx State Implementation Plan (SIP) Call by the U.S. EPA (U.S. EPA, 2004) which came into effect in 2003. The biogenic emission fields were processed inline using BEIS v3.09. Being an off-line model, AURAMS is driven by a regional weather forecast model simulation (GEM; Côte et al., 1998). For this study, GEM version 3.2.0, with the additional recent parameterization for anthropogenic heat islands (Makar
Modelling Regional Aerosols: Impact of Cloud Processing on Gases and Particles
541
et al., 2006), was used in a regional configuration with a 15 km resolution in its uniform “core” cantered over North America. Time-independent chemical lateral boundary conditions were used for this simulation. The gridded monthly ozone climatology of Logan (1998) was used for the initial and boundary condition for O3; the monthly MOPITT data (http://neo.sci. gsfc.nasa.gov/Search.html?group=35) was used for CO; and for the rest of the gases species and particle components the initial and boundary conditions were constructed from an existing AURAMS annual simulation (Moran et al., 2007) and observational data (e.g., Macdonald et al., 2006).
3. Model Evaluation 3.1. Surface ozone and PM2.5 Model predicted ground-level ozone and PM2.5 mass concentrations are compared to the hourly observations from the AIRNow network (http://airnow.gov/index. cfm). The AIRNOW observations are reported as hourly averages, whereas the model outputs are hourly, valid at the top of the hour. No attempt is made here to process model output in order to obtain equivalent hourly averages. Standard statistical measures are calculated at each AIRNow site (without spatial interpolation) within the area as in McKeen et al. (2005, 2007) over the 36-day simulation period (excluding the first seven-day spin-up). The spatial distribution of the correlation coefficient (r) and the mean bias (MB) are shown in Figure 1 for daily 1-hour maximum ozone and daily average PM2.5. The mean and median of these statistical measures are also included in the figure.
Fig. 1 Spatial distribution of (a) correlation coefficient (r), (b) mean bias (MB) for 1-hour daily maximum ozone, (c) r and (d) MB for daily mean PM2.5 over the 36-day simulation period
The model correlated relatively well with the observation of O3 over southern Ontario and the U.S. eastern seaboard, but the O3 comparison was relatively poor over the Ohio River Valley and the Appalachians. The mean correlation coefficient of 0.54 is a marginal improvement from 0.51 from the real-time run for the same period. There is a significant positive bias that averages at 13.5 ppbv. This amounts to a considerable increase from the real-time run mainly due to the change in biogenic emission processing and vegetation data. The simple representation of the U.S. SIP Call’s NOx control resulted in an average of 2 ppbv reduction in the ozone mean bias.
W. Gong et al.
542
As for PM2.5, the mean correlation coefficient of 0.51 is a noticeable improvement from the previous 0.47 from the real-time run. The overall bias is small. The spatial distribution of model performance (particularly in terms of r) is similar to that for ozone.
3.2. Speciated PM2.5
7/16/04 30
7/26/04
8/5/04
20
10
0 7/16/04
7/26/04
8/5/04
Quaker City, OH
Observed
10
0 7/26/04
8/5/04
30
10
0 7/16/04
8/15/04
7/26/04
8/5/04
8/15/04
Model (base case) 20
10
0
8/15/04
20
7/16/04
30
Model (base case)
20
8/15/04
Connecticut Hill, NY Observed
30
PM2.5, SO42.5, NO32.5, OM2.5 (ug m)
0
30
PM2.5, SO42.5, NO32.5, OM2.5 (ug m-3)
PM2.5, SO42.5, NO32.5, OM2.5 (ug m-3)
Observed
10
PM2.5, SO42.5, NO32.5, OM2.5 (ug m-3)
PM2.5, SO42.5, NO32.5, OM2.5 (ug
-3
)
Cape Cod, MA 20
PM2.5, SO42.5, NO32.5, OM2.5 (ug m-3)
Measurements of speciated PM2.5 at 23 sites from the IMPROVE network (http:// vista.cira.colostate.edu/views/Web/Data/DataWizard.aspx) are available for compareson with the model results. The measurements at these sites are 24-hour averages (midnight-to-midnight, local daylight saving time) every three days. The hourly model outputs are therefore averaged over the same 24-hour periods and also sampled every three days accordingly to be paired up with the observations. The model-observation comparison of speciated PM2.5 at three sites is shown in Figure 2.
7/16/04
7/26/04
8/5/04
8/15/04
Model (base case)
20
10
0 7/16/04
7/26/04
8/5/04
8/15/04
Fig. 2 Comparison with speciated PM2.5 measurements (left column – measurement, right column – model) at three sites from the IMPROVE network: (top) Cape Cod, (middle) Connecticut Hill, and (bottom) Quaker City; red – sulphate, blue – nitrate, and green – organics; pluses denote total PM2.5 mass
In general, the model tends to overpredict sulphate close to the sources (e.g. Quaker City, OH). The overprediction decreases moving away from source regions (e.g. Connecticut Hill, NY), and over the east coast region the model in general under-predicts sulphate (e.g. Cape Cod, MA). Nitrate2.5 is generally overpredicted by the model. There is a significant underprediction of the organic aerosol component by the model. This can be largely attributed to the fact that the implementation of Jiang’s SOA scheme (Jiang, 2003) in this study only considered the existing secondary organic aerosol as “seed” particles for the partitioning. A recent sensitivity simulation with several SOA parameterization updates (partitioning of SOA products to both primary and secondary OA, updates to partitioning and stoichio-metric coefficients, and inclusion of isoprene as a precursor to SOA) yielded an order of magnitude increase in SOA. Figure 3 shows the model-vs-observation scatter-plots of total PM2.5, sulphate2.5, nitrate2.5, and sulphate-to-PM2.5 ratio for all available sites in terms of averages over the simulation period (or, 12 24-hour samples). In this case, the correlations
Modelling Regional Aerosols: Impact of Cloud Processing on Gases and Particles
543
between model and observation are good for both PM2.5 and sulphate2.5. The modelled and observed mean values are also comparable (particularly in the case of sulphate; a bias of ~ –1.3 ȝg m-3 for PM2.5). As mentioned above, nitrate is overpredicted by the model and this is reflected here also. The overestimate of sulphate fraction by the model is often a result of the significant underprediction of organic component. 12
12 Y = 1.47 * X - 6.39 # of data points = 22 Average X = 10.84 Average Y = 9.56 R2 = 0.71
8 4 0
Model
Model
16
8
Y = 1.53 * X - 2.47 # of data points = 23 Average X = 4.46 Average Y = 4.33 R2 = 0.71
4 0
0
4
8 12 16 20 24 Observation
0
4 8 12 Observation
16
Nitrate2.5 (ug m-3)
2
(b)
Y = 0.88 * X - 1.45 # of data points = 23 Average X = 4.46 Average Y = 2.45 R2 = 0.70
0.7
(c)
Sulphate2.5/PM2.5
(d)
0.6
1.6
0.5
1.2 Y = 2.48 * X + 0.11 # of data points = 23 Average X = 0.22 Average Y = 0.66 R2 = 0.17
0.8 0.4
Model
Sulphate2.5 (ug m-3)
16
(a)
20
Model
PM2.5 (ug m-3)
24
Y = 0.77 * X + 0.18 # of data points = 19 Average X = 0.40 Average Y = 0.49 R2 = 0.25
0.4 0.3 0.2
Acadia
Cape Cod Martha's Vineyard
0.1
0 0
0.4
0.8 1.2 1.6 Observation
2
0.1 0.2 0.3 0.4 0.5 0.6 0.7 Observation
Fig. 3 Model-vs-observation scatter-plots of (a) PM2.5, (b) sulfate2.5, (c) nitrates2.5, and (d) sulphate fraction. Both model results and observations are averaged over the simulation period. Also plotted in (b) are results from a sensitivity run with in-cloud oxidation turned off (blue)
3.3. Comparison with aircraft observations During ICARTT 2004, 23 research flights were flown with the NRCC Convair 580 based at Cleveland, Ohio. Measurements onboard the aircraft included trace gases, aerosol particle size distribution and chemistry, and cloud microphysics and chemistry (see Hayden et al., 2007; Leithead et al., 2007; Zhang et al., 2007) for details on the measurement and instrumentation). Complementary to the surface-based observations, these airborne measurements provide information on both primary and secondary pollutants aloft which is valuable for improving our knowledge of the atmospheric processes and for evaluating model representations of these processes.
Flt 17 Flt 16
(a)
Flt 16
Flt 17
(b)
Fig. 4 Model predicted (a) O3 and (b) SO2 at 1,500 m for 21 Z August 10, 2004, overlaid with aircraft-observed O3 and SO2 along the flight tracks from Flt 16 and 17
Here we focus on two of the flights, Flt 16 and 17, conducted on August 10, 2004, between Lake Chicago and Lake Erie in the air mass ahead of an advancing cold front in an attempt to study cloud processing of industrial and urban plumes down-wind Chicago. Figure 4 shows the modelled ozone and SO2 at 1,500 m (close to the altitude for the in-cloud flight legs) overlaid with the in-situ measurements. As seen the model predicted ozone is considerably higher than the aircraft observations. On the other hand, the model is capturing the SO2 plumes reasonably
W. Gong et al.
544
well particularly for Flt 16. The plumes observed by the aircraft are more intense than the model, which is reasonable given that the model resolution is at 42 km. 4000
4000
3000
3000
3000
3000
3000
1000 0
0.01 0.1 1 10 100 SO2 (ppbv)
Altitude (m)
4000 3000 2000 1000
aircraft obs. model Altitude (m)
(b)
0
1000 0
0.01 0.1 1 10 100 NO2 (ppbv)
2000 1000 0
50
150 250 350 CO (ppbv)
1000 0
0
20 40 60 80 O3 (ppbv)
0
4000
4000
4000
3000
3000
3000
3000
2000 1000
2000 1000 0
0.01 0.1 1 10 100 NO2 (ppbv)
2000 1000
150 250 350 CO (ppbv)
4 8 12 SO4= (ug/m3)
10-min. PILS bulk CW samples (air equiv.) model
2000 1000 0
0 50
10-min PILS bulk CW samples (air equiv.) model
2000
4000
0 0.01 0.1 1 10 100 SO2 (ppbv)
2000
Altitude (m)
0
2000
Altitude (m)
1000
Altitude (m)
2000
Altitude (m)
4000
Altitude (m)
aircraft obs. 4000 model Altitude (m)
4000
Altitude (m)
Altitude (m)
(a)
0
20 40 60 80 O3 (ppbv)
0
4 8 12 SO4= (ug/m3)
Fig. 5 In-situ comparison of observed and modelled gaseous species and particle sulphate (a) fro Flt 16 and (b) for Flt 17. Measurements of the gaseous species (SO2, NO2, CO, and O3) are at 1-s frequency; measurements of inorganic ions are from the 10-minute integrated PILS; also included is air equivalent cloud-water sulphate from bulk cloud water samples (varying in sampling duration)
A series of in-situ comparisons between modelled and observed gaseous species (SO2, NO2, CO, and O3) and particle sulphate is presented in Figure 5. The comparison is done by plotting the model results and the observations along the flight track against height in a form of “vertical profiles”. This is in an attempt to get a sense of whether the model comes close to the aircraft observation over the flight area in terms of the range and vertical structure, since a true point-to-point in-situ comparison is not appropriate given the model resolution. The sampling through the modelled fields is done given the grid locations along the flight path over the entire flight period. The comparison is surprisingly good for SO2, NO2 and, to a lesser degree, CO. The excursions to higher values from the observations reflect the encountering of plumes during horizontal legs at two different altitudes. The 1-s (or ~100 m) observations are able to resolve much sharper peaks than the model at 42km resolution. Consistent with the evaluation above, the modelled ozone shows a significant positive bias (more pronounced at lower levels), although the observed vertical structure seems to be captured by the model. The model overpredicted particle sulphate in comparison to the 10-minute integrated measurement from PILS (Particle-in-Liquid-Sampler) particularly at lower levels (though the PILS measurements for Flt 17 are uncertain due to solution problems on that flight). Note that no screening for in-cloud sampling of PILS is done here, and the PILS samples in clouds are more difficult to interpret. Also included in the plot are air-equivalent cloud-water sulphate measurements in air-equivalent concentration from the bulk
Modelling Regional Aerosols: Impact of Cloud Processing on Gases and Particles
545
cloud-water samples with varying durations. Much higher sulphate concentrations are observed in clouds. If these clouds remained non-precipitating the sulphate would remain in the atmosphere after the evaporation of the clouds.
4. Impact of In-Cloud Oxidation The model simulation was repeated with the in-cloud oxidation turned off to assess its impact on aerosols over eastern North America and its outflow to the North Atlantic. Figure 6a shows the model predicted sulphate2.5 at 1,810 m for July 23, 2004 at 20 Z from the base case simulation and Figure 6b the difference (delta) between the base case (i.e., with in-cloud oxidation) and the run without in-cloud oxidation in modelled particle sulphate. The period of July 22–24, 2004 corresponded to an outflow event when, as seen from Figure 6a, the plumes from the Ohio valley and the New York-Boston area are merging and being transported to the north Atlantic under south-westerly to westerly flow. Based on Figure 6, for this snapshot, the in-cloud oxidation is responsible for just under 50% the particle sulphate over eastern North America. (a)
(b)
(c)
(d)
Fig. 6 (a) Model predicted sulpahte2.5 at 1,810 m overlaid with wind at the same level for 20Z July 23, 2004; (b) the corresponding difference in model predicted sulphate2.5 with vs without incloud oxidation; (c) modelled sulphate2.5 column loading (ȝg m-2) averaged over July 14–August 18, 2004; (d) difference in the modelled average sulphate column loading with vs without in-cloud oxidation
To assess the overall impact of in-cloud oxidation for the field study period, column loadings of particle mass (total and speciated components) were computed based on the hourly model output and averaged over the 36-day simulation period. Figure 6c presents the average sulphate2.5 column loading over the simulation period from the base case and Figure 6d the difference in the column loading when in-cloud oxidation is turned off. Similar to the case of the single snapshot above, the in-cloud oxidation contributes to about half of the average column loading during the ICARTT period. Not shown here, the impact of in-cloud oxidation on modelled PM2.5 is somewhat smaller due to the increase in nitrate (i.e., more nitrate will partition to particle phase when sulphate is reduced when everything else remains the same). The results from the run without the in-cloud oxidation are also included in Figure 3b. It is seen that, without the in-cloud oxidation, there is a negative mean bias of ~2 ug m-3 (or roughly 50%) in modelled sulphate, although the slope is closer to 1 than the base case.
546
W. Gong et al.
Acknowledgments The authors would like to acknowledge the IMPROVE network and the AIRNow group and cooperating agencies in U.S. and Canada for making their data available. The National Research Council of Canada Institute for Aerospace Research Lab and the Meteorological Service of Canada Aircraft facility are also acknowledged for aircraft operation and support.
References Bouchet VS, Crevier L-P, Cousineau S, Ménard S, Moffet R, Gong W, Makar P, Moran M, Pabla B (2004) Realtime regional air quality modelling in support of the ICARTT 2004 campaign. Proc. 27th NATO/CCMS ITM on Air Pollution Modelling and Its Application, Banff, Canada, October 25–29. Côté J, Desmarais J-G, Gravel S, Méthot A, Patoine A, Roch M, Staniforth A (1998) The operational CMC/MRB Global Environmental Multiscale (GEM) model. Part 1: Design considerations and formulation, Mon. Wea. Rev., 126, 1373–1395. Fehsenfeld FC, Ancellet G, Bates T, Goldstein A, Hardesty M, Honrath R, Law K, Lewis A, Leaitch R, McKeen S, Meagher J, Parrish DD, Pszenny A, Russell P, Schlager H, Seinfeld J, Trainer M, Talbot R (2006) International Consortium for Atmospheric Research on Transport and Transformation (ICARTT): North America to Europe: Overview of the 2004 summer field study, J. Geophys. Res., 111, D23S01, doi:10.1029/2006JD007829. Gong W, Dastoor AP, Bouchet VS, Gong S, Makar PA, Moran MD, Pabla B, Ménard S, Crevier L-P, Cousineau S, Venkatesh S (2006) Cloud processing of gases and aerosols in a regional air quality model (AURAMS), Atmos. Res., 82, 248–275. Hayden KL, Macdonald AM, Gong W, Toom-Sauntry D, Anlauf KG, Leithead A, Li S-M, Leaitch WR, Noone K (2007) Cloud Processing of Nitrate, J. Geophys. Res. (submitted). Jiang W (2003) Instantaneous secondary organic aerosol yields and their comparison with overall aerosol yields for aromatic and biogenic hydrocarbons, Atmos. Environ., 37, 5439–5444. Leithead A, Macdonald AM, Li SM, Gong W, Anlauf KG, Toom-Sauntry D, Hayden KL, Leaitch WR (2006) Investigation of carbonyls in cloud-water during ICARTT, J. Geophys. Res. (submitted). Logan JA (1998) An analysis of ozonesonde data for the troposphere: Recommendations for testing 3D models, and development of a gridded climatology for tropospheric ozone, J. Geophys. Res., 104, 16115–116149. Macdonald AM, Anlauf KG, Leaitch WR, Liu P (2006) “Multi-year chemistry of particles and selected trace gases at the Whistler high elevation site”, Eos. Trans. AGU, 87(52), Fall Meet. Suppl., Abstract A53B-0179. Makar PA, Gravel S, Chirkov V, Strawbridge KB, Froude F, Arnold J, Brook J (2006) Atmos. Environ., 40, 2750–2766.
Modelling Regional Aerosols: Impact of Cloud Processing on Gases and Particles
547
McKeen S, Wilczak J, Grell G, Djalalova I, Peckham S, Hsie E-Y, Gong W, Bouchet V, Ménard S, Moffet R, McHenry J, McQueen J, Tang Y, Carmichael GR, Pagowski M, Chan A, Dye T (2005) Assessment of an ensemble of seven real-time ozone forecasts over Eastern North America during the summer of 2004, J. Geophys. Res., 110, D21307, doi:10.1029/2005JD005858, 16 pp. McKeen S, Chung SH, Wilczak J, Grell G, Djalalova I, Peckham S, Gong W, Bouchet V, Moffet R, Tang Y, Carmichael GR, Mathur R, Yu S (2007) Evaluation of several real-time PM2.5 forecast models using data collected during the ICARTT/NEAQS 2004 field study, J. Geophys. Res., 112, D10S20, doi:10.1029/2006JD007608, 20 pp. Moran MD, Zheng Q, Samaali M (2007) Long-term multi-species performance evaluation of AURAMS for first 2002 annual run. EC internal report, Toronto, Ontario (in preparation). (Available from first author: email [email protected]) Tarasick DW, Moran MD, Thompson AM, Carey-Smith T, Rochon Y, Bouchet VS, Gong W, Makar PA, Stroud C, Ménard S, Crevier L-P, Cousineau S, Pudykiewicz JA, Kallaur A, Moffet R, Ménard R, Robichaud A, Cooper OR, Oltmans SJ, Witte JC, Forbes G, Johnson BJ, Merrill J, Moody JL, Morris G, Newchurch MJ, Schmidlin FJ, Joseph E (2007) Comparison of Canadian air quality forecast models with tropospheric ozone profile measurements above mid-latitude North America during the IONS/ICARTT campaign: evidence for stratospheric input, J. Geophys. Res., 112, D10S20, doi:10.1029/2006JD007782 (In press). U.S. EPA (2004) NOx Budget Trading Program: 2003 Progress and Compliance Report, Clean Air Markets Program, Office of Air and Radiation, U.S. Environmental Protection Agency, Washington, DC, August, 28 pp. + App. (Available from http://www.epa.gov/airmarkets/progress/progress-reports.html: viewed 11 June 2007) Zhang J, Gong W, Leaitch WR, Strapp JW (2007) Evaluation of modeled cloud properties against aircraft observations for air quality applications, J. Geophys. Res., 112, D10S16, doi:10.1029/2006JD007596.
Discussion P. Builtjes: Is the overestimation of nitrates due to problems in the chemical, and the dependence on temperature? W. Gong: The nitrate over-prediction may partly be attributable to overprediction of gas-phase nitric acid, but the uncertainty in the gasparticle partition of total nitrate (via inorganic heterogeneous chemistry, which is temperature dependent) is a major contributing factor. For example, it has been shown that whether or not treating particles as metastable aqueous droplets in the equilibrium can have an important impact on the partitioning. This is definitely an area needing more investigation.
5.8 On the Role of Ammonia in the Formation of PM2.5 C. Mensink and F. Deutsch
Abstract We studied the formation and composition of PM2.5 using the EUROS model. This model contains comprehensive modules (CACM, MADRID) for the formation of secondary atmospheric aerosols and their precursors. Some spatial and temporal patterns in which ammonia emissions can be associated with elevated PM2.5 and PM10 concentrations are analysed. Especially the episode of 15–16 April 2007 revealed some interesting features, e.g. the importance of the impact of temperature, relative humidity and hygroscopic water on PM2.5 and PM10 concentrations. A hypothesis is formulated in which it is stressed that ammonia can be a provider of an abundant amount of condensation nuclei in the form of ammonium nitrate and ammonium sulphate which, under favourable meteorological conditions, attract hygroscopic water, leading to rapid increase in the PM2.5 mass fraction.
Keywords Ammonia, ammonium, Belgium, hygroscopic water, nitrate, PM2.5 1. Introduction In the proposal for a new directive of the European Parliament and of the Council on ambient air quality and cleaner air for Europe, a target value of 20 µg/m³ to be met by 1 January 2010 and a limit value of 20 µg/m³ to be met by 1 January 2015 are proposed for PM2.5. This is the first time in Europe that exposure to ambient air concentrations of PM2.5 is considered legally. From measurements and model studies it is known that the contribution to PM2.5 of secondary inorganic compounds like ammonium, nitrate and sulphate is considerable, with a mean value of 40% for 6 European cities (Sillanpää et al., 2006). Van Grieken et al. (2003) showed that 40% of the PM2.5 measured at five different locations in Belgium consists of Secondary Inorganic Aerosols (SIA). About 20% of the PM2.5 was found to be elementary carbon (EC) and of organic origin (OA). They also concluded that the other 40% could not be identified, e.g. potentially being soil dust components, sea salt, particles of biogenic origin and water. Maenhaut et al. (2006) showed however, that sea salt and soil dust are rather found in the PM10–PM2.5 fraction, at least for a busy road in Antwerp, where they performed their measurements. They found large amounts of nitrate, sulphate and ammonium in the PM2.5 fraction (contributing more than 50%). These SIA compounds were found to C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
548
On the Role of Ammonia in the Formation of PM2.5
549
be more present in winter time than in summer time. This was especially true for nitrate. In this paper we study the role of ammonia in the formation of PM2.5 and PM10 in Belgium. We analyse its spatial and temporal patterns, by using the extended EUROS model and by making comparisons with observations. The Belgian version of the EUROS model (Delobbe et al., 2002) has been extended to model fine particulate matter (PM) by implementing the Caltech Atmospheric Chemistry Mechanism (CACM, Griffin et al., 2002) and the Model of Aerosol Dynamics, Reaction, Ionization and Dissolution (MADRID 2, Zhang et al., 2004). Currently, the modelling system is able to model mass and chemical composition of aerosols in two size fractions (PM2.5 and PM10–2.5) and seven components: ammonium, nitrate, sulphate, elementary carbon, primary inorganic compounds, primary organic compounds and secondary organic compounds (SOA). The model has been applied and validated for Belgium (Deutsch et al., 2008a, b, c). Once emitted in the air, ammonia (NH3) is converted to ammonium sulphate and ammonium nitrate, contributing to the PM2.5 (and the PM10) fraction: and
2NH3 + H2SO4 Æ (NH4)2SO4 or NH3 + H2SO4 Æ NH4 HSO4 NH3 + HNO3 ÅÆ NH4 NO3
(1) (2)
Since sulphuric acid (H2SO4) has a low vapour pressure, the sulphate will be available in the particle phase and will easily be neutralised to form ammonium sulphate or ammonium bisulphate (1). Ammonia can also react in the gas phase with nitric acid (HNO3) to form ammonium nitrate (2), but this is an equilibrium reaction and its formation depends strongly on temperature, humidity and the individual concentrations. As reported by Erisman and Schaap (2004), the amount of nitrate will increase with decreasing temperatures, provided that there is more ammonium available than needed to neutralise the sulphate. Or as stated by Pinder et al. (2007): in terms of emission control strategies, a reduction in SO2 and hence a reduction in sulphate will increase the available free NH3 and a portion of this free NH3 will react to form ammonium nitrate. In this way a part of the ammonium sulphate is replaced by ammonium nitrate. The relation between NH3 emissions and the formation of PM2.5 and PM10 is studied in Section 2, where we briefly address spatial patterns of NH3 emissions and particle concentrations in Belgium. In Section 3 we study a remarkable event occurring on the evening and the night of 15–16 April 2007, showing a sharp increase in PM2.5 and PM10 concentrations in Brussels, attaining levels of 150 µg/m³ within a couple of hours. In order to analyse this event in more detail we performed a sensitivity calculation using the EUROS model and increasing the ammonia emissions by a factor of 5. Results show that the formation mechanism expressed by (1) and (2) can only partially explain the enormous rise in PM2.5 concentration observed in the monitoring station. This will be further discussed in Section 4.
C. Mensink, F. Deutsch
550
2. Spatial Distribution of NH3 and PM10 over Belgium Figure 1 shows the modelled NH3 concentrations for April 2007. It gives a typical picture of the spatial distribution of the NH3 emissions and concentrations over Belgium. More than 90% of the NH3 emissions originate from agricultural activities (cattle breeding, pig farms and manure spreading). These activities are concentrated in West Flanders, a province in the north western part of Belgium. Fig. 1 Ammonia (NH3) concentrations over Belgium, northern France, western Germany and the south of the Netherlands as modelled by the EUROS model for the period 1–20 April 2007
Legend [µg/m³]
Legend PM10 [µg/m³]
error
error
1 - 10
1 - 18
11 - 12
19 - 21
13 - 14
22 - 23
15 - 17
24 - 25
18 - 20
26 - 27
21 - 23
28 - 29
24 - 26
30 - 31
27 - 29
32 - 34
30 - 33
35 - 38
> 33
> 38
Fig. 2 Maps of the observed annual mean NO2 (left) and annual mean PM10 (right) for Belgium in 2002
Fig. 3 Annual mean PM10-concentration in Belgium as modelled (left) and measured (right) in 2003 (www.irceline.be)
On the Role of Ammonia in the Formation of PM2.5
551
Concentration [µg/m³] 120
2,5
ML01
ML02
R841
Model
100
80
60
40
20
0 1/01/2003 1/02/2003 1/03/2003 1/04/2003 1/05/2003 1/06/2003 1/07/2003 1/08/2003 1/09/2003 1/10/2003 1/11/2003 1/12/2003
Fig. 4 Modelled and the measured PM2.5 concentrations in 2003 for three monitoring stations in Mechelen
Referring to mechanisms (1) and (2) in the previous section, these high emissions and concentrations of NH3 are expected to lead to the formations of ammonium sulphate and ammonium nitrate. For the formation of these SIA, both SO2 and NO2 are available from long range transport as well as from local sources. Figure 2 shows a map of the measured NO2 and measured PM10 concentrations over Belgium in 2002. The NO2 map clearly reflects the traffic patterns and industrial activities over the country, where one can immediately point out the urban areas of Brussels, Antwerp, Ghent, Charleroi and Liège. In contrast, the PM10 map shows remarkable elevated concentrations in West Flanders. These high concentrations can be associated with the high agricultural activities leading to high NH3 emissions (Figure 1). Through mechanisms (1) and (2), elevated PM2.5 and PM10 concentrations are expected. Figure 3 shows a comparison of the modelled and the measured PM10 concentrations in 2003, clearly showing the elevated PM10 concentrations in West Flanders. Unfortunately PM2.5 concentration maps are not yet available, because of lack of sufficient monitoring stations for PM2.5. Figure 4 shows a comparison of the modelled and the measured PM2.5 concentrations in 2003 for three monitoring stations in the area of Mechelen, some 30 km north of Brussels. The figure shows a good comparison between the modelled and measured PM2.5 concentrations, although peak values are underestimated.
3. Temporal Patterns of PM2.5 on 15–16 April 2007 In the evening and night of 15 and 16 April 2007, a sharp increase in PM2.5 and PM10 concentrations was observed between 18h00 and 3h00 CEST in several monitoring stations in Brussels. Figure 5 shows both PM2.5 and PM10
552
C. Mensink, F. Deutsch
Fig. 5 Observed PM2.5 and PM10 concentrations and relative humidity (RH) in Sint-JansMolenbeek (Brussels), for 15–16 April 2007
concentrations as observed by TEOM FDMS equipment in the station Sint-JansMolenbeek (41R001). It can be seen that PM2.5 and PM10 concentrations are about equal, indicating that the increase concerns the smaller fraction of the PM mass and possibly a sudden increase of SIA. During this episode the strong ratio of increase in PM2.5 coincided with a strong ratio of increase in relative humidity (RH). Likewise the temperature dropped sharply from 25°C at 18h00 CEST to 10°C at 2h00 CEST. During this period the wind was coming from westerly directions (260°–310°). Wind speeds were rather low (<1 m/s) and the high concentrations were only observed in the Brussels area for TEOM-FDMS stations, so it is not very likely that transport of polluted air masses have caused the high concentrations. In order to analyse the possible contributions of SIA, we performed a model study for this episode. Figure 6 shows the modelled PM2.5 concentrations and a breakdown of the chemical composition for Sint-Jans-Molenbeek. As we see, the total amount of PM2.5 modelled at the peak moment in the early hours of 16 April 2007, is less than 50 µg/m³, which is three times less than observed. Boogaard and Duyzer (1998) showed that during the months March and April, the ammonia emissions are expected to be higher than the annual average which is actually applied in the model. They estimated that in the Netherlands NH3 emissions are six times higher in March and three times higher in April. Taking this into account and observing intensive manure spreading activities in this period in Flanders, we performed a sensitivity analysis increasing the NH3 emissions with a factor of 5. Results are shown in Figure 7. By comparing Figures 6 and 7 we observe an increase in PM2.5 of only ±7 µg/m³, which is due to an increase of the nitrate concentration. It shows that there is enough ammonia in the atmosphere to form ammonium nitrate. It also shows that the sharp increase in NH3 cannot explain the sharp increase in the observed PM2.5 concentrations. It is only when we allow for the contribution of hygroscopic H2O (Figure 8) that we are able to explain a sharp increase in PM2.5 and obtain the observed levels of 150 µg/m³. Hygroscopic water is water that forms a thin film around individual particles. It is non-mobile and can only be removed through heating. It has the ability to absorb water and therefore accelerates the condensation process. Its role will be further discussed in Section 4.
On the Role of Ammonia in the Formation of PM2.5
553
Chemical composition of PM2.5 in Sint-Jans-Molenbeek
70
EC ammonium
60
prim. inorg. sulfate
prim. org. SOA
nitrate
50 40 30 20 10
20 /0 4/ 20 07
01
01
01
19 /0 4/ 20 07
01
18 /0 4/ 20 07
17 /0 4/ 20 07
16 /0 4/ 20 07
15 /0 4/ 20 07
01
01
01
01
14 /0 4/ 20 07
12 /0 4/ 20 07
13 /0 4/ 20 07
01
01
01
11 /0 4/ 20 07
01
10 /0 4/ 20 07
09 /0 4/ 20 07
01 08 /0 4/ 20 07
07 /0 4/ 20 07
01
0
Fig. 6 Modelled PM2.5 concentrations and a breakdown of the chemical composition for Sint-JansMolenbeek for the period 1–20 April 2007 Chemical composition of PM2.5 in Sint-Jans-Molenbeek; NH3 emissions *5
70 60
prim. org. sulfate
prim. inorg. ammonium
EC nitrate SOA
50 40 30 20 10
01 20 /0 4/ 20 07
19 /0 4/ 20 07
01
01
18 /0 4/ 20 07
01 17 /0 4/ 20 07
01 16 /0 4/ 20 07
01
01 14 /0 4/ 20 07
15 /0 4/ 20 07
01
01 13 /0 4/ 20 07
11 /0 4/ 20 07
12 /0 4/ 20 07
01
01 10 /0 4/ 20 07
09 /0 4/ 20 07
01
01 08 /0 4/ 20 07
07 /0 4/ 20 07
01
0
Fig. 7 Modelled PM2.5 concentrations and a breakdown of the chemical composition for Sint-JansMolenbeek for the period 1–20 April 2007, increasing NH3 emissions with a factor of 5
01
01
01 20 /0 4/ 20 07
19 /0 4/ 20 07
01
01
01
18 /0 4/ 20 07
17 /0 4/ 20 07
16 /0 4/ 20 07
01
15 /0 4/ 20 07
14 /0 4/ 20 07
01
nitrate H2O
13 /0 4/ 20 07
12 /0 4/ 20 07
01
prim. org. SOA
01 11 /0 4/ 20 07
01
01
prim. inorg. sulfate
10 /0 4/ 20 07
01
09 /0 4/ 20 07
01 07 /0 4/ 20 07
Chemical composition of PM2.5 in Sint-Jans-Molenbeek EC ammonium
08 /0 4/ 20 07
300 280 260 240 220 200 180 160 140 120 100 80 60 40 20 0
Fig. 8 Modelled PM2.5 concentrations and a breakdown of the chemical composition including H2O for Sint-Jans-Molenbeek for the period 1–20 April 2007
4. Discussion The drastic temperature drop and the sharp increase in RH shown in Figure 5, together with the abundance of condensation nuclei due to the presence of ammonium sulphate and ammonium nitrate (shown in Figure 6) will enhance the
554
C. Mensink, F. Deutsch
condensation process. This is observed and reported on in literature by means of the aerosol growth factor (GF). The GF is defined as the ratio between the diameter of an aerosol at a set RH and the diameter of the aerosol at a RH = 10% (dry diameter). For ammonium sulphate, it was found that, during the period of deliquescence, GF can rise to 1.4 at RH = 80% and to 1.8 at RH = 90%. For ammonium sulphate the deliquescence process starts at a RH = 40%. For ammonium nitrate it starts much later at a RH = 74.3%. Since these are the humidity percentages observed in the night of 15–16 April, it can be argued from a theoretical point of view that the hygroscopic water may have contributed considerably to the PM2.5 aerosol mass. Note that at RH = 90%, mass is expected to grow with a factor of almost 6, due to the volume increase which goes with the third power of the diameter. The strong correlation of the rate of increase of PM2.5 with RH (Figure 5) also seems to indicate that a PM2.5 mass increase due to hygroscopic water is very likely. The point is however, to what extend hygroscopic water is (to be) measured. The European guideline 2003/37/EC on PM 2.5 monitoring suggests that, in accordance with guideline 1999/30/EC, gravimetric methods are preferred and are considered as the reference method. This means that PM2.5 is to be filtered and measured at surrounding conditions and that heating should be avoided. Although several monitoring methods are still investigated by the CEN Working Group 15, it was already shown that TEOM FDMS equipment gives results that are very close to the reference method. If one would conclude that the measurements should contain as much of the hygroscopic and semi-volatile compounds as possible, this would mean that hygroscopic water should be taken into account as part of the modelled PM2.5 as well, i.e. in case one wants to carry out a proper validation.
5. Summary The distribution of NH3 emissions in Belgium coincides with an observed pattern of elevated PM10 concentrations in West Flanders. Through the formation of ammonium sulphate and ammonium nitrate, this NH3 contributes to the PM2.5 fraction and thus to PM10. On average, 40% of the PM2.5 measured at different locations in Belgium consists of SIA. During a measured episode of high PM2.5 concentrations in Brussels on 15–16 April we could observe that: (i) the measured PM2.5 concentration was almost equal to the measured PM10 concentration; (ii) the rapid increase of PM2.5 up to 150 µg/m³ coincides with a rapid increase in RH and a sharp decrease in temperature; (iii) the model shows that only 30% of this increase can be attributed to an increase in SIA, even after an increase in the NH3 emissions by a factor of 5; (iv) only if we allow for hygroscopic water in the total mass of PM2.5, we can explain both the rapid increase in mass and the concentration level obtained (150 µg/m³). From these observations we can formulate the hypothesis that ammonia can be a provider of an abundant amount of condensation nuclei in the form of ammonium nitrate and ammonium sulphate which, under favourable meteorological conditions,
On the Role of Ammonia in the Formation of PM2.5
555
attract hygroscopic water, leading to rapid increase in the PM2.5 mass fraction. It still has to be investigated to what extend the various measurement methods (mentioned in 2003/37/EC) are taking hygroscopic water into account.
References Boogaard A, Duyzer J (1998) A comparison between results from measurements and calculations of ammonia concentrations in ambient air on a scale smaller than 5 km. TNO MEP report 97/423 (in Dutch). Delobbe L, Mensink C, Schayes G, Brasseur O, Passelecq C, Passelecq D, Dumont G, Demuth C (2002) BelEUROS: Implementation and extension of the EUROS model for policy support in Belgium, in: Global Change and Sustainable Development, Federal Science Policy Office, Brussels. Deutsch F, Janssen L, Vankerkom J, Lefebre F, Mensink C, Fierens F, Dumont G, Roekens E (2008a) Modelling changes of aerosol compositions over Belgium and Europe, International Journal of Environment and Pollution 32, 162–173. Deutsch F, Vankerkom J, Janssen L, Lefebre F, Mensink C, Fierens F, Dumont G, Blommaert F, Roekens E (2008b) Extension of the AURORA and EUROS integrated air quality models to fine particulate matter by coupling to CACM/ MADRID 2, Environmental Modeling and Assessment in press, available online, doi:10.1007/s10666–007–9100–2. Deutsch F, Mensink C, Vankerkom J, Janssen L (2008c) Application and validation of a comprehensive model for PM10 and PM2.5 concentrations in Belgium and Europe, Applied Mathematical Modelling 32, 1501–1510. Erisman JW, Schaap M (2004) The need for ammonia abatement with respect to secondary PM reductions in Europe, Environmental Pollution 129, 159–163. Griffin RJ, Dabdub D, Seinfeld JH (2002) Secondary organic aerosol 1. Atmospheric chemical mechanism for production of molecular constituents, J. Geophys. Res. 107(D17), 4332, doi:10.1029/2001JD000541. Maenhaut W, Chi X, Wang W (2006) Analyses for EC/OC and ions in fine particulate matter, 2005–2006; Report for the Flemish Environment Agency, University of Ghent (in Dutch). Pinder RW, Adams PJ, Pandis SN (2007) Ammonia emission controls as a costeffective strategy for reducing atmospheric particulate matter in eastern United Sates, Environ. Sci. Technol., 41, 380–386. Sillanpää M, Hillamo R, Saarikoski S, Frey A, Pennanen A, Makkonen U, Spolnik Z, Van Grieken R, Branis M, Brunekreef B, Chalbot M-C, Kuhlbusch T, Sunyer J, Kerminen V-M, Kulmala M, Salonen RO (2006) Chemical composition and mass closure of particulate matter at six urban sites in Europe, Atmos. Environ. 40, S212–S223. Van Grieken R, De Hoog J, Bencs L, Spolnik Z, Deutsch F, Khaiwal R, Stranger M, Berghmans P, Bleux N (2003) Measurements of PM2.5 in Flanders, 2001–2003; Final report for the Flemish Environment Agency, University of Antwerp/VITO/ VMM.
556
C. Mensink, F. Deutsch
Zhang Y, Pun B, Vijayaraghavan K, Wu S-Y, Seigneur C, Pandis SN, Jacobson MZ, Nenes A, Seinfeld JH (2004) Development and application of the Model of Aerosol Dynamics, Reaction, Ionization, and Dissolution (MADRID), J. Geophys. Res. 109, D01202, doi:10.1029/2003J
Discussion M. Astitha: Have you looked at the possibility of long-range transport that gave the high concentrations for that specific day? C. Mensink: We did check this. Both by analysing the local meteorological conditions in Brussels during the event and by simulating this longrange transport through the model calculations (EUROS model). The atmospheric conditions were very calm and stable during the event, which makes the impact of advection of polluted masses not very likely. The EUROS model does take into account long-range transport, using reanalysed ECMWF data for its meteorological input. The model was not able to simulate the event on 16 April 2007. Based on this result and on the observation that the PM10 and PM2.5 concentrations increased very rapidly and only in a limited number of monitoring stations, we concluded that the phenomenon was not caused by long-range transport. P. Builtjes: C. Mensink:
I do not know of any chemical composition observations in Europe show more than H2O in PM. The question is then of course to what extend H2O is measured. There is a difference between hygroscopic water which is bound to the particles and free water which is measurable as H2O. If the current measurement equipment only measures the free water, then it is obvious that the observations show a lower content of H2O in the composition. The model tells us that the rapid mass increase and the very high levels of PM10 and PM2.5 can be explained only by taking into account hygroscopic water.
R. Hill: Do you measure the chemical composition of PM2.5 again at this time of year as the episode may? C. Mensink: This is a valuable suggestion. At the moment the chemical composition is analysed only once a day in Brussels and includes sulphate, nitrate, ammonia and a “volatile fraction”. This fraction could be further analysed. A more detailed measurement campaign during the event (with samples taken on an hourly base) could certainly support further investigations.
5.1 Predicting Air Quality: Current Status and Future Directions Gregory R. Carmichael, Adrian Sandu, Tianfeng Chai, Dacian N. Daescu, Emil M. Constantinescu and Youhua Tang
Abstract Air quality prediction plays an important role in the management of our environment. As more atmospheric chemical observations become available chemical data assimilation is expected to play an essential role in air quality forecasting. In this paper the current status of air quality forecasting is discussed and illustrated by comparison of predictions with observations. The future directions are also discussed, with an emphasis on data assimilation. Applications of the four dimensional variational method (4D-Var) and the ensemble Kalman filter (EnKF) approach are presented and discussed.
Keywords Air quality modeling, forecasting chemical weather, ozone pollution
1. Introduction Predicting air quality is of growing importance to society. The chemical composition of the atmosphere has been (and is being) significantly perturbed by emissions of trace gases and aerosols associated with a variety of anthropogenic activities. This changing of the chemical composition of the atmosphere has important implications for urban, regional and global air quality, and for climate change. Chemical transport models (CTMs) have become an essential tool for providing science-based input into best alternatives for reducing urban pollution levels, for designing cost-effective emission control strategies, for the interpretation of observational data, and for assessments into how we have altered the chemistry of the global environment. The use of CTMs to produce air quality forecasts has become a new application area, providing important information to the public, decision makers and researchers. Currently hundreds of cities world-wide are providing real time air quality forecasts. In addition, national weather services throughout the world are broadening their traditional role of mesoscale weather prediction to also include prediction of other environmental phenomena (e.g., plumes from biomass burning, volcanic eruptions dust storms, and urban air pollution) that could potentially affect the health and welfare of their inhabitants. For example, the U.S. National Weather Service (NWS) has recently started to provide mesoscale numerical model forecast guidance for short-term air quality predictions, beginning with next-day C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
481
482
G. Carmichael et al.
ozone (O 3 ) forecasts for the northeastern, and plans to expand this air quality capability over the next ten years to include the entire U.S., to lengthen the forecast period to three-days, and to add fine particulate matter (PM2.5) to the forecasts. While significant advances in CTMs have taken place, predicting air quality remains a challenging problem due to the complex processes occurring at widely different scales and by their strong coupling across scales. Air quality predictions also have large uncertainties associated with: incomplete and/or inaccurate emissions information; lack of key measurements to impose initial and boundary conditions; missing science elements; and poorly parameterized processes. Improvements in the analysis capabilities of CTMs require them to be better constrained through the use of observational data. Borrowing lessons learned from the evolution of numerical weather prediction (NWP) models, improving air quality predictions through the assimilation of chemical data holds significant promise (but is yet realized). In this paper we present a view of the current status of air quality prediction (with a focus on air quality forecasting) using CTMs, and illustrate some of the current activities focused on improving the prediction capabilities. Further details regarding the current state and future research needs in air quality forecasting can be found in (Dabberdt et al., 2004) and references therein.
2. Air Quality Forecasting Air quality forecasts built upon CTM predictions (in contrast to other techniques such as statistical methods) contain components related to emissions, transport, transformation and removal processes. Since the four-dimensional distribution of pollutants in the atmosphere is heavily influenced by the prevailing meteorological conditions, air quality models are closely aligned with weather prediction. Air quality forecast models are driven by meteorological models (global and/or mesoscale), and this coupling is done in off-line or on-line modes, referring to whether the air quality constituents are calculated within the meteorological model itself (on-line, e.g., WRF/Chem (Grell et al., 2005) or calculated in a separate model, which accepts the meteorological fields as inputs (off-line). Air quality forecasting differs in important ways from the problem of weather forecasting. One important difference is that weather prediction is typically focused on severe, adverse weather conditions (e.g., storms), while the meteorology of adverse air quality conditions frequently is associated with benign weather. Boundary-layer structure and wind direction are perhaps the two most poorly determined meteorological variables for air quality prediction. Meteorological observations are critical to effectively predict air quality, yet meteorological observing systems are typically designed to support prediction of severe weather and not the subtleties of adverse air quality. Research needs associated with the meteorological elements of air quality prediction have been recently assessed (Dabbert et al., 2004). Air quality predictions also differ from weather forecasting due to the additional processes associated with emissions, chemical transformations, and removal. Because many
Predicting Air Quality: Current Status and Future Directions
483
important pollutants (e.g., ozone and fine particulate sulfate) are secondary in nature (i.e., formed via chemical reactions in the atmosphere), air quality models must include a rich description of the photochemical oxidant cycle. The capabilities of current CTMs used in air quality forecasting are illustrated using results from the STEM model application during the NASA TRACE-P (Tropospheric Atmospheric Chemistry Pacific Experiment (Jacob et al., 2003) study (Carmichael et al., 2003). The model results are compared to the 5-min merged data set for observations taken on-board the DC-8 aircraft. The model was sampled every 5 minutes along each fight-track for the period when the aircrafts were operating in the western Pacific (4 March–2 April 2001). Model results were interpolated to the aircraft location and time (using tri-linear and linear interpolation). The correlation coefficients for a variety of meteorological and chemical variables are plotted in Figure 1. As shown the meteorological parameters are modeled most accurately. The better skill in predicting these meteorological variables reflects the large amount of observational data ingested into the analysis of the large-scale meteorological fields. The correlation coefficients for the trace species are lower than those for the meteorological parameters. In general the predictive skill for the chemical species decreases with distance above the surface, reflecting the increased uncertainty associated with vertical transport processes.
Fig. 1 Correlation coefficients between observed and modelled meteorological and species for the <1 km, 1–3 km, and >3 km altitude bins for the P3-B and DC-8 aircraft data
Currently most air quality forecasts are based on limited area models in order to achieve the requisite temporal and spatial representation. For such applications an additional source of uncertainty is the boundary conditions. One current approach for forecasting is to drive the limited area forecasts with boundary conditions (BCs) provided by a global chemical weather forecasts. The sensitivity of the predictions of a limited area model to various treatments of the boundary conditions is available in (Tang et al., 2006), where ozone predictions using BCs from three different global model were shown to vary by more than 20%.
G. Carmichael et al.
484
3. Directions for Better Air Quality Forecasts
3.1. Ensemble forecasts An important technique to deal with the underlying uncertainty in air quality modeling is to make forecasts using ensembles of predictions. Ensemble techniques are commonly used to improve the forecast ability of weather models (Kalnay, 2003). The application of ensemble techniques to air quality forecasts is very recent. As part of a collective, informal model verification project within the ICARTT/ NEAQS-2004 study, forecasts of several key meteorological, radiation, and gasphase atmospheric constituents were gathered in near real time (typically 4- to 10hour computational delay) from seven CTMs and used to prepare and evaluate forecast skill for predicting surface ozone (McKeen et al., 2005). The major intent of this study was to critically examine the usefulness of the ensemble forecast relative to its individual members, and to provide a reference for future real-time AQ ensemble forecasts. The ensemble predictions based on the mean of the seven models was found to have significantly more temporal correlation to the observed daily maximum 1-hour average and maximum 8-hour average O 3 concentrations than any individual model. Ensembles using simple bias correction algorithms were also evaluated and found to improve the forecast skill. Illustrative results are shown in Figure 2. One CTM model
Ensemble ((8 models)
46
46
44
44
42
42
40 38 36
85
40
r coeffient (1 hr max 03) < .57 0.57 to 0.71 0.71 to 0.82 0.82 to 1
80
75
70
r coeffient (1 hr max 03) < .57 0.57 to 0.71 0.71 to 0.82 0.82 to 1
38 36
65
85
80
75
70
65
Fig. 2 Comparison of forecast skills for the summer of 2004. Shown are calculated correlation coefficients for predicted and observed surface ozone (hourly) at the Air Now monitoring sites. (a) The forecasts of the STEM model; (b) forecasts skill using an eight member ensemble
3.2. Chemical data assimilation For the predictive capabilities of CTMs to improve, they must be better constrained through the use of observational data. The close integration of observational data is recognized as essential in weather/climate analysis, and it is accomplished by a
Predicting Air Quality: Current Status and Future Directions
485
mature experience/infrastructure in data assimilation – the process by which models use measurements to produce an optimal representation of the state of the atmosphere. This is equally desirable in CTMs. Data assimilation combines information from three different sources: the physical and chemical laws of evolution (encapsulated in the model), the reality (as captured by the observations), and the current best estimate of the distribution of trace gas or aerosol species in the atmosphere (all with associated errors). As more chemical observations in the troposphere are becoming available, chemical data assimilation is expected to play an essential role in air quality forecasting, similar to the role it has in numerical weather prediction. Advanced assimilation techniques fall within the general categories of variational (3D-Var, 4D-Var) and Kalman filter—based methods, which have been developed in the framework of optimal estimation theory. The variational data assimilation approach seeks to minimize a cost functional that measures the distance from measurements and the “background” estimate of the true state. In the 3D-VAR method the observations are processed sequentially in time. The 4D-Var generalizes this method by considering observations that are distributed in time. These methods have been successfully applied in meteorology and oceanography, but they are only just beginning to be used in nonlinear atmospheric chemical models (Menut et al., 2000; Elburn et al., 2000; Sandu et al., 2005; Chai et al., 2006). When chemical transformations and interactions are considered, the complexity of the implementation and the computational cost of the data assimilation are highly increased. A discussion of current approaches follows. 3.2.1. Chemical Data Assimilation Using 4D-Var
In the 4D-Var approach an optimal solution is sought by adjusting chosen parameters according to available measurements in the analysis time interval. Such parameters are often called control variables and they may include initial concentrations, emission rates, concentrations and fluxes at domain boundaries, and other physical or chemical parameters. The optimal solution of the control variables minimizes a cost functional that is generally defined as < =
1 0 y y b0 2
1 >h ( y ) y obs 2
>
@
T
>
@
B 1 y 0 y b0
@T O 1 >h ( y )
y obs
1 > p p b @T P 1 > p p b @ 2
(1)
@
where B , P , and O are error covariance matrices in discrete spaces for back0 ground initial values yb , parameters pb , and observations yobs , respectively. h is a projection operator, used to calculate the observation vector y obs (t ) from the model state variables. The optimal solution depends on the uncertainty of both observations and control variables, which are represented by the error covariance matrices, B , P , and O . As observational errors are often uncorrelated with each other, O is
G. Carmichael et al.
486
typically assumed as a diagonal matrix. The complexity of P depends on the choices of control parameters. The background error covariance matrix B is often correlated in space and between different species. An accurate estimation of the background error covariance matrix is difficult to provide and, given its huge dimensionality, simplifying approximations are required for the practical implementation. Information on the error statistics may be obtained using differrences between forecasts with different initialization time (NMC method) or ensemble methods based on a perturbed forecast-analysis system (Fisher, 2003). The data assimilation problem is then formulated as an optimization problem
min < p, y 0
for
p PADMISSIBLE, y 0 YADMISSIBLE
(2)
The minimization process is computationally demanding, but can be efficiently n implemented using adjoint modeling to compute the gradients o < and y m p < of the cost functional. Illustrative results on the application of this data assimilation capability using the ICARTT ozone measurements from different sources to improve the predicted ozone distributions are presented in (Chai et al., 2007). In this paper we demonstrated how the modeling and measurement activities of ICARTT can be used in the data assimilation framework. The ICARTT experiments produced comprehensive observation data sets and intense modeling applications upon which to study important aspects of data assimilation. The data used included in-situ ozone from the DC8 and P3, lidar observations from the DC3 and DC8, ozonsodes, MOZAIC profiles, and surface observations (AiRNow and AIRMAP). The model error correlation was constructed using the NMC approach. It is implemented into a 4DVar regional chemical data assimilation system with a truncated SVD regularization method is introduced. The observational (Hollingworth-Lonnberg) method was used to calculate the weighting between observations and model backgrounds in 4D-Var. It should be noted the increase of the computational time is minimal using the current approach, compared to using a diagonal matrix for the background error covariance. The weighting between the model and observations in determining the final optimal analysis depends on the both the background and observational error covariance matrices, which are objectively approximated in the current application. Ozone observations by different platforms during the ICARTT field experiment were assimilated into the regional CTM. It is found with little exception that assimilating observations from each individual platform improves the model predictions against the withheld observations. For example shown in Figure 3 are the domainaveraged vertical profiles (with standard deviation) constructed using the observations and the corresponding predictions for Case 9 (all the data from the sources listed in Figure 3 are used in the assimilation) and the base case (with out data assimilation). It clearly shows that the model biases both below and above 3,000 m
Predicting Air Quality: Current Status and Future Directions
487
Fig. 3 Example of the impact of data assimilation on predicted ozone distributions. Shown are forecasted (base-case) ozone fields for July 20, 2004. The results after the assimilation of ozone from aircraft (in-situ and remote-sensed), surface-based, and ozonesondes are also shown (case-9). Domain averaged vertical profiles and quantile plots are shown for comparison
are substantially reduced for Case 9. The predicted values for Case 9, now show a negative bias at low to mid- altitudes and a positive bias at high altitudes. Also shown are the quantile-quantile (q-q) plots of the ozone observations versus the corresponding predictions, for the base case and Case 9. Each point in a quantilequantile plot shows the values from two data sets that have the same quantile, i.e. the fraction of data points that fall below the given value. The q-q plot of the base case clearly shows the predictions are biased high overall. After assimilation, Case 9 generates a predicted ozone field that has a very similar population distribution as the observations. The q-q plot of Case 9 also indicates that the model has difficulty to generate low ozone concentrations (<20 ppbv) in the data assimilation time period. This is largely due to the coarse model resolution. Analysis where all the observations by the different platforms were simultaneously assimilated, resulted in a reduction in bias to –1.5 ppbv from 11.3 ppbv for the base case without assimilation. A reduction of 10.3 ppbv in RMS error was also seen. The information content of the observations depends on the model resolution and it can be approximated by the number of four-dimensional (in space and time) grid cells that the observations spread over. In addition, we also evaluated the impact of this assimilation to improve ozone forecast skills. The potential of improving air quality forecasts with chemical data assimilation was demonstrated by a case where the near surface ozone predictions better match observations out to 48 hours after assimilation. Specifically we analyzed the ozone observations on-board the Ron Brown. This data was not used in the data assimilation process so it also serves as a data set for evaluation of the model predictions. Figure 4 shows the UV absorbance ozone measurements and the
G. Carmichael et al.
488
corresponding model predictions by the base case and Case 9. In the data assimilation time window (the first 20 hours), the UV absorbance ozone measurements serve as another validation data set. It is clearly seen that after assimilation the model predictions for the Ron Brown better match the observations. The model using the new initialization caused as the result of the data assimilation during the 20 hour assimilation window, was then run in forecast model for 48 hours. Results show that the positive effect of assimilation continues beyond the data assimilation time window (even out to 48 hours). This demonstrates the potential of improving air quality forecasts by chemical data assimilation. 3.2.2. Chemical Data Assimilation Using Ensemble Kalman Filters
The ensemble Kalman filter (EnKF) approach to data assimilation has recently received considerable attention in meteorology. The Kalman filter (Kalman, 1960; Evensen and van Leeuwen, 2000) works under the assumptions that the model is k 1 is normally distributed linear, and the model analysis state at previous time t k 1 k 1 with mean ya and covariance matrix Pa . The Extended Kalman Filter (EKF) allows for nonlinear models and observations by assuming the error propagation is linear (through the tangent linear model) and by linearizing the observation k k k operators, yobs = H k y H obs . However, the (extended) Kalman Filter is impractical for large systems due to the high cost of propagating covariance matrices. A practical approach is provided by the ensemble Kalman Filter (EnKF), which estimates covariances through sampling the state space.
Fig. 4 UV absorbance ozone measurements on NOAA research vessel Ronald H. Brown and corresponding model predictions by the base case (left) and Cast 9 (right). Measurement uncertainties (±(2% + 1 ppbv)) are shown as error bars. Model prediction uncertainties are given as (±10 ppbv), indicted by a band width of 20 ppbv. The assimilation window is the first 20 hours; the forecast period is the next 48 hours k 1
k 1
Consider an ensemble of N states { ya [i]}1di d N at t . Each of the ensemble states is evolved in time using model equation to obtain a forecast k ensemble at t ,
Predicting Air Quality: Current Status and Future Directions
489
y kf >i @ = M p, yak 1 >i @ , 1 d i d N
(3)
The mean and the covariance of the forecast PDF are approximated by the ensemble statistics:
y kf =
1 N
N
¦ y kf >i@, Pfk i =1
i, j
=
1 N k ¦ y f >i@ y kf N 1 i =1
y >i@ k f
y kf
T
(4)
k
An ensemble of observation vectors { yobs [i ]}1di d N is constructed by adding to k the most recent observation vector yobs perturbations drawn from a normal distribution with zero mean and covariance Ok . Each member of the ensemble is k assimilated using the EKF to obtain the ensemble of analyzed states { ya [i ]}1di d N :
yak >i @ = y kf >i @ Pfk H kT Ok H k Pfk H kT
y >i @ H y >i @ 1
k obs
k
k f
(5)
The ensemble mean and covariance describe the PDF of the assimilated field. The cost of updating the covariance matrix is that of N model evaluations. The ensemble implicitly describes a density function that can be non-Gaussian. Experience gained in numerical weather prediction indicates that relatively small ensembles (50–100 members) are sufficient to accurately capture this density function. Extensions of this approach proposed in the literature include the Ensemble Kalman Smoother (Evensen and Leeuwen, 2000), the 4D-EnKF method, the Ensemble Transform Kalman Filter, the hybrid approach and ensemble nonlinear filters. The application of EnKF presents several challenges: (1) the rank of estimated covariance matrix is (much) smaller than its dimension; (2) the random errors in the statistically estimated covariance decrease only by the square-root of the ensemble size; (3) the subspace spanned by random vectors for explaining forecast error is not optimal; and (4) the estimation and correct treatment of model errors is possible but difficult. In addition, a careful implementation is required for efficiency. In spite of these challenges, EnKF has many attractive features including: (1) it is able to propagate the PDFs through highly nonlinear systems; (2) it does not require additional modeling efforts such as the construction of tangent linear model and its adjoint; and (3) the method is highly parallelizable. The performance of EnKF applied to chemical data assimilation has recently been reported (Sandu et al., 2005; Constantinescu et al., 2007a, b) for the ICARTT study discussed above (Figure 3). The observations used for data assimilation are ground-level ozone (O 3 ) measurements taken by the 340 EPA AirNow stations. These observations are available hourly in the assimilation window (0–23 EDT, July 20, 2004). After assimilation the model is allowed to evolve in forecast mode for another 24 hours. Results for the EnKF are shown in Table 1. For these calculations the “perturbed observations” implementation of the filter was employed. EnKF adjusts the concentration fields of 66 “control” chemical species in each grid point of the domain every hour in the assimilation window. The ensemble size was chosen to be 50 members to provide a good balance between accuracy and computational efficiency. An autoregressive model of background errors was used, which accounted for spatial correlations, distance decay,
490
G. Carmichael et al.
and chemical lifetime. The “textbook application” of EnKF to this particular scenario lead to filter divergence and a decreasing ability towards the end of the assimilation window. We explored several ways to “inflate” the ensemble covariance in order to prevent filter divergence were investigated. These included: additive inflation (addition of uncorrelated noise to model results), multiplicative inflation (each member’s deviation from the ensemble mean is multiplied by a constant), and model-specific inflation (obtained through perturbing key model parameters like the wind field velocities, boundary conditions, and emissions). The results found that model-specific inflation best preserves the correlations between various chemical species. Illustrative results are shown in Table 1. Shown are results in the analysis window for initial conditions and for joint state (initial and boundary conditions and emissions) analysis. The performance of each data assimilation experiment is measured by the R 2 correlation factor. The correlation between the observations 2 and the model solution in the assimilation window is R = 0.24 for the non2 assimilated solution, R = 0.52 for 4D-Var (results not shown), and 2 R | 0.8 0.9 for EnKF (with various forms of covariance inflation and localization). The time evolution of ozone concentrations at selected ground stations (not shown) shows how the assimilated ozone series follow the observations much closer than the non-assimilated ones in the analysis window. The impact of data assimilation on the forecast skill is also shown in Table 1. The period from 24 to 48 hours represents the forecast. The forecast skill is increased; R2 increased from 0.28 for no-assimilation to 0.34–0.42 for the assimilation cases. The effect of assimilation of surface ozone on forecast improvements at particular sites is mixed. At some stations the effects are significant, while at others the effects are slight. This is due in part to the fact that only ground level observations are assimilated and the vertical profiles are not constrained at all. This is also due to the fact that near surface ozone levels are strongly dependent on chemical production/destruction processes involving a variety of precursor species. Table 1 Model-observations agreement (R2 and RMS [ppbv]) for the EnKF data assimilation of only the initial state and of joint state (ST), emissions (EM) and lateral boundary conditions (BC) parameters. Visible improvements in both analysis and forecasts are obtained.
Predicting Air Quality: Current Status and Future Directions
491
4. Further Improvements in Forecasting One of the main differences between weather and chemical-weather forecasting is the strong local forcing due to emissions, which as discussed earlier are typically the largest source of uncertainty in the predictions. Thus an increase in air quality prediction skill requires better estimates of emissions. This is challenging due to the complexity and uncertainties associated with the bottom-up emission estimates (i.e., inventories built on activity information, e.g., emission factors, fuel use and type, control technologies, and their regional variation), the transient nature of some emissions (fires and dust storms), and their ever changing nature (e.g., trends due to policy and/or technology changes) (Streets et al., 2006). Improvements in emissions also will come from the closer integration of observations and models. The same data assimilation techniques discussed above can encompass emission estimates. (Pan et al., 2007). Another direction that is promising is to incorporate emissions as control variables, along with initial conditions and with boundary conditions, in the assimilation cycle. Unlike the traditional inverse modeling approach where the emissions are adjusted and then used in subsequent model runs, in this manner the emissions are adjusted each assimilation cycle. The importance of including the emission estimates as a control variable in order to improve prediction skill in air quality has been reported by Elbern et al. 2000, where they show a marked improvement in air quality forecast skill when emissions and initial conditions are simultaneously treated as controls. As we have discussed throughout this paper, improved predictions require a closer integration of measurements with models. The weather forecast system is supported by a comprehensive observing system designed to improve forecasting skill. No such system exists to support air quality forecasts. The chemical observations presently available were designed largely for environmental compliance and not to enhance predictive skill. However that opens the question as to what chemical data is needed to improve the predictions? The chemical data assimilation techniques can be used to help address this issue as well (Daescu and Carmichael, 2003; Daescu and Navon, 2004).
5. The Road Forward The importance of air quality prediction in the management of our environment continues to grow. The recent developments in atmospheric chemical observations and modeling are also leading to more effective linkages of air pollution issues on different scales from urban to global. For example it is now recognized that in urban air pollution studies it is important to consider also the regional and global contributions, and in global air pollution, the effects of megacities and hemispheric transport. In addition, the emergence of chemical weather forecasting as an important activity places greater importance on linking pollution detection and prediction
492
G. Carmichael et al.
capabilities. The close integration of observational data is recognized as essential in weather/climate analysis and forecast activities, and this is accomplished by a mature experience/infrastructure in meteorological data assimilation. Borrowing lessons learned from the evolution of numerical weather prediction (NWP) models, improving air quality predictions through the assimilation of chemical data holds significant promise. As more atmospheric chemical observations become available chemical data assimilation is expected to play an essential role in air quality forecasting, similar to the role it has in NWPs (and may benefit weather forecasting as well). Advances in our predictive capabilities will require a better matching of the observational capabilities of the community with chemical weather forecast needs. This will require closer interactions between the observing and the modeling communities. One important activity will be the use of chemical data assimilation systems to help design the observing systems needed to produce better forecasts. We need to rigorously quantify the value-added to a forecast by: adding observations of additional species; extending surface coverage; including observations above the surface; and enhancing observations from satellites. Advances will also require a growth in activities related to chemical data assimilation techniques and algorithms. While there is much to build upon from the expertise and experiences in assimilation of weather, there are significant differences and challenges related to chemical weather. As we have illustrated in this paper 4DVar and EnKF are powerful techniques, and there are exciting possibilities in combining their strengths in hybrid data assimilation methods. However, there is relatively little experience in applying modern data assimilation techniques to real atmospheric chemistry problems, and much work needs to be done before their true impact on air quality prediction is felt. Furthermore, feedbacks between the meteorological and air quality components – which have mostly been studied as separate systems – are also critical to improve AQ forecasting. Many challenging and important questions remain to be addressed; including: What is the relationship between mixing depth heights and near surface concentrations? What is the role of ambient aerosols in influencing the surface energy budgets and in altering the moisture fields via cloud interactions? How do these feedbacks impact weather and AQ forecasts? To what extent will the assimilation of chemical data lead to improvements in weather forecasting? Sensitivity studies are needed to quantify these feedbacks, which in turn can help prioritize future research efforts. However, this will require a closer integration of meteorological and air quality models, and ultimately the evolution to a tightly coupled combined meteorological and air quality forecasting and data assimilation systems. These aspects are being explored in projects such as the European Union Global and Regional Earth-System (Atmosphere) Monitoring using Satellite and in-situ data (GEMS) project (reference). The integration of enhanced observing systems with modeling tools for use in air quality and climate change is a priority area within the Global Earth Observing System of Systems (GEOSS) and the Integrated Global Atmospheric Chemistry Observations (IGACO) (reference) frameworks. IGACO is a highly focused strategy for bringing together ground-based, aircraft
Predicting Air Quality: Current Status and Future Directions
493
and satellite observations using atmospheric forecast models that assimilate not only meteorological observations but also chemical constituents. Finally, the requirements to use CTMs for urban to global scale applications, to couple chemistry with weather and climate, and to incorporate data assimilation, place even more demands for computational efficiency and accuracy. The growing importance of chemical weather forecasting to society should help stimulate significant advances in the field over the next decade. Acknowledgments This work has been supported by NSF though the award ITR AP&IM 0205198, and by grants from NOAA Global Change program and NASA. The work of A. Sandu and E.M. Constantinescu has also been supported in part by NSF CAREER ACI-0413872, NSF CCF-0515170, and by the Houston Advanced Research Center (HARC) awards H45C and H59.
References Chai T, Carmichael GR, Sandu A, Tang YH, Daescu DN (2006) Chemical data assimilation of transport and chemical evolution over the pacific (TRACE-P) aircraft measurements. J. Geophys. Res., 111(D02301), doi:10.1029/ 2005JD005883. Chai T, Carmichael GR, Sandu A, Hardesty M, Pilewskie P, Whitlow S, Browell E V, Avery MA, Thouret V, Nedelec P, Merrill JT, Thompson AM (2007) Four dimensional data assimilation experiments with ICARTT (International Consortium for Atmospheric Research on Transport and Transformation) ozone measurements. J. Geophys. Res., 112, D12515, doi:10.1029/2006JD007763. Constantinescu EM, Sandu A, Chai T, Carmichael GR (2007a) Ensemble-based chemical data assimilation. I: general approach. Quart. J. Roy. Met. Soc., 133, 1229–1243. Constantinescu EM, Sandu A, Chai T, Carmichael GR (2007b) Ensemble-based chemical data assimilation. II: covariance localization. Quart. J. Roy. Met. Soc., 133, 1245–1256. Dabberdt WF, Carroll MA, Baumgardner D, Carmichael G, Cohen R, Dye T, Ellis J, Grell G, Grimmond S, Hanna S, Irwin J, Lamb B, Madronich S, McQueen J, Meagher J, Odman T, Pleim J, Schmid HP, Westphal DL (2004) Meteorological research needs for improved air quality forecasting – report of the 11th prospectus development team of the us weather research program. Bull. Amer. Meteorol. Soc., 85(4):563+. Daescu DN, Carmichael GR (2003) An adjoint sensitivity method for the adaptive location of the observations in air quality modeling. J. Atmos. Sci., 60(2):434– 450. Daescu DN, Navon IM (2004) Adaptive observations in the context of 4D-Var data assimilation. Meteorol. Atmos. Phys., 85(4):205–226.
494
G. Carmichael et al.
Elbern H, Schmidt H, Talagrand O, Ebel A (2000) 4d-variational data assimilation with an adjoint air quality model for emission analysis. Environ. Mod. Software, 15:539–548. Evensen G, van Leeuwen PJ (2000) An ensemble kalman smoother for nonlinear dynamics. Mon. Wea. Rev., 128(6):1852–1867. Fisher M (2003) Background error covariance modelling. In Proceedings of the ECMWF Workshop on Recent Developments in Data Assimilation for Atmosphere and Ocean, Reading, UK. Grell GA, Peckham SE, Schmitz R, McKeen SA, Frost G, Skamarock WC, Eder B, Petron G, Granier C, Khattatov B, Yudin V, Lamarque JF, Emmons L, Gille J, Edwards DP (2005) Fully coupled “online” chemistry within the wrf model. Geophys. Res. Lett., 39(37):6957–6975. Jacob DJ, Crawford JH, Kleb MM, Connors VS, Bendura RJ, Raper JL, Sachse G W, Gille JC, Emmons L, Heald CL (2003) Transport and chemical evolution over the pacific (trace-p) aircraft mission: design, execution, and first results. J. Geophys. Res., 108(D20):1–19. Kalman RE (1960) A new approach to linear filtering and prediction problems. Trans. ASME, Ser. D: J. Basic Eng., 82:35–45. Kalnay E (2003) Atmospheric modeling, data assimilation, and predictability. Cambridge University Press, Cambridge/New York. McKeen SA, Wilczak J, Grell G, Djalalova I, Peckham S, Hsie E, Gong W, Bouchet V, Menard S, Moffet R, McHenry J, McQueen J, Tang Y, Carmichael GR, Pagowski M, Chan V, Dye T, Frost V, Lee, P, Mathur R (2005) Assessment of an ensemble of seven real-time ozone forecasts over eastern North America during the summer of 2004. J. Geophys. Res., 110(D21):Art. No. D21307, December 2005. Menut L, Vautard R, Beekmann M, Honore C (2000) Sensitivity of photochemical pollution using the adjoint of a simplified chemistry-transport model. J. Geophys. Res., 105(D12):15379–15402. Pan L, Chai T, Carmichael GR, Tang Y, Streets D, Woo J, Friedli HR, Radke LF, Top-down estimate of mercury emissions in China using four-dimensional variational data assimilation (4D-Var). Atmos. Environ. (in review). Sandu A, Constantinescu EM, Liao WY, Carmichael GR, Chai TF, Seinfeld JH, Daescu D (2005) Ensemble-based data assimilation for atmospheric chemical transport models. In Computational Science – ICCS 2005, Pt 2, volume 3515 of Lecture Notes in Computer Science, pp. 648–655. Springer-Verlag Berlin,. Streets DG, Zhang Q, Wang L, He K, Hao J, Wu Y, Tang Y, Carmichael G (2006) Revisting China’s CO emissions after transport and chemical evolution over the pacific (TRACE-P): synthesis of inventories, atmospheric modeling and observations. J. Geophys. Res., 111(D14):Art. No. D14306. Tang YH, Carmichael GR, Thongboonchoo N, Chai T, Horowitz LW, Poerce RB, Al-Saadi JA, Pfister G, Vukovich M, Avery MA, Sachse GW, Ryerson TB, Holloway JS, Atlas EL, Flocke FM, Weber RJ, Huey LG, Dibb JE, Streets DG, Brune WH (2006) The influence of lateral and top boundary conditions on regional air quality prediction: a multi-scale study coupling regional and global chemical transport models. J. Geophys. Res. (in review).
Predicting Air Quality: Current Status and Future Directions
495
Discussion A. Baklanov: What are your future plans with data-assimilation and when would it be operational? G. Carmichael: We are continuing to develop various data assimilation tools. We are continuing with the 4dvar approach, where we are focusing on in-situ and satellite observations for both forecast and inverse modeling of emission applications. We are working to provide ad joints for CMAQ as well as WRF/Chem applications. A first version of the CMAQ adjoint is now available and the WRF/Chem adjoint we hope to have in one to two years. We are also working on ensemble Kalman filter methods as well as hybrid methods.
5.4 Regional Coverage Modelling of Marine Aerosols Concentration in French Mediterranean Coastal Area Romain Blot, Gilles Tedeshi and Jacques Piazzola
Abstract The present study focuses on the extension of the predictions of the Mediterranean coastal aerosol model (based on the Navy Aerosol Model) to a regional scale, using a mesoscale meteorological model to better take into account the details of the topography and shoreline of the coast for calculations of the wind field and the fetch. The aerosol and the meteorological models are coupled and the spatial variation of the aerosol size distributions is determined in the whole study area. The results show a non-homogeneous spatial coverage of the aerosol concentrations over the northern Mediterranean, with wake due to the shoreline for a continental wind. Another dataset recorded in the Mediterranean in 1995 is used to validate the coupling for steady conditions. The results are found in correct agreement. A discussion is then held about the influence of unsteady conditions on the aerosol concentration is coastal zone.
Keywords: Aerosols, coastal area, fetch, rams, sea salt
1. Introduction The coastal area induces particular specificities for aerosol properties. Inshore and offshore aerosol sources, sinks and surface properties are quite different. At either side of the shoreline the aerosol properties will change gradually due to the mixture of continental and marine aerosol and depending on the direction the wind is blowing from, the fetch length, the sea state…. The aerosol concentration in the atmosphere is thus very variable both in time and in space. Very few relevant models for the aerosol size distributions were published during the last decades. One of the most used is the Navy Aerosol Model, NAM (Gathman, 1983) which provides the particle size distribution at a height of 10 m above sea level. Although this model gives reasonable predictions over the open ocean, experimental evidences show that it is often less reliable in coastal regions (Piazzola et al., 2000). To include coastal effects in the model for the prediction of aerosol concentrations, (Piazzola et al., 2003) proposed an extension of the Navy
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
514
Regional Coverage Modelling of Marine Aerosols Concentration in French Mediterranean
515
Aerosol Model to coastal areas. It was based on an extensive series of measurements in a Mediterranean coastal zone. The present paper focuses on the extension of the predictions of the Mediterranean coastal aerosol model to a regional scale, using a mesoscale meteorological model, RAMS (Cotton et al., 2003). It allowed to better take into account the details of the topography and shoreline of the coast for calculations of the wind field and the fetch, which are important inputs of the aerosol model. First, this paper presents the coupling between RAMS and the aerosol model into the study area. The spatial variation of the modelled aerosol size distributions is then determined in the whole study area. Another dataset recorded in the Mediterranean in 1995 (Piazzola and Despiau, 1998) is used to validate the coupling for steady conditions. As the coupling provides least performance for unsteady conditions compared to steady cases, a discussion is held about the influence of such conditions.
2. Field Site and Instrumentation The study area is the Toulon-Hyères bay (Figure 1) located on the French Riviera. The region is exposed to air masses coming from the open sea and to air masses coming from the European mainland, which case corresponds to continentally polluted conditions.
Fig. 1 Detailed view of the study area (the opened circles represent the locations of the aerosol concentrations measurements of the 1995 dataset)
The Mediterranean coastal aerosol model is based on an extensive series of measurements which were recorded on the island of Porquerolles during the 2000 and 2001 years. The measurement station was located west of the Porquerolles Island (Figure 1). The stations were equipped with meteorological sensors which were fixed at the top of a 10 m height mast and recorded wind speed and direction, air temperature, relative humidity and pressure. In addition, optical counters allow aerosol size distribution measurements. Size distributions in the 0.21–42.5 µm range were obtained using two Particle Measuring Systems: the classical scattering spectrometer probes CSASP-200 and CSASP-100HV. For validation of the model, a second dataset was used which was previously recorded on a French Navy ship “Albacore” in different locations of the Toulon bay in 1995 (see Figure 1) by Piazzola and Despiau (1998). The aerosol concentrations in the 0.1–20 µm size ranges were recorded using two Particle Measuring Systems, the active scattering spectrometer probe (ASASP) and the classical scattering spectrometer probe (CSASP).
516
R. Blot et al.
3. Aerosol and Meteorological Models As said previously, the Mediterranean coastal aerosol model is based on measurements recorded during 2000 and 2001. This extended period allowed observation of a large variety of aerosol size distributions for different meteorological conditions. The experimental data from Porquerolles were statistically analyzed to develop an empirical coastal aerosol model. The Mediterranean coastal aerosol model is a modification of the Navy Aerosol Model (NAM). The particle size distribution dN ( r ) dr is calculated as the sum of modified lognormal functions, but the coefficients of the various modes are parameterized as functions of not only the wind speed but also the fetch (Piazzola et al., 2000). As stated above, the wind speed and the fetch are the main inputs of the coastal aerosol model (with the relative humidity). The wave field growth depends on the duration of the wind and the fetch. In practice, the waves conditions are either fetch limited or duration limited. Steady state of the wave field correspond to fetch limited conditions for which we assume that the wind have blown constantly long enough in the same direction for wave heights at the end of the fetch to reach equilibrium (Hsu, 1986). These conditions include constant wind speed and direction, i.e., a steady state of the meteorological data. The occurrence of unsteady conditions and their implications on the predictions of the model are specifically commented in the paragraph 5. RAMS is a meteorological model used for numerical simulation of atmospheric processes. It employs multiple grid nesting and resolves (among others) the equations of motion, heat, moisture and mass using a staggered Arakawa-C grid and a terrain-following coordinate system. The kind of soil, of surface and variable SST are taken into account. For boundary conditions a 4DDA is used, allowing the atmospheric fields to be nudged toward large-scale data, as well as to local stations or radiosonde data. The topography and the vegetation cover (variable surface characteristics) used are both issued from the United States Geological Survey (USGS) 30” high resolution model and fitted to the different grids. RAMS has been already used for the simulation of local winds in the study area (Pezzoli et al., 2004; Guenard et al., 2006) and the results were found in agreement with experimental data. The model RAMS has been then implemented in the study area, and topography and surface type were fitted to the grids. Two two-way nested grids have been used, with 4 km (grid 1) and 1 km (grid 2) horizontal resolutions, covering respectively a 250 × 202 km and a 600 × 320 km over the French Mediterranean coast. A nudging was made every 6 hours using the ECMWF reanalyzed pressure levels data as well as the experimental recordings acquired on the Porquerolles Island to account for the very local coastal influence. Time steps for numerical integration were 4 s for grid 1 and 1 s for grid 2. As an example, the Figure 2 shows the wind field (10 m above surface level) simulated by the meteorological model at 1200 UTC 17 November 2000 for the finest grid. For a clearer view, only 5% of the vectors have been plotted. This case corresponds to a high intensity wind blowing from northwest to west direction on
Regional Coverage Modelling of Marine Aerosols Concentration in French Mediterranean
517
the study area, which induces short fetch conditions for the aerosol measurements recorded in Porquerolles.
Fig. 2 Wind field simulated by RAMS for the northwest wind episode, 1200 UTC 17 November 2000
The validation of the simulation was made with wind data acquired at the Porquerolles station. As these data have been already used for the nudging, this is not a “‘real” validation (which is not the aim of the study), but rather a check that the meteorological field can be used accurately for input in the coastal aerosol model. 20
360 340
Exps RAM S
16 14 12
320 Wind direction ( °)
Wind speed (m/s)
18
10 8 6 4 2 0 17/11/2000 0000 UTC
300 280 260
RAMS Exps
240 220 200
17/11/2000 1200 UTC
18/11/2000 0000 UTC Time
18/11/2000 1200 UTC
19/11/2000 0000 UTC
180 17/11/2000 0000 UTC
17/11/2000 1200 UTC
18/11/2000 0000 UTC
18/11/2000 1200 UTC
19/11/2000 0000 UTC
T ime
Fig. 3 Comparison between simulated and experimental wind speed and direction at Porquerolles station location (17–18 November 2000)
4. Simulation of the Aerosol Concentration One of the interests of coupling the meteorological model with the aerosol model is to be able to re-calculate the inputs (wind and fetch) of the aerosol model for each cell of the grid, i.e. to obtain a spatial distribution. This is particularly interesting in a view of the determination of a relevant spatial cover of aerosols at mesoscale. The meteorological model also allows determination of the fetch conditions (duration limited, fetch limited conditions or fully developed sea), as various fetches can be found for different wind directions depending on the location of the study area. Two cases are presented here for simulation/experiment comparisons. The first one (at Porquerolles Island, 0000 UTC 18 November 2000) is part of the dataset used for the development of the Mediterranean coastal aerosol model. Even if the dataset covers a several months period and thus the model is not especially fitted to this short time case, it has been found preferable to use a second test case. This
R. Blot et al.
518
latter corresponds to measurements made for two locations in the Toulon bay, the 16 May 1995. 1,E+04
1,E+04
1,E+03
1,E+03 Exps Model
1,E+02
1,E+01 -1
1,E+00 1,E-01
-3
1,E-01
Exps Model
1,E+02
dN/dr(cm Pm )
-1
1,E+00
-3
dN/dr(cm P m )
1,E+01
1,E-02 1,E-03 1,E-04
1,E-02 1,E-03 1,E-04
1,E-05
1,E-05
1,E-06
1,E-06
1,E-07
1,E-07
1,E-08
1,E-08
0,1
1,0
10,0 D (Pm)
100,0
0,1
1
10
100
D(Pm)
Fig. 4 Comparison between simulated and measured aerosol size distributions in Porquerolles island (18 November 2000, left side) for a 25 km fetch and for a wind speed of 10 m/s and at two different locations in the Toulon bay (16 May 1995, right side) during a northwest wind of 7 m/s
The Figure 5 shows the simulation of the aerosol concentration in the study area for a constant wind blowing from the northwest direction at 1400 UTC 17 November 2000. The analysis was focused on particles of 5 and 1 µm diameter. These sizes were chosen because the behavior of these particles is representative for the marine and continental contributions to the aerosols, respectively. In the present analysis they can be used as tracers for the influence of the fetch on the production of marine aerosol and on the deposition of the continental particles.
Fig. 5 Spatial coverage of the 5 µm (left) and 1 µm (right) particle concentrations for a wind from NW direction and a velocity of 15 m/s at the Porquerolles station, 1400 UTC 17 November 2000
We can notice a non-homogeneous concentration field which is the consequence of the non-homogeneous wind field near the Mediterranean coast and more especially the consequence of the variation of the fetch length along the southern coast. This induces a spatial gradient of aerosol concentrations along the coasts. In particular, we can see the “sheltering” character of the coasts resulting in the occurrence of a “wake” which corresponds to lower aerosol concentrations.
Regional Coverage Modelling of Marine Aerosols Concentration in French Mediterranean
519
5. Influence of Unsteady Conditions The present modelling is limited by the occurrence of unsteady conditions of the wave field and its effect on the production of marine aerosols. The wave field is in steady conditions when equilibrium is reached with the wind input. Before (or after) this state of equilibrium, the amplification (or attenuation) of the wave energy corresponds to unsteady conditions. In this case, the whitecap coverage, and hence the sea spray production, can be small even for high wind speeds before the equilibrium between waves and wind is reached. In the same way, the aerosol generation can be large for low wind speeds during a period of wave attenuation. The time for the waves to reach equilibrium with the wind (fetch limited conditions) can be estimated using the criterion for duration-limited growth proposed by Hsu (1986):
g t / U10
2 68.8 gX U10
23
(1)
where t is the minimum duration for the waves to be in equilibrium with the wind, X the fetch length and U10 the wind speed measured at a 10 m height. In order to analyse the error due to the unsteady conditions, a comparison is made between the temporal variation of the measured and the computed aerosol concentrations. On Figure 6 is plotted the error of the model for the aerosol concentrations (5 µm particles) recorded at Porquerolles island just after a change in the wind direction and intensity. The fetch length is 25 km. 25
Error (%)
20 15 10 5 0 17/11/00 17/11/00 17/11/00 17/11/00 17/11/00 17/11/00 17/11/00 18/11/00 18/11/00 10:00 UTC 12:00 UTC 14:00 UTC 16:00 UTC 18:00 UTC 20:00 UTC 22:00 UTC 0:00 UTC 2:00 UTC Time
Fig. 6 relative errors between computed and recorded aerosol concentrations measured at Porquerolles island the 17 November 2000
Three temporal periods can be identified (see also Figure 3): – Between 0000 UTC 17 November 2000 and 1200 UTC, the wind direction is changing from NW to W while the wind speed is increasing. The relative error is high (more than 20%) around 1100 UTC and decreasing continuously until 1400 UTC, as the atmospheric conditions are becoming more stable (see below). – Between 1200 UTC and 1700 UTC, the atmospheric conditions are roughly stable: the direction is W and the speed has reached its maximum. This induces a minimal error (less than 10%) with a 2 hours shift (around 1400– 1800 UTC).
R. Blot et al.
520
–
After 1600 UTC, the direction is changing again towards NW while the wind speed is decreasing continuously. The error is then increasing with a relative maximum (15%) 3 hours later (around 1900–2000 UTC). We can notice that the time to reach the minimal error is in accordance with the time for the waves to be in fetch limited conditions (3 hours for U10 | 15m / s ), as reported in Eq. (1). A more statistical study has been held. For five different modes (see Table 1) corresponding to diameters between 0.2 and 20 Pm, the accuracy of the model has been checked for steady-state cases only and for all cases (including unsteady conditions). For that, the parameter corresponding to a 68% confidence interval:
1
n
V
(2)
m
has been computed, where n is the number of records for a specific condition, m the mean value of records and V the standard-deviation. Table 1 Model accuracy (Eq. (2)) for a 68% confidence interval and for different particle sizes: comparison between steady conditions and all conditions (including unsteady). D (µm) 0.2 0.5 5 15 20
Steady conditions only 2.15 1.48 1.31 1.42 1.86
All conditions 2.80 1.70 1.49 1.50 2.00
The model accuracy is better for the second and the third modes (0.5 and 5 Pm particles) for the two cases and the largest error is found for the first mode (0.2 Pm particles). For steady-state cases, the model predicts the concentration of 0.5 µm and 5 µm particles within a factor of 1.5 and 1.3, respectively (with a 68% confidence level), which can be considered as a very valuable performance. As expected, when taking unsteady conditions into account the model performance is decreasing, especially for the smaller diameters for which the aerosols travel in the atmosphere is long. Their concentration depends then on atmospheric conditions as they were several hours sooner, that can be different from actual (at recording time) conditions. On the contrary, large particles do not stay a long time in the atmosphere and are less influenced by unsteady conditions.
6. Summary and Conclusion The present paper deals with the extension of the modelling of the Mediterranean coastal aerosol model to a regional scale. It has been coupled with a mesoscale meteorological model (RAMS) to take into account the details of the topography of the coast and the shoreline. The simulations have been validated using two different datasets recorded in the Mediterranean. The results show a good agreement between the modelled and the recorded aerosol concentrations at different locations
Regional Coverage Modelling of Marine Aerosols Concentration in French Mediterranean
521
of the study area, under steady-state conditions. A non-homogeneous spatial coverage of the aerosol concentration has been found. In particular, a “sheltering effect” of the coast has been noticed. The present work, resulting in a 2D-space simulation of the aerosol concentration is of great interest for electro-optical apparatus performance. The least agreement observed between the modelled and the recorded aerosol concentration is mainly due to the occurrence of unsteady conditions of the wave field and its effect on the production of marine aerosols. Calculation of the model accuracy (for a 68% confidence interval) shows that the model predicts the concentration of the 5 µm diameter particles within a factor of 1.3 if only the data recorded during steady conditions are selected, whereas this factor is 1.5 for the whole dataset (including unsteady conditions). This confirms the steady-state character of the Mediterranean coastal aerosol model. It has been seen that the time to reach the minimal error between the modelled and the recorded data corresponds to the time for the waves to reach an equilibrium with the wind, and then to the time necessary for the aerosol concentrations to be in equilibrium. The relevance of the criterion proposed by Hsu for the determination of this time delay is confirmed by our data. One of the future objectives of this study will be to determine if it is possible to take unsteady conditions into account.
References Cotton WR et al. (2003) RAMS 2001: current status and future directions, Meteor. Atmos. Phys., 82, 5–29. Gathman SG (1983) Optical properties of the marine aerosol as predicted by the Navy aerosol model, Opt. Eng., 22, 57–62. Guenard V, Drobinski P, Caccia JL, Tedeschi G, Currier P (2006) Dynamics of the MAP IOP-15 severe Mistral event: observations and high-resolution numerical simulations, Quart. J. R. Met. Soc., 132, 757–777. Hsu SA (1986) A mechanism for the increase of wind stress (drag) coefficient With wind speed over water surfaces: a parametric model, J. Phys. Oceanogr. 16, 144–150. Pezzoli A, Tedeschi G, Resch F (2004) Numerical simulation of strong wind situations near the Mediterranean French coast: comparison with FETCH data, Appl. Meteorol., 43, 7, 997–1015. Piazzola J, Despiau S (1998) The vertical variation of extinction and atmospheric transmission due to aerosol particles close above the sea surface in Mediterranean coastal zone, Opt. Eng., 22, 57–62. Piazzola J, Van Eijk AMJ, De Leeuw G (2000) An extension of the Navy Aerosol Model to coastal areas, Opt. Eng., 39, 1620–1631. Piazzola J, Bouchara F, Van Eijk AMJ, De Leeuw C (2003) Development of the Mediterranean extinction code MEDEX, Opt. Eng., 42, 4, 912–924.
5.6 The Origins and Formation Mechanisms of Aerosol during a Measurement Campaign in Finnish Lapland, Evaluated Using the Regional Dispersion Model SILAM Marje Prank, Mikhail Sofiev, Marko Kaasik, Taina Ruuskanen, Jaakko Kukkonen and Markku Kulmala
Abstract This paper is intended to clarify the geographical extent of processes leading to a nucleation event and the role of atmospheric transport in it. The study is based on the inverse (adjoint) runs of atmospheric advection-diffusion model SILAM and general knowledge on basic mechanisms and time scales of nanometer particle formation in the atmosphere. Results of an aerosol measurement campaign carried out in Värriö, Finland, Eastern Lapland, April–May 2003, were used as sensitivity source data for backward tracing. The footprint areas of three observed nucleation events suggest that (1) spatial scale of a nucleation event may reach about 1,000 km, and (2) impact of atmospheric transport to the aerosol processes recorded by a Eulerian (ground-based) observer may be significant. Formation of an intense event over extensive forested areas supports the theory on the role of biogenic VOC emissions. Need for coupling the models of atmospheric transport and aerosol dynamics was stressed.
Keywords Adjoint model, atmospheric transport, backward tracing, nucleation burst
1. Introduction Most of observations and modelling studies of aerosol formation are based on homogeneous volume assumption, i.e., it is supposed that properties of the passing air mass do not vary substantially during the observation of a certain event at fixed measurement point. However, the meteorological conditions can substantially change on a time-scale of hours, or in specific cases, on a time-scale of minutes. The chemical composition of air is also changed when it is transported over different source and sink areas, and the pollutants can be scavenged by precipitation. Based only on limited experimental data, it is commonly not possible to deduce, whether such changes have possibly occurred recently. This can potentially lead to serious
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
530
The Origins and Formation Mechanisms of Aerosol during a Measurement Campaign
531
misinterpretations of aerosol dynamics. Thus, there is an obvious need to include detailed atmospheric dynamics into aerosol studies. Although atmospheric nucleation is obviously almost permanently present in the atmosphere, specific conditions and presence of necessary aerosol precursors are needed for nucleation bursts (e.g. Kulmala et al., 1998), when a large number of nanometer-size particles is produced during a short time period. After that newly formed particles grow due to condensation of vapours and reach the size of Aitken mode (30–100 nm) within a few hours. Although precise mechanism and all conditions for the nucleation burst are not known, a low concentration of large aerosol particles (i.e. low condensation sink) is typically necessary (see e.g. Kulmala et al., 2005). Thus, nucleation events occur predominantly during dry and sunny weather in presence of a clean air mass and sufficient solar radiation. Due to biogenic emissions the boreal forest is an important source of fine aerosol (e.g. Tunved et al., 2006). It is found that nucleation events in boreal forest occur even in late winter with snow cover, as life activities in evergreen tree crowns appear with sufficient solar radiation (Kulmala et al., 2004). This paper is intended to clarify the geographical extent of processes leading to nucleation burst and the role of atmospheric transport in it. The basic tools for that are (1) an atmospheric advection-diffusion model, and (2) results of an aerosol measurement campaign as the input data for that model. rw
ay
Murmansk Nikel Monchegorsk Apatity Värriö
en
No
Sw
ed
Russia
Finland
St. Petersburg
Helsinki Tallinn Estonia
Fig. 1 Location map: empty circles – major cities; small filled circles – major industrial pollution sources (metal smelters); cross – measurement location
We have applied the source apportionment modelling techniques based on adjoint formalism in order to trace the air masses backward from the receptor point and to follow the dispersion of potentially aerosol-forming species from anthropogenic and natural sources. The adjoint modelling techniques have substantial advantages compared to the previously more widely applied trajectory analyses. Atmospheric dispersion and removal processes can be properly taken into account – this also
532
M. Prank et al.
facilitates the separate modelling of various aerosol size modes. Solution of the adjoint dispersion equation provides a prediction of the geographic areas, where the sources responsible for the observed concentrations are located, and quantitative estimates of their contributions to the measured concentrations. In case of trajectory analyses, merely a qualitative picture of the transport path of a single air parcel can be modelled, as driven by prevailing airflows. The trajectories for various particle size modes would be identical. Trajectory analyses also do not yield any information on the most probable source areas as a function of distance along the trajectory.
2. Materials and Methods The field experiments were carried out at the SMEAR I site (Station for Measuring Forest Ecosystem – Atmosphere Relation, 67q 46’ N, 29q 35’ E), located in Värriö nature park in eastern Lapland, less than 10 km from the border of Russia, 100–200 km far from major pollution sources at Kola Peninsula (Figure 1). Campaign included measurements of aerosol particle size distributions with EAS (electric aerosol spectrometer, Tammet et al., 2002) from April 28 to May 11, 2003. A description of campaign design and results (records of particle size distributions) is given by Ruuskanen et al. (2007). The EAS was used to measure the aerosol size distribution in the aerodynamic diameter range of 3 nm–10 Pm with spectral resolution of eight fractions per decade (i.e., 28 fractions in total) and time resolution of 10 minutes. This study is focused on the nucleation (3–24 nm, based on exact EAS fractions) and the Aitken mode (24–100 nm) particles. The SILAM model version 3.7 used in this study is based on a Lagrangian particle Monte-Carlo dynamics. The model has been developed at the Finnish Meteorological Institute and it has been validated against the ETEX meso-scale dispersion experiment, the results of various field measurement campaigns, and the long-term air quality observations of EMEP (Sofiev et al., 2006a). For aerosol, the SILAM considers advective and turbulent transport, and size-dependent dry (including sedimentation) and wet deposition. The SILAM model has two modes of operation: forward and adjoint (Sofiev et al., 2006b). In the forward mode, the input data contains the emissions from specified sources, meteorological fields produced by numerical weather prediction models, and land use. The output of the forward simulations consists of the 3D spatial concentration patterns developing in time and 2D dry and wet deposition fields. In the adjoint mode, the model input in the specific case of this study contains measured aerosol concentrations (the so-called sensitivity source function), and the meteorological fields produced by the ECMWF numerical weather prediction model. The output (the so-called sensitivity distribution) is a 4D probability field for the sources of observed concentrations. This output specifies the probability that the measured concentration is originated from a specific location or region; it could also be called the footprint of the observations.
The Origins and Formation Mechanisms of Aerosol during a Measurement Campaign
533
We have performed the adjoint simulations separately for each size mode. The dry and wet deposition processes are included; however, the current model setup does not include adjoint aerosol dynamics or chemistry. Generally, a predicted positive sensitivity distribution value can be interpreted as the presence of an aerosol or its precursor (if chemical transformation is taken into account) in the specific volume of air at the specific time moment that has contributed to the corresponding observed concentration. Pollution sources or concentrations that are not predicted as positive sensitivity distribution values have not contributed to the observed concentration peaks. Forward model runs were also performed in order to investigate the contribution of anthropogenic and natural emissions of sulphur, primary particulate matter (PM2.5) and sea salt to the aerosol detected at Värriö. Anthropogenic emission data is based on the EMEP database. The horizontal resolution of model output was 20 km, and the concentration and sensitivity distribution fields were saved each 15 minutes. Output in the vertical direction consisted of five layers up to the height of 3,150 m above the ground surface. For the numerical results presented here, the output fields were averaged over the two lowest layers (in-total from the ground level up to a height of 150 m). The wind vectors overlaid in the concentration maps are also averaged over the two lowest layers, corresponding to a layer from the ground level to approximately a height of 150 m.
3. Results The measured mass concentrations of the nucleation mode and Aitken nuclei are presented in Figure 2. The nucleation-mode peaks occurred on 30th of April (event 1), and on 5th and 9th of May (events 2 and 3, respectively). The highest peak of Aitken nuclei (also, coarser particles) was observed at April 30, just after event 1.
Fig. 2 Hourly average mass concentrations of aerosol in nucleation and Aitken modes measured by the EAS during the Värriö campaign in April–May 2003
534
M. Prank et al.
Before event 1, the concentrations of all size modes were anomalously low and, after a few hours, highly polluted air masses were observed with large amount of Aitken nuclei and remarkable amount of larger particles. Also, a remarkable amount of nucleation mode particles appeared again a few hours after change in air mass (Figure 3). The first nucleation event is interesting due to the dramatic change of air masses at Värriö that interrupted its observation. It is rather complicated, on the basis of the performed SILAM run, to highlight the origin of its precursors than in the case of first event. We can expect either emissions from the Arctic Ocean or biogenic emissions from coniferous forests of Lapland during a few hours of transport over land areas before reaching the measurement site. Although the area was still covered with snow and the temperatures were slightly below zero, solar heating of tree crowns might induce some vegetation activity. As the air masses advected very lose to the Nikel metallurgy factory (Russia), we cannot exclude some triggering influence of gaseous emissions from there. According to the forward run with a local correction in EMEP-based database (Kaasik et al., 2007), the polluted air mass (incl. the peak of Aitken nuclei) originated from the Nikel area.
Fig. 3 Plot of fractional number concentrations of aerosol particles in Värriö, April 30
The second nucleation event started on 5th of May at 11 a.m. GMT, when a large number of nanometer particles grew to Aitken sizes in late evening. Inverse computations show that the air masses spent two previous days over the continental areas of northern Sweden and central Finland (most of time over boreal forest) and were transported over the Botnian Bay (Figure 4). When the nucleation event began at Värriö, a well-defined footprint was formed that extended over the Finnish Lapland from north-east to south-west (Figure 4b). During and shortly after this nucleation event the forward model computations did not show any advection of substantial concentrations of sulphate over Värriö. Sunny and relatively warm weather during these days conditioned accumulation of biogenic aerosol precursors in the air mass. Considering the typical time scale of nucleation, the observed nano-particles were formed during the morning hours of May 5 over the southern Lapland.
The Origins and Formation Mechanisms of Aerosol during a Measurement Campaign
535
The nucleation event 2nd of May 5 (Figure 4) was followed by a long “tail” of Aitken nuclei. When the event started, the source area of Aitken nuclei extended along the western coast of Finland to the Baltic Sea and reached the coast of Latvia (Figure 5). The maximum of Aitken mode particle concentration at Värriö occurred a few hours later than that of the nucleation mode particles, in the late evening of May 5 with remarkable concentrations of Aitken nuclei observed until the evening of May 6. Thus, we suppose that these Aitken nuclei were the aged particles from the nucleation event that took place during May 5 all over the western Finland and was observed at Värriö as two sequential peaks of concentration – first of the nucleation-mode particles and then of Aitken ones. Considering the part of “tail” over the Baltic Sea, some contribution of sea salt is possible. a)
c)
b)
+
+
V ä r r iö
+
V ä r r iö
V ä r r iö
Fig. 4 Time evolution of the sensitivity distribution (footprint) of the nucleation mode particle mass concentration for the nucleation event 2: (a) May 4, 19 p.m. GMT, (b) May 5, 11 a.m., and (c) May 6, 2 a.m. a)
c)
b)
+
V ä r r iö
+
V ä r r iö
+
V ä r r iö
Fig. 5 Sensitivity distributions of Aitken mode particle mass concentration, nucleation event 2 (Figure 2): (a) May 5, 12 a.m. GMT, nucleation started at Värriö, Aitken “cloud” approaching; (b) May 6, 0 a.m., nucleation mode is vanishing, long “tail” of Aitken mode footprint is approaching Värriö; (c) May 6, 18 p.m., end of the elevated concentration of Aitken-mode particles
M. Prank et al.
536
Looking at fractional aerosol concentrations during the nucleation event 2, it is evident that uniform condensation growth of particles is disturbed (Figure 6). Continuous growth of particles is seen in size range of 3–10 nm, but a small secondary maximum, appearing about 3.5 hours after the main one, becomes equal to first one for particles about 20 nm in diameter. Both maximums propagate through fractions very fast, with seeming growth rate up to 27 nm/h (Kaasik et al., 2006). Neither such a structure nor growth rate can be explained by condensation growth only. A likely explanation is an advective effect: condensation growth in the air mass approaching later might start earlier; the gap in time series may appear due to air mass that had no significant nucleation due to some reason. a)
b)
Fig. 6 Evolution of number concentration in fraction 1 (3.2–4.2 nm) and fraction 7 (18–23 nm) during the nucleation event 2: data points and polynomial fit (used for determining the growth rate of particles, see Kaasik et al., 2006). Local winter time: GMT + 2 hours
In the case of nucleation event on 8th of May, the air masses were transported from the Norwegian Sea. A well-defined footprint was formed that extended over the Finnish and Norwegian Lapland from north-west to south-east. Thus, the leading role of marine emissions is expected.
4. Conclusions Detailed inverse dispersion modelling provides new insights and improves the understanding of processes related to new particle formation (nucleation): it localises the areas contributing to emission of both aerosol particles and their precursors, and provides a detailed time schedule of the new particle formation event. The nucleation process, where the biogenic VOC are present cannot be assumed to take place at a single place: the nucleation and particle-growth event can extend over hundreds and sometimes thousands of kilometres. The need for incorporating the aerosol formation and development processes into an atmospheric transport model is recognised and corresponding cooperation project between the Finnish Meteorological Institute, University of Tartu and University of Helsinki is initiated.
The Origins and Formation Mechanisms of Aerosol during a Measurement Campaign
537
Acknowledgments This study was supported by Nordic Research Board (NordForsk) – cooperation network NetFAM, Maj and Tor Nessling Foundation (Finland) and the Estonian Science Foundation, research grant 7005.
References Kaasik M, Sofiev M, Prank M, Ruuskanen T, Kukkonen J, Kulmala M (2006) Model-delineated origin and growth of particles during the nucleation events observed in Värriö campaign in 2003. In: Proceedings of BACCI, NEC and FcoE activities in 2005 Report Series in Aerosol Science, 81, 221–226. Kaasik M, Prank M, Kukkonen J, Sofiev M (2007) A suggested correction to the EMEP database, regarding the location of a major industrial air pollution source in Kola Peninsula (in this volume). Kulmala M, Toivonen A, Mäkelä JM, Laaksonen A (1998) Analysis of the growth of nucleation mode observed in Boreal forest. Tellus B, 50B, 449–462, 1998. Kulmala M, Boy M, Suni T, Gaman A, Raivonen M, Aaltonen V, Adler H, Anttila T, Fiedler V, Grönholm T, Hellen H, Herrmann E, Jalonen R, Jussila M, Komppula M, Kosmale M, Plauškaite K, Reis R, Savola N, Soini P, Virtanen S, Aalto P, Dal Maso M, Hakola H, Keronen P, Vehkamäki H, Rannik Ü, Lehtinen K E J, Hari P (2004) Aerosols in boreal forest: wintertime relations between formation events and bio-geo-chemical activity. Boreal Environment Research, 9, 63–74. Kulmala M, Petäjä T, Mönkkönen P, Koponen IK, Dal Maso M, Aalto PP, Lehtinen KEJ, Kerminen V-M (2005) On the growth of nucleation mode particles: source rates of condensable vapour in polluted and clean environments. Atmospheric Chemistry and Physics, 5, 409–416. Ruuskanen TM, Kaasik M, Aalto PP, Hõrrak U, Vana M, Mårtensson EM, Yoon YJ, Keronen, P, Mordas G, Ceburnis D, Nilsson ED, O’Dowd C, Noppel M, Alliksaar T, Ivask J, Sofiev M, Prank M, Kulmala M (2007) Concentrations and fluxes of aerosol particles during the LAPBIAT measurement campaign in Värriö field station. Atmospheric Chemistry and Physics, 7, 3683–3700. Sofiev M, Siljamo P, Valkama I, Ilvonen M, Kukkonen J (2006a) A dispersion modelling system SILAM and its validation against ETEX data. Atmospheric Environment 40, 674–685. Sofiev M, Siljamo P, Ranta H, Rantio-Lehtimäki A (2006b) Towards numerical forecasting of long-range air transport of birch pollen: theoretical considerations and a feasibility study. International Journal on Biometeorology, doi:10 1007/s00484-006-0027-x, 50, 392–402. Tammet H, Mirme A, Tamm E (2002) Electrical aerosol spectrometer of Tartu University. Atmospheric Research 62, 315–324. Tunved P, Hansson H-C, Kerminen V-M, Ström J, Dal Maso M Lihavainen HY, Viisanen Y, Aalto PP, Komppula M, Kulmala M (2006) High natural aerosol loading over boreal forests. Science, 312, 5771, 261–263, doi:10.1126/science.1123052.
538
M. Prank et al.
Discussion Gryning: The events sub-parts footprints. You said they are about the same, but they look very different. Please comment. M. Prank: The footprints shown are for the same moment in time, while the subparts of the event happen at different times, so latest one has several hours more to reach the measurement station. Also, as the landuse in all the areas is the same, the minor differences in the subfootprints are not enough to explain the huge drop in the measured concentrations. B. Fisher: Could you tell us what the biogenic nucleation material consists of? M. Prank: By biogenic emissions I mean precursors, biogenic VOCs like terpenes and Į-pinene
6.2 Impacts of Climate Change on Air Pollution Levels in the Northern Hemisphere with Special Focus on Europe and the Arctic Gitte B. Hedegaard, Jørgen Brandt, Jesper H. Christensen, Lise M. Frohn, Camilla Geels, Kaj M. Hansen and Martin Stendel
Abstract The evolution in air pollution levels and spatial distribution in the 21st century is investigated with respect to climate change. The coupled atmosphereocean general circulation model ECHAM4-OPYC3 is providing meteorological fields for two time slices (1990s and 2090s) to the chemical long-range transport model DEHM-REGINA. The dominating impacts from climate change on a large number of the chemical species are related to the predicted temperature increase since most of the reaction rates of the involved species are temperature dependent. The ECHAM4-OPYC3 projects a global mean temperature increase of 3 K with local maxima up to 11 K in the Arctic. As a consequence of this temperature increase, the temperature dependent biogenic emission of isoprene is predicted to increase significantly over land by the DEHM-REGINA model simulation. This leads to an increase in the ozone production and together with an increase in water vapour to an increase in the number of free OH radicals. Furthermore an increase in the number of radicals contributes to a significant change in the typical life times of many species, since hydroxyl radicals are participating in a large number of chemical reactions.
Keywords Air pollution, biogenic emissions, chemical transport model, climate change, coupled models, isoprene
1. Introduction Recently, there has been a growing interest in the effects of climate change on future air pollution levels. It is well known that the composition of the atmosphere will change due to changes in anthropogenic emissions. According to the newly released IPCC report (Solomon et al., 2007) some meteorological parameters will change in the future due to the man-made changes of the composition of the atomsphere. E.g. a general temperature increase will affect many if not all other
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
568
Impacts of Climate Change on Air Pollution Leve
569
meteorological parameters and since the distribution of air pollution is highly dependent on the meteorology, it is hypothesized, that the air pollution levels and spatial distribution even with unchanged emissions will be changed in a warmer climate. Here the hemispheric chemical long-range transport model DEHM-REGINA (Danish Eulerian Hemispheric Model – REGIonal high resolutioN Air pollution model) is used to investigate the future air pollution levels and distribution in the northern hemisphere.
2. Experimental Design The general circulation model ECHAM4-OPYC3 (see Roeckner et al., 1996, 1999; Stendel et al., 2002 for model descriptions) is providing meteorology for the 21st century and part of the 20th century based on the IPCC SRES A2 scenario (Nakicenovic et al., 2000). The meteorology is saved every 6 hours and here after used as input to the chemical long-range transport model DEHM-REGINA (see Christensen, 1997; Frohn et al., 2002, 2003; Frohn, 2004 for a full model description). In order to save computing time the experiment is carried out as a timesliced experiment instead of simulating the 21st century in one continuous run. The time-slices are the two decades 1990s and the 2090s. In Figure 1 the model setup is sketched. The six-hourly climate data is used as a one-way input to the DEHM-REGINA model. The DEHM-REGINA model also needs an emission input. In the experiment carried out here the anthropogenic emissions are conserved at a 1990 emission level. The emissions consist of a combined data set from the EDGAR, GEIA and EMEP data bases (cf. Hedegaard, 2007). The chemical transport model keeps track of the transport, chemistry, depositions and emissions of 63 chemical species and the model includes 120 of the most important chemical reactions between these species. Horizontally the model has a resolution of 150 × 150 km. The ECHAM4 model on the other hand is defined in a spectral grid with truncation T42 (T42 corresponding roughly to a 2.8˚ × 2.8˚ transformed grid). Vertically ECHAM4 is defined in a hybrid sigma-pressure coordinate system and divided into 19 layers extending from the surface of the earth to the 10 hPa pressure level. The DEHM-REGINA is defined in sigmapressure coordinate system and is divided into 20 irregularly distributed layers extending from the earth surface to 100 hPa pressure level. Therefore in order to use the meteorological fields from ECHAM4-OPYC3 as input data to the DEHMREGINA model a transformation has been carried out (cf. Hedegaard, 2007; Hedegaard et al., 2007).
G. Hedegaard et al.
570
ECHAM4 Sea ice surface/ mixed layer
A2 Emissions EMEP-GAIA-EDGAR
OPYC3
interior
3D advection 3D dispersion Chemistry
Emissions (nat.+ant)
DEHMREGINA
Wet and dry deposition
Fig. 1 Off-line setup of the ECHAM4-OPYC3 and the DEHM-REGINA model. In this setup the emissions are kept constant at a 1990 level in order to separate out the effect of climate change. The meteorological input originates from the climate model ECHAM4-OPYC3 which is a coupled atmosphere-ocean model and in the simulations used here forced with the IPCC A2 emission scenario
3. Results and Discussion The chemical species analyzed are, sulphur (SO2), sulphate (SO4), ozone (O3), nitrogen dioxide (NO2), PM10 (particular matter with diameters below 10 ȝm), sea salt, hydroxyl radical (OH) and isoprene (C5H8). Also ammonium (NH4), ammonia (NH3), nitrate (NO3), nitrogen oxides (NOx), TSP (total suspended particles) and PM2.5 (particular matter with diameter below 2.5 ȝm) have been treated in this analysis. We will discuss a few of these species here. The result of the total analysis is given in Hedegaard (2007).
3.1. Changes in ozone and isoprene In Figure 2 (left figure) the difference in the average 2 m temperature of the two time slices (2090s -1990s) is shown. The temperature is increasing everywhere in the domain. This general temperature increase with local hot spots over Southern Europe and the Arctic is similar to other model results (Stendel et al., 2002). The difference in the ten-year average ozone concentration of the two decades is displayed in Figure 3 (left plot). A latitudinal dependence is evident in this figure. Ozone concentrations increase in the future and the increase gets stronger with increasing latitude. North of approximately 30ºN the increase is highly significant (cf. right plot of Figure 3). South of 30ºN difference in ozone concentration changes and in the equatorial areas the ozone concentrations levels tends to
Impacts of Climate Change on Air Pollution Leve
571
decrease significantly between the two decades. However, also a blurred landocean contrast in the ozone increase is evident and the ozone concentration generally increases less over the ocean.
Fig. 2 Left: 2 m temperature difference between the mean values of the two decades 2090s -1990s
in K˚. Right: The statistical significance of the changes of mean values between the two decades according to the t-test. The threshold value for significance is chosen to be within the 0.95 fractile corresponding to the 10% significance level (which is the same as the 1,734 percentile value in the plot). Dark colours (positive percentile values) indicate a significant increase and light shaded colours (negative percentile values) indicate a significant decrease. The projected temperature increase is highly significant everywhere in the domain (black everywhere)
Fig. 3 Left: The difference in ozone concentration between the mean values of the two decades 2090s 1990s in percent. Right: The statistical significance of the changes of mean values between the two decades according to the t-test. Threshold values as in Figure 2
572
G. Hedegaard et al.
The ozone production is strongly dependent on the presence of the precursors NOx and Volatile Organic Compounds (VOC). In the experiment analyzed here, the anthropogenic emissions are kept constant. However, VOCs also have biogenic emitters, which can alter their emissions due to changes in meteorology. The only natural VOC emitter included in the DEHM-REGINA model is isoprene. Isoprene is through participation in chemical reactions with OH acting as a sink for hydroxyl radicals (For further details see Hedegaard, 2007; Hedegaard et al., 2007). In DEHM-REGINA, the submodel BEIS (Biogenic Emissions Inventory System) is included to account for the biogenic isoprene emissions (Guenther et al., 1995). Isoprene is emitted from trees and other plants and therefore primarily existent over land. From Figure 4 (left plot) it can be seen that the concentrations of isoprene is expected to increase where there are emitters present and this increase is highly significant (right plot). The general increase in isoprene concentration over land due to the temperature increase can contribute to explain the increase in ozone, which posses a blurred land-sea contrast in the distribution field. The projected observed level of isoprene will alter the ozone production in a positive direction and thereby enhancing the ozone level.
Fig. 4 Left: the difference in isoprene concentration between the mean values of the two decades 2090s–1990s in ppbV. Right: the statistical significance of the changes of mean values between the two decades according to the t-test. Threshold values as in Figure 2
Langner et al. (2004) used the regional chemistry/transport/deposition model MATCH to simulate the distribution of surface ozone in the future. The Rossby Centre regional atmospheric climate model (RCA) version 1 provided the projected meteorology in these simulations. Langner et al. (2004) found a general increase in the surface ozone concentration over southern and Central Europe. They calculated the domain-total emission of isoprene to increase with 59% due to the predicted temperature increase. This is generally consistent with the results found in the current experiment.
Impacts of Climate Change on Air Pollution Leve
573
Langner et al. (2004) state that the predicted changes found in surface ozone concentrations are substantial and if the climate scenario (IPCC, IS92a) is representative for the future climate, the increase in surface ozone due to the predicted warming would be significant compared to the expected reductions resulting from the emission reduction protocols currently in force. This is in line with Tuovinen et al. (2002) who made a sensitivity analysis of which factors will effect the surface ozone concentration in Europe. They found that the increased biogenic VOC emissions significantly will counteract the effects of reduced anthropogenic emissions. Murazaki and Hess (2006) have also studied the contribution from climate change on the future ozone levels and spatial distribution with the global chemical transport model MOZART-2. Substantially different from the simulations carried out here, Murazaki and Hess (2006) kept both the anthropogenic and biogenic emissions constant (Personal e-mail correspondence with P. Hess, 2006). Murazaki and Hess (2006) divide the surface ozone into two contributions; locally produced ozone and background ozone. In the high emission areas the local increase in ozone is expected to exceed the decrease in background ozone resulting in a net increase. On the contrary a net decrease in ozone is predicted away from these high-emission zones. In the work carried out here the ozone concentration is predicted to increase everywhere over the United States (cf. Figure 3). This difference relative to results of Murazaki and Hess (2006) is probably due to the lack of the temperature dependence of the biogenic emissions in the experiment carried out by Murazaki and Hess (2006). These emitters are as earlier mentioned ozone precursors and by the results of this work they contribute to a relatively large increase in ozone concentration over land.
3.2. Changes in typical life times In this simulation the specific humidity (not shown) is predicted to increase and thereby increasing the number of H2O molecules in the atmosphere. When ozone already is present, more H2O molecules will lead to more hydroxyl radicals through the process shown in the chemical reaction scheme Eq. (1).
O3 hQ o O2 O O H 2 O o 2OH
(1)
It is predicted that the temperature, specific humidity and ozone concentration (only north of approximately 30º N) will increase from the 1990s and to the 2090s. By the reaction scheme Eq. (1) these projected increases must lead to an increase in the number of hydroxyl radicals. The predicted increase by the reasoning above, are confirmed from the concentrations plots of hydroxyl radicals (not shown) (cf. Hedegaard, 2007); Hedegaard et al., 2007). By the projected general increase in the ozone, it can be concluded that DEHM-REGINA predicts an increase in the
574
G. Hedegaard et al.
reaction rates in a great number of chemical reactions over sea and at higher altitudes due to the resulting increase in hydroxyl radicals (Hedegaard et al., 2007) which will have a great influence on the life times of many chemical species. In Hedegaard (2007) and Hedegaard et al. (2007) it is shown that the life times of for example nitrogen dioxide will be reduced and lead to an increased level in nitrate (NO3) and nitric acid (HNO3). Also the sulphate production through the conversion of sulphur will increase in the future.
4. Conclusions The predicted general temperature increase throughout the 21st century results in an increase in the biogenic emissions of isoprene. Isoprene is an important ozone precursor and the ozone production is projected to increase significantly by these simulations. This increase in ozone together with an increase in specific humidity is found to enhance the chemical reaction rates of a great number of chemical reactions. The humidity and ozone increase results in an increase in the number of hydroxyl radicals, which are the activating agent in many chemical processes. For example an indication of a enhanced sulphur to sulphate conversion and decreased life times of some primary species were found.
References Christensen JH (1993) Testing Advection Schemes in a Three-Dimensional Air Pollution Model, Mathematical and Computational Modelling 18 (2), 75–88. Christensen JH (1997) The Danish Eulerian Hemispheric Model – A threedimensional air pollution model used for the Arctic, Atmospheric Environment 31 (24), 4169–4191. Frohn LM (2004) A study of long-term high-resolution air pollution modelling, PhD thesis, University of Copenhagen and National Environmental Research Institute, Aarhus University, Denmark, pp 203. Frohn LM, Christensen JH, Brandt J (2002) Development of a High-Resolution Nested Air Pollution Model. The Numerical Approach, Journal of Computational Physics 179 (1), 68–94. Frohn LM, Christensen JH, Brandt J, Geels C, Hansen KM (2003) Air pollution modelling using a 3D hemispheric nested model, Atmospheric Chemistry and Physics Discussions 3, 3543–3588. Guenther A, Hewitt CN, Erickson D, Fall R, Geron C, Graedel T, Harley P, Klinger L, Lerdau M, McKay WA, Pierce T, Scholes B, Steinbrecher R, Tallamraju R, Taylor J, Zimmerman P (1995) A global-model of natural volatile organiccompound emissions, Journal of Geophysical Research 100, 8873–8892. Hedegaard GB (2007) Impacts of climate change on air pollution levels in the Northern Hemisphere, Technical Report 240, National Environmental research
Impacts of Climate Change on Air Pollution Leve
575
Institute, Aarhus University, Frederiksborgvej 399, P.O. Box 358, 4000 Roskilde, Denmark. Hedegaard GB, Brandt J, Christensen JH, Frohn LM, Geels C, Stendel M (2007) Impacts of climate change on air pollution levels in the Northern Hemisphere with special focus on Europe and the Arctic, Atmospheric Chemistry and Physics Discussions, submitted October 2007. Langner J, Bergstrøm R, Foltescu V (2004) Impact of climate change on surface ozone and deposition of sulphur and nitrogen in Europe, Atmospheric Environment 39, 1129–1141. Murazaki K, Hess P (2006) How does climate change contribute to surface ozone change over the United States, Journal of Geophysical Research 111 (D05301). Nakicenovic N, Alcamo J, Davis G, de Vries B, Fenhann J, Gaffin S, Gregory K, Grübler A, Jung TY, Kram T, La Rovere EL, Michaelis L, Mori S, Morita T, Pepper W, Pitcher H, Price L, Riahi K, Roehrl A, Rogner HH, Sankovski A, Schlesinger A, Shukla P, Smith S, Swart R, van Rooijen S, Victor N, Dadi Z (2000) Special Report on Emission Scenarios, Cambridge University Press, Cambridge, United Kingdom/New York, , pp. 570. Roeckner E, Arpe K, Bengtsson L, Christoph M, Clausen M, Dümenil L, Giorgetta M, Schlese U, Schulzweida U (1996) The atmospheric general circulation model ECHAM-4: Model description and simulation of present-day climate 218, pp. 167, Max-Planck-Institut für Meteorologie, Hamburg, Germany. Roeckner E, Bengtsson L, Feichter J (1999) Transient climate change simulations with a coupled atmosphere-ocean GCM including the tropospheric sulfur cycle, Journal of Climate 12, 3004–3032, Max-Planck-Institut für Meteorologie, Hamburg, Germany. Solomon S, Qin D, Manning M, Chen Z, Marquis M, Averyt KB, Tignor M, Miller HL (2007) Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, pp. 996, Cambridge University Press, Cambridge, United Kingdom/New York. Stendel M, Schmidt T, Roeckner E, Cubasch U (2002) The Climate of the 21st century: transient simulations with a coupled atmosphere-ocean circulation model, Danish climate centre (02-1), pp. 50, Copenhagen, Denmark. Tuovinen JP, Simpson D, Mayerhofer P, Lindfors V, Laurila T (2002) Surface ozone exposures in Northern Europe in changing environmental conditions. In: Hjort J, Raes F, Angeletti G (eds), A Changing Atmosphere: Proceeding of the 8th European Symposium on the Physico-Chemical Behaviour of the Atmospheric Pollutants, European Commission, DG Research, Joint Research Centre, CD-ROM, Paper Ap6, pp. 6.
576
G. Hedegaard et al.
Discussion D. Stein: Does the climate model capture correctly the frequency of occurrence of weather types conducting to degraded air quality? G. Hedegaard: So far only the average weather parameters predicted by the climate model have been validated against average weather from the MM5 meteorological model. We have not yet investigated the frequency of any particular air-quality-degrading weather type. However, this is one of the aims in the future validation process. P. Kishcha: Some peculiarity around Novilski has been mentioned in the presentation. Could you provide some more details about the effect? G. Hedegaard: The Siberian industrial city, Norilsk is a city based on metal extraction and production and a very large amount of sulphur dioxide is emitted from this production every year. In the simulations here a general increase in sulphate concentration is found in the surroundings of Norilsk due to an increase in the conversion rate of sulphur dioxide to sulphate (the OH concentration is predicted to increase). This increase differs from the Siberian and Arctic areas in general where the overall trend is observed to be a decrease in the concentration of sulphur dioxide and sulphate. P. Builtjes: Did you consider changes in land use in the future; the forest might not be anymore where it is now. G. Hedegaard: It is true that the forest and vegetation distribution in general might change in the future and changes like these are not accounted for in the current simulations. T. Dore: Does the model assume increases in emissions of SO2 from shipping? G. Hedegaard: No. All anthropogenic emissions are kept constant at a 1990 level in order to separate out the signal from climate change in the future air pollution levels.
6.1 Linking Global and Regional Models to Simulate U.S. Air Quality in the Year 2050 Chris Nolte, Alice Gilliland and Christian Hogrefe
Abstract The potential impact of global climate change on future air quality in the United States is investigated with global and regional-scale models. Regional climate model scenarios are developed by dynamically downscaling the outputs from a global chemistry and climate model and are then used by the Community Multiscale Air Quality (CMAQ) model to simulate climatological air quality. The CMAQ model is first applied to a five-year period representing current climate and evaluated by comparison against measurements of chemically speciated fine particulate matter (PM2.5) concentrations in the U.S. Next, the model is applied to a simulated climate for the year 2050 based on the A1B scenario developed by the Intergovernmental Panel on Climate Change (IPCC). Two five-year future simulations are conducted, one with anthropogenic emissions held at 2001 levels, and one with anthropogenic emissions reduced to emulate the A1B scenario for the developed world. In both future simulations, biogenic and other climate-sensitive emissions are varied with the simulated climate. Results for the future simulation with current emissions indicate modest decreases of 1–2 Pg m-3 PM2.5 in most of the eastern U.S., but large decreases exceeding 10 Pg m-3 PM2.5 are predicted for the future reduced emissions case.
Keywords Climate change, CMAQ, particulate matter 1. Introduction Currently, regional-scale air quality models are being used to test proposed emission controls for management of air quality without regard to interannual meteorological variability or the possibility of climate change. In cases where emission controls are implemented over several decades, (e.g., U.S. Clean Air Interstate Rule), taking climate change into account could potentially lead to a different conclusion as to an optimal control strategy. Recently, a number of studies have been conducted exploring the impact of climate change on future air quality (Hogrefe et al., 2004; Stevenson et al., 2006; Liao et al., 2006; Racherla and Adams, 2006; Cooter et al., 2007; Wu et al., 2007). Nolte et al. (2008) described a study in which downscaled regional climate scenarios are created from outputs of a global climate and chemistry model and are used by the CMAQ model to simulate air quality over the U.S. under both current and future (ca. 2050) climatologies. In Nolte et al. (2008), C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
559
C. Nolte et al.
560
modeled current ozone concentrations are evaluated against current observations and compared with predicted future concentrations. As a follow up, this study adopts the same approach in comparing modeled current particulate matter concentrations with observations and with predicted future concentrations.
2. Modeling System The modeling system used for this study is comprised of a global climate model (GCM) linked to regional-scale climate and air quality models. Each component of this modeling system is briefly described below. 2.1. Climatological meteorology
The GCM used is derived from the GISS 2’ model as described by Mickley et al. (2004), coupled to the Harvard tropospheric ozone-NOx model as in Mickley et al. (1999). The GCM has a horizontal resolution of 4° latitude and 5° longitude and nine vertical layers in a sigma coordinate system extending from the surface to 10 hPa. The global climate simulation covers the period 1950–2055, with greenhouse gas concentrations updated annually using observations for 1950–2000 (Hansen et al., 2002) and the A1B scenario from the IPCC for 2000–2055 (IPCC, 2000). The radiation scheme assumes present-day climatological values for ozone and aerosol concentrations and has no feedbacks due to future pollutant concentration changes. A regional climate model based on the Penn State/NCAR Mesoscale Model (MM5) was used to downscale the GCM outputs to a 36 km grid for 1995–2005 and for 2045–2055 (Leung and Gustafson, 2005). Lateral boundary conditions from the GCM outputs were applied at 6 hours intervals without assimilation of observational data. The regional climate model outputs were archived hourly and used to provide meteorological conditions for both emissions and air quality models. 2.2. Emissions
The Sparse Matrix Operator Kernel Emissions (SMOKE) modeling system was used to prepare emissions inputs consistent with the simulated meteorology, as both evaporative emissions and plume rise are functions of temperature. Meteorologically-driven biogenic emissions were computed using the Biogenic Emissions Inventory System (BEIS; Hanna et al., 2005). Three five-year sets of daily emissions inputs were prepared, as listed in Table 1. In the first set, Table 1 Description of air quality simulations. Simulation name
Modeling period
Anthropogenic emissions
CURR
1999–2003
2001
FUT1
2048–2052
2001
FUT2
2048–2052
2050
CURR, anthropogenic emissions were based on the U.S. Environmental Protection Agency National Emission Inventory for 2001, modulated by the climatology
Linking Global and Regional Models to Simulate U.S. Air Quality in the Year 2050
561
simulated for the current period, and were merged with biogenic emissions computed using the same simulated current climatology. For the second set, FUT1, the same underlying anthropogenic emissions were used, though they were modulated by the simulated future climatology and merged with biogenic emissions computed for that future period. For the third set, FUT2, the same biogenic emissions were used as in FUT1, but anthropogenic emissions were scaled according to the A1B 2050 projections by the Asian Pacific Integrated Model (AIM) for the developed world (see Table 2). Table 2 Scaling factors applied for all anthropogenic emission sectors in simulation FUT2, relative to FUT1. Species
Factor
NOx
0.52
SO2
0.37
VOCs
0.79
CO
1.5
PM
1
NH3
1
2.3. Air quality simulations Air quality simulations were performed with the Community Multiscale Air Quality (CMAQ) model (Byun and Schere, 2006) version 4.5. A continuous fiveyear CMAQ simulation was run for each of the three emissions scenarios listed in Table 1. Chemical boundary conditions for ozone, NOx, and related VOCs were taken from monthly averaged outputs of the Harvard tropospheric chemistry module coupled to the GISS GCM. For each time period, mean aerosol boundary conditions were computed from outputs of a related simulation conducted with the modeling system of Liao et al. (2003).
3. Results and Discussion 3.1. Current period evaluation
Total mass and speciated 24-hour measurements of PM2.5 concentrations are collected every third day at sites in the Interagency Monitoring of Protected Visual Environments network (IMPROVE; see http://vista.cira.colostate.edu/improve). Summer and winter differences between modeled (CURR) average PM2.5 concentrations and 2000–2004 observations are shown in Figure 1. For the summer, modeled concentrations at sites in the Pacific Northwest and in the central U.S. are positively biased by 2–6 Pg m-3, while most sites in the northeast exhibit an equivalent negative bias. For the winter, however, positive biases exist at nearly every site, exceeding 6 Pg m-3 at most sites in the central and eastern U.S.
562
C. Nolte et al.
Fig. 1 Differences between average modeled PM2.5 concentrations and measurements at IMPROVE monitoring sites for summer (left) and winter (right)
Similarly, difference plots between modeled and measured sulfate concentrations are shown in Figure 2. During the summer, predicted sulfate concentrations exhibit a positive bias greater than 1.5 Pg m-3 in much of the central U.S. and somewhat weaker negative biases in the northeast and in California. During the winter, sulfate predictions are unbiased (within 0.5 Pg m-3) at most monitoring sites.
Fig. 2 Differences between average modeled sulfate concentrations and measurements at IMPROVE monitoring sites for summer (left) and winter (right)
Measured “soil” concentrations (a derived quantity based on measurements of certain trace elements) and modeled “other unspeciated PM” concentrations are shown in Figure 3. During both summer and winter, a large positive bias in modeled “other PM” concentrations is evident at nearly every monitoring site, making it the largest contributor to the errors in total PM2.5. Preliminary investigations into this bias suggests that it is due to unrealistically high levels of dust (summer average 2 Pg m-3; winter average 8 Pg m-3) in the global model coming into the CMAQ modeling domain via the northern boundary.
Linking Global and Regional Models to Simulate U.S. Air Quality in the Year 2050
563
Fig. 3 Differences between average modeled “soil” concentrations and measurements at IMPROVE monitoring sites for summer (left) and winter (right)
3.2. Current-future differences
Differences between the average PM2.5 concentrations for the two future period simulations and the CURR simulation are shown in Figure 4. For FUT1, summer average PM2.5 concentrations decrease by 1–3 Pg m-3 throughout most of the central and eastern U.S. and in California. In the winter, PM2.5 decreases by 2–5 Pg m-3 in the northern part of the domain. The decrease is even more substantial for the A1B-scaled emissions case FUT2, with average decreases from 3 to 9 Pg m-3 in the eastern third of the U.S during both summer and winter.
Fig. 4 Changes from CURR in five-year-average summer (left) and winter (right) PM2.5 concentrations for FUT1 (top) and FUT2 (bottom)
564
C. Nolte et al.
Fig. 5 Changes from CURR in five-year-average summer (left) and winter (right) sulfate concentrations for FUT1 (top) and FUT2 (bottom)
Differences between average sulfate concentrations for the two future period simulations and the CURR simulation are shown in Figure 5. For FUT1, summer average sulfate concentrations decrease by 0.5–1.5 Pg m-3 throughout most of the central and southern U.S., while there is a slight increase of 0.2–0.4 Pg m-3 in a portion of the Midwest. The decrease is larger for the A1B-scaled emissions case FUT2, with average sulfate decreases from 3 to 5 Pg m-3 in the eastern third of the U.S during summer and 1–2 Pg m-3 during the winter.
Fig. 6 Changes from CURR in five-year-average summer (left) and winter (right) “other PM” concentrations for FUT1
Differences between the summer and winter average “other PM” concentration for FUT1 and CURR simulation are shown in Figure 6. Dust emissions and boundary conditions were the same for FUT2 as for FUT1, so FUT2 “other PM” concentrations are virtually identical to those for FUT1 and are not shown. Relative to CURR, future summer average “other PM” concentrations decrease by 0.5–1.0 Pg m-3 in the western U.S. and in parts of the central U.S., while large decreases exceeding 2 Pg m-3 are evident during the winter. Since “other PM” is a chemically
Linking Global and Regional Models to Simulate U.S. Air Quality in the Year 2050
565
inert species in CMAQ, the large decrease in future concentrations is due to changes in boundary conditions from the global model; these were unrealistically high for the current period as noted in Section 3.1.
4. Summary Three sets of five-year air quality simulations for the continental U.S. have been conducted using downscaled meteorology and chemical boundary conditions from a global climate model. Model predictions for current period PM2.5 concentrations are in reasonable agreement with recent observations, with the error dominated by overpredictions in dust concentrations from the global model. Comparison of model results for the current period with those for the future period with current anthropogenic emissions shows decreases in sulfate of 0.5–1.5 Pg m-3 during the summer, while summer sulfate concentrations decrease 3–5 Pg m-3 in the reduced emissions case. Future work for this study will explore the relative impact of changing meteorological variables and changes in chemical boundary conditions on PM2.5 concentrations. A coupled climate and air quality model is under development, which will integrate feedbacks from pollutants on radiative forcing to better understand the relationships between climate change and air quality. Acknowledgments The authors wish to thank Loretta Mickley and Pavan Racherla for conducting the GCM simulations and Ruby Leung of the Pacific Northwest National Laboratory for performing the MM5 downscaling of the GCM data. William Benjey, Robert Gilliam, and Steven Howard of NOAA’s Atmospheric Sciences Modeling Division and Allan Beidler and Ruen Tang of the Computer Sciences Corporation assisted with processing emissions and meteorological data and with the air quality simulations. Disclaimer The research presented here was performed under the Memorandum of Understanding between the U.S. Environmental Protection Agency (EPA) and U.S. Department of Commerce’s National Oceanic and Atmospheric Administration (NOAA) and under agreement number DW13921548. This work constitutes a contribution to the NOAA Air Quality Program. Although it has been reviewed by EPA and NOAA and approved for publication, it does not necessarily reflect their policies or views.
References Byun D, Schere KL (2006) Science algorithms of the EPA Models-3 Community Multiscale Air Quality (CMAQ) Modeling System, Applied Mechanics Reviews 59, 51–77. Cooter EJ, Gilliam R, Benjey W, Nolte C, Swall J, Gilliland A (2007) Examining the impact of changing climate on regional air quality over the U.S. In:
566
C. Nolte et al.
Developments in Environmental Science, vol. 6, C. Borrego and E. Renner, eds. Elsevier, Amsterdam. Hanna SR, Russell AG, Wilkinson JG, Vukovich J, Hansen DA (2005) Monte Carlo estimation of uncertainties in BEIS3 emission outputs and their effects on uncertainties in chemical transport model predictions, Journal of Geophysical Research 110, D01302. Hansen J et al. (2002) Climate forcings in Goddard Institute for Space Studies SI2000 simulations, Journal of Geophysical Research 107 (D18), 4347. Hogrefe C, Lynn B, Civerolo K, Ku JY, Rosenthal J, Rosenzweig C, Goldberg R, Faffin S, Knowlton K, Kinney PL (2004) Simulating changes in regional air pollution over the eastern United States due to changes in global and regional climate and emissions, Journal of Geophysical Research 109, D22301. Intergovernmental Panel on Climate Change (2000), Special Report on Emissions Scenarios, N. Nacenovic and R. Swart, eds., Cambridge University Press, New York. Available on the Web at http://www.grida.no/climate/ipcc/emission Leung LR, Gustafson WI Jr (2005) Potential regional climate change and implications to U.S. air quality, Geophysical Research Letters 32, L16711. Liao H, Adams PJ, Chung SH, Seinfeld JH, Mickley LJ, Jacob DJ (2003) Interactions between tropospheric chemistry and aerosols in a unified general circulation model, Journal of Geophysical Research 108 (D1), 4001. Liao H, Chen WT, Seinfeld JH (2006) Role of climate change in global predictions of future tropospheric ozone and aerosols, Journal of Geophysical Research 111 (D12), D12304. Mickley, LJ, Murti PP, Jacob DJ, Logan JA, Koch DM, Rind D (1999) Radiative forcing from tropospheric ozone calculated with a unified chemistry-climate model, Journal of Geophysical Research 104 (D23), 30153–30172. Mickley, LJ, Jacob DJ, Field BD, Rind D (2004) Effects of future climate change on regional air pollution episodes in the United States, Geophysical Research Letters 31, L24103. Nolte CG, Gilliland AB, Hogrefe C, Mickley LJ (2008) Linking global to regional models to assess future climate impacts on surface ozone concentrations in the United States, Journal of Geophysical Research, submitted. Racherla PN, Adams PJ (2006) Sensitivity of global tropospheric ozone and fine particulate matter concentrations to climate change, Journal of Geophysical Research 111, D24103. Stevenson, DS, Johnson CE, Collins WJ, Derwent RG, Edwards JM (2006) Multimodel ensemble simulations of present-day and near-future tropospheric ozone, Journal of Geophysical Research 111, D08301. Wu S, Mickley LJ, Leibensperger EM, Jacob DJ, Rind D, Streets DG (2008) Effects of 2000-2050 global change on ozone air quality in the United States, Journal of Geophysical Research, doi:10.1029/2007JD008917, 113, D06302.
Linking Global and Regional Models to Simulate U.S. Air Quality in the Year 2050
567
Discussion J. Baldasano: What were the reasons why you chose the year 2050? C. Nolte: This study is part of a group of studies investigating the impact of future climate change on air quality. One of the reasons for agreeing on a common time period among these studies is to provide an ensemble of air quality projections. The year 2050 was chosen based on consensus of several groups. Climate signals at 2030 could be too small to detect within interannual variability. Climate modelers preferred 2100 for the same reason, but emission scenarios for 2100 are so uncertain as to be untenable. Hence, 2050 was a good compromise. P. Kishcha: Were future land use changes considered in the model predictions under discussion? C. Nolte: No. Future land use categories were assumed to be unchanged. This represents an important uncertainty in our modeling system. A. Aulinger: Did you compute statistics on peak concentrations of PM and O3 in order to assess the number of days with increased health risks due to climate change or changes in precursor concentrations? C. Nolte: Yes. We have computed the number of days per year at each site where the maximum 8-hour average ozone and PM2.5 exceeded threshold values of 80 ppb and 35 Pg m-3. The spatial pattern of change in the number of exceedances generally follows the pattern of the changes in the means. However, our ozone concentration predictions under current climate conditions are positively biased by 10–15 ppb in parts of the U.S., which hinders our ability to predict with accuracy exceedances above a given threshold.
6.3 Regional Climate Change Impacts on Air Quality in CECILIA EC 6FP Project Tomas Halenka, Peter Huszar and Michal Belda
Abstract Recent studies show considerable effect of atmospheric chemistry and aerosols on climate on regional and local scale. For the purpose of qualifying and quantifying the magnitude of climate forcing due to atmospheric chemistry/aerosols on regional scale, the development of coupling of regional climate model and chemistry/aerosol model has been started recently on the Department of Meteorology and Environmental Protection, Faculty of Mathematics and Physics, Charles University in Prague, for the EC 6FP Project QUANTIFY and finally for EC 6FP Project CECILIA. One of the project objectives, aiming to study climate change impacts in Central and Eastern Europe based on very high resolution simulations using regional climate models (RCM) in 10 km grid, is dealing with climate change impacts on and interaction to air quality. For this coupling, existing regional climate model and chemistry transport model are used. Climate is calculated using model RegCM and ALADIN-Climate while chemistry is solved by model CAMx. Climate change impacts on large urban and industrial areas modulated by topographical and land-use effects which can be resolved at the 10 km scale, are investigated by CECILIA as well. Meteorological fields generated by RCM drive CAMx transport, chemistry and a dry/wet deposition. A preprocessor utility was developed for transforming RegCM provided fields to CAMx input fields and format. As the first step, the distribution of pollutants can be simulated off-line for long period in the model couple. There is critical issue of the emission inventories available both for present and scenarios runs as well as cross-boundary transport for regional simulations. The next step is the inclusion of the radiative active agents from CAMx into RCM radiative transfer scheme to calculate the changes of heating rates. Only the modification of radiative transfer due to atmospheric chemistry/ aerosols is taken into account first, the indirect effect of aerosols will be studied later. Ten years time slices for present, control and scenarios runs for mid- and end of century are supposed in framework of the project. Some sensitivity runs will be run in present climate.
Keywords Air-pollution, air-quality, climate change, air pollution modelling
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
577
578
T. Halenka et al.
1. Introduction In decision making process there is significant problem arising from the weak link between global climate change information and impact studies necessarily based on real local conditions. Global Circulation Models (GCMs) can reproduce reasonably well climate features on large scales (global and continental), but their accuracy decreases when proceeding from continental to regional and local scales because of the lack of resolution. This is especially true for surface fields, such as precipitation, surface air temperature and their extremes, which are critically affected by topography and land use. However, in many applications, particularly related to the assessment of climate-change impacts, the information on surface climate change at regional to local scale is fundamental. To bridge the gap between the climate information provided by GCMs and that needed in impact studies, especially when aiming the interactions of climate and air-quality issues, dynamical downscaling, i.e., nesting of a fine scale limited area model (or Regional Climate Model, RCM) within the GCM is the most convenient tool. In the region of Central and Eastern Europe the need for high resolution studies is particularly important. This region is characterized by the northern flanks of the Alps, the long arc of the Carpathians, and smaller mountain chains and highlands in the Czech Republic, Slovakia, Romania and Bulgaria that significantly affect the local climate conditions. A resolution sufficient to capture the effects of these topographical and associated land-use features is necessary. That is why 10 km resolution has been introduced in the project CECILIA of EC FP6. The main aim of the project dealing with climate change impacts and vulnerability assessment in targeted areas of Central and Eastern Europe is the application of regional climate modelling studies at a resolution of 10 km for local impact studies in key sectors of the region. The project is covering studies on hydrology, water quality, and water management (focusing at medium-sized river catchments and the Black Sea coast), agriculture (crop yield, pests and diseases, carbon cycle), and forestry (management, carbon cycle), as well as air quality issues in urban and industrialized areas (e.g. Black Triangle – a polluted region around the common borders of the Czech Republic, Poland and Germany). Climate change impacts on large urban and Industrial areas modulated by topographical and land-use effects which can be resolved at the 10 km scale are investigated by CECILIA as well. The concentration of air pollutants depends on both anthropogenic and climate factors. A main issue is the quantity of emissions of primary pollutants as well as of precursors of secondary pollutants. Long range transport to the target regions will be taken into account by simulation for the whole Europe, driven by RCM with a grid resolution of 50 × 50 km. These simulations will be used to constrain nested higher resolution runs (10 × 10 km) for a smaller domain focusing in CEE both for present and future climate. The key species will be ozone, sulphur and nitrogen as well as PM, which have a central role in tropospheric chemistry as well as the strong health impacts.
Regional Climate Change Impacts on Air Quality in CECILIA EC 6FP Project
579
2. Models Involved It is now well established that climatically important (so called radiatively active) gases and aerosols can have substantial climatic impact trough their direct and indirect effects on radiation, especially on regional scales (Qian and Giorgi, 2000; Qian et al., 2001, Giorgi et al., 2002). The study of these effects requires coupling of regional climate models with atmospheric chemistry/aerosols to assess the climate forcing to the chemical composition of the atmosphere and its feedback to the radiation, eventually other components of the climate system. For this coupling, existing regional climate model and chemistry transport model are used. At our Department climate is calculated using model RegCM while chemistry is solved by model CAMx, for the projects the attempt is done to develop and to use the couple ALADIN-Climate and CAMx as well. The model RegCM used here was originally developed by Giorgi et al. (1993a, b) and then has undergone a number of improvements described in Giorgi et al. (1999), and, finally, Pal et al. (2005). The dynamical core of the RegCM is equivalent to the hydrostatic version of the mesoscale model MM5. Surface processes are represented via the Biosphere-Atmosphere Transfer Scheme (BATS) and boundary layer physics is formulated following a non-local vertical diffusion scheme (Giorgi et al., 1993a). Resolvable scale precipitation is represented via the scheme of Pal et al. (2000), which includes a prognostic equation for cloud water and allows for fractional grid box cloudiness, accretion and re-evaporation of falling precipitation. Convective precipitation is represented using a mass flux convective scheme (Giorgi et al., 1993b) while radiative transfer is computed using the radiation package of the NCAR Community Climate Model, version CCM3 (Giorgi et al., 1999). This scheme describes the effect of different greenhouse gases, cloud water, cloud ice and atmospheric aerosols. Cloud radiation is calculated in terms of cloud fractional cover and cloud water content, and the fraction of cloud ice is diagnosed by the scheme as a function of temperature. For more details on the use of the model see Elguindi et al. (2006). CAMx is an Eulerian photochemical dispersion model developed by ENVIRON Int. Corp. (Environ, 2006). Currently in version 4.40 CAMx is used for air quality modeling in more than 20 countries by government agencies, academic and research institutions, and private consultants for regulatory assessments and general research. It is available for free in the form of the source code with various supporting programs. CAMx can use environmental input fields from a number of meteorological models (e.g., MM5, RAMS, CALMET) and emission inputs from many emissions processors. CAMx includes the options of two-way grid nesting, multiple gas phase chemistry mechanism options (CB-IV, SAPRC99), evolving multisectional or static two-mode particle size treatments, wet deposition of gases and particles, plume-in-grid (PiG) module for sub-grid treatment of selected point sources, Ozone and Particulate Source Apportionment Technology, mass conservative and consistent transport numerics, parallel processing. It allows for integrated “one-atmosphere” assessments of gaseous and particulate air pollution (ozone, PM2.5, PM10, air toxics) over many scales ranging from sub-urban to continental.
580
T. Halenka et al.
CAMx simulates the emission, dispersion, chemical reaction, and removal of pollutants in the troposphere by solving the pollutant (Eulerian) continuity equation for each chemical species on a system of nested three-dimensional grids. These processes are strongly dependent on the meteorological conditions, therefore CAMx requires meteorological input from a NWP model or RCM for successful run.
Fig. 1 Average concentration of NO2 (upper left), O3 (upper right) and SO2 (bottom panel) for year 2000 for 10 × 10 km resolution central Europe domain in ppbv
3. Preprocessor and Settings Meteorological fields generated by RegCM drive CAMx transport and dry/wet deposition. A preprocessor utility was developed for transforming RegCM fields to CAMx input fields and formats. For fields not provided by the meteorological model the diagnostic formulas are used, cloud/rain water content and cloud optical depth are gained from the rain rates and the vertical profile of water vapour content and temperature. Vertical diffusion coefficients are calculated following O’Brien (1970). As the first step, the distribution of pollutants can be simulated for long period in the model couple. There are problems with the anthropogenic emission inventories available, at this stage emissions from EMEP 50 × 50 km database are interpolated. We are testing VOC speciation technique, biogenic emissions of isoprene and monoterpenes calculated as a function of 2 m temperature, global radiation and landuse by Guenther et al. (1993, 1994). We use 23 vertical ı-levels reaching up to
Regional Climate Change Impacts on Air Quality in CECILIA EC 6FP Project
581
70 hPa, with time step of 150 s, at 45 km resolution in preliminary experiments for RegCM configuration, the same horizontal grid for CAMx. Initial and boundary conditions are set to CAMx’s top concentrations (independent of time) (Simpson et al., 2003) for 45 km resolution run, the results are used for driving the same couple of RegCM-CAMx in 10 km resolution on smaller “CECILIA” region. In our setting CB-IV chemistry mechanism is used (Gery et al., 1989).
Fig. 2 Comparison of simulated ten-days running average concentration of NO2 for selected stations in year 2000 (ppbv). Grey line for 45 km resolution, black line for 10 km resolution
4. Preliminary Results Some examples of the high resolution integration for year 2000 are presented in Figure 1 for selected species. There is much more local features seen in this simulation compare to less resolution run (not shown), especially for O3 the effect of high resolution land use which provides basis for biogenic emission computation is well pronounced in the concentration fields, even more in summer (with respect to limited extent not shown again). More interesting comparison of the driving 45 km
582
T. Halenka et al.
resolution run with 10 km high resolution run for selected stations can be seen in Figure 2 in terms of time series of NO2 simulated concentrations. Reasonable agreement can be seen during spring and summer, some decreases appear in winter and autumn seasons of high resolution simulation. Mainly in cold season the values of high resolution runs are significantly lower in Bohemian region on Kosetice and Svratouch station. More reliable analysis can be done by comparison of both simulations of O3 concentration with real data in Figure 3.
Fig. 3 Comparison of simulated and measured ten-days running average concentration of O3 for selected stations in year 2000 (ȝg/m3). Grey line for 45 km resolution, black line for 10 km resolution, light grey for measurement
Underestimation of the ozone concentration by the model especially during warm season appears for some stations of the Central Europe whereas overestimation is presented in comparison for Ispra mainly in cold period of the year. Basically, high resolution runs brings slight improvement of the results for selected stations.
Regional Climate Change Impacts on Air Quality in CECILIA EC 6FP Project
583
5. Future Outlooks To produce more reliable conclusions longer runs are essential. At present we are preparing the longer experiment in very high resolution of 10 km driven by 50 km run based on off-line simulation using CAMx with the meteorological data from ICTP RegCM simulation driven by ERA40 reanalyses done for EC FP6 IP ENSEMBLES. Period of ten years simulation covering years 1991–2000 will provide the comparison on reliable data both on input (emission databases) and measurement side. Further in framework of the CECILIA Project three time slices of ten years are supposed to be completed in high resolution of 10 km with the couple RegCM-CAMx: control (1991–2000), middle of the century (2041–2050) and end of century (2091–2100), using A1B scenario. The next step of the couple development will be the inclusion of the radiative active agents from CAMx into RegCM radiative transfer scheme to calculate the changes of heating rates. Only the modification of radiative transfer due to atmospheric chemistry/aerosols will be taken into account first, the indirect effect of aerosols will be taken into account later, there are still many uncertainties in understanding of this issue and possibility of inclusion of appropriate processes into the model. The feedback of chemistry/aerosols on climate will be studied in terms of monthly and yearly averages of 2 m temperatures and of the top-of-the-atmosphere (TOA) radiative forcing, the results will provide the estimate of the effect of interactive atmospheric chemistry and aerosols on climate in regional and local scales. Acknowledgments This work is supported in framework of EC FP6 STREP CECILIA (GOCE 037005), partially by EC FP6 Integrated project QUANTIFY (GOCE 003893) as well as under local support of the grant of Programme Informacni spolecnost, No. 1ET400300414 and Research Plan of MSMT under No. MSM 0021620860.
References Elguindi N, Bi X, Giorgi F, Nagarajan B, Pal J, Solmon F, Rauscher S, Zakey A, (2006) RegCM Version 3.1 User’s Guide. PWCG Abdus Salam ICTP. ENVIRON Corporation (2006) CAMx Users’ Guide, version 4.40 Gery MW, Whitten GZ, Killus JP, Dodge MC (1989) A photochemical kinetics mechanism for urban and regional scale computer modeling. J. Geophys. Res., 94, 925–956. Giorgi F, Marinucci MR, Bates GT (1993a) Development of a second generation regional climate model (RegCM2). Part I: boundary layer and radiative transfer processes. Mon. Wea. Rev., 121, 2794–2813.
584
T. Halenka et al.
Giorgi F, Marinucci MR, Bates GT, DeCanio G (1993b) Development of a second generation regional climate model (RegCM2). Part II: convective processes and assimilation of lateral boundary conditions. Mon. Wea. Rev., 121, 2814–2832. Giorgi F, Huang Y, Nishizawa K, Fu C (1999) A seasonal cycle simulation over eastern Asia and its sensitivity to radiative transfer and surface processes. Journal of Geophysical Research, 104, 6403–6423. Giorgi F, Bi X, Qian Y (2002) Direct radiative forcing and regional climatic effects of anthropogenic aerosols over East Asia: a regional coupled climate-chemistry/ aerosol model study. J. Geophys. Res., 107, 4439, doi:10.1029/2001JD001066. Guenther AB, Zimmerman PR, Harley PC, Monson RK, Fall R (1993) Isoprene and monoterpene rate variability: model evaluations and sensitivity analyses. Journal of Geophysical Research, 98, No. D7, 12609–12617. Guenther A, Zimmerman P, Wildermuth M (1994) Natural volatile organic compound emission rate estimates for U.S. woodland landscapes. Atmospheric Environment, 28, 1197–1210. O’Brien JJ (1970) A note on the vertical structure of the eddy exchange coefficient in the planetary boundary layer. Journal of Atmospheric Science, 27, 1213–1215. Pal JS, Small EE, Eltahir EA (2000) Simulation of regional-scale water and energy budgets: Representation of subgrid cloud and precipitation processes within RegCM. Journal of Geophysical Research, 105, 29579–29594. Pal JS, Giorgi F, Bi X, Elguindi N, Solmon F, Grimm A, Sloan L, Syed F, Zakey A, (2005) The ICTP Regional Climate Model version 3 (RegCM3). Benchmark simulations over tropical regions. Submitted to the Bull. Amer. Meteorol. Soc. QianY, Giorgi F (2000) Regional climatic effects of anthropogenic aerosols? The case of Southwestern China. Geophysical Research Letters, 27(21), 3521–3524, 10.1029/2000GL011942. Qian Y, Giorgi F, Huang Y, Chameides WL, Luo C (2001) Simulation of anthropogenic sulfur over East Asia with a regional coupled chemistry/climate model. Tellus, Series B, 53, 171–191. Simpson D, Fagerli H, Jonson J, Tsyro S, Wind P (2003) Transboundary Acidification, Eutrophication and Ground Level Ozone in Europe PART I, Norwegian Meteorological Institute.
Discussion G. Kallos: How did you perform your regional scale model simulations? They are always biases in surface parameterization (ex soil moisture) Heat strongly affects meteorology (ex latent heat flux/sensible heat and therefore cloud and precipitation processes heat effect air quality and deposition. how you managed these issues? T. Halenka: Of course there are always biases of many kinds in any simulation. Especially, this might be of great importance when using these
Regional Climate Change Impacts on Air Quality in CECILIA EC 6FP Project
585
results for driving air-quality simulations. However, when developing the tools for estimation of future development under different scenarios of climate change, there is no other way than to cope with these uncertainties. In presented results for experiment with one year simulations there is not so much material for systematic analysis, but in framework of our CECILIA project ten years period is supposed to be run in framework of “perfect” boundary conditions (based on reanalysis) as well as the control simulation driven by global model before the time slices for the middle and the end of this century. Careful statistical analysis of mainly reanalysis run and comparison to control both in air-quality results and climate ones can reveal the role of individual sources of biases and their impact on air-quality simulations.
7.3 A Modeling Methodology to Support Evaluation of Public Health Impacts on Air Pollution Reduction Programs Vlad Isakov and Halûk Özkaynak
Abstract Environmental public health protection requires a good understanding of the types and locations of pollutant emissions of health concern and their relationship to environmental public health indicators. Therefore, it is necessary to develop the methodologies, data sources, and tools for assessing the public health impact of air pollution reduction programs, also referred to as accountability analysis. Since air quality models are among the main tools that can be used to evaluate the impacts from emissions changes, either due to growth or implementtation of source control strategies, these approaches play a vital role in most air accountability studies. In this study, we present a modeling methodology to estimate concentrations for multiple pollutants that include both local features (hot spots) and regional transport. The local impacts from mobile sources and significant stationary sources are estimated using a dispersion model (AERMOD). These local details are combined with regional background estimates computed by a photochemical grid model (CMAQ) in a “hybrid” approach to derive total concentrations required for the subsequent human exposure analysis. We demonstrate an application of this methodology in New Haven, Connecticut. The city of New Haven has implemented a comprehensive Clean Air Initiative, which includes a number of federally mandated and voluntary air pollution programs. This project is a collaborative effort with state and local agencies including government, academia, and the New Haven community, to apply and evaluate air quality and human exposure models that can be used with health data and to assess the feasibility of using this information to conduct an air accountability study. Although this study is based in one city, the methodologies developed through this project can have broad application to other areas within the United States and internationally.
Keywords Air quality, exposure, modeling
1. Introduction Air quality has improved substantially in the United States in recent decades, in large part due to increasingly stringent federal and state air quality regulations. While C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
614
A Modeling Methodology to Support Evaluation of Public Health impacts on Air Pollution
615
many studies have documented links between better air quality and improvements in a variety of human health metrics, direct evidence is lacking about the extent to which specific control measures have improved health. Over the past decade, various epidemiological studies have examined the relationships between acute and chronic health outcomes and measured ambient particulate matter, ozone and other co-pollutant concentrations. In the context of air pollution health effects, results from recent epidemiological studies indicate the importance of determining the key sources and constituents of indoor, outdoor pollution, and personal exposures to PM, ozone and other air pollutants. However, understanding the magnitude and nature of human exposure is clearly the first step in assessing the occurrence of adverse effects that could follow upon contact with environmental pollutants. One of the ways to access the human exposure is through the use of exposure models such as EPA’s Hazardous Air Pollutant Exposure Model (HAPEM), the Air Pollutant Exposure Model (APEX), or Stochastic Human Exposure and Dose Simulation (SHEDS). Since predicted concentrations from air quality models are key drivers for human exposure models, it is essential to improve the accuracy and precision of spatial and temporal characterization of results from these models. However, complex interactions between interventions over time can make it difficult to isolate the environmental impacts and associated health effects of any one regulation. For example, a regulatory action may have varying effects on emissions depending upon compliance and the real-world effectiveness of the interventions applied. Often, the connection between emissions and ambient air quality depends on complex atomspheric and chemical transformations. Environmental public health protection requires a good understanding of the types and locations of pollutant emissions of health concern and their relationship to environmental public health indicators. Therefore, it is necessary to develop the methodologies, data sources, and tools for assessing the public health impact of air pollution reduction programs, and accounttability analysis. Since air quality models are the principal predictive tool for assessing the impacts of potential emissions control strategies on future-year concentrations, we describe here a modeling approach to support air accountability studies.
2. Air Quality Modeling Approach Environmental health studies require detailed information on air quality. Therefore, air quality modeling should include local-scale features, long-range transport, and photochemistry to provide the best estimates of air concentrations. There are several available modeling approaches capable of assessing pollutant concentration gradients at a fine resolution (Touma et al., 2006) and these can be categorized into two major types of air quality models: source-based dispersion models and Eulerian grid-based chemical transport models. Chemical transport models, such as the Community Multi-scale Air Quality (CMAQ, Byun and Schere, 2006), are used to simulate the transport and formation of ozone, acid rain, particulate matter (PM)
616
V. Isakov, H. Özkaynak
and other pollutants formed by chemical reactions among precursor species that are emitted from hundreds or thousands of emission sources. Such models may be set up to apply to a wide range of scales ranging from global to urban. However, regional-scale grid-based models can address photochemistry effects, but not locallevel gradients. CMAQ provides volume-average, hourly concentration values for each grid cell in the modeling domain. Emissions are assumed to be instantaneously well-mixed. While grid-models are the model platform of choice for simulation of chemicallyreactive airborne pollutants, source-based dispersion models such as AERMOD (Cimorelli et al., 2005) that have been developed to simulate pollutant concentrations within a few hundred meters or a few kilometers from the source are typically used for local scales. These models generally do not take into account atmospheric chemical reactions or they do so using simplified representations such as first-order pollutant decay. They provide detailed resolution of the spatial variations in hourly-average concentrations. It would be desirable to combine the capabilities of grid-models and dispersion models into one model, but this is a yet evolving area of research and development. One option is a hybrid approach (Isakov et al., 2007), where a regional grid model and a local plume model are run independently. To illustrate how air quality models can be used to provide inputs to human exposure models, we focus on a 20 by 20 km area encompassing New Haven, Connecticut that includes many stationary sources emitting toxic pollutants and several major roadways as indicated in Figure 1. The city of New Haven, with population of approximately 125,000, is a recipient of one of EPA’s nationally funded Community Air Toxics projects. Through this project, New Haven has implemented a comprehensive Clean Air Initiative, which includes a number of voluntary air pollution programs. Along with local and state efforts, there are also several Federal regulations that have either recently been or soon will be implemented (e.g., Clean Air Interstate Rule). This project presents an opportunity to assess the feasibility of using air quality and human exposure models that can be used with health data to conduct an air accountability study. Resolving fine scale pollutant gradients to identify local concentration hot spots from both stationary and mobile sources is critical for exposure assessments. For example, individuals who spend more time near busy highways are likely to be exposed to higher levels of air pollution. To account for this near-road exposure we modeled ambient air quality concentrations for multiple pollutants resulting from roadway emissions. There are multiple modeling techniques to simulate near-road dispersion from mobile sources (Borrego et al., 2006; Cook et al., 2006). In this study, we used the AERMOD dispersion model which treats individual road links as area sources to simulate hourly concentrations of various pollutants near the road. AERMOD also simulates near-source impacts from stationary sources. Contributions to photochemical interactions are provided as a background concentration level from CMAQ, a regional grid model.
A Modeling Methodology to Support Evaluation of Public Health impacts on Air Pollution
617
Fig. 1 Modeling domain showing locations of emission sources and model receptors
A hybrid approach (Isakov et al., 2007) is a logical and efficient way to combine regional grid and local plume models. Results of both model simulations are combined to provide the total ambient air pollutant concentrations. The hybrid approach uses the appropriate modeling tools to describe different types of sources, making its application computationally efficient. Furthermore, since local dispersion models are not resource intensive, this methodology allows the study of local concentration variability due to changes in several model inputs and physical parameters, helping to gain confidence in the simulation results by encompassing a range of model outcomes. This constitutes a clear advantage of the hybrid approach, since performing a local concentration variability estimation using a nested grid model alone would be an impractical task, especially over larger urban areas. A schematic of the hybrid approach is shown in Figure 2, where the CMAQ model was used to estimate regional background concentrations and AERMOD was used to estimate local-scale details for stationary and mobile sources.
618
V. Isakov, H. Özkaynak
Fig. 2 Application of the hybrid approach in New Haven, Connecticut
In this study, CMAQ was used to simulate ambient concentrations of several air toxics (Luecken et al., 2006). The CMAQ modeling system was run for an annual period in a nested mode at 36 and 12 km horizontal grid dimensions using the 1999 National Emission Inventory and meteorological outputs from 2001 using the MM5 meteorological model. The CMAQ results were extracted for the New Haven modeling domain to provide regional background concentration values. This regional background was combined with local concentrations predicted by the AERMOD dispersion model. The application of this hybrid approach is illustrated in Figure 3 which displays modeled annual average outdoor carbon monoxide concentrations in New Haven, Connecticut. Predicted CMAQ concentrations at 12 by 12 km resolution are combined with estimates at 200 m receptor resolution from AERMOD.
A Modeling Methodology to Support Evaluation of Public Health impacts on Air Pollution
619
Fig. 3 Annual average CO concentrations in New Haven, Connecticut: (a) impact of stationary sources; (b) impact of mobile sources; (c) regional background; (d) combined concentrations using hybrid approach
3. Modeling Approach for Accountability Studies The methodologies developed under this project can be applied to future projects in other areas to simulate air quality impacts for various controls scenarios. For example: (1) what happens if emissions from some specific stationary sources are reduced by “x” percent? (2) what happens if emissions from mobile sources could be reduced by “y” percent? (3) what is the impact of local controls? (4) what is the impact of regional/national controls resulting in reduction of regional background? Figure 4 provides a hypothetical example of the relative impacts of various control strategies on ambient concentrations. These examples include: reducing emissions from mobile sources, controlling emissions from stationary sources, and reducing impact of the regional background. This example helps determine which control options are most effective in reducing ambient concentrations. In order to link these air quality estimates to health effects associated with human exposures to environmental pollutants, we will use exposure modeling. When combined with exposure models, the control strategies can be assessed to optimize the impacted population, or population subset, such as children, people with respiratory problems, etc. In this study, we are evaluating alternative techniques for estimating cumulative exposures to selected air toxics, PM, and ozone using probabilistic cumulative exposure models: HAPEM 6 and SHEDS-Air Toxics, time-series based models, using human activity pattern data, modeled/measured concentrations, and exposure factors.
620
V. Isakov, H. Özkaynak
Fig. 4 Example of modeling application to investigate the effect of various control strategies
4. Summary At the present, most Federal and State air quality implementation plans rely heavily on ambient modeling study results for targeting emissions reductions. However, the complexity in the spatial and microenvironmental variation of exposures among the different population subgroups, especially in the context inter- and intra-urban analysis of air pollution health effects, could pose several challenges. Thus, integrated air quality – human exposure modeling provides the means to evaluate the potential health risks from air pollution exposures and the basis to determine optimum risk management strategies, while considering scientific, social and economic factors. Ideally, emission control strategies not only aim at reducing the emissions from principal sources of targeted pollutants but also to identify those sources and microenvironments that contribute to greatest portion of personal or population exposures. Recent advances in exposure modeling tools and better information on time-activity, commuting and exposure factors data provide unique opportunities for improving the assignment of exposures during the course of future accountability and community health studies. Moreover, the combination of sophisticated air quality and exposure models will improve the accuracy of present air quality and exposure forecasts, and help us better quantify the health and economic benefits of emissions reductions programs, as part of air accountability studies.
A Modeling Methodology to Support Evaluation of Public Health impacts on Air Pollution
621
Acknowledgments and Disclaimer The research presented here was performed under the Memorandum of Understanding between the U.S. Environmental Protection Agency (EPA) and the U.S. Department of Commerce’s National Oceanic and Atmospheric Administration (NOAA) and under agreement number DW13921548. This work constitutes a contribution to the NOAA Air Quality Program. It does not necessarily reflect Agency policies or views.
References Borrego CA, Tchepel O, Costa AM, Martins H, Ferreira J, Miranda AI (2006) Traffic-related particulate air pollution exposure in urban areas. Atmospheric Environment 40, 7205–7214. Byun DW, Schere K (2006) Review of the governing equations, computational algorithms, and other components of the Models-3 Community Multiscale Air Quality (CMAQ) modeling system. Applied Mechanics Review 59, 51–77. Cimorelli AJ, Perry SG, Venkatram A, Weil JC, Paine RJ, Wilson RB, Lee RF, Peters WD, Brode RW (2005) AERMOD: a dispersion model for industrial source applications. Journal of Applied Meteorology 44, 682–693. Cook R, Touma JS, Beidler A, Strum M (2006) Preparing highway emissions inventories for urban scale modeling: a case study in Philadelphia. Transportation Research Part D: Transport and Environment 11, 396–407. Isakov V, Irwin JS, Ching J (2007) Using CMAQ for exposure modeling and characterizing the sub-grid variability for exposure estimates. Journal of Applied Meteorology and Climatology 46, 1354–1371. Luecken DJ, Hutzell WT, Gipson GJ (2006) Development and analysis of air quality modeling simulations for hazardous air pollutants. Atmospheric Environment 40, 5087–5096. Touma JS, Isakov V, Ching J, Seigneur C (2006) Air quality modeling of hazardous pollutants: current status and future directions. Journal of Air and Waste Management Association 56, 547–558.
Discussion B. Fisher: Do you tell your users about uncertainty in the high resolution concentration fields, which can arise from errors in emissions models etc? V. Isakov: Yes, assessing uncertainties is an integral part of the health risk assessment process. It is, therefore, desirable to incorporate some treatment of uncertainties in the entire modeling process: emissions and meteorological inputs, model formulation, monitoring data, and
622
V. Isakov, H. Özkaynak
exposure and risk. In this study, a major contribution to the uncertainty in the model simulation results originated from the model inputs rather than from the model formulation. Therefore, in order to reduce uncertainty in the high resolution concentration fields, it is important to improve spatial allocation of emissions. For mobile sources, we have developed a practical, readily adaptable methodology to create spatially-resolved, link-based highway vehicle emission inventory. This methodology takes advantage of geographic information system (GIS) software to improve the spatial accuracy of the activity information obtained from a Travel Demand Model. An example of application of this methodology in New Haven, CT is shown in Cook et al. (2008), Journal of Air and Waste Management Association 58, 451–461.
7.4 Evaluating the Effects of Emission Reductions on Multiple Pollutants Simultaneously Deborah Luecken, Alan Cimorelli, Cynthia Stahl and Daniel Tong
Abstract Modeling studies over the Philadelphia metropolitan area have examined how emission control strategies might affect several types of air pollutants simultaneously. NOx reductions in July are predicted to increase ozone in the urban core and decrease it elsewhere, decrease PM2.5 and formaldehyde, and slightly increase acetaldehyde and 1,3-butadiene. In January, NOx reductions increase ozone, formaldehyde and acetaldehyde everywhere. VOC reductions decrease aldehydes but have little effect on ozone in this domain. A combination of VOC and NOx reductions reflects the cumulative behavior of each of the emission reductions separately, and minimizes disbenefits for both HAPs and ozone. A comparison of these changes in terms of their effect on health shows that differing behavior of PM2.5 and ozone can counterbalance each other to some extent. While changes in HAPs are affected by changes to reduce ozone and PM2.5, their effect on health impacts is smaller than PM2.5 and ozone. This study supports considering effects of multiple pollutants in determining optimum pollution control strategies. Keywords Emission control, HAPs, multipollutant, ozone 1. Introduction Many areas around the world have air quality problems with simultaneously high concentrations of one or more pollutants, including ozone (O3), particulate matter (PM2.5), oxides of nitrogen (NOx) and/or hazardous air pollutants (HAPs). There is a growing awareness that pollution control should be considered for its overall benefit to pollutants, rather than on a single pollutant basis (Scheffe et al., 2007). There has been some discussion on the potential effect of volatile organic hydrocarbon (VOCs) and NOx control on species other than ozone, such as other oxygenated nitrogen species and secondary organic aerosol (National Research Council, 1991; Russell et al., 1988; Blanchard et al., 2007), but few comprehensive, multi-pollutant studies, especially considering effects on HAPs. Ozone and the secondarily-produced portion of PM2.5 and HAPs are interrelated through complex atmospheric photochemistry, so control strategies for PM2.5 and ozone might C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
623
D. Luecken et al.
624
decrease or increase HAP concentrations. There are ongoing efforts to reduce concentrations of important pollutants to meet health standards for ozone, NOx and particles, but these reductions will affect the concentrations of radicals and VOCs that produce and destroy HAPs. We need to understand the effect of these controls on HAP concentrations in order to calculate the full economic benefit of control strategies or compare alternative strategies. In this study, we use a three-dimensional air quality model, the Community MultiScale Air Quality (CMAQ) modeling system, to examine the effect of emission control strategies on concentrations of ozone, PM2.5, and four important HAPs: formaldehyde, acetaldehyde, 1,3-butadiene and benzene. The objective of this paper is to begin to address the question of how control strategies formulated for pollutants such as ozone and PM might benefit or disbenefit other pollutants.
2. Model Formulation and Application Model simulations are centered on the Philadelphia metropolitan area, with a 4-km horizontal grid size, (76 by 82 cells), and 15 vertical layers, nested within a continental-scale simulation with a 36-km horizontal grid resolution (Figure 1). We used CMAQ v4.5 (Byun and Schere, 2003; Community Modeling and Analysis System, 2006). Within this domain, we selected four grids representing different chemical characteristics, noted as A, B, C and D. Grid A is located in urban central Philadelphia, B is upwind of the urban area, C and D are on the same latitude as A, but more rural. Emissions for 1999 were used, with a refined mobile source inventory for Philadelphia. The MM5 v3.6.1 model provided meteorological fields (MM5 Community Model, 2007). We performed base case model simulations for January and July, 2001. The SAPRC-99 chemical mechanism (Carter, 1990), modified to include 26 explicit air toxics (Luecken et al., 2006) characterizes the chemistry. 50
45
-76
-74
40
C
35
B
A D
40
39
30
25 -120
-110
-100
-90
-80
-70
Fig. 1 Simulation domain with nested area and location of grid cells for detailed analysis
38
Evaluating the Effects of Emission Reductions on Multiple Pollutants Simultaneously
625
To examine how pollutant concentrations might be affected differently by emission reductions, we performed sensitivity studies with across-the-board anthropogenic emission reductions. The studies simulated two periods, July 14–25, 2001 and January 10–21, 2001. While these reductions do not reflect “realistic” control scenarios, they mimic potential future emission reductions. The three scenarios are (1) 50% reduction in nitric oxide and nitrogen dioxide emissions (NOx-only); (2) 20% reduction in VOC emissions (VOC-only); and (3) 50% reduction in NOx and 20% VOC reduction in VOC emissions (NOx + VOC). Biogenic emissions constitute a significant portion of VOC emissions, so the overall cut in the VOC-only scenario is less than 20%. To account for reduction of pollutant transport, we reduced NOx and VOCs at the boundaries by 50% or 20% of the anthropogenic portion. We compare strategies at the four different grid points shown in Figure 1.
3. Results
3.1. Concentrations Changes in ozone and PM2.5 for the July simulation are shown in Figure 2. Ozone values are 12-day averages of the daily maximum 8-hour average concentrations, and PM2.5 values are calculated as 12-day averages. This domain has an urban corridor (represented by grid A) that is largely VOC-sensitive but surrounding areas that are NOx-sensitive. Ozone in the urban corridor is predicted to increase when NOx is reduced, while other grids show ozone decreases in the NOx-only scenario in July. The VOC-only scenario has a small effect on decreasing ozone, and the NOx +VOC simulations show larger benefit and less disbenefit than the NOx-only simulations. 3 50% NOx
20% VOC
50%NOx +20%VOC
50% NOx
20% VOC
50%NOx +20%VOC
8
2
4
1
0
0 PM2.5 12-day avg
-4
-8
-12
Ozone avg of 1-hr max
Site A (urban) Site B (upwind) Site C (west of city) Site D (east of city)
-1
-2
Concentration change (base minus control),Pg/m3
Concentration change (base minus control), ppb
12
-3
Fig. 2 Change in concentrations of ozone and PM2.5 for each sensitivity scenario during July
D. Luecken et al.
626
The NOx-only reductions have a small effect on the PM2.5 concentrations (4–12% reductions) due to decrease in formation of aerosol nitrate. Nitrate comprises a small portion of the total PM2.5 (1–17%) and we predict the largest PM2.5 decreases where the fraction of nitrate in PM2.5 is largest (grid C, where ammonia emissions are large) and the smallest effect in grid D, where nitrate is low. There is a small change in PM2.5 because changes in ozone and OH affect formation of organic aerosol (increases at grid A, decreases elsewhere) and aerosol sulfate. For the January episode, the ozone increases in the NOx-only scenario for all grids, slightly less so in the VOC + NOx scenario. The PM2.5 changes are less than 0.5%. Figure 3 shows changes in the 12-day average concentrations of formaldehyde, acetaldehyde, 1,3-butadiene and benzene for each of the emission reduction scenarios, calculated as (base – control). Formaldehyde concentrations are reduced by a small amount and acetaldehyde concentrations are increased on average when NOx emissions are reduced, but the variability of the data is also large. The 1,3butadiene increases in C and D and decreases in A and B. The aldehydes are primarily produced in the atmosphere from other VOCs, so changes in ozone, OH, hydroperoxy radical and organic radicals resulting from emission cuts affect concentrations of formaldehyde and acetaldehyde in a complex manner. Although all four HAPs that we examine are VOCs, the concentrations of aldehydes do not change linearly with the VOC reductions. Benzene, on the other hand, shows an approximately linear reduction with VOC emission reductions. In January, both formaldehyde and acetaldehyde increase with the NOx-only scenario but decrease in all others, benzene has negligible change, and 1,3-butadiene decreases for all. Site A (urban) Site B (upwind) Site C (west of city) Site D (east of city)
Concentration change (Base - Control), Pg/m3
0.7 0.6
50% NOx
50%NOx 50% NOx + 20% VOC
20% VOC
50%NOx 50% NOx + 20% VOC 20% VOC
50%NOx 50%NOx 50% NOx + 20% VOC 20% VOC + 20% VOC 20% VOC
0.5 0.4 0.3 0.2 0.1 0
benzene
-0.1 -0.2 -0.3
formaldehyde
acetaldehyde
1,3-butadiene
Fig. 3 Change in concentrations of four HAPs for each sensitivity scenarios during July
Evaluating the Effects of Emission Reductions on Multiple Pollutants Simultaneously
627
3.2. Comparison of changes in terms of health days lost Differing responses to pollutant reductions by ozone, PM2.5 and HAPs make it difficult to determine an overall, optimum control strategy. What types of strategies should be pursued if a strategy increases some pollutants but decreases others? To compare relative changes for different types of pollutants, we attempt to normalize them by converting into health-related effects. There are several different ways to characterize health effects. Exposure modeling, as described in Georgopoulos et al. (2005), provides one of the most accepted and complete ways to estimate overall health effects by accounting for indoor and outdoor pollutant sources, movement of people, local peaks, etc. While these models are appropriate for small-scale studies with finely-resolved concentration gradients, they are more difficult to apply and require fine-scale inputs. Another option for estimating changes in health endpoints is concentrationresponse functions derived from relationships between ambient concentrations and health effects (US EPA, 1999). These are useful for analyzing regional-scale control strategies where only ambient sources are controlled. Because the goal of this study is to estimate relative changes in PM2.5, ozone and HAP concentrations, we use concentration-response functions to quantify the direction and approximate changes in health effects due to changes in each pollutant. While not as comprehensive as exposure models, the results can screen potential control scenarios which can be followed up with more detailed exposure modeling. To include effects in both summer and winter, we summed January and July responses. For ozone and PM2.5, we follow the method used for ozone by Tong et al. (2006). Short-term mortality, hospital admissions, and ER visits for respiratory conditions were included for ozone. For PM2.5, long-term mortality for adults 30 years and older was included. For ozone and PM2.5 concentration changes, 'Ci, we calculate change in the base rates of each health endpoint, 'Hi, using a log-linear equation:
'Hi
>expEi * 'Ci @
(1)
where E = 0.00052 for daily ozone short-term mortality (Bell et al., 2006), E = 0.00631 for hospital visits (Burnett et al., 2001), and E =0.0035 for emergency room visits for asthma (Stieb et al., 1996). For PM2.5, we use E = 0.006 for annually-averaged PM2.5 (Pope et al., 2002) for adults older than 30 years. We use a 25 ppb threshold value for ozone, and no threshold for PM2.5. We report the total value in terms of change in health days, HDi from the base rate for each population age group and the days lost for each mortality, Dm, with a median lifetime of 77 years:
HDi
Hbase , i * 'Hi 1 * population * Dm
(2)
There is not a consensus on how to compare health effects from HAPs with those from ozone and PM2.5 because the health effects differ. For HAPs examined in this
D. Luecken et al.
628
study, carcinogenic effects are of most concern. To calculate health days lost due to mortality from carcinogenic effects of HAPs, we use linear equations of the form
'Hi HDi
Ei * 'Ci / 70
(3)
'Hi * population * Dm * 0.44
(4)
Values of E are Unit Risk Estimates reported in the US EPA Integrated Risk System (US EPA, 2007), with E = 1.3E-5 for formaldehyde, E = 2.2E-6 for acetaldehyde, E = 3.0E-5 for 1,3-butadiene and E = 5.0E-6 for benzene. The changes in health days are based on a 70 year lifetime, consistent with National Air Toxic Assessment (US EPA, 2006), a 44% fatality rate for all cancers (American Cancer Society, 2001) and the base mortality distribution. Figure 4 shows the overall change in health days for each scenario at the four grids for ozone, PM2.5, the sum of HAPs and the overall sum. At some grids for some scenarios, the PM2.5 and ozone have opposite effects on health days, and the overall sum reflects those counteracting effects. The HAPs examined in this study show small contributions to the changes in health days. Because only four out of hundreds of recognized HAPs were included in this analysis, the total effects of all HAPs, especially with other high risk HAPs such as diesel PM and acrolein, would be larger than the values displayed here. Since some HAPs decrease under the NOx-controlled scenarios and some increase, the health effects balance out somewhat when computing the sum of effects from HAPs. Grid A
Change in Number of Health Days
8000
4000
4000
0
0
-4000
-4000 50% NOx 20% VOC 50% NOx+20%VOC
-8000
O3 (24-hr)
PM2.5
HAPs
50% NOx 20% VOC 50% NOx+20%VOC
-8000
O3 (24-hr)
Sum
Grid C
8000 Change in Number of Health Days
Grid B
8000
PM2.5
4000
0
0
Sum
Grid D
8000
4000
HAPs
-4000
-4000 50% NOx 20% VOC 50% NOx+20%VOC
-8000
O3 (24-hr)
PM2.5
HAPs
Sum
50% NOx 20% VOC 50% NOx+20%VOC
-8000
O3 (24-hr)
PM2.5
HAPs
Sum
Fig. 4 Change in number of health days for three emission change scenarios at four grids
Evaluating the Effects of Emission Reductions on Multiple Pollutants Simultaneously
629
4. Summary and Conclusion Modeling studies over the Philadelphia metropolitan area have examined the potential for emission control strategies to affect several types of air pollutants simultaneously, through both direct and indirect effects of emission reductions. Over this domain, a 50% NOx reduction in July increase ozones in the urban core and decreases ozone elsewhere, decreases PM2.5 and formaldehyde, and slightly increases acetaldehyde. In January, NOx-only reductions increase ozone, formaldehyde and acetaldehyde everywhere, to a significant fractional extent. When considering VOC-only reductions, we predict that a 20% reduction in VOCs decreases aldehyde concentrations everywhere, although the decreases are less than 20%, but has little effect on ozone in this domain. A combination of VOC and NOx reductions reflects the cumulative behavior of each of the emission reductions separately, and minimizes disbenefits for both HAPs and ozone. Comparing these changes in terms of their effect on health allows us to initially rank emissions change scenarios and to compare different scenarios in terms of their overall potential effect on health. The differing behavior of species supports the need to consider effects of multiple pollutants in determining optimum pollution control strategies. While changes in HAPs, including secondarily-produced ones, are affected by changes to reduce ozone and PM2.5, their effect on health impacts is smaller than PM2.5 and ozone. We note that uncertainties in concentrationresponse functions, in HAPs risk estimates, and in base rates of mortality could change the conclusions, and future work should be done to explore the effect of these uncertainties on the identification of optimum control strategies. Acknowledgments We gratefully acknowledge technical assistance of Bill Hutzell, and the support of EPA’s Regional and Applied Research (RARE) Program. Disclaimer The research presented here was performed under the Memorandum of Understanding between the U.S. Environmental Protection Agency (EPA) and the U.S. Department of Commerce’s National Oceanic and Atmospheric Administration (NOAA) and under agreement number DW13921548. Although it has been reviewed by EPA and NOAA and approved for publication, it does not necessarily reflect their policies or views.
References American Cancer Society (2001) Cancer Facts and Figures, 2001. http://www. cancer.org/downloads/STT/F&F2001.pdf Bell ML, McDermott A, Zeger SL, Samet JM, Dominici F (2004) Ozone and shortterm mortality in 95 US urban communities, 1987–2000. Journal of the American Medical Association, 292, 2372–2378. Blanchard C, Tanenbaum, S, Hidy GM (2007) Effects of sulfur dioxide and oxides of nitrogen emission reductions on fine particulate matter mass concentrations:
630
D. Luecken et al.
regional comparisons, Journal of the Air and Waste Management Association, 57, 1337–1350. Burnett RT, Smith-Doiron M, Stieb D, Raizenne ME, Brook JR, Dales RE, Leech JA, Cakmak S. Krewski D (2001) Association between ozone and hospitallization for acute respiratory diseases in children less than 2 years of age, American Journal of Epidemiology, 153 (5), 444–452. Byun D, Schere KL (2003) Review of the governing equations, computational algorithms, and other components of the Models-3 Community Multiscale Air Quality (CMAQ) modeling system, Applied Mechanics Reviews, 59, 51–77. Carter WPL (2000) Implementation of the SAPRC-99 Chemical Mechanism into the Models-3 Framework; Report to the United States Environmental Protection Agency, http://www.cert.ucr.edu/~carter/absts.htm#s99mod3 Community Modeling and Analysis System (2006) University of North Carolina, http://www.cmascenter.org/help/documentation.cfm?temp_id=99999 Georgopoulos PG, Wang S-W, Vikram M, Vyas VM, Sun Q, Burke J, Vedantham R, Mccurdy T, Ozkaynak H (2006) A source-to-dose assessment of population exposures to fine PM and ozone in Philadelphia, PA, during a summer 1999 episode, Journal of Exposure Analysis and Environmental Epidemiology, 15, 439– 457. Luecken DJ, Hutzell WT, Gipson G (2006) Development and Analysis of Air Quality Modeling Simulations for Hazardous Air Pollutants, Atmospheric Environment, 40, 5087–5096. MM5 Community Model (2007) Pennsylvania State University and National Center for Atmospheric Research, http://box.mmm.ucar.edu/mm5 National Research Council (1991) Rethinking the Ozone Problem in Urban and Regional Air Pollution, National Academy Press, Washington, DC. Pope C.A, Burnett RT, Thun MJ, Calle EE, Krewski D, Ito K, Thurston GD (2002) Lung cancer, cardiopulmonary mortality, and long-term exposure to fine particulate air pollution, Journal of the American Medical Association 287(9), 1132– 1141. Russell AG, McCue KF, Cass GR (1988) Mathematical modeling of the formation of nitrogen-containing pollutants. 2. Evaluation of the effect of emission controls, Environmental Science and Technology 22, 1336–1347. Scheffe R, Hubbell B, Fox T, Rao V, Pennell W (2007) The rationale for a multipollutant, multimedia air quality management framework, Environmental Manager, May 2007, 14–20. Stieb, DM, Burnett RT, Beveridge RC, Brook JR (1996) Association between ozone and asthma emergency department visits in Saint John, New Brunswick, Canada, Environmental Health Perspectives 104(12), 1354–1360. Tong D, Muler NZ, Mauzerall DL, Mendelsohn RO (2006) Integrated assessment of the spatial variability of ozone impacts from emissions of nitrogen oxides, Environmental Science and Technology, 40, 1395–1400. US EPA (2007) Integrated Risk Information System, http://www.epa.gov/iris US EPA (2006) 1999 National Scale Air Toxics Assessment, http://www.epa.gov/ ttn/atw/nata1999/nsata99.html
Evaluating the Effects of Emission Reductions on Multiple Pollutants Simultaneously
631
US EPA (1999) The Benefits and Costs of the Clean Air Act 1990 to 2010, EPA Report to Congress, http://www.epa.gov/oar/sect812/prospective1.html
Discussion S.T. Rao: In addition to the concentration-response functions, are you looking at the exposure-response functions is assessing the impact of emission reduction strategies? D. Luecken: We have used concentration-response functions to perform our initial analyses because we are looking at relative changes in health effects. However, our plans for follow-up work on the next project include using exposure modelling as a more detailed way to compare overall health impacts in the Baltimore, MD area.
7.2 Long-Term Regional Air Quality Modelling in Support of Health Impact Analyses C. Hogrefe, B. Lynn, K. Knowlton, R. Goldberg, C. Rosenzweig and P.L. Kinney
Abstract This paper investigates the use of long-term regional scale meteorological and air quality simulations for tracking changes in air quality and for supporting public health assessments. For this purpose, year-round simulations with the MM5/CMAQ modelling system have been performed over the northeastern United States for 1988–2000. Emission inputs for the CMAQ simulations were prepared with the SMOKE processing system and were based on 1990 and 1996– 2000 inventories. During this period, significant reductions in anthropogenic emissions of VOC, NOx, and SO2 have occurred in the point source and mobile source sectors. Model evaluation results show that the modelling system performs better in capturing temporal than spatial patterns. Moreover, the modelling system captured the effects of SO2 emission reductions on SO4 concentrations in ambient air and rain water. Finally, examples are provided for how these CMAQ simulations could be used in combination with observations to study the link between air quality and public health.
Keywords Acid deposition, emission trends, health impacts, regional-scale air quality modelling
1. Introduction To date, most studies investigating the link between air pollution and human health have relied on monitoring data to characterize ambient pollutant concentrations (e.g. Bell et al., 2004; Samoli et al., 2005, and references therein). More recently, studies have begun to explore the potential benefits of incorporating concentration fields simulated by grid-based air quality modelling systems into health impact assessments (Bell, 2006). In this paper, we present results from a study aimed at performing long-term air quality simulation over the northeastern portion of the U.S. and assessing the usefulness of these simulations for health impact studies. We describe the design of the model simulations in Section 2, present evaluation results for these simulations in Section 3, and finally provide examples for the potential use of these model simulations and discuss next steps in Section 4.
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
605
C. Hogrefe et al.
606
2. Description of Modelling System and Databases 2.1. Modelling system Meteorological conditions for the time period from January 1, 1988–December 31, 2000 were simulated with the MM5 meteorological model (Grell et al., 1994). The simulations were performed on two nested grids with 36 and 12 km horizontal grid spacing, respectively. Among the physics options chosen for the simulation are the ETA scheme for representing the planetary boundary layer, the Kain-Fritsch cloud scheme, and the RRTM radiation scheme. Throughout the model simulation, MM5 was nudged towards NCEP reanalysis fields using four-dimensional data assimilation. Annual anthropogenic emission inventories for area, nonroad and point sources were obtained from the EPA National Emission Trends database for 1990 and 1996–2000 (U.S. EPA, 2005). The 1990 emissions were used for the simulation of 1988–1990, while emissions from 1991–1995 were estimated by interpolation between 1990 and 1996 for these source categories. Onroad mobile source emissions were estimated with the MOBILE6 model using annual county-level vehicle miles travelled (VMT) and MM5 temperatures from 1988 to 2000. Biogenic sources for 1988–2000 were estimated with the BEIS3.12 model taking into account MM5 temperature, radiation, and precipitation. All emissions processing including mobile sources and biogenic sources was performed within the SMOKE system (Houyoux et al., 2000). Table 1 provides a summary of domain-total annual anthropogenic emissions for 1990 and 2000. It is evident that significant emission reductions for all pollutants have occurred between 1990 and 2000. An analysis of the time series of the annual total anthropogenic emissions from 1988 to 2000 utilized in these simulations reveals a gradual decline of NOx emissions and a the steep reduction of SO2 emissions in the mid-1990s. This reduction is due to Title IV of the Clean Air Act Amendments and its impact on observed and simulated SO4 concentrations is discussed in Section 3. Using the meteorological and emission fields described above, hourly gridded fields of concentrations, wet deposition, and dry deposition for a large number of gas phase and aerosol species were simulated with the Community Multiscale Air Quality (CMAQ) model (Byun and Schere, 2006), version 4.5.1. The simulations were performed with two nested grids of 36 and 12 km, corresponding to the MM5 grids except for a ring of buffer cells. The boundary conditions for the 36 km grid Table 1 Annual total anthropogenic emissions over the 12 km modelling domain processed by SMOKE for 1990 and 2000. All emissions are shown in kilotons. Area + nonroad Mobile Point
1990 2000 1990 2000 1990 2000
NOx
VOC
CO
SO2
NH3
PM2.5
1,933 1,757 2,895 2,180 3,443 1,884
3,580 3,353 2,877 1,461 1,209 334
10,379 11,627 31,924 18,190 7,743 1,190
1,018 691 67 65 8,725 5,088
502 513 42 68 15 11
790 661 77 41 172 328
Long-Term Regional Air Quality Modelling in Support of Health Impact Analyses
607
correspond to climatological values, while the boundary conditions for the 12 km grid were derived from the 36 km simulation. Gas phase chemistry was represented by the CB-IV mechanism while aerosol chemistry was simulated with the aero3 module. For all subsequent analyses, only simulations from the 12 km CMAQ grid were utilized.
2.2. Observations Observed daily maximum 8-hour ozone concentrations for the period 1988–2000 were determined from hourly ozone observations at surface monitors from the U.S. EPA’s AQS database. In order to be included in the analysis, monitors had to be (a) located within the 12 km CMAQ domain, and (b) have at least 40% non-missing days during each year between 1988 and 2000. The application of these screening criteria resulted in the selection of 112 monitors. For the evaluation of SO4 aerosol concentrations, observed weekly-average air concentrations were extracted at eight monitors located within the 12 km CMAQ domain from the Clean Air Status and Trends Network (CASTNet) database. Furthermore, weekly-average precipitationweighted rainwater SO4 concentrations were extracted from the National Acid Deposition Program (NADP) database for 25 monitors located within the 12 km CMAQ domain.
3. Model Evaluation Figure 1a, b show time series of model performance statistics for daily maximum 8-hour ozone calculated for each ozone season (May–October) from 1988–2000. For each ozone season, all daily observations – model pairs located in the modelling domain were utilized to compute these statistics, therefore, the metrics measure the model’s ability to capture the total temporal and spatial variability of ozone observations. Figure 1a indicates that there is year-to-year variability but no obvious trend in correlation coefficients, model bias, and the ratio of simulated to observed standard deviations. On the other hand, the root mean square error (RMSE) shows a downward trend throughout the analysis time period from about 18 ppb for the 1988 simulation to about 16 ppb for the 2000 simulation. Following the definition by Willmot (1982), Figure 1b shows the decomposition of the total RMSE into its systematic and unsystematic components and reveals that most of the reduction is driven by the unsystematic RMSE with smaller changes in the systematic RMSE.
608
C. Hogrefe et al.
While the analyses shown above measured the total spatial and temporal variability and indicated that model performance for daily maximum 8-hour ozone for all summers was within typical ranges found for regional-scale photochemical models (Hogrefe et al., 2007), it is of interest to separately investigate model performance in time and space. The rationale is that in health impact applications, models can potentially be used to estimate pollutant concentrations both during unmonitored time intervals and at unmonitored locations, but the skill of the model between these two tasks may differ. Figure 2 shows distributions of correlation coefficients, total RMSE, systematic RMSE, and unsystematic RMSE for daily maximum 8-hour ozone for both time series and spatial patterns. The time series distributions for a given quantity (e.g. total RMSE) were constructed by calculating this quantity for each observed and simulated time series for each station and each ozone season, therefore, the time series distributions consist of 1,456 datapoints (13 years times 112 stations). In contrast, the spatial pattern distributions for a given quantity were constructed by calculating this quantity for each observed and
Fig 1 (a) Time series of correlation coefficients, bias, ratio of simulated to observed standard deviations, and total RMSE calculated from observation and model predictions for each May 1– October 31 time period for each year. (b) Same as in (a), but for total, systematic, and un systematic RMSE
Fig 2 Distributions of correlation coefficients (upper left), total RMSE (upper right), systematic RMSE (lower left), and unsystematic RMSE (lower right) for simulated time series (solid lines) and spatial patterns (dotted lines)
Long-Term Regional Air Quality Modelling in Support of Health Impact Analyses
609
simulated map of daily maximum 8-hour ozone for each ozone season day and each year, therefore, the time series distribution consists of 2,379 datapoints (13 years times 183 days). Results show that correlation coefficients between observed and simulated time series are substantially higher and have a much narrower distribution than correlation coefficients between observed and simulated spatial patterns. For total RMSE, both the time series and spatial pattern distributions show similar medians but a larger spread for the spatial pattern distribution. While most of the temporal variability in observations and model predictions typically is due to synoptic and seasonal scale fluctuations (Rao et al., 1997), changes in anthropogenic emissions such as those reported in the NEI and shown in Table 1 also play a role. Moreover, since these changes are typically caused by control programs designed to improve public health, the model’s ability to capture the effects of such emission control program is an important consideration when applying models for health impact studies. Here, we investigate the impacts of the steep reductions in SO2 emissions in 1995 due to the implementation of Title IV of the Clean Air Act Amendments. Other studies have shown that these emission reductions led to decreases in acid deposition over the Eastern U.S. (e.g. Lynch et al., 2000). To assess whether CMAQ captures this signal, Figure 3 shows time series of observed and CMAQ aerosol SO4 concentrations in ambient air averaged over eight CASTNet monitors in the modelling domain and precipitation-weighted SO4 and NO3 concentrations measured in rainwater averaged over 25 NADP monitors. For the CASTNet time series, a moving average window of 15 days iterated five times was applied to illustrate seasonal patterns and trends. For the NADP time series, observed weekly samples and corresponding CMAQ values were weighted by the weekly precipitation amount and were then aggregated to compute annual average concentrations for each year from 1988 to 2000 at each site; results were averaged over all sites for presentation in this figure. The CASTNet results show that CMAQ general captures the magnitude and seasonal fluctuations of observed aerosol SO4 concentrations as well as the decrease after 1995. The comparison of the observed and simulated annual-average rainwater SO4 concentrations at the NADP sites shows that CMAQ captures the reduction of concentrations in the mid 1990s and also tends to capture interannual variability in observed concentrations.
Fig. 3 Time series of observed and CMAQ aerosol SO4 concentrations in ambient air averaged over 8 CASTNet monitors in the modelling domain (left) and precipitation-weighted SO4 and NO3 concentrations measured in rainwater averaged over 25 NADP monitors (right). Note, a moving average window of 15 days iterated five times was applied to the CASTNet time series to illustrate seasonal patterns and trends, while annual average values are shown for the NADP time series
610
C. Hogrefe et al.
Separate analysis shows that the change between the 1989–1994 average and the 1995–2000 average concentration is –0.53 mg/l (–23%) for the observations and –0.63 mg/l (–24%) for CMAQ. Moreover, this separate analysis shows that CMAQ not only captures the overall magnitude of the change but also its spatial pattern.
4. Discussion and Next Steps As discussed above, two of the motivations for using photochemical modelling systems such as CMAQ to enhance the characterization of air quality in health impact studies are the potential ability to estimate concentrations in unmonitored locations and to estimate concentrations during unmonitored time periods and for unmonitored species. In this section, we provide examples of both applications. For estimating values at unmonitored locations, it is clear that the model errors at monitored locations discussed above need to be taken into account. One possible approach to achieve this objective is to estimate differences between observed and CMAQ concentrations at each monitor for each day, spatially interpolate these differences using inverse-distance weighting, and then add the interpolated differrence field to the gridded CMAQ output. This approach is illustrated in Figure 4 for daily maximum 8-hour ozone concentrations for July 12, 1999. It can be seen that the overpredictions of the original CMAQ output in the eastern part of the modelling domain are corrected in the combined surface. Furthermore, as intended the combined surface provides estimates of ozone concentrations in unmonitored regions. To evaluate this methodology for 8-hour daily maximum ozone, leave-one-out cross validation was performed at each site for each day during the ozone seasons of 1988–2000, and the average absolute cross-validation error was 0.2 ppb. Other potential methods for accomplishing the same objective would be to construct weighted averages of observation and model simulations or employ a hierarchial Bayesian approach as described in McMillan et al. (2007). The second potential application for CMAQ in health impact studies is to estimate total and speciated PM2.5 during time periods when measurements are not available. In the U.S., such time periods include the years prior to 1999 in which hardly any PM2.5 measurements were performed in the U.S., and the estimation of PM2.5 on days when filter samplers which typically follow a one-in-three days schedule are not operating. Figure 5 shows the CMAQ simulated composition of PM2.5 over the entire simulation time period at the location of CASTNet monitors as stacked time series. It can be seen that SO4 is the major component of PM2.5 in this modelling domain but that its importance has decreased as a result of the SO2 emission reductions discussed earlier. There also is a clear seasonal variation in the composition of PM2.5. While the accuracy of the simulated species concentrations shown in these figures cannot be evaluated due to the lack of detailed speciated PM2.5 for most of the simulation time period, the limited analysis of aerosol SO4 concentrations from CASTNet and SO4 concentrations in rainwater from the NADP network in the previous section indicates that the seasonal-scale fluctuations and trends of this important constituent are well captured by CMAQ, building
Long-Term Regional Air Quality Modelling in Support of Health Impact Analyses
611
25
20
15
10
OM EC NH4 NO3 SO4
5
0 Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan- Jan88 89 90 91 92 93 94 95 96 97 98 99 00
Fig. 4 Example illustration for combining observations and CMAQ simulations. The maps show daily maximum 8-hour ozone concentrations for July 12, 1999 for observations (upper left), CMAQ (upper right), spatially interpolated differences (lower left), and the combined observation/CMAQ surface (lower right)
Fig. 5 Example for estimating total and speciated PM2.5 concentrations for unmonitored time periods. The figure shows a stacked time series of CMAQsimulated PM2.5 spcecies concentrations (bottomto-top: SO4, NO3, NH4, elemental carbon – EC, and organic mass – OM) for January 1, 1988 – December 31, 2000 at the location of CASTNet monitors after the application of a 15-day moving average filter iterated five times
confidence in the use of these estimates for health impact studies. However, it is also necessary to point out that the evaluation of NO3 and organic aerosols simulated for more recent time periods by various studies revealed larger uncertainties compared to SO4 (e.g. Eder and Yu, 2006). Therefore, once CMAQ simulations are completed for more recent time periods covered by speciated observations, methods will be explored to determine species-specific bias adjustments approaches that take into account seasonal and spatial (urban vs rural) fluctuations in model performance. Despite these uncertainties, it is expected that the information about PM2.5 mass and speciation derived from CMAQ (i.e. based upon our current understanding of the interactions between emissions, meteorology, and atmospheric processes) for unmonitored time period will help to study the link between PM2.5 concentrations and health impacts. In the future, daily maps of ambient ozone and PM2.5 concentrations constructed from observations and CMAQ simulations will be used to study the link between air quality and health. More importantly, these results will be compared to health impact studies that relied solely on ambient measurements of air pollution (e.g. Bell et al., 2004; Samoli et al., 2005) to assess the utility of long-term CMAQ simulations in enhancing the characterization of ambient air quality for health impact studies. Furthermore, it will of interest to develop methodologies that take into account uncertainties in the estimated air pollution surfaces during the health analysis. Finally, the eventual goal is to create a framework in which the impact of emission control programs is quantified both in terms of changes in ambient pollutant concentrations and in terms of changes in health outcomes through the integrated use of observations and photochemical models.
612
C. Hogrefe et al.
Acknowledgments The work presented here was supported by the National Oceanic and Atmospheric Administration under award NAO40AR4310185185, but it has not been subjected to its required peer and policy review. Therefore, the statements, findings, conclusions, and recommendations are those of the authors and do not necessarily reflect the views of the sponsoring agency and no official endorsement should be inferred. Christian Hogrefe also gratefully acknowledges partial support for this work through a research fellowship from the Oak Ridge Institute for Science and Education (ORISE).
References Bell ML (2006) The use of ambient air quality modeling to estimate individual and population exposure for human health research: a case study of ozone in the Northern Georgia Region of the United States. Environ. Intern. 32, 586–593, doi:10.1016/j.envint.2006.01.005 Bell ML, McDermott A, Zeger SL, Samet JM, Dominici F (2004) Ozone and shortterm mortality in 95 US urban communities, 1987–2000. J. Am. Med. Assoc. 292, 2372–2378. Byun DW, Schere KL (2006) Review of the governing equations, computational algorithms, and other components of the Models-3 Community Multiscale Air Quality (CMAQ) modeling system. Appl. Mech. Rev., 59, 51–77. Eder B, Yu S (2006) A performance evaluation of the 2004 release of Models-3 CMAQ. Atmos. Environ., 40, 4811–4824 Grell GA, Dudhia J, Stauffer D (1994) A description of the fifth-generation Penn State/NCAR Mesoscale Model (MM5), NCAR Technical Note, NCAR/TN-398 + STR. Hogrefe C, Ku J-Y, Sistla G, Gilliland A, Irwin JS, Porter PS, Gégo E, Kasibhatla P, Rao ST (2007) Has the performance of regional-scale photochemical modelling systems changed over the past decade?, preprints, NATO 28th ITM, Aveiro, Portugal, September 25–29, 2007. Houyoux, MR, JM Vukovich, CJ Coats, Jr, NJM. Wheeler, P. Kasibhatla (2000) Emission inventory development and processing for the seasonal model for regional air quality. J. Geophys. Res., 105, 9079–9090. Lynch JA, Bowersox VC, Grimm JW (2000) Changes in sulfate deposition in eastern USA following implementation of Phase I of Title IV of the Clean Air Act Amendments of 1990. Atmos. Environ., 34, 1665–1680. McMillan, NJ, Holland DM, Morara M (2007) Combining Different Sources of Fine Particulate Data Using Bayesian Space-Time Modeling, in preparation. Rao ST, Zurbenko IG, Neagu R, Porter PS, Ku JY, Henry RF (1997) Space and time scales in ambient ozone data. Bull. Amer. Meteor. Soc., 78, 2153–2166. Samoli E, Analitis A, Touloumi G, Schwartz J, Anderson HR, Sunyer J (2005) Estimating the exposure–response relationships between particulate matter and mortality within the APHEA multicity project. Environ. Health Perspect., 113, 88–95.
Long-Term Regional Air Quality Modelling in Support of Health Impact Analyses
613
U.S. EPA (2005) Criteria pollutant emissions summary files extracted from the national emissions inventory (NEI) database, available online at http://www. epa. gov/ttn/chief/net/critsummary.html Willmott CJ (1982) Some Comments on the Evaluation of Model Performance. Bull. Amer. Meteor. Soc., 63, 1309–1313.
Discussion S.T. Rao: With regard to model’s ability to better capture the temporal variability than the spatial variability, would the spatial homogeneity in the monitors (more monitors in urban than in rural areas) contribute to the lower correlation in space? Also urban influences may have to be accounted for incoming models and observed variables? C. Hogrefe: The uneven spacing of monitors between urban and rural areas may contribute to the lower correlations of observed-vs-simulated spatial maps of daily maximum 8-hour ozone compared to the correlations of observed-vs-simulated time series of daily maximum 8-hour ozone. However, the lower correlations persist even when spatial and temporal moving average windows are applied to the raw data, thereby smoothing out most of the urban/rural differences. Therefore, I do believe that most of the lower correlations for spatial patterns are due to errors in the placement of plumes caused by uncertainties in wind fields and emission patterns, while the higher correlations for time series are at least partially attributable to the use of four-dimensional data assimilation in the meteorological model that ensures the correct timing of synoptic events.
7.5 Modelling of the Exposure of Urban Populations to PM2.5, NO2 and O3, and Applications in the Helsinki Metropolitan Area in 2002 and 2025 J. Kukkonen, P. Aarnio, A. Kousa, A. Karppinen, K. Riikonen, B. Alaviippola, M. Kauhaniemi, J. Soares, T. Elolähde and T. Koskentalo
Abstract A mathematical model EXPAND (EXposure model for Particulate matter And Nitrogen oxiDes) can be used to evaluate human exposure to air pollution in an urban area. The model combines the predicted concentrations and the information on the time use of the population on an hourly basis. The model allows for the exposure in residences, workplaces and traffic, and partly also in other activities, such as recreational facilities. The model has been integrated to an urban dispersion modelling system of the Finnish Meteorological Institute. The computed results are processed and visualised using the GIS system MapInfo. The numerical results contain the predicted hourly spatial concentration, time activity and exposure distributions of PM2.5, NO2 and O3 in 2002 and 2025 in the Helsinki Metropolitan Area.
1. Introduction Ambient air pollution has been associated to excess mortality and morbidity at the current urban levels (e.g. Pope and Dockery, 2006; WHO, 2006). Air pollution is an additional risk factor that increases the statistical probability of death and other adverse health effects caused primarily by cardio-vascular and respiratory diseases. Most of the epidemiological studies have been based on air pollution concentrations at fixed ambient air quality monitoring sites. However, the measurement data from these stations does not necessarily represent areas beyond their immediate vicinity, as the concentrations of pollutants in urban areas may vary by orders of magnitude on spatial scales varying from tens to hundreds of metres. Therefore there is a need to model the population exposures to pollutants. In order to evaluate comprehensively the adverse health effects, it would be ideal to evaluate both the exposure of selected individual citizens, and the spatial and temporal variations of the exposures of the whole urban population. However, this is not commonly possible, due either to the lack of relevant input data, or the physical and computational limitations of exposure models, or both reasons. Exposure can be modelled by deterministic or probabilistic models (the latter have C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
632
Modelling of the Exposure of Urban Populations to PM2.5, NO2 and O3
633
been discussed by, e.g., Hänninen et al., 2003). The probabilistic models predict the exposure distributions; however, the results cannot be presented spatially, using GIS techniques. Jensen (1999) and Gulliver and Briggs (2005) have developed exposure models which can combine time-activity, dispersion modelling, and GIS techniques. In both of these models, the exposure of a few individuals is evaluated in different microenvironments during the day; this approach does not therefore allow for the evaluation of the population exposure. We have developed a primarily deterministic mathematical model, EXPAND (EXposure model for Particulate matter And Nitrogen oxiDes), for the determination of human exposure to ambient air pollution in an urban area (Kousa et al., 2002). The model can be used to evaluate the spatial and temporal variation of the average exposure of the urban population to ambient air pollution in different microenvironments. This paper describes a new model version, in which the time activity sub-model has been refined. The refined model includes a detailed treatment of the time use of population in various traffic modes, including cars and buses, trains, trams, metro, pedestrians and cyclists. The model has also been extended to contain a simple treatment of the infiltration of pollutants from outdoor to indoor, and the model can treat separately various population sub-groups. The emission and atmospheric dispersion modules of the system have been refined to include an improved treatment of fine particulate matter (PM2.5).
2. Methodology 2.1. Emission modelling We have modelled the traffic flows in the street network of the Helsinki Metropolitan Area using the EMME/2 interactive transportation planning package. The model generates traffic demand on the basis of given scenarios, and allocates the activity over the links (i.e. segments of road or street) of this network, according to predetermined set of rules and individual link characteristics (Laurikko et al., 2003). According to the link characteristics and number of vehicles, the software computes the average speed of vehicular traffic for each link on given time of the day. Furthermore, both weekly and seasonal variations of the traffic density are taken into account. For modelling purposes, the profiles of vehicle speed and vehicle numbers are then computed for each link for each hour of the day (separately for weekdays, Saturdays and Sundays), and further aggregated over the year. Emissions are computed for each link using average speed-dependent functions, determined separately for each vehicle category. Vehicle categories for 14 different vehicle types are passenger cars, divided to petrol cars with or without a catalytic converter, and diesel-fuelled vehicles, as well as busses and heavy duty vehicles. The division of the vehicles within the passenger car category is based on the registration statistics. The traffic demand generated by the model is governed by the assumed socioeconomic urban structure and location of the main activities, such as residential
634
J. Kukkonen et al.
areas and workplaces, as well as the usage rate of public transport. Local urban bus routes are directly integrated in the model, but incoming/outgoing coach traffic is generated separately. The activities at two major ports in the city of Helsinki also enhance heavy duty vehicle traffic into the arteries in the road and street network.
2.2. Dispersion modelling Dispersion modelling is based on the combined application of the Urban Dispersion Modelling system (UDM-FMI) and the road network dispersion model (CAR-FMI), developed at the Finnish Meteorological Institute (FMI) (e.g., Karppinen et al., 2000, 2004; Kousa et al., 2001). Clearly, the main limitation of such so-called second generation Gaussian dispersion models is that they do not allow for the detailed structure of buildings and obstacles. Approximately 5,000 road and street links were included in the computations. The model uses meteorological parameters from the FMI database and computes ambient air pollution concentrations for each hour over the whole year. The concentrations were computed in an adjustable grid, in which the spatial resolution is dependent on the distance of receptor points to the nearest traffic sources; these range from 5 m in the immediate vicinity of major roads to 1 km in the least polluted areas.
2.3. Activity modelling We obtained the information on the location of the population from the data set that is collected annually by the municipalities of the Helsinki metropolitan area. This data set contains data on the dwelling houses, enterprises and agencies located in the area. The data set provides geographic information on the total number and age distribution of people living in a particular building or the total number of people working at a particular workplace. The information on the number and location of people in shops, restaurants and other recreational activities is also based on this data set. The location of people in traffic was evaluated using the computed traffic flow information; this information is available separately for buses, cars, trains, trams, metro, pedestrians, and cyclists for each street and rail section on an hourly basis. However, this information does not identify individual persons. The time-microenvironment activity data is based on the time use survey by Statistics Finland. The time activity data were collected from 813 randomly selected over ten-year old inhabitants in the Helsinki metropolitan area. For our model the time-activity of the population was divided into four main categories: home, workplace, traffic, and other activities.
Modelling of the Exposure of Urban Populations to PM2.5, NO2 and O3
635
2.4. Exposure modelling The residential co-ordinates are combined with the information on the number of inhabitants at each building and the time spent at home during the day. Correspondingly, for the workplace co-ordinates, the number and age distribution of the personnel, and the time spent at the workplace are combined. The population activities at other locations (shops, cinemas, theatres, opera, libraries, restaurants, cafes, pubs, etc.) are also evaluated using statistical information of leisure time (City of Helsinki Urban Facts, 2003). The number of persons in traffic is evaluated based on the predicted traffic flows. In the case of buses, trains, metro, trams and also pedestrians and cyclists, the number of persons and the time they spend in each street or rail section is estimated using the traffic-planning model EMME/2. In the case of private cars, the EMME/2 model predicts the number of cars; we assumed that the number of passengers in each car is equal to the average value in the area, i.e., 1.33 (Hellman, 2004). The ratio of pollutant concentrations in indoor and outdoor air (I/O) is also included in the model. The I/O ratio data is based on the results of EXPOLIS study (Hänninen et al., 2004). The I/O ratios of 0.59 for PM2.5 and 0.71 for NO2 were used for buildings, and 1.0 for traffic (both for PM2.5 and NO2). Clearly, this is a very simple approach; for a more detailed evaluation of indoor concentrations, one would need reliable data on the I/O ratios in terms of the building characteristics, and especially for various traffic modes. In the model, the air pollutant concentrations are interpolated on to a rectangular grid. The data regarding population activities (number of persons * hour) is also transformed to the same grid. The GIS system MapInfo is subsequently utilised in the post-processing and visualisation of this information.
3. Results and Discussion 3.1. Predicted concentrations and their comparison with measured data The concentrations of PM2.5, NO2 and O3 were predicted for 2002, and also for selected scenarios for 2025. The model predictions have also been compared with the data measured at urban monitoring stations. Both the daily and hourly averaged predicted PM2.5 concentrations agreed fairly well with the measured data both at an urban roadside and urban background station. For instance, the correlation coefficient squared between the measured and predicted sequential daily averaged PM2.5 concentrations were 0.54 and 0.60 for these two stations. For a more detailed description of the concentration computations regarding PM2.5, and their comparison to measured data, the reader is referred to Kauhaniemi et al. (2007).
636
J. Kukkonen et al.
3.2. Definition of the scenarios for the future The scenario for 2025 is based on the so-called transport system plan that has been projected by the Helsinki Metropolitan Area Council. The transport system plan takes into consideration all aspects of the transport system within the metropolitan area. It defines common objectives and focal points for the long-term development of the transport system, allowing also for the conceivable development of the regional transport policies. For instance, it is assumed that there will be 1,133,500 inhabitants, 662,400 working places and approximately 456 cars per 1,000 inhabitants in the Helsinki Metropolitan Area in 2025. The corresponding figures for 2000 are 928,950 inhabitants, 540,401 working places and 348 cars per 1,000 inhabitants. This plan also estimates, how land use and transport will evolve up to 2025 (e.g., it assumes new underground and railway lines, and new dwelling areas).
Fig. 1 a, b The predicted number of persons that is exposed to the concentration of nitrogen dioxide exceeding 30 µg/m3 during the morning rush hour in 2002 in the Helsinki Metropolitan Area, on a characteristic winter day (upper figure), and during a day with a prevailing surface temperature inversion (lower figure)
Modelling of the Exposure of Urban Populations to PM2.5, NO2 and O3
637
3.3. Examples of the spatial and temporal variation of exposure
The number of working age population
The spatial distribution of population that is exposed to a NO2 concentration exceeding a specified value during a morning rush hour on a characteristic day in winter has been presented in Figure 1a. The ‘characteristic’ winter day was selected to represent meteorological dispersion conditions that are common in winter, especially regarding the values of temperature, wind speed and stability. As expected, during the morning rush hour the working age population is mostly located on major traffic routes and in the residential areas. e.g., during the morning rush hour, two thirds of the working age population was in the environments where they exposed to the nitrogen dioxide concentrations less than 20 µg/m3. These environments were mostly homes or workplaces. People were exposed to the highest concentrations in traffic, near the busy traffic lanes and in the city centre. For comparison purposes, the corresponding results have been presented in Figure 1b for an episodic day, during which high concentrations were mainly caused by the local vehicular emissions due to a ground-based temperature inversion. There are substantial differences both in the magnitude and the spatial distribution of the exposures of the working age population for these two cases. For instance, during the episodic case, the number of population exposed to nitrogen dioxide concentration larger than 30 µg/m3 is more than 2.2 times higher than that during the typical winter day. The distributions of the number of persons in terms of concentrations are presented in Figure 2. 200000 180000 160000
Winter
140000
Inversion
120000 100000 80000 60000 40000 20000 0 <10
10-20 20-30 30-40 40-50 NO2 concentration (µg/m3)
>50
Fig. 2 The distributions of the number of the working age population exposed to the selected ranges of the NO2 concentrations in the Helsinki Metropolitan Area during the morning rush hour in 2002 for two cases: (i) a characteristic winter day (legend “Winter”), and (ii) an episodic day with a prevailing ground-based temperature inversion (legend “Inversion”)
3.4. Predicted results for the scenario for 2025 The transport system plan contains several alternative scenarios for the future. We have applied here only the basic “business-as-usual” –type scenario. The working age population was predicted to be exposed to considerably lower concentrations
638
J. Kukkonen et al.
in 2025, compared with the corresponding results for 2002. According to the basic 2025 scenario, over 80% of the working age population would be exposed to the nitrogen dioxide concentrations less than 10 µg/m3, whereas the corresponding percentage for the working age population in 2002 was approximately 25%. The substantially lower exposures of NO2 in 2025 are mainly caused by the projected lower NOx emissions from local vehicular traffic.
4. Conclusions The numerical results contain the predicted hourly spatial concentration, time activity and exposure distributions of PM2.5, NO2 and O3 in 2002 and 2025 in the Helsinki Metropolitan Area. The results illustrate the most problematic areas and time periods with concurrent high population activity and concentration values. The highest population exposures occur especially in the centre of Helsinki, and along the major traffic routes. As expected, the population exposures are also temporally highly variable diurnally, seasonally and in terms of specific meteorological conditions. The methodologies developed, and the EXPAND model itself, are available to be utilised also in other European urban areas (e.g., Baklanov et al., 2007), and within other integrated modelling systems (e.g., Sokhi et al., 2007). The model, including the GIS-based methodology, could also be extended on a regional scale. It would be especially important to evaluate the exposure of children; however, the data on their time use is scarce. The projections of the time activities for the future also involve major uncertainties. For instance, it is conceivable that both the magnitude and types of time activities during the leisure time will change dramatically in the future. Acknowledgments The study was supported by the EU-funded FUMAPEX and OSCAR projects, and it is part of the CLEAR cluster of air quality projects. We also wish to acknowledge the funding of the Academy of Finland (TERVE, project no 53246).
References Baklanov A, Hänninen O, Slørdal LH, Kukkonen J, Bjergene N, Fay B, Finardi S, Hoe SC, Jantunen M, Karppinen A, Rasmussen A, Skouloudis A, Sokhi RS, Sørensen JH, Ødegaard V (2007) Integrated systems for forecasting urban meteorology, air pollution and population exposure. Atmos. Chem. Phys., 7, 855–874, www.atmos-chem-phys.net/7/855/2007/ City of Helsinki Urban Facts (2003) Statistical Yearbook of the City of Helsinki. Gummerrus Kirjapaino Oy, Jyväskylä. Gulliver J, Briggs D (2005) Time-space modelling of journey-time exposure to traffic-related air pollution using GIS. Ph.D. thesis. Environmental Research 2005, 97 (1), 10–95.
Modelling of the Exposure of Urban Populations to PM2.5, NO2 and O3
639
Hellman T (2004) The average burden of the private vehicles in Helsinki in 2004. City Planning Department, City of Helsinki. Memo 21.6.2004. Helsinki (in Finnish). Hänninen O, Kruize H, Lebret E, Jantunen M (2003) EXPOLIS simulation model: PM2.5 application and comparison with measurements in Helsinki. J. Exp. Anal. Environ. Epidemiol. 13, 74–85. Hänninen O, Lebret E, Ilacqua V, Katsouyanni K, Künzli N, Sram R, Jantunen M (2004) Infiltration of ambient PM2.5 and levels of indoor generated non-ETS PM2.5 in residences of four European cities. Atmos. Environ. 38, 6411–6423 Jensen SS (1999) A Geographic Approach to Modelling Human Exposure to Traffic Air Pollution using GIS. Ph.D. thesis. National Environmental Research Institute, Denmark. Karppinen A, Kukkonen J, Elolähde T, Konttinen M, Koskentalo T, Rantakrans E (2000) A modelling system for predicting urban air pollution, Model description and applications in the Helsinki metropolitan area. Atmos. Environ. 34–22, pp. 3723–3733. Karppinen A, Härkönen J, Kukkonen J, Aarnio P, Koskentalo T (2004) Statistical model for assessing the portion of fine particulate matter transported regionally and long-range to urban air. Scand. J. Work Environ. Health, 30 suppl. 2: 47–53. Kauhaniemi M, Karppinen A, Härkönen J, Kousa A, Koskentalo T, Aarnio P, Kukkonen J (2007) Refinement and statistical evaluation of a modelling system for predicting fine particle concentrations in urban areas. In: R.S. Sokhi, M. Neophytou (eds), Proceedings of the 6th International Conference on Urban Air Quality, Limassol, Cyprus, 27–29 March 2007, CD-disk: ISBN 978-1-90531346-4, University of Hertfordshire and University of Cyprus (pp. 68–71). Kousa A, Kukkonen J, Karppinen A, Aarnio P, Koskentalo T (2001) Statistical and diagnostic evaluation of a new-generation urban dispersion modelling system against an extensive dataset in the Helsinki Area. Atmos. Environ., Vol. 35/27, pp. 4617–4628. Kousa A, Kukkonen J, Karppinen A, Aarnio P, Koskentalo T (2002) A model for evaluating the population exposure to ambient air pollution in an urban area. Atmos. Environ. 36, 2109–2119. Laurikko J, Kukkonen J, Koistinen K, Koskentalo T (2003) Integrated modelling system for the evaluation of the impact of tranport-related measures to urban air quality. In: Proceedings of the 12th symposium “Transport and Air Pollution”, 16–18 June 2003, Avignon, France. Pope CA, Dockery DW (2006) Health effects of fine particulate air pollution: lines that connect. J. Air Waste Manage. Assoc. 56, 709–742. Sokhi, R.S, Hongjun Mao, Srinivas TG, Srimath, Shiyuan Fan, Nutthida Kitwiroon, Lakhumal Luhana, Jaakko Kukkonen, Mervi Haakana, K Dick van den Hout, Paul Boulter, Ian S McCrae, Steinar Larssen, Karl I Gjerstad, Roberto San Jose, John Bartzis, Panos Neofytou, Peter van den Breemer, Steve Neville, Anu Kousa, Blanca M Cortes, Ari Karppinen and Ingrid Myrtveit (2007). An integrated multi-model approach for air quality assessment: development and evaluation of the OSCAR air quality Assessment system. Environ. Mod. Software (in print). WHO (2006) Air quality guidelines for particulate matter, ozone, nitrogen dioxide and sulphur dioxide. Global update 2005. Summary of risk assessment.
7.1 Models of Exposure for Use in Epidemiological Studies of Air Pollution Health Impacts Michael Brauer, Bruce Ainslie, Michael Buzzelli, Sarah Henderson, Tim Larson, Julian Marshall, Elizabeth Nethery, Douw Steyn and Jason Su
Abstract Observational epidemiological studies have had an important role in understanding the public health impacts of air pollution. In such studies, accurate assessment of exposure remains a major challenges, especially in studies involving large populations. Here we review state-of-the-art approaches to assessment of population exposure in epidemiological studies with a focus on approaches applied in the Border Air Quality Study (www.cher.ubc.ca\baqs.htm). The strengths and limitations of these methods are discussed and future research needs identified.
Keywords Air quality, epidemiology, exposure assessment, health effects, land use regression, vehicle emissions, wood smoke
1. Introduction Observational epidemiological studies have assumed an important role in understanding the public health impacts of air pollution. Although laboratory toxicologycal studies and controlled human exposure experiments continue to provide new knowledge, especially regarding biological mechanisms, epidemiological studies are increasingly the main measure upon which policies, standards, guide-lines and regulations are based. Exposure assessment is a major challenge in epidemiological studies of air pollution health effects, especially those involving large study populations from which population health inferences can be made. While measurement of environmental concentrations of air pollutants is reasonably routine, especially for common air pollutants, assessing exposure – the intersection over time and space between humans and concentrations in air – incorporates additional layers of complexity. Importantly, in addition to spatial and temporal variability in ambient concentrations, exposure contains a behavioural component that is also complex and variable (Nieuwenhuijsen, 2003). Nieuwenhuijsen, in an adaptation from an earlier National Research Council Report, describes general approaches to exposure assessment for epidemiological studies which have been modified here (Figure 1) to describe air pollution exposure assessment:
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
589
590
M. Brauer et al.
Fig. 1 Approaches to assess air pollution exposure for epidemiological studies
Although continuous measurements of personal exposure for all study participants throughout the period of observation provides the best measure of exposure, and has been accomplished in several intensive studies of health impacts of shortterm exposure, this is seldom possible due to financial and logistical constraints and not realistic for studies of long term exposures. Further, the very act of wearing a personal air pollution monitoring device has been shown to modify individual activities and therefore likely to lead to biased measures of exposure. In contrast, in most urban areas, ambient monitoring networks provide continuous measurements of air quality for a small number of air pollutants at one or more locations, providing a ready-made and easy to exploit source of potential exposure information. In the following sections, we describe the main approaches to assess exposure for common air pollution epidemiological study designs. We emphasize, in particular, the approaches used in and developed for the Border Air Quality Study (www.cher.ubc.ca\baqs.htm).
2. Components of Exposure To simplify estimation of exposure, researchers often sub-divide exposure into different components. For example, exposure has both temporal and spatial components both of which may be sub-divided into increasing levels of resolution depending upon, for example, the health impact of interest or the spatial and temporal variability of a particular pollutant. Although outdoor air pollution is generally the area of interest from a policy perspective, an understanding of the time-activity patterns of individuals has led to an increased emphasis on exposures experienced in specific indoor microenvironments (home, work/school, in-transit) and the extent of infiltration of ambient pollutants into these environments. The microenvironment with greatest overall impact on exposure is usually the home environment, although in some cases exposures (to ambient pollutants) encountered while in transit or while outdoors may be important contributors.
Models of Exposure for Use in Epidemiological Studies of Air Pollution Health Impacts
591
3. Epidemiologic Study Designs Acute (short-term exposures) Time series studies have been conducted in hundreds of urban areas throughout the world. The general approach is to collect daily regulatory ambient air quality monitoring data for all available pollutants and to assess associations with counts of specific health outcomes, such as daily mortality, or hospital admissions. In this design the ambient monitoring data are typically averaged for the entire study area or, less frequently, counts are apportioned to data from individual monitors. Exposure on any given day is assumed to be the same for the entire study population. Although it is well-known that this assumption is poor, the relevant issue for the purposes of interpretation of the epidemiologic studies is the extent to which exposures are correlated to the overall average exposure. A number of studies have demonstrated that, for particulate matter, reasonably high correlations over time are found between ambient concentrations measured at central (urban background) locations and measured personal exposures (Ebelt et al., 2000; Janssen et al., 1999). These findings generally support the use of ambient monitoring network data in such time series studies of the short-term impacts of air pollution. For ozone, correlations are lower (Sarnat et al., 2000). In both cases the level of the correlations does appear to vary across individuals and is related to the ventilation properties of the indoor environment. Several studies have used additional information from personal monitoring to “correct” for the impact of non-ambient exposures (Ebelt et al., 2005; Koenig et al., 2005; Strand et al., 2006) and have shown that associations with health outcomes are more closely related to the component of exposure derived from ambient source pollution, compared to total measured exposure which is composed of particulate matter of indoor and outdoor origin or compared to particulate matter of indoor origin. While these studies show the importance of accurately accounting for indoor sources of exposure and the ventilation properties of indoor environments, new approaches need to be developed to account for these factors in studies of large populations where such data are not readily available. One promising area of current research is the development of models of infiltration (Allen et al., 2003). Using these models, information on building characteristics that may be readily available from property assessment data can be used to calculate building-specific factors to characterize the indoor infiltration of pollutants (Setton et al., 2005). Adjustment factors can then be applied to studies of long-term exposure if data on subject mobility (residential history, work/school locations) is known. Chronic (long-term) exposure While the health impacts of short term exposures have been repeatedly demonstrated in numerous epidemiological studies and in many cases supported by controlled human exposure studies, the population health impacts of long-term exposures are generally believed to have greater public health impacts (Kunzli et al., 2001). The study of health impacts associated with long term exposures relies mainly upon spatial comparisons between areas of differing air pollutant concentrations (Pope et al., 2002). Until recently most comparisons were
M. Brauer et al.
592
based upon comparisons between cities where health impacts and other risk factors (e.g. diet, smoking, age, occupation, etc.) are measured at the level of the individual while air pollution exposure has been based upon ambient monitoring network data (single monitors or an average of all available monitors in the study area). While this approach is reasonable in that it compares concentrations between cities (regional and local sources) as surrogates of exposure, it masks any within-city differrences in concentrations (local sources) that may exist. Moving from an assignment of subjects to the monitor nearest to their residence was reported to increase the measurement of risk associated with air pollution (Miller et al., 2007). Interpolation of available monitors to produce individual-specific estimates of exposure has also enabled studies of within-city contrasts in concentrations. In one example, estimates of risk based upon within-city contrasts were three times larger than those based upon between-city comparisons in concentrations (Jerrett et al., 2005b). Increasingly, measurement studies highlighting spatial differences in air pollutant concentrations (Gilbert et al., 2003; Zhang et al., 2004), combined with citizen concern regarding neighborhood sources of air pollution such as traffic, have led to an increase in studies designed to estimate the impact of spatial contrasts in concentrations on health effects. Given that most regional air pollution is wellcharacterized by monitoring networks, local air pollution concentration differrences require alternate approaches for characterization (Jerrett et al., 2005a). Identification of within-city contrasts in concentrations of air pollutants also has implications for the design of new measurement programs and for air quality management in general. Hoek et al. used a combination of measurement data to characterize regional and urban components of air pollution, and measures of proximity to major roads to characterize neighborhood differences related to traffic where the estimated individual exposure was a sum of regional, urban contribution and roadway contributions (Hoek et al., 2001): Exposure = Cregional + Curban + Croad
(1)
where Cregional is derived from interpolation of network monitoring, Curban is estimated from a regression relationship between air pollution and level of urbanization and Croad is assigned for those living within 100 m of a freeway or within 50 m of a major urban street. Other studies have used simple surrogates of exposure such as road proximity or traffic level on the nearest major road as surrogates of local-scale differences in pollutant concentrations (Venn et al., 2000).
4. The Border Air Quality Study In developing the exposure assessment strategy for the Border Air Quality Study, we built upon the study of Hoek et al. to enhance the assessment of traffic sources using a land use regression model (described below), adding characterization of additional sources of importance in the local airshed (woodsmoke and industrial point sources) and developing approaches to incorporate the impact of meteorology
Models of Exposure for Use in Epidemiological Studies of Air Pollution Health Impacts
593
and street canyons to more accurately characterize sources of variability in air pollution concentrations within the airshed and their impact on health outcomes: Exposure = Cregional + Curban + Ctraffic + Cwoodsmoke + Cpoint sources
(2)
Land Use Regression models Recent studies have measured and reported considerable spatial variability in the concentrations of traffic-related pollutants within urban areas. These “neighborhood scale” intra-urban differences tend not to be well-characterized by air quality monitoring networks. Land use regression (LUR) was first developed to address this shortcoming and has recently gained attention in the air quality management and urban planning communities. There is no standard method for conducting LUR (Figure 2), but detailed descriptions of can be found elsewhere (Jerrett et al., 2005a; Briggs et al., 1997; Henderson et al., 2007). In brief, a pollutant is measured at multiple sites specifically selected to capture the complete intra-urban range of its concentrations. Geographic attributes that might be associated with those concentrations (e.g. surrounding land use, population density, and traffic patterns) are measured around each measurement site in a Geographic Information System (GIS). Linear regression is used to correlate measured concentrations with the most predictive variables, and the resulting equation is used to estimate pollutant concentrations anywhere that all of the predictors can be measured. Concentration maps with high spatial resolution can be generated by rendering the regression model in GIS. Fig. 2 The LUR modelling procedure
Traffic sources Land use regression was initially developed in Europe to estimate individual-level exposure to traffic-related air pollutants for epidemiological studies. This need arose from (1) the infeasibility of collecting individual measurements for large populations and (2) inaccuracies inherent to crude surrogates such as distance to nearest road, or data from the nearest regulatory monitoring locations. With LUR, researchers were able to estimate individual exposures from statistical models that combined the predictive power of several surrogates based on their relationship with measured concentrations. Although interest in traffic-related health effects has favoured the development of LUR for traffic-related pollutants, the method is now being explored for other sources.
M. Brauer et al.
594
Published results are summarized in Table 1. Most of these studies were undertaken to provide exposure assessment for epidemiological research. Comparison of R2 values across study areas and pollutant types in Table 1 suggests that LUR produces consistent results regardless of location. Table 1 Summary of previous LUR studies. Investigator, Year
Henderson, 2007
Study location
Vancouver
Domain size (km2)
Mean NO2 (SD) (ppb)
R2 for NO2 (N sites)
Mean ABS (SD) (10-5 m-1)
16.2 (5.6)
0.56–0.60 (114)
0.84 (0.47)
0.39–0.41 (25)
–
–
1.28 (0.83)
0.56–0.65 (39)
2200
Larson, 2007
R2 for ABS (N sites)
Sahsuvaroglu, 2006
Hamilton
1400
16.4 (3.7)
0.76 (101)
–
–
Jerrett, 2006
Toronto
900
32.7 (10.5)
0.69 (95)
–
–
Gilbert, 2004
Montreal
1200
11.6 (3.0)
0.54 (67)
–
– 0.75 (24)
Ryan, 2007
Cincinnati
1600
–
–
0.67 (0.29)*
Ross, 2005
San Diego
2100
14.8 (5.7)
0.77 (39)
–
–
El Paso
800
20.6 (7.1)
0.81 (20)
–
–
Western Germany
3300
13.7
0.89 (40)
1.71
0.81 (40)
Netherlands Rotterdam Stockholm Munich
38000 200 150 80
15.4 (4.9) 17.5 (3.9) 10.1 (4.0) 15.2 (4.1)
0.85 (40) 0.79 (18) 0.73 (42) 0.62 (40)
1.64 (0.58) 1.79 (0.56) 1.29 (0.35) 1.84 (0.43)
0.81 (40) 0.77 (18) 0.66 (42) 0.67 (40)
Amsterdam
30
Huddersfield
300
0.63 (80) 0.61 (80) 0.72 (80)
–
–
Prague
50
Gonzales, 2005 Hochadel, 2006 Hoek, 2002 Brauer, 2003 (TRAPCA)
Briggs, 1997 (SAVIAH)
20.1–28.6 (3.4–6.7) 14.1–26.3 (5.2–7.8) 12.3–21.9 (5.7–9.9)
LUR Versus Dispersion Modeling One alternative to LUR is dispersion modeling, where emissions parameters are input into models that use physical and chemical equations to predict pollutant concentrations at individual receptors. While this is a common approach in risk assessment and air quality management evaluation, it is rarely used for epidemiological studies because dispersion models require specific inputs (traffic volume, motor vehicle fleet makeup, street configurations, industrial emissions, local meteorology, etc.) that may not be available for all areas. Even where complete input data exist, dispersion model operation requires considerable time, resources and expertise. In comparison, LUR allows flexibility in terms of inputs, resource requirements, and outputs. Land use regression models can be built on a location-by-location basis with whatever data are available. Sampling can be conducted at a flexible number of sites over a flexible period of time using a wide range of instrumentation.
Models of Exposure for Use in Epidemiological Studies of Air Pollution Health Impacts
595
Once data collection is complete the analyses can easily be conducted by individuals with a background in statistics and GIS. Final models can be rendered into high-resolution pollution maps. Because LUR is a stochastic approach that uses actual measurements, model estimates tend to be realistic. Dispersion models use estimated emission factors that can result in considerable disparity between model output and actual concentrations. On the other hand, dispersion models can easily be used to evaluate different emissions scenarios. Within the TRAPCA project (Brauer et al., 2003; Cyrys et al., 2007), results for LUR and dispersion models of NO2 concentrations were compared in Stockholm and Munich. In Stockholm the R2 for estimates made with the AIRVIRO1 model and measured concentrations of NO2 was 0.69, with greater correlations observed for sites located in street canyons. The LUR model had an R2 value of 0.76. The TRAPCA study concluded that AIRVIRO and LUR had similar predictive power, but the applicability of LUR in the absence of emission inventories was an attractive advantage. This finding was supported in a recent study by Cyrys et al. that compared dispersion (IMMIS net2) and LUR estimates of NO2 and PM2.5 concentrations for their study population in Munich, Germany and concluded that both methods performed equally well in estimating exposures of their study population. More recently, Briggs et al. (2006) compared LUR with a state-of-the-art dispersion model (ADMS-Urban) for NO2 and PM10 at a limited number of measurement sites (N = 18 for PM10, N = 8 for NO2) in London, England. The LUR estimates had correlations (Pearson’s coefficient, r) of 0.61 for NO2 and 0.88 for PM10 compared to the annual mean. The ADMS estimates had correlations of 0.72 and 0.81 for NO2 and PM10, respectively. These results suggest that LUR pollutant concentration estimates are of equal or better accuracy than those from dispersion models. Beyond its aforementioned flexibility, another important advantage of LUR is its applicability to specific components of particulate matter, such as elemental carbon or source-specific tracers. In contrast, sophisticated dispersion models are only available for a limited set of pollutants. In BAQS we built upon previous traffic-related pollutant LUR models in developing approaches to be used in our epidemiological analyses (Henderson et al., 2007). Briefly, passive samplers to collect NO and NO2 were deployed for two-14 day periods at 116 sites in the study area. Mean concentrations during these two periods were highly correlated with, and closely approximated annual averages from regulatory monitoring network data. Sites were selected by a locationallocation algorithm that used a crude LUR model based upon regulatory monitoring stations as a demand surface. The algorithm optimally locates samplers to maximize their ability to characterize variability in the demand surface. More samplers are located in areas predicted to have higher spatial variability. The model output was then weighted by population density to ensure adequate numbers of samplers in residential areas. In addition, PM2.5 mass was measured at a subset of 25 locations during a two-month sampling period. Integrated one-week average PM2.5 samples were collected on Teflon filters using Harvard Impactors. For a 1 http://www.indic-airviro.smhi.se/ 2 http://www.ivu-umwelt.de/e/index.html
596
M. Brauer et al.
subset of 36 sites, we measured particle absorbance (Black Carbon) using a Particle Soot Absorption Photometer in a mobile monitoring platform and adjusted these measurements for temporal variation based upon repeated measurements at a centrally-located site. For each of the 116 (and the subsets of 25 and 36) measurement sites, 55 variables were generated in a Geographic Information System (GIS) and linear regression models of NO, NO2 and Black Carbon were built with the most predictive covariates. Models were developed using both road classifications and traffic density as potential predictors. Using road classifications (to allow the model to be applied to areas without traffic density measures), for NO, the model had an R2 of 0.62 and included the number of major roads within 100 and 1,000 m radius circular buffers of the measurement sites, the number of secondary roads within a 100 m buffer, population density within a 2,500 m radius and elevation. For NO2, the model (R2 = 0.56) included the same variables as well as the amount of commercial land use within 750 m. For PM.5 the model (R2 = 0.52) included the amount of commercial and industrial land use within 300 m, the amount of residential land use within 750 m and elevation. For Black Carbon the model (R2 = 0.56) included the number of secondary roads within a 100 m buffer, distance to the nearest highway and the amount of industrial land use within 750 m. Models developed using traffic density did not show significantly higher correlations but had marginally improved agreement in evaluation analyses. The model surfaces clearly showed differences in spatial extent of primary vs secondary pollutants. For applications to epidemiological analyses, we used the models to generate smooth spatial surfaces of predicted (annual average) concentrations for the entire study area at a resolution of 10 m. The surfaces were then smoothed to remove abrupt changes and edge effects so as to more accurately reflect the measured effect of proximity to roadways (Gilbert et al., 2003). For each LUR model, the corresponding monitoring network data for each pollutant were fit with a monthly dummy variables and a covariate for linear trend (Times Series Forecasting System, SAS v 9.1). For Black Carbon, the PM2.5 trend was used as there were no corresponding regulatory monitoring network data. From these models, month-year adjustment factors were applied to each surface to estimate monthly average concentrations. Using these we then computed individual subject exposures for the same exposure windows as described above for the monitor-based approaches. Wood Smoke Residential wood smoke can be an important local source of ambient particulate matter during winter months but its distribution is often not well-characterized by regulatory monitoring networks due, in part, to the sparselylocated sources in residential areas. We developed a novel land use regression model for wood smoke based upon a mobile monitoring campaign to map the impact of wood smoke particulate matter across the BAQS study area (Larson et al., 2007). Briefly, we first identified potential hotspots for wood-burning based on property assessment data, a telephone survey on woodburning practices conducted for a local emissions inventory and topography (Su et al., 2007). A network of fixed monitors was located at potential hotspots and control sites to collect twoweek samples of PM2.5 and levoglucosan, a biomass combustion tracer compound.
Models of Exposure for Use in Epidemiological Studies of Air Pollution Health Impacts
597
The monitoring of levoglucosan was conducted to confirm that PM2.5 concentrations assessed by mobile monitoring and used to model spatial patterns were associated with wood smoke. On 19 cold, clear nights (9pm–1am) during the heating season we conducted mobile sampling in a vehicle equipped with a logging GPS and light-scattering nephelometer. Two routes were pre-selected to cover target areas of predicted variability in woodsmoke concentrations, traverse populated areas, and circumnavigate the fixed-location monitoring sites. These campaigns generated more than 12,000 pairs of geospatial coordinates and light-scattering coefficients (bsp) that were temporally-adjusted and merged into a single, high-resolution file for LUR analysis. To generate data for linear regression the model domain was divided into ~50 air catchments (based upon hydrological catchment basins), assuming that a given location is systematically downwind of uphill sources under stable meteorological conditions (e.g. cold, clear nights). The bsp values and predictive variables were averaged at the catchment level, and all uphill catchments within an 8km radius were assumed to contribute to the mean bsp of the downhill catchment. Variables describing the population, ethnic composition, economic status, buildings, and wood-burning appliance usage in each catchment produced an R2 value of 0.64. A similar mobile monitoring campaign was also conducted in a second geographically distinct region and comparable model results were obtained. For epidemiological analyses, the catchment area values of woodsmoke PM concentrations were assigned to tertiles with the highest tertile areas considered to be woodsmoke-exposed. Since woodsmoke is emitted seasonally, the surface was “activated” for epidemiological analyses only during the heating season. We calculated Heating Degree Days (HDD) (using 18ºC as the base value) for each of twelve two-week monitoring periods during which we collected measurements of levoglucosan and regressed the mean levoglucosan concentrations (from all six sites) against the mean HDD for the corresponding period (R2 = 0.60). From this regression, we estimated that a HDD of 12.0 was equivalent to approximately 115 ng/m3 of Levoglucosan which is above a concentration of ~100 ng/m3 at which woodsmoke-impacted areas and woodsmoke events can be differentiated from other areas and periods. Accordingly, days with HDD>12.0 were considered “woodsmoke days” and the spatial surface was applied on these days only. Point Sources The approach taken for point-source (i.e., industrial) emissions was a simple alternative to more complex dispersion/photochemical models. The metric is a proxy for residential exposure to industrial point-source emissions but no actual concentrations are estimated. We built upon the work of Yu et al. (2006) in which fixed size and shape (3 km radius and 90q angle) wedges that partially account for wind direction were used to estimate exposures from petrochemical facilities. We simplified the approach to use uniform circles, thus ignoring wind direction, to account for the hundreds of industrial point sources in our study region. The point-source exposure metric is a proximity-weighted summation of relative emissions within a given radius of each postal code (PC) centroid. The metric incurporates point source emissions and locations, postal code locations and a cut-off threshold distance. Point-source emissions and locations, used as inputs into the
M. Brauer et al.
598
CMAQ model, were obtained for CO, NOX, VOC, NH3, SO2, PM10, PM2.5, and coarse mode PM (i.e., PM10-PM2.5). If the distance between a PC and an emission source is greater than a certain value, that emission source is ignored when evaluating that PC. Given the nature of air dispersion, the variability among sources in stack height and plume rise, and the influences of complex topology and meteorology, it is not possible to identify a single value as the “correct” cut-off threshold distance. Two cut-off threshold distance values, 10 and 40 km, were separately employed to represent the approximate distance needed for a point-source plume to fully mix throughout the atmospheric mixing height. Both values (10 and 40 km) were derived from PasquillGifford curves and offer an order-of-magnitude estimate of the “impact zone” for a point source. The true size of an impact zone will vary widely over time and among sources, based on parameters such as emissions, stack height, exit velocity, meteorology, and topography. The point-source exposure metric is calculated using the following formula: Wi
¦ ^ `
j ; d ij ! x
Ej
dij
(3)
where Wi is the point-source exposure metric for postal code i, Ej is the relative emissions for point source j, and dij is the distance between postal code i and point source j. The summation is made for all point sources within distance x of the PC. To calculate relative emissions (Ej) for a point source j, emissions for each point source are first converted from a raw emission rate (tons per day) into the percentile of that source among all emitting point sources. This step is repeated for each of four pollutants (PM2.5, SOx, NOx, and VOCs). For example, a point source that does not emit SOx is assigned a percentile of zero; a source whose SOx emissions are at the 85th percentile (i.e., 85% of the SOx-emitting point sources have an emission rate that is less than this source) is assigned a SOx value of 0.85. Next, percentile scores for the four pollutants are summed to yield the relative emissions for a specific point source. The largest relative emission rate is 3.96, for a specific point source in the 99th emission percentile for all four pollutants. The lowest relative emission rate is zero, representing sources with no emissions of the four pollutants. In the Vancouver study area, each postal code has an average of 173 point sources within 10 km and 753 point sources within 40 km. Mean (st dev) values for the point-source exposure metric are 21.6 (21.8) for x = 10 km, and 41.5 (27.6) for x = 40 km. Geometric means (GSD) are 12.7 (3.5) for x = 10 km, and 30.3 (2.5) for x = 40 km. The correlation between the two exposure metrics is very high (r = 0.91). For epidemiological analyses the raw values of the point source metric were assigned to each postal code of interest as a continuous variable. Meteorology The main strength of land use regression (LUR) models is the empirical structure of the regression mapping and its relatively simple inputs/low cost (Jerrett et al., 2005a). However, the method is case- and area-specific (Briggs et al., 1997), has to date not been temporally resolved, and is aimed at long-term
Models of Exposure for Use in Epidemiological Studies of Air Pollution Health Impacts
599
average concentrations. Also there is little theoretical-physical basis behind its application, particularly the use of circular buffers to extract local covariates for exposure estimates. By contrast dispersion models have the potential advantage of incorporating both spatial and temporal variation of air pollution within a study area. However, dispersion models often unrealistically assume a Gaussian distribution of dispersion patterns, require extensive cross-validation with monitoring data and have relatively costly data input (Jerrett et al., 2005a). Some have addressed both of these issues, such as the use of monthly exposure fixed-sized “wedges” to identify potentially exposed areas using the monthly prevailing wind directions (Yu et al., 2006). Using wind fields, Arain et al. (2007) marginally increased NO2 concentration precision for locations downwind of major highways, while we used hydrological catchment area and uphill search algorithms rather than circular buffers for modeling residential woodsmoke PM2.5 concentrations (Larson et al., 2007). These models represent a step forward in terms of conceptualizing the physical pathways of exposure by contrast with LUR, for which a more general theoretical basis has still not been developed. To enhance the tools available for assessment of exposure to local sources, a source area analysis (SA) model (Ainslie et al., 2008) incorporating spatially and temporally refined components of ambient exposures was developed and applied to LUR covariates. This model accounts for wind speed and direction in modifying the zone of influence over receptor sites. By incorporating varying meteorological parameters while maintaining the same land use covariates, we also introduce a temporal component that allows estimation of ambient concentrations/exposures on shorter time scales. Briefly, the model uses a 3D wedge to estimate pollutant concentration. The radius (i.e., length of a wedge) is given by the distance travelled by wind in one hour and wind speed, direction, cloud cover/insolation and time of day are also used to estimate the atmospheric stability classes, wedge angle and wedge depth. Circular buffers are used for stability class A, B, E and F, and a 90o wedge for classes C and D. Wedge depth equals atmospheric mixing height and takes on a constant value at night and in the early morning for convective (A and B) classes. During the day the wedge depth of the convective classes is assumed to linearly increase reaching a peak value late in the afternoon before rapidly dropping in the early evening. For neutral (C and D) and stable (E and F) classes, the wedge depth is assumed, respectively, proportional to and square root of the wind speed. An evaluation test of the source-area and simple buffer model was performed using measurements taken at the 116 field sites used to build the LUR model (Henderson et al., 2007) and is described in detail elsewhere (Su et al., 2008). When the source area models are assigned the same radii of influence as in the LUR models, the two approaches yield similar amounts of explained variance. To investigate whether the performance of the source area model was limited by the need to interpolate windspeed and direction data to the 116 sampling sites and the aggregation of hourly data to seasonal averages, we conducted a similar comparison for the 19 regulatory monitoring sites with available meteorological data. When both hourly measurements and estimations are aggregated to seasonal averages the unmeasured hourly variability is eliminated. Accordingly the prediction powers of the source area model increases significantly and performs much better
600
M. Brauer et al.
than the LUR model using circular buffers, a result similar to the initial comparison described above. The new models have (compared to LUR) an increase of variance explained from 60% to 81% for NO prediction and from 75% to 87% for NO2. Using hourly meteorological data from a single location, reduced the source area model’s predictive power, but still showed improvement over the circular buffer approach. The better prediction powers of the source area model over LUR circular buffer model for the 19 monitoring stations demonstrates the potential strength of the new approach when we scale up temporal variations that are lost with a more typical cross-section approach in LUR. Since the LUR model’s covariates do not rely on a Gaussian assumption and do not need background concentrations, less computing power than for a dispersion model is required.
5. Conclusions Exposure assessment approaches for epidemiological studies of air pollution health impacts have progressed dramatically in recent years from traditional reliance on available regulatory ambient monitoring network data. In particular, improved approaches to characterize high resolution spatial concentration differences have been developed and applied in a number of studies. In most situations these models have focused on characterizing the impacts of traffic sources, but applications to point sources and residential wood combustion have also been developed. A limitation of existing spatial models is their emphasis on long-term average concentrations and their imperfect characterization of temporal variability in air pollution concentrations. New approaches to incorporate time-varying meteorological data into high resolution spatial concentration models are being developed but have seen only limited applications to date.
References Ainslie B, Steyn DG, Su J, Buzzelli M, Brauer M, Larson TV & Rucker M (2008) “A Source Area Model Incorporating Simplified Atmospheric Dispersion and Advection at Fine Scale for Population Air Pollutant Exposure Assessment”, Atmos Environ. 42, 2394–2404. Allen R, Larson T, Sheppard L, Wallace L, Liu LJ (2003) “Use of real-time light scattering data to estimate the contribution of infiltrated and indoor-generated particles to indoor air”, Environ Sci Technol, 37, 16, 484–3492. Arain MA, Blair R, Finkelstein N, Brook JR, Sahsuvaroglu T, Beckerman B, Zhang, L, Jerrett M (2007) “The use of wind fields in a land use regression model to predict air pollution concentrations for health exposure studies”, Atmos Environ, 41, 16, 3453–3464. Brauer M, Hoek G, van Vliet P, Meliefste K, Fischer P, Gehring U, Heinrich J, Cyrys J, Bellander T, Lewne M, Brunekreef B (2003) “Estimating long-term
Models of Exposure for Use in Epidemiological Studies of Air Pollution Health Impacts
601
average particulate air pollution concentrations: application of traffic indicators and geographic information systems”, Epidemiology, 14, 2, 228–239. Cyrys J, Hochadel M, Gehring U, Hoek G, Diegmann V, Brunekreef B, Heinrich J. GIS-based estimation of exposure to particulate matter and NO 2 in an urban area: stochastic versus dispersion modeling. Environ. Health Perspect. 2005 Aug; 113(8):987–92. Briggs DJ, Collins S, Elliott P, Fischer P, Kingham S, Lebret E, Pryl K, VAnReeuwijk H, Smallbone K, VanderVeen A (1997) “Mapping urban air pollution using GIS: a regression-based approach”, Int J Geogr Inform Sci, 11, 7, 699–718. Briggs D, de Hoogh K, Gulliver J (2006) “Matching the metric to need: modelling exposures to traffic-related air pollution for policy support.”, NERAM V (http://www.irr-neram.ca/about/Colloquium.html (accessed July 24, 2007). Ebelt ST, Fisher TV, Petkau AJ, Vedal S, Brauer M (2000) “Exposure of chronic obstructive pulmonary disease (COPD) patients to particles: relationship between personal exposure and ambient air concentrations”, J Air Waste Manage Assoc, 50, 174–187. Ebelt ST, Wilson WE, Brauer M (2005) “Exposure to ambient and nonambient components of particulate matter: a comparison of health effects”, Epidemiology, 16, 3, 396–405. Gilbert NL, Goldberg MS, Beckerman B, Brook JR, Jerrett M (2005) “Assessing spatial variability of ambient nitrogen dioxide in Montreal, Canada, with a landuse regression model”, J Air Waste Manage Assoc, 55, 8, 1059–1063. Gilbert NL, Woodhouse S, Stieb DM, Brook JR (2003) “Ambient nitrogen dioxide and distance from a major highway”, Sci Total Environ, 312, 1–3, 43–46. Gonzales M, Qualls C, Hudgens E, Neas L (2005) “Characterization of a spatial gradient of nitrogen dioxide across a United States-Mexico border city during winter”, Sci Total Environ, 337, 1–3, 163–173. Henderson SB, Beckerman B, Jerrett M, Brauer M (2007) “Application of land use regression to estimate long-term concentrations of traffic-related nitrogen oxides and fine particulate matter”, Environ Sci Technol, 41, 7, 2422–2428. Hochadel M, Heinrich J, Gehring U, Morgenstern V, Kuhlbusch T, Link E, Wichmann HE, Kramer U (2006) “Predicting long-term average concentrations of traffic-related air pollutants using GIS-based information”, Atmos Environ, 40, 3, 542–553. Hoek G, Fischer P, Van Den Brandt P, Goldbohm S, Brunekreef B (2001) “Estimation of long-term average exposure to outdoor air pollution for a cohort study on mortality”, J Expo Anal Environ Epidemiol, 11, 6, 459–469. Janssen NA, Hoek G, Harssema H, Brunekreef B (1999) “Personal exposure to fine particles in children correlates closely with ambient fine particles”, Arch Environ Health, 54, 299192099, 95–101. Jerrett M, Arain A, Kanaroglou P, Beckerman B, Potoglou D, Sahsuvaroglu T, Morrison J, Giovis C (2005a) “A review and evaluation of intraurban air pollution exposure models”, J Expo Anal Environ Epidemiol, 15, 2, 185–204.
602
M. Brauer et al.
Jerrett M, Burnett RT, Ma R, Pope CA, Krewski D, Newbold KB, Thurston G, Shi Y, Finkelstein N, Calle EE, Thun MJ (2005b) “Spatial analysis of air pollution and mortality in Los Angeles”, Epidemiology, 16, 6, 727–736. Koenig JQ, Mar TF, Allen RW, Jansen K, Lumley T, Sullivan JH, Trenga CA, Larson T, Liu LJ (2005) “Pulmonary effects of indoor- and outdoor-generated particles in children with asthma”, Environ Health Perspect, 113, , 499–503. Kunzli N, Medina S, Kaiser R, Quenel P, Horak F Jr, Studnicka M (2001) “Assessment of deaths attributable to air pollution: should we use risk estimates based on time series or on cohort studies?”, Am J Epidemiol, 153, 11, 1050– 1055. Larson T, Su J, Baribeau AM, Buzzelli M, Setton E, Brauer M (2007) “A spatial model of urban winter woodsmoke concentrations”, Environ Sci Technol, 41, 7, 2429–2436. Miller KA, Siscovick DS, Sheppard L, Shepherd K, Sullivan JH, Anderson GL, Kaufman JD (2007) “Long-term exposure to air pollution and incidence of cardiovascular events in women”, New Engl J Med, 356, 5, 447–458. Nieuwenhuijsen MJ (ed) (2003) Exposure Assessment in Occupational and Environmental Epidemiology, First edn, Oxford University Press, Oxford, England. Pope CA 3rd, Burnett RT, Thun MJ, Calle EE, Krewski D, Ito, K, Thurston GD (2002) “Lung cancer, cardiopulmonary mortality, and long-term exposure to fine particulate air pollution”, JAMA, 287, 9, 1132–1141. Ross Z, English PB, Scalf R, Gunier R, Smorodinsky S, Wall S, Jerrett M (2006) “Nitrogen dioxide prediction in Southern California using land use regression modeling: potential for environmental health analyses”, J Expo Sci Environ Epidemiol, 16, 2, 106–114. Sahsuvaroglu T, Arain A, Kanaroglou P, Finkelstein N, Newbold B, Jerrett M, Beckerman B, Brook J, Finkelstein M, Gilbert NL (2006) “A land use regression model for predicting ambient concentrations of nitrogen dioxide in Hamilton, Ontario, Canada”, J Air Waste Manage Assoc, 56, 8, 1059–1069. Sarnat JA, Koutrakis P, Suh HH (2000) “Assessing the relationship between personal particulate and gaseous exposures of senior citizens living in Baltimore, MD”, J Air Waste Manage Assoc, 50, 720395124, 1184–198. Setton EM, Hystad PW, Keller CP (2005) “Opportunities for using spatial property assessment data in air pollution exposure assessments”, Int J Health Geogr, 4, 26. Strand M, Vedal S, Rodes C, Dutton SJ, Gelfand EW, Rabinovitch N (2006) “Estimating effects of ambient PM(2.5) exposure on health using PM(2.5) component measurements and regression calibration”, J Expo Sci Environ Epidemiol, 16, 1, 30–38. Su JG, Larson T, Baribeau A, Brauer M, Rensing M, Buzzelli M (2007) “Spatial modeling for air pollution monitoring network design: Example of residential woodsmoke.”, J Air Waste Manag Assoc, 57, 8, 893–900. Su JG, Brauer M, Ainslie B, Steyn D, Larson TV, Buzzelli M (2008). Comparing source area analysis with land use regression models for exposure analysis. Sci Total Environ. 390, 520–529.
Models of Exposure for Use in Epidemiological Studies of Air Pollution Health Impacts
603
Venn A, Lewis S, Cooper M, Hubbard R, Hill I, Boddy R, Bell M, Britton J (2000) “Local road traffic activity and the prevalence, severity, and persistence of wheeze in school children: combined cross sectional and longitudinal study”, Occup Environ Med, 57, 320277145, 152–118. Yu CL, Wang SF, Pan PC, Wu MT, Ho CK, Smith TJ, Li Y, Pothier L, Christiani DC, Kaohsiung Leukemia Research Group (2006) “Residential exposure to petrochemicals and the risk of leukemia: using geographic information system tools to estimate individual-level residential exposure”, Am J Epidemiol, 164, 3, 200–207. Zhang KM, Wexler AS, Zhu YF, Hinds WC, Sioutas C (2004) “Evolution of particle number distribution near roadways. Part II: the ‘road-to-ambient’ process”, Atmos Environ, 38, 38, 6655–6665.
Discussion S.T. Rao: Given the spatial homogeneity of sulphate levels, why do you see differences in health outcomes between within city and between cities? Since sulphate level has been decreasing due to SO2 emission reduction, do you see a trend the health outcomes? Regarding the BAQ study, do you see a commonality between pollutants in Seattle vs Vancouver and associated health outcomes? What seems to be primary driver for the health impacts? M. Brauer: There is a misunderstanding in our use of sulphate to assess air pollution exposures. Because sulphate has no known major indoor sources, we use it as an indicator of the amount of ambieny particulate matter that infiltrates indoors from outside. This allows us to more accurately assess the amount of ambient particles that people are exposed to, and when we apply this in studies of health effects we get stronger relationships compared to when we estimate exposure by just using the measured ambient concentrations. Regarding the question about the BAQ study, in general we do see similarities between our studies in Vancouver and Seattle and one common aspect is the importance of primary, combustion-source pollutants – especially those related to traffic and traffic proximity. It is however, difficult to quantitatively compare the findings since we are relaying on administrative data and there are differences in they type of data that are available. For example, in Canada we have health statistics for essentially the entire population for all encounters with the health care system whereas in Seattle much of our data are limited to those based on hospital admissions and vital statistics (births, deaths).
604
M. Brauer et al.
J. Pleim: Have land use regression models been combined with high resolution meteorology models? Would it help to include such meteorology data in the regression? M. Brauer: There have been some attempts to incorporate output from meteorological models as additional predictor variable into land use regression models – in general these have shown relatively small improvements in the model predictions, in part due to the limited spatial resolution of the meteorological models relative to the other land use regression input data and the fact that most of the land use regression models have been developed for long term (annual) estimates. Certainly in developing land use regression models for shorter time periods, incorporating meteorological information (from measurements or high resolution models) will be very helpful. P. Suppan: The presentation shows the very important link between epidemiology and air quality. Some slides show the NO distribution provided by the land use model and CMAQ. But as the results from CMAQ are far too coarse for epidemiological studies the question arises if also results from micro scale models can be introduced, or can be provided in order to improve the results. M. Brauer: Land use regression models are one approach to incorporate spatial variability in air pollution models into epidemiology. Other types of high resolution spatial models can certainly be just as useful if estimates can be developed over large study areas and can even improve over land use regression models if they are able to incorporate temporal variability in pollutant concentrations and spatial dispersion due to meteorology and/or variable emissions.
7.6 The Importance of Exposure in Addressing Current and Emerging Air Quality Issues Tim Watkins, Ron Williams, Alan Vette, Janet Burke, B.J. George and Vlad Isakov
Abstract The air quality issues that we face today and will face in the future are becoming increasingly more complex and require an improved understanding of human exposure to be effectively addressed. The objectives of this paper are (1) to discuss how concepts of human exposure and exposure science and should be applied to improve air quality management practices, and (2) to show how air quality modeling tools can be used to improve exposures estimates used for understanding associations between air quality and human health. Data from a large human exposure monitoring study is presented to demonstrate the value of exposure in understanding important air quality issues, such as health effects associated with exposure to components of particulate matter (PM), to PM of different size fractions (coarse and ultrafine), and to air pollution in near roadway environments. Various approaches for improving estimates of exposure via application of air quality modeling are discussed and results from example modeling applications are presented. These air quality modeling approaches include: the integration of regional scale eulerian air quality models with local scale gaussian dispersion models; the fusion of modeled estimates with air quality observations; the integration of air quality and human exposure modeling tools; and the use of exposure factors, such as housing ventilation, to adjust modeled estimates of ambient air quality.
Keywords Air quality, exposure, modeling, particulate mater, toxic air pollutants 1. Introduction Existing air quality standards and regulations are in place to protect public health and the environment. However, there remain uncertainties regarding whether existing standards/regulations should be adjusted to be more protective or perhaps to more effectively target air quality management activities. For particulate matter (PM) standards, there are questions regarding whether current mass based standards for PM10 and PM2.5 should be revised to address specific components or sources of PM or different PM size fractions (NRC, 2004a; U.S. EPA, 2005). There are also concerns about potential disproportionate health effects associated with air pollution “hotspots.” These “hotspot” concerns are often related to environmental C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
640
The Importance of Exposure in Addressing Current and Emerging Air Quality Issues
641
justice issues (NRC, 2004b). For these air quality management issues, understanding human exposure is critical. This paper discusses why this is the case, presents data from human exposure monitoring and modeling studies to further highlight the importance of air pollution exposure issues, and discusses how air quality models may be used to improve exposure assessment. While air pollution impacts both humans and ecological resources, this paper focuses on human health outcomes.
2. Exposure and Exposure Science Although air quality standards to protect public health are most often based on levels of a pollutant in the ambient air, people experience health impacts from the pollutants in the air they breathe, i.e., from their exposure. The United States Environmental Protection Agency defines exposure as contact (of an environmental pollutant) with the exterior of the person (U.S. EPA, 1992). The critical factors to characterize and understand human exposure to air pollution include the following. x
Spatial and temporal variability of ambient pollutant concentrations – When ambient air concentrations are relatively homogeneous in space and time, then human exposures may be more closely approximated by ambient concentrations. However, when there is significant spatial and temporal variability in ambient air concentrations, ambient levels of pollutants will more poorly represent human exposure.
x
Concentrations of ambient pollutants in microenvironments – Micro-environments are locations where people spend their time (e.g., indoors at home, indoors at work/school, in-vehicle, outdoors at home). A person’s exposure to ambient air pollution will depend greatly on concentrations of ambient pollutant in microenvironments, which in turn depend upon exposure factors such as proximity to sources, air exchange rates, penetration rates, indoor air chemistry, and indoor decay rates and removal mechanisms.
x
Human Activities – A person’s daily activities play a significant role, if not the most significant role, in characterizing human exposure. Where a person spends their time and how much time he/she spends in each location will impact that person’s exposure.
Human exposure is the critical link between ambient air concentrations and human health outcomes. The field of exposure science includes research to measure and model factors and human activities that influence magnitude, frequency, and duration of exposure to air pollutant concentrations in various microenvironments. Understanding human exposures requires an understanding of the factors that influence the spatial and temporal variability of ambient air concentrations, which in turn requires an understanding of air pollution sources, fate and transport of air pollutants, and ambient air concentrations. Therefore, the field of exposure science
642
T. Watkins et al.
also includes aspects of research related to source emission characterization, atmospheric processes, and ambient air measurement and modeling.
3. The Growing Importance of Exposure Understanding actual human exposures is critical to addressing the current and emerging air quality management issues mentioned in the introduction of this paper. Human exposure is the link between ambient concentrations and health outcomes. Many existing air quality management policies are based upon studies that associated ambient concentrations with health impacts by inferring that ambient concentrations are equivalent to actual human exposures. However, for current and emerging air quality management issues this inference may not be appropriate and understanding exposure may in fact be the critical factor to developing and implementing effective air quality management policies for these issues. The following discussion provides examples of why this is the case.
3.1. Particulate matter components, size fractions, and sources The uncertainties surrounding existing PM standards are largely based on whether specific PM characteristics, such as composition or size fraction, lead to a greater proportion of observed health impacts (NRC, 2004a; U.S. EPA, 2005). Existing standards for PM2.5 and PM10 are largely based upon epidemiological studies that found associations between ambient concentrations of PM and observed health impacts. These studies often used a central site monitor to estimate exposures to PM, which is reasonable for PM2.5 because the variability of PM2.5 across many urban areas is relatively homogeneous. However, the spatial variability of specific PM components or PM of different size fractions (e.g., ultrafine PM or coarse PM) is greater than that for PM2.5 (U.S. EPA, 2004, 2005). In addition, there are significant uncertainties regarding the microenvironmental concentrations of PM components and PM size fractions (U.S. EPA, 2004). Therefore, any epidemiological evidence or risk assessment for PM components or PM size fractions will require an improved exposure assessment due to the spatial heterogeneity of ambient concentrations. Related to the issues of PM components and size fractions is the issue of whether PM standards should be targeted at sources of PM that may be disproportionately responsible for observed health impacts. Characteristics of PM emissions vary by source, thus there is the potential that the relative toxicity of PM from different sources may also vary. Exposure will play an important role in addressing this issue as well. One approach to address this issue is to evaluate the inter-city variability of PM characteristics. The composition and characteristics of PM in different urban areas varies because the contributing sources of PM vary (McMurry et al., 2004), therefore studies that evaluate the differential exposures and health
The Importance of Exposure in Addressing Current and Emerging Air Quality Issues
643
impacts across multiple urban areas may provide insights into the issue regarding standards for particular PM sources.
3.2. Air pollution hotspots and environmental justice issues Recent concerns have emerged regarding whether existing regulations provide ample protection for certain subpopulations that may be vulnerable due to elevated exposures in “hotspots.” In many cases, these issues are centered around environmental justice and toxic air pollutants (NRC, 2004b). An example of such an issue is near roadway exposures and health effects. Exposure assessment will be central to addressing these issues because improved characterization of the variability of ambient concentrations, microenvironmental concentrations, and human activities will all be required to evaluate environmental policies to address potential hotspots and environmental justice issues.
4. The Detroit Exposure and Aerosol Research Study The US Environmental Protection Agency has conducted an exposure study that is generating data to confirm the importance of exposure in addressing the air quality management issues above and to provide insight for future air quality policy decisions. The Detroit Exposure and Aerosol Research Study (DEARS) was designed to describe the relationships between concentrations at a central site and residential/ personal concentrations for PM components, PM size fractions, PM from specific sources (mobile and point), and air toxics. To accomplish this, the DEARS included extensive field work conducted over three years and six seasons. The DEARS data collection sites included a central ambient air monitoring site and six exposure measurement areas (EMA) that were selected to evaluate the impact of local sources.
PM2.5
Eenzene
Ind
Ind ust r Di ial ese Tra l ffic Hi ghw Re ay gio nal
Sulfur
ust r Di ial ese Tra l ffic Hi ghw Re ay gio nal
5 4 3 2 1 0
5 4 3 2 1 0
5 4 3 2 1 0
Ind ust ria Di l ese Tra l ffic Hi ghw Re ay gio nal
Increased Variability
Fig. 1 Ratio of residential outdoor concentrations to central site concentrations for DEARS season 1
Fig. 2 Relationships between concentrations at residential outdoor sites and the central site for DEARS EMAs
T. Watkins et al.
644
mg m mg m
3
Fig. 3 Comparisons of personal, indoor, outdoor residential and ambient central site measurements for DEARS (seasons 1–4)
3
Figures 1 and 2 present data on the spatial variability of pollutants by comparing data collected at the central site to data collected at residential outdoor sites in each EMA. Figure 1 presents the ratios of concentrations at residential outdoor locations to concentrations at the central site for sulfur, PM2.5, and benzene. The ratios for sulfur, a PM component, are all very close to 1. The ratios for total PM2.5 are also very close to 1, but exhibit a bit more variability than sulfur. The ratios for sulfur and PM2.5 confirm the relative spatial homogeneity for these pollutants. However, the ratios for benzene are generally greater than 1, which shows the influence of sources, particularly in industrial and mobile source influenced EMAs. Figure 2 provides additional data to show how spatial variability differs by pollutant. For PM2.5, there is a relatively strong relationship between residential outdoor sites and the central site. For sulfur, the correlation is also very strong plus seasonal variation is observed. Again, the results for PM2.5 and sulfur are not surprising given the regional nature of these pollutants. However, there is much more variability between the central site and residential outdoor sites for 1,3 butadiene and iron. For 1,3 butadiene, the influence of local sources can be seen particularly at EMA 6, which is impacted by highway sources. For iron, the influence of local sources is even greater, particularly at EMAs 1 and 3. The DEARS also provides insights into how microenvironmental factors and human activities impact human exposures to various pollutants. Figure 3 compares measurements of various air toxics taken at the central monitoring site, at outdoor residential locations, at indoor residential locations, and personal exposure measurements. The influence of micro-environmental concentrations (including indoor sources) and human activities can be seen as total personal exposure is generally higher and exhibits more variability than indoor, outdoor, and ambient concentrations. 16 14 12 10 8 6 4 2 0 40
Benzene
m,p-Xylene
30 20 10
mg m
3
0 8 6 4 2 0 0
Acrolein
70 60 50 40 30 20 10 0 14 12 10 8 6 4 2 0 6 5 4 3 2 1 0
Toluene Personal Indoor Outdoor Ambient
o-Xylene
12 10 8 6 4 2 0 4
Ethylbenzene
1,3-Butadiene
3 2 1
1,3,5-Trimethylbenzene
0 6 5 4 3 2 1 0
4-Ethyltoluene
5. Improving Exposure Assessment with Air Quality Models Addressing the air quality management issues above will require improved human exposure assessments. Improved exposure estimates could be obtained through more spatially, temporally, and compositionally refined air quality monitoring and
The Importance of Exposure in Addressing Current and Emerging Air Quality Issues
645
through conducting human exposure monitoring studies. However, increased monitoring is cost prohibitive. Air quality modeling tools offer an alternative to increased monitoring. For example, air quality models can provide more spatially, temporally, and compositionally refined estimates of ambient concentrations. They also can provide scientific insights to improve the understanding and characterization of the atmospheric processes that impact spatial and temporal variability of pollutants. When air quality models are integrated with each other and with other data sources even more opportunities to improve exposure estimates arise. For example, output from air quality models can be used to fill in the spatial and temporal gaps in existing ambient monitoring data to improve the estimates of air quality and exposure estimates for health studies (Bell, 2006). Another way air quality models can be used to improve exposure estimates is by combining results from regional scale air quality models, such as CMAQ, and local scale dispersion models, such as AERMOD. Figure 4 demonstrates this concept from a modeling analysis done in New Haven, CT. The regional air quality model provides a regional background concentration upon which the influence of local stationary and mobile sources can be added. The result is a more spatially refined ambient air quality estimate that may be used to improve exposure assessments. Combining air quality modeling output with monitoring data and integrating regional and local scale air quality models provide more spatially and temporally refined estimates of ambient air quality that can be used to improve human exposure estimates. However, while these approaches address the spatial and temporal variability exposure factor, they do not address other exposure factors such as microenvironmental concentrations and human activities. Linking air quality modeling outputs with human exposure models provides an approach to address these exposure factors. Using ambient concentration inputs from air quality models, human exposure models estimate actual human exposures by modeling factors, such as pollutant penetration rates, that impact pollutant concentrations in microenvironments and then integrating human activity data including time spent in each microenvironment. Figure 5 shows results from linking air quality and human exposure models in Philadelphia, PA (Isakov et al., 2006). As shown in Figure 5, while the spatial patterns are similar, the actual human exposures to benzene are greater than ambient concentrations, most likely due to exposures experienced in high exposure microenvironments, such as in-vehicle, at gas stations, or in attached garages.
646
T. Watkins et al.
Fig. 4 Approach for combining regional scale and local scale air quality models (New Haven, CT)
Fig. 5 Results from linking air quality and human exposure models in Philadelphia, Pa (Isakov et al., 2006)
6. Summary and Conclusions To effectively address the current and emerging air quality management issues that we face today will require an improved understanding of human exposures. This paper provided examples and case studies demonstrating how improved exposure assessment will be an important tool to support environmental policy decisions on whether to revise existing or develop new air quality standards and regulations. Improved exposure assessments can also inform other air quality management activities such developing and evaluating alternative emissions control strategies and evaluating whether air quality regulations have met anticipated goals to protect human health. Air quality modeling tools offer tremendous promise for improving exposure estimates needed for air quality management activities. To date, air quality models have been used sparingly in health studies. However, as air quality modeling approaches become more sophisticated, the opportunities to use air quality models to enhance exposure assessments in health studies will grow and potentially lead to improved air quality policies. Disclaimer The views expressed in these Proceedings are those of the individual authors and do not necessarily reflect the views and policies of the United States Environmental Protections Agency (EPA). Scientists in the EPA have prepared the
The Importance of Exposure in Addressing Current and Emerging Air Quality Issues
647
EPA sections and those sections have been reviewed in accordance with EPA’s peer and administrative review policies and approved for presentation and publiccation.
References Bell ML (2006) The use of ambient air quality modeling to estimate individual and population exposure for human health research: a case study of ozone in the Northern Georgia Region of the United States, Environment International, 32 (5), 586–593. Isakov V, Graham S, Burke J, Ozkaynak H (2006) Linking Air Quality and Exposure Models, EM, September 26–29. McMurry P, Shepard M, Vickery J (eds) (2004) Particulate Matter Science for Policy Makers: A NARSTO Assessment. New York: Cambridge University. National Research Council (2004a) Research Priorities for Airborne Particulate Matter IV – Continuing Research Progress. Washington DC: National Academies Press. National Research Council (2004b) Air Quality Management in the United States. Washington DC: National Academies Press. U.S. EPA (1992) Guidelines for Exposure Assessment. U.S. Environmental Protection Agency, Risk Assessment Forum, Washington, DC, 600Z-92/001. U.S. EPA (2004) Air Quality Criteria for Particulate Matter (October 2004). U.S. Environmental Protection Agency, Washington, DC, EPA 600/P-99/002aF-bF. U.S. EPA (2005) Review of the National Ambient Air Quality Standards for Particulate Matter: Policy Assessment of Scientific and Technical Information, OAQPS Staff Paper (June 2005). U.S. Environmental Protection Agency, Washington, DC, EPA-452/R-05-005.
Discussion P. Builtjes: Do you see a future for bio-monitoring, by for example measuring radicals in blood? T. Watkins: Yes, I do believe that bio-monitoring holds promise for providing valuable information about human exposures. However, there are currently limited biomarkers routinely available for many air pollutants, particularly for particulate matter. Also, while bio-monitoring information is certainly useful, biomarkers indicate whether an exposure has occurred, but they do provide other important information such the source, route, duration or intensity of exposure. Therefore other tools, including modelling tools, are needed to more fully characterize exposures.
P4.4 A Construction and Evaluation of Eulerian Dynamic Core for the Air Quality and Emergency Modelling System SILAM Mikhail Sofiev, Michael Galperin and Eugene Genikhovich
Abstract The paper presents a new dynamic core of the SILAM modelling system. It is based on the original Eulerian advection algorithm combined with extended resistance scheme for vertical diffusion. Apart from the standard advantages of the Eulerian environment, it has several unique features: (i) exactly zero numerical viscosity and a possibility to utilise the sub-grid information on mass location inside a grid cell; (ii) robustness to sharp gradients of concentrations and their preservation during the transport, (iii) applicability at high Courant numbers, (iv) options for prescribing or dynamically evaluating the horizontal diffusion; (v) vertical diffusion with thick adaptive layers that utilises the sub-grid information of advection for refining the flux values.
Keywords Advection-diffusion schemes, Eulerian models, numerical modelling
1. Structure of SILAM and the New Items Historically, most of emergency-response systems are based on Lagrangian advection, often with the Monte-Carlo random-walk diffusion (Saltbones et al., 1996; Stohl et al., 1998; Sørensen et al., 2000; Sofiev et al., 2006). Systems for air quality assessments are mainly based on Eulerian approaches, such as the widely used scheme of Bott (1989) and its numerous variations. For diffusion, the K-theory and its Crank-Nicholson three-diagonal solution are the most universal. The SILAM system (Figure 1, Sofiev et al., 2006) is a flexible environment made for a wide variety of tasks, including emergency response, air quality, observation analysis, data assimilation and inverse-problem applications. It is a modular system with object-oriented code. Lagrangian and Eulerian dynamic cores utilise the same supplementary routines including meteorological, emission, and I/O servers. At present, the cores are not connected and the user has to select the type of dynamics used for the run. The advection routine (Galperin, 2000) of the new Eulerian core is connected with the adaptive vertical diffusion algorithm of Sofiev (2002), which makes use of the advection-controlled sub-grid variable (the first moment of the mass located in
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
699
M. Sofiev et al.
700
the grid cell) for dynamic adaptation of the vertical. The tests of the advection scheme (examples in Figure 2) show that is has zero numerical viscosity and is capable of operating at very high Courant numbers. Vertical diffusion scheme serves for the whole column and incorporates dry deposition and re-evaporation via extended resistive algorithm. Its parameterization is based on K-theory after Genikhovich et al. (2004). Acknowledgments The study was supported by TEKES-KOPRA and EU-GEMS projects, and the Network scale Atmospheric Modelling NETFAM. Control unit Dispersion interface Lagrangean pollution cloud
Eulerian data buffer
Dispersion dynamics
Emission composer
Fig. 1 Schematic picture of the main SILAM components
Area Bomb Point emission
Processing
Dispersion physics & chemistry
Interface Data storage Control flow Data flow
a
b
Fig. 2 Tests of the advection scheme: a point-type and rectangular masses in a circular vortex; 1D jet-stream with Courant number (grey) up to 9; relative concentrations are shown in black
References Bott A (1989) A positive definite advection scheme obtained by non-linear renormalization of the advection fluxes. Mon. Wea. Rev., 117, 1006–1012. Galperin MV (2000) The Approaches to Correct Computation of Airborne Pollution Advection. In: Problems of Ecological Monitoring and Ecosystem Modelling, vol. XVII, St. Petersburg, Gidrometeoizdat, pp. 54–68. Genikhovich E, Sofiev M, Gracheva I (2004) Interactions of meteorological and dispersion models at different scales. In Air Polution Modelling and its Applications XVII (eds. C Borrego, A-LNorman), Springer 2007, pp. 158–166, ISBN-10: 0-387-28255-6
A Construction and Evolution of Eluerian Dynamic Core for the Air Quality
701
Saltbones J, Foss A, Bartnicki J (1996) A real time dispersion model for severe nuclear accidents, tested in the European tracer experiment. Syst. Anal. Model. Simul. 25, 263–279. Sofiev M (2002) Extended resistance analogy for construction of the vertical diffusion scheme for dispersion models. J. of Geophys. Research-Atmosphere, 107, D12, doi:10.1029/2001JD001233. Sofiev M, Siljamo P, Valkama I, Ilvonen M, Kukkonen J (2006) A dispersion system SILAM and its evaluation against ETEX data. Atmos. Environ., 40, 674–685. Sørensen JH, Mackay DKJ, Jensen CØ and Donaldson AI (2000) An integrated model to predict the atmospheric spread of foot-and-mouth disease virus. Epidemiol. Infect. 124, 577–590. Stohl A, Hittenberger M, Wotawa G (1998) Validation of the Lagrangian particle dispersion model FLEXPART against large scale tracer experiments. Atmos. Environ. 24, 4245–4264.
P7. Air quality and human health
P7.1 A Multi-Objective Problem to Select Optimal PM10 Control Policies Claudio Carnevale, Enrico Pisoni and Marialuisa Volta
Abstract To implement efficient air quality policies Environmental Agencies require integrated systems allowing the evaluation of both the effectiveness and the cost associated to different emission reduction strategies. These tools are even more useful when considering atmospheric PM10 concentrations, a strongly nonlinear secondary pollutant. The classical approaches of cost-benefit and cost-effectiveness analysis create unique solutions, hiding possible stakeholders conflicts. In this work the formulation of a multi-objective problem to control particulate matter is proposed, defining: (a) control objectives, namely the air quality indicators and the cost functions; (b) decision variables and their constraints; (c) source-receptor models, describing the cause-effect relation between air quality indicators and decision variables. The multi-objective problem results obtained for Northern Italy are analyzed in terms of not-dominated solutions. 1. Introduction Due to nonlinearities of secondary pollutants, it is very challenging to develop sound policies considering both air quality improvements and implementation costs. In literature, to solve this control problem, multi-objective analysis (Guariso et al., 2006; Carnevale et al., 2007) has been rarely used, and only for ozone control. In this paper, a multi-objective optimization methodology for PM10 control is proposed. The nonlinear relations between decision variables and PM10 exposure index, defining the air quality objective, are described by neuro-fuzzy models, identified processing long-term simulations of GAMES multiphase modelling system performed in the CityDelta project (Cuvelier et al., 2007).
2. Methodology and Results The target of this study is to control particulate matter exposure at ground level. This issue can be attained by optimizing both air quality indicators and emission abatement costs. The emission reduction rates (decision variables) are computed by C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
715
C. Carnevale et al.
716
a two-objective mathematical programming algorithm as the ones corresponding to the most efficient strategies, with respect to both the considered objectives. The problem can be formalized as follows: min J ( T ) J ( T ) [ AQI ( E ( T )); CPI ( E ( T ))]
T 4 where E represents the precursor emissions for the reference case, T are the decision variables (namely the emission reductions) constrained to assume values in 4 , AQI(E( T )) and CPI(E( T )) are the Air Quality Index and Cost of Policy Index respectively, both depending on precursor emissions and emission reductions. Optimization methodology results suggest: (a) efficient solutions of the PM10 control problem, depicting the Pareto boundary; (b) emission reduction priorities needed to obtain a particular result; (c) costs of implementation of the different emission reduction policies. I.e. in Figure 1 the Pareto boundary (left) is shown, stressing point A (the point associated to no reductions), point B (that reduces PM10 to 32 Pg/m3 with a cost of roughly 700 Meuro), and point C (maximum PM10 reduction with maximum cost). The macrosector costs for point B and C are shown in Figure 1 (right).
Fig. 1 Pareto boundary solution of the optimisation problem (left) and emission reduction costs associated to point B and C in the Pareto Boundary, for each CORINAIR macrosector
References Carnevale C, Pisoni E, Volta M (2007) Selecting effective ozone exposure control policies solving a two-objective problem, Ecological Modelling 204, 93–103. Cuvelier C et al. (2007) CityDelta: a model intercomparison study to explore the impact of emission reductions in European cities in 2010, Atmospheric Environment, 41, 189–207. Guariso G, Pirovano G, Volta M (2004) Multi-objective analysis of ground level ozone concentration control, Journal of Environmental Management 71, 25–33.
P7.5 Air Pollution Assessment in an Alpine Valley Peter Suppan, Klaus Schäfer, Stefan Emeis, Renate Forkel, Markus Mast, Johannes Vergeiner and Esther Griesser
Abstract ALPNAP (Monitoring and Minimization of Traffic-Induced Noise and Air Pollution along Major Alpine Transport Routes) is an ongoing research project focussing on corresponding effects along several major transit routes across the European Alps. Assisting regional authorities with appropriate output, a unique cooperation of scientists within the alpine region was setup to better assess and predict the spatial and temporal distribution of air pollution and noise close to major alpine traverses. A methodology for measurement strategies and model simulations for air pollutants will be demonstrated for the Brenner traverse. Results of a field measurement campaign give detailed insights of the complexity of the atmospheric conditions and the distribution of air pollutants in the Inn valley. First results of the air quality simulations with the regional meteorology-chemistry model MCCM (MM5/chem) on a coarse resolution show the importance of a detailed emission inventory in an alpine valley. Within ALPNAP comprehensive measuring campaigns as well as detailed modelling simulation are foreseen to describe the air pollution situation in an alpine valley. In a first step a measurement campaign in the lower Inn-valley was designed to determine cross-valley air pollution and meteorological information as well as vertical profiles to determine flow regimes (valley, slope winds), mixing layer height, stability in the boundary layer and emission sources at specific locations. Covering the major part of the winter season, this data will be used for enhanced analysis as well as for the set-up and validation of corresponding models merging the wealth of different remote sensing and in-situ measurements. In a second step air quality simulations will be carried out, in order to receive detailed 3D information of the distribution of air quality parameters. In the frame work of ALPNAP, a field campaign in the Inn valley was performed between November 2005 and January 2006. The most important goals for the field campaign have been to study the spatial variation and distribution of air pollutants within a cross-section along the northern and southern slopes and to evaluate the best use of an existing slope temperature profile for mixing height and stability analyses and finally to use the field data for the set-up and validation of analysis and modelling inter-comparison studies.
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
723
724
P. Suppan et al.
Results The NO concentrations are clearly dominated by the traffic volume and therefore are the highest at the highway followed by the valley floor in 800 m distance to the highway. The exceedance at the valley floor as compared to the slope station is basically related to a stable layering of the valley atmosphere during nearly all the time. Outstanding high pollution episodes were found during: 20–24 December, 2005, 09–22 January and 25 January–02 February, 2006, when the detailed structure of the valley atmosphere was captured by the additional intensive measurements periods. Moreover, the data indicate that the ratio of concentrations near the highway to the background is systematically higher for NO2 (more than factor 2) than for NO. On a local scale the data allow e.g. to demonstrate the impact of nearby construction work imposing on the CO, NO and NOx measurements. The temporal variations of air pollution concentration (CO, PM10, and NO2) at the valley ground are clearly dominated by the mentioned weather conditions and emissions. As for the NO2 and CO emissions, the main source is road traffic at the highway. The NO2 and CO maxima in the morning and in the evening correspond to the local traffic maxima. In the early afternoon mixing height rises and the concentrations decrease, whereas the ozone concentrations increase. During midnight the CO concentrations show a maximum and PM10- and NO2-concentrations show a minimum. However, even detailed measurements like these cannot reflect the full temporal and spatial variability of the complex flow regimes or the horizontal and vertical distribution of chemical parameters in alpine valleys. Future numerical simulations will contribute there and the corresponding validation process will greatly benefit from the available data. Thus setting up different models in a most sophisticated way will enable for process oriented studies, model inter-comparisons and impact studies. Further information about the ongoing work can be obtained at the ALPNAP web site http://www.alpnap.org/ Acknowledgments The project ALPNAP is implemented through financial assistance from funds of the European Community Initiative Programme “Interreg III B Alpine Space”.
P4.2 Air Pollution Dispersion Modelling Arround Thermal Power Plant Trbovlje in Complex Terrain – Model Verification and Regulatory Planning Marija Zlata Božnar, Primož Mlakar, Boštjan Grašiþ and Gianni Tinarelli
Abstract The paper shows extensive verification of Lagrangian particle model Spray coupled with Minerve 3D mass consistent diagnostic wind field model in extremely complex terrain of Zasavje region in Slovenia. On-line measured emission and ambient data was used to reconstruct air pollution situation across the area for one-year time interval.
1. Introduction and Campaign Description Zasavje region is a highly industrial area located in a river canyon in central part of Slovenia. Thermal Power Plant (TPP) Trbovlje, Cement factory Trbovlje and Glass factory Hrastnik are main sources of air pollution in the area. TPP and Cement factory have just installed wet desulphurisation plants that decreased the previous level of SO2 pollution significantly. A study (Božnar et al., 2006) was done to reconstruct the current air pollution situation in the area, to quantify the expected reduction of SO2 pollution by desulphurisation plants and to model the future scenarios (new planned gas powered TPP). One year of on-line meteorological, air pollution and emission data was analysed. Across the area of interest mainly SO2 and NO2 pollution exceed the regulation limits. High pollution of the area is caused by high emissions, but it is also emphasized by local microclimatological conditions (low wind speeds, calm situations and strong thermal inversions) as the area is a highly complex terrain (canyon with steep slopes of approximately 45q, several valleys perpendicular to main canyon, hills of relative height over 1,000 m, see GoogleMaps, search keyword “Trbovlje, Slovenia”. Zasavje region therefore represents an excellent study case for the evaluation of dispersion modelling in complex terrain. In the area there exist a dense monitoring of meteorological parameters (including one SODAR profile) environmental concentrations (nine stations) and one on-line emission station on the existing TPP stack. On line measured emission and ambient data for one-year time interval from 01.07.2005 till 31.06.2006 was used to reconstruct air pollution situation across the area. C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
695
M.Z. Božnar et al.
696
2. Results and Conclusions To determine the optimal height of the new planned TPP a comparison of reconstructed ground level concentrations for stacks with different heights was performed. Extensive verification was performed to qualify the results of reconstruction. Two very good reconstructed air pollution situations are depicted on Figure 1 were measured and reconstructed values at locations Dobovec and Kovk are compared. First situation occurred before and second after the TPP desulphurisation plant installation. 1000,00 900,00 800,00
200,00
Dobovec
SO2 concentration
180,00
[µg/m3]
160,00
Kovk Measured Reconstructed
140,00
700,00 600,00
Measured
120,00
500,00
Reconstructed
100,00
400,00
80,00
300,00
60,00
200,00
40,00
100,00
20,00
0,00 12.09.2005 00:00
SO2 concentration [µg/m3]
12.09.2005 06:00
12.09.2005 12:00
12.09.2005 18:00
13.09.2005 00:00
13.09.2005 06:00
13.09.2005 12:00
0,00 12.06.2006 00:00
12.06.2006 06:00
12.06.2006 12:00
12.06.2006 18:00
13.06.2006 00:00
13.06.2006 06:00
13.06.2006 12:00
Fig. 1 Comparison of reconstructed and measured SO2 concentrations at stations Dobovec and Kovk
Good behaviour and shortcomings of Lagrangian particle model approach were identified and discussed. Extensive verification showed that foundation for good reconstruction is good database without any corrupted data measurements. Modelling of the future scenarios (operating of two desulphurisation plants and possible new TPP) will be used by the decision makers to allow or reject building of new TPP (or other objects) in the area. As the area is considered to be highly polluted, these results may also help to prepare other remediation programs for better air quality in the area.
References Božnar M, Grašiþ B, Tinarelli G (2006) Thermal power plant Trbovlje air pollution impact modelling in complex terrain., Sixth Annual Meeting of the European Meteorological Society (EMS) [and] Sixth European Conference on Applied Climatography (ECAC): Ljubljana, Slovenia, 4–8 September 2006 (EMS annual meeting abstracts, volume 3). Ljubljana: European Meteorological Society: Agencija RS zaokolje
P2.13 Analysis of Atmospheric Transport of Radioactive Debris Related to Nuclear Bomb Tests Performed at Novaya Zemlya Jørgen Saltbones, Jerzy Bartnicki, Tone Bergan, Brit Salbu, Bjørn Røsting and Hilde Haakenstad
Abstract Tropospheric transport calculations of debris from nuclear bomb tests in the atmosphere at Novaya Zemlya in October 1958 and November 1962 have been performed and compared with daily measured radioactivity in air at Norwegian monitoring sites. Our analysis of potential vorticity (PV) anomaly strongly indicate that episodic intrusion of stratospheric air into the troposphere is the most probable transport mechanism for the peaks in radioactivity measured in Norway. The first hypothesis tested was that episodic increases in radioactivity in air sampled close to the ground in Norway were caused by a direct tropospheric transport of radioactive debris from the nuclear bomb tests at Novaya Zemlya. Forward trajectory computations from Novaya Zemlya have been performed, starting at different heights in the troposphere and for these periods, just a few of the bomb tests could possibly have sent radioactive debris via a tropospheric direct route to Norway. This indicated that a direct tropospheric transport from Novaya Zemlya to the Norwegian monitoring sites was not a very likely transport mechanism (Bartnicki et al. 2004; Bergan et al., 2005). However, when calculating the age of the radioactive debris, by observing ratios between radio isotopes with different decay rates, it seems likely that radioactive debris arrived at Norwegian monitoring sites 10–40 days after the detonations; – (Bergan et al., 2005). This supports earlier observations of quite long residence times for radioactive debris in the polar stratosphere. The nuclear bomb tests in these periods had variable total yields up to several kt and can be expected to have sent radioactive debris high up into the stratosphere just after the blast (STANAG, 1994). To investigate more closely the transport mechanism responsible for peaks in the measurements, we have worked out the following screening procedure: Using ERA-40 re-analyses data from ECMWF as a start, we have performed analysis/ short prognosis of the weather situation – using the Norwegian HIRLAM–10 km model, which is operational at the Norwegian Meteorological Institute. Pattern indicating intrusion of stratospheric air into the troposphere are, e.g. low relative humidity (Rh) and high values of potential vorticity (PV). Results from the HIRLAM model for 20 October 1958 indicate just that. As shown in Figure 1, high values of PV and low values of Rh, dipping down as a tongue just west of C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
683
684
J. Saltbones et al.
Norway, just at the place and at the time of the peak in total beta activity measured in Bergen. This is a strong indication of intrusion of stratospheric air.
Fig. 1 Vertical crossection north of Bergen showing potential vorticity (blue lines), potential temperature (red lines) and relative humidity (green lines) for 20 October 1958 at 12 UTC
Acknowledgments This study was supported and partly financed by Norwegian Research Council for Science and Humanities and by Norwegian Radiation Protection Authority.
References Bartnicki J, Foss A, Saltbones J (2004) Analysis of trajectories related to nuclear bomb tests performed in Novaya Zemlya. Met. no Note 9. Norwegian Meteorological Institute. Oslo, Norway. Bergan DT, Bartnicki J, Dowdal M, Foss A, Saltbones J (2005) Analysis of trajectories related to nuclear bomb tests performed at Novaya Zemlya. In: Proceedings from The 6th International Conference on Environmental Radioactivity in the Arctic & Antarctic, 2–6 October 2005 in Nice, France (P. Strand, P. Børretzen, T. Jølle, eds). Norwegian Radiation Protection Authority, Østeraas, Norway 2005. pp. 77–86. STANAG (1994) STANAG 2103, APT-45 Vol I/II: ‘Reporting Nuclear Detonations, Biological and Chemical Attacks, and Predicting and Warnings of Associated Hazards and Hazard Areas (NATO Declassified), June 1994.
P2.2 Application of Back Trajectories Using Flextra to Identify the Origin of 137Cs Measured in the City of Barcelona Delia Arnold, Arturo Vargas, Petra Seibert and Xavier Ortega
Abstract In order to determine the possible origin of the 137Cs detected at the Energy Technologies Institute (INTE) station from the Technical University of Catalonia (Barcelona, NE Spain), a study by means of trajectory analysis has been carried out and compared with results from Physikalisch-Technische Bundesanstalt (PTB) station in Braunschweig (Germany). A relation between high detection and incoming air from highly contaminated regions is clearly visible for the PTB site but not for Barcelona. Before the Chernobyl accident, the source of 137Cs was just from past nuclear weapons tests. This 137Cs was dispersed all over the globe and the measured values were therefore similar in different locations. However, after the Chernobyl accident 137 Cs activities increased significantly in Europe and especially near the Chernobyl region. Some works (Seibert and Frank, 2004; Swanberg and Hoffert, 2001; Wershofen and Arnold, 2005) have related measured values in Germany and Scandinavia and atmospheric transport simulations, indicating that the origin of 137 Cs is found in many cases in the highly contaminated areas of Ukraine, Belarus and Russia near Chernobyl. 137 Cs is measured in weekly seven-days samples at the INTE using a highvolume air sampler. To study its possible origin, ten-day backward trajectories computed by the FLEXTRA trajectory model (Stohl et al., 1995) have been analysed for the period 2001–2004. The results are compared with those found for the PTB station at Braunschweig (Germany) for the years 1998–2003 with the same method. At Barcelona approximately the 10% of the weekly measurements give 137Cs values above the detection level. Therefore, it has been decided to consider as episodes with high 137Cs measured values those 10% with highest measured values. For these selected periods, the percentage of the total residence times below the mixing height, which is associated with measurable 137Cs upon arrival of the trajectories, has been computed in a 1º × 1º grid covering Europe. In fair agreement with Wershofen and Arnold (2005) it has been found that an increase in 137Cs concentration at the German site is related to air masses coming from the East (see Figure 1, left). High values of residence times are found near the Chernobyl region, indicating that this is a possible source for the 137Cs measured at that site. From the visual inspection of trajectories, the highly contaminated areas of the Scandinavia C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
661
662
D. Arnold et al.
appear to be a possible source. This shall be studied in the future. For Barcelona, no clear influence of 137Cs transported from Belarus and Ukraine regions is found (Figure 1, right). The main reason is that almost no trajectories passed over the contaminated areas. It appears therefore likely that the contribution of locally or regionally resuspended 137Cs makes an equal or even dominant contribution to the measured 137Cs as compared to contributions from the surroundings of Chernobyl. Thus, even a simple tool such as trajectory residence times can be used as a first approach.
Fig. 1 Residence time per grid cell (%) for the PTB (left) and the INTE (right) stations. Crosses are station locations and the circled black dot is Chernobyl. White areas have zero residence times
Acknowledgments This work is included in the Spanish Education and Science Ministry project CGL2005-04182/CLI. Authors would like to thank the PTB to allow the use of data appeared in PTB-Ra-45 report and ECMWF for data access through the Special Project MOTT. P. Seibert acknowledges the support of the European Commission through the FP6 Network of Excellence ACCENT.
References Stohl A, Wotawa G, Seibert P, Kromp-Kolb H (1995) Interpolation errors in wind fields as a function of spatial and temporal resolution and their impact on different types of kinematic trajectories. J. Appl. Meteor. 34, 2149–2165. Seibert P, Frank A (2004) Source-receptor matrix calculation with a Lagrangian particle dispersion model. Atmos. Chem. Phys. 4, 51–63. Swanberg EL, Hoffert S (2001) Using atmospheric 137Cs measurements and Hysplit to confirm Chernobyl as a source of 137Cs in Europe. 23rd Seismic Research Review: Worldwide monitoring of nuclear explosions, October 2–5. Wershofen H, Arnold D (2005) Radionuclides in the Ground-level Air in Braunschweig – Report of the PTB Trace Survey Station from 1998 to 2003. Report Radioaktivität PTB-Ra-45, ISBN 3-86509-431-7, Braunschweig.
P1.3 Assessment of the Breathability in Urban Canyons Through CFD Simulations and Its Application to Sustainable Urban Design Mário Tomé, Ricardo J. Santos, António Martins and Mário Russo
Abstract According to the European Commission´s own Impact Assessment, every year, 369,980 people die prematurely because of air pollution. CFD simulations can be used to assess the advection and turbulent diffusion of pollutants emitted by moving sources on urban canyons. Although these tools need more development and validation, they are already sufficiently robust to support the urban design process, maximising the pollutants (and other scalars such as energy) cleaning inside the canyon. On the other hand, the urban designers and planners are aware of the importance of urban air quality but do not have enough knowledge to support their planning decisions as no relevant guidelines are available. In order to evaluate the capacity that an urban canyon has to exchange pollutant in its virtual ceiling top, CFD simulations were carried-out based on both eulerian and lagrangian approaches for both stationary and dynamic urban canyons with different aspect ratios. Moreover, the breathability conditions inside the canyons are defined using classical approaches borrowed from chemical reactor theory such as pollutant residence time distribution and dimensionless concentrations maps. CFD tools can contribute to understand and help design better urban landscapes for strong cleaning of the pollutants released inside the urban canyons (Xiaomin et al., 2006). Besides many remaining questions about the difficulties of CFD to model the open atmospheric boundary layer, namely its roughness and stability (Blocken et al., 2007), there are other important issues that need more studies. We are interested to compare eulerian with lagrangian approaches for recirculating flows that arise in urban canyons of aspect ratio of 1. The 2D simulation domain consists of an infinite array of buildings of 16 × 16 m separated from each others with 16 m. The domain height is 100 m. The infinite array was modelled using the periodic boundary condition of the Fluent CFD software applied to the singular repeated domain of 32 m. A simple fortran code extrapolated this periodic solution for a six canyons domain (6 × 32 m = 192 m). In this larger domain the flow is not computed because it has previously converged in the 32 m wide periodic domain with second order accuracy. The 192 m wide domain was used to compute the eulerian dis-persion of a passive pollutant emitted by a surface with 8 m wide by 1 m height. This virtual road has an emission rate of 0.001 kg pollutant/(m2s–1). The same emission is used by a lagrangian approach in which these same area of road emits 220 discrete particles in each lagrangian time step (=0.05 s). Thus, each C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
655
M. Tomé et al.
656
numerical particle represents 1.818 × 10-6 kg of pollutant. The flow field was initialized by an exponential profile with wind velocity of 5 m s-1 at the height of 26 m (10 m + displacement = 16 m). This initial wind field does not parameterize the solution because the periodic boundary is not dependent of this initial solution. The computed wind field has a maximum horizontal velocity of 8.36 m s-1 (at the domain top) and an air mass flow rate of 720 kg s-1. The turbulence model used was the standard k-e. Due to the strong vortex the lagrangian approach with stochastic numbers revealed to be much more time consuming and also it requires more attention from the modeller as the particles can be recirculating nearly for endless. Thus, eulerian and lagrangian results can produce quite different outcomes unless the lagrangian simulation is very “well done”, which means having appropriate (small) time steps, a very long global temporal simulation time period in order to achieve “lagrangian stability”, which can be easily assessed by the particles escaping rate from the domain being equal to the number o injected particles (220) per time step. A RTD analysis, from chemical engineering, is applied in order to interpret the results of a constant flux tracer (pollutant) emitted in the second canyon. The RTD curves (Figure 2) show an average time of pollutant residence in the domain much higher than the wind passage time. The wind should approximately transport materials: to x = 64 m in approximately 4 s; to x = 128 m in approximately 12 s; to x = 160 m in approximately 16s and to x = 192 m in approximately 20 s. The RTD curves at the different proving location keep approximately the same relative distance, but the typical value is much higher: to x = 64 m E(t)max = 30 s; to x = 128 m E(t)max = 45 s; to x = 160 m in E(t)max = 50 s and to x = 192 m E(t)max = 55 s. The Eulerian approach clearly shows that the steady recirculation vortices between buildings, seen to trap the particles, are dead volumes with very small mass transfer across the virtual ceiling, even considering turbulent diffusion. Probably LES results will yield less stable vortices between buildings, and thus account more precisely for mass transfer, representing the next envisaged step in this work. Results and further discussion of these simulations are presented in the poster (available [email protected]). 100
Pollutant mass fraction (-)
7.00E-03 5.00E-03 2.00E-03 1.00E-03 5.00E-04 2.00E-04 1.00E-04 5.00E-05 3.00E-05
Y (m)
79
58
37
zoom
16
8
24
40
56
72
88
104
X (m)
120
136
152
168
184
Fig. 1 Eulerian concentration (mass fraction) field of pollutant and pathlines
Fig. 2 RTD analysis, with the system outlet boundary at different distances, from the “frozen” wind field and a step tracer emission curve
Acknowledgments R. Santos is grateful for the funding from POCI/N010/2006.
References Blocken B, Tathopoulos T, Carmeliet J (2007) CFD. Atmos. Environ. 41, 238–252.
P4.5 BOLCHEM Air Quality Model: Performance Evaluation over Italy Alberto Maurizi, Mihaela Mircea, Massimo D’Isidoro, Lina Vitali, Fabio Monforti, Gabriele Zanini and Francesco Tampieri
Abstract The modeling system BOLCHEM for air quality simulations has been run to study the evolution of tropospheric ozone over the Italian peninsula during 1999. The comparison of measured and modeled ozone time series shows that BOLCHEM predicts well the ozone concentrations. The summer cases are better simulated than the winter ones. The model configuration using SAPRC90 meets always the US-EPA criteria for the statistical indexes UPA, MNBE and MANGE.
The Italian peninsula has a very complex topography, therefore the separation of meteorology and chemistry in offline simulations can lead to a loss of potentially important information about atmospheric processes, which often have a much smaller time scale than the meteorological output frequency. Here, we show the ability of a new developed air quality model, BOLCHEM, to reproduce the observed ozone concentrations for four clear sky periods (one from winter, the others from summer season) selected based on Meteosat images of Europe. The calculated O3 concentrations were compared with measurements made at rural or semi-rural stations. The statistical measures recommended by the U.S. Environmental Protection Agency (US-EPA, 2005) for air quality model validations were also computed for hourly values of ozone concentration. BOLCHEM couples the meteorological model BOLAM (Buzzi et al., 2003) to SAPRC90 (Carter, 1990), and CB-IV (Gery et al., 1989) as alternative photochemical mechanisms. The mechanisms were chosen since they adopt different criteria for grouping the organic gases: CB-IV groups the organics according to bond type, while SAPRC90 groups them according to molecule type. Figure 1 shows that the agreement between simulated and measured ozone concentrations is good for both summer and winter. Generally, the model has difficulties in reproducing low ozone concentrations observed during the winter and during the night. A more extensive discussion of the results can be found in Mircea et al. (2007). Table 1 shows the statistical indexes recommended by US-EPA (USEPA, 2005). It can be seen that generally, UPA is lower than 35%, MNBE is lower than 15%, MANGE is lower than 30–35%.
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
702
BOLCHEM Air Quality Model: Performance Evaluation over Italy
703
Montelibretti 300 3
2 P g/m )
250 200
SAPRC90
150
CB4
100
obs
50 1999080717
1999080709
1999080701
1999080617
1999080609
1999080601
1999080517
1999080509
1999080501
1999070417
1999070409
1999070401
1999070317
1999070309
1999070301
1999070217
1999070209
1999070201
1999070117
1999070109
1999070101
1999060317
1999060309
1999060301
1999060217
1999060209
1999060201
1999060117
1999060109
1999060101
1999012417
1999012409
1999012401
1999012317
1999012309
1999012301
1999012217
1999012209
1999012201
1999012117
1999012109
1999012101
1999012017
1999012009
1999012001
0
Fig. 1 Simulated (SAPRC90-red line and CB-IV-pink line) and observed (green line) surface ozone concentrations at Montelibretti during one winter and three summer periods Table 1 Quantitative model performance statistics for hourly surface ozone over the whole domain during summer. MNBE and MANGE were calculated with 40 ppb cutoff to the ozone concentrations.
1–3 June
UPA SAPRC90 –20.45
CB-IV –37.15
MNBE SAPRC90 –6.98
CB-IV –19.55
MANGE SAPRC90 22.53
CB-IV 24.61
1–4 July 5–7 August
9.3 –13.96
–17.59 –28.74
–4.37 –2.37
–19.38 –16.46
25.60 22.84
27.71 23.81
The model performances are better during summer, when the photochemistry is active, than during winter. During summer, the validation exercise shows that the model configured with SAPRC90 always meet the US-EPA criteria for UPA, MNBE and MANGE, while the MNBE calculated with CB-IV ozone concentration is sometimes higher than the recommended values. Acknowledgments This work was supported by the European Commission through the Network of Excellence ACCENT and the project GEMS, and by the Italian Ministry of Environment through the Program Italy-USA Cooperation on Science and Technology of Climate Change.
References Buzzi A, D’Isidoro M, Davolio S (2003) A case-study of an orographic cyclone south of the Alps during MAP SOP, Quart. J. Roy. Met. Soc., 129, 1795–1818. Carter WPL (1990) A Detailed Mechanism for the gas-phase atmospheric reactions of organic compounds, Atmos. Environ., 24A, 481–518. Gery MW, Witten GZ, Killus JP, Dodge MC (1989) A photochemical kinetics mechanism for urban and regional scale computer modeling. J. Geophys. Res., 94 (D10), 12925–12956. Mircea M, d’’Isidoro M, Maurizi A, Vitali L, Monforti F, Zanini G, Tampieri F (2007) A comprehensive performance evaluation of the air quality model BOLCHEM over Italy, doi:10.1016/j.atmosenv.2007.10.043.
704
A. Maurizi et al.
US-EPA, U.S. Environmental Protection Agency (2005) Guidance on the use of models and other analyses in attainment demonstrations for 8-hour ozone NAAQS. EPA report EPA-454/R-05-002, 128pp.
P3. Data assimilation and air quality forecasting
P3.1 Detection of a Possible Source of Air Pollution Using a Combination of Measurements and Inverse Modelling Borivoj Rajkovic, Zoran Grsic and Mirjam Vujadinovic
Abstract Detection of a possible source of air pollution as a combination of measurements and inverse modelling based on Bayesian statistics has been proposed. The simplicity of the approach and its numerical efficiency qualifies it for the problem, especially in the operational mode. Detection of an air pollution point source, during an accidental release, requires time efficient and relatively accurate solution. These two characteristics can be fulfilled with an approach based on the Bayesian statistical method (Fuentes and Raftery, 2001; Tarantola, 2005, among others). Relative positions of the assumed source and measurement points are presented in the Figure 1, upper right panel. The first step, in finding source’s position, was to generate field of passive substance concentration by its release during 120 minutes, using a PUFF model (Grsic and Milutinovic, 2000). The grid had 301 × 301 points with spacial distance of 60 m. The time interval between two consecutive puffs was 1 minute. Values at the measurement positions where then randomly pertubatied by 5%, thus mimicking the measurement errors. The second step was to create the cluster of points, possible sources, with center positioned in the measuring point with the highest concentration observed. In the first iteration we had relatively high cluster resolution of 30 grid points. From the cluster points we calculated probability density function (pdf) of the source position, assuming that it is Gaussian (Tarantola, 2005). In the second iteration, we translated the center of a cluster, to the possition of the first pdf’s maximum. Again, we calculated pdf and its maximum and translated the center to the possition of the new maximum. Since now cluster member with maximum pdf was at the inside area of the previous cluster, we decided also to halve the spread among third cluster’s members. In the next two iterations we again translated the cluster and halved the distance between its members. The first four, consecutive, positions of the cluster are presented in Figure 1, left panel. The right lower panel shows the last cluster position and calculated source location (Ms). The error in the obtained possition, after five iterations, was 120 m (two grid points). C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
689
690
B. Rajkovic et al.
Fig. 1 The right upper panel shows relative positions of the source and observation points. The left panel presents changes of the cluster position during four iterations, while the position of the final, fifth, cluster is shown at the right lower panel, where Ms denotes modelled position of a source
Combination of measurements and Bayesian statistical approach through several iterations could result in a very accurate and yet efficient method of detecting a possible, point like, source of a pollution. Acknowledgments This work has been partialy funded by Republic of Serbia, Ministry of Science, Technology and Environment, grant no. 1197, Italian Ministry of Environment and Territories through its two projects, SINTA and ADRICOSMSTAR. The second author is partially funded through the project of the decommission of the nuclear reactor in Vinca.
References Grsic Z, Milutinovic P (2000) Air Pollution Modelling and Its Application, XIII: Automated meteorological station and the appropriate software for air pollution distribution assessment. Fuentes M, Raftery A (2001) Model evaluation and spatial interpolation by combining observations with outputs from numerical models via Bayesian Melding, Technical report no. 403, Department of Statistics, University of Washington, Seattle, WA. Tarantola A (2005) Inverse problem theory and methods for model parameter estimation, SIAM, Philadelphia.
P2.14 Development and Application of a New Model for the Atmospheric Transport and Surface Exchange of Semi-Volatile Organics Using the CMAQ Model Framework Fan Meng, Baoning Zhang, Fuquan Yang and James Sloan
Abstract PCBs and PCDD/Fs are toxic, persistent pollutants that can bioaccumulate in the food chain and become serious health hazards. Although the manufacture of materials such as PCBs has been banned in most parts of the world for some time, they are still found in significant concentrations in the environment. In view of this, it has become important to understand not only their sources, but also the mechanisms responsible for their transport in the environment. Since they are semi-volatile, it is necessary to consider their transport by atmospheric particulate matter as well as in the gas phase. We have developed a capability to simulate the atmospheric behaviour of PCBs and PCDD/Fs within the framework of the CMAQ modelling system. To describe transport on particulate matter, we have added two gas/particle partitioning models – the Junge-Pankow adsorption model and the KOA absorption model to the basic CMAQ system. We have also included gas phase chemistry of these semi-volatile organic materials as well as their atmosphere/water surface exchange processes. Using this modified model system, we have conducted simulations of the atomspheric behaviour of these materials for the years 2000 and 2002 on a domain covering most of North America. Validation studies show that both partitioning models give reasonable results when compared with available measurements of deposition rates and air concentrations. The simulations confirm that long range transport occurs by both gas phase and heterogeneous mechanisms. This causes these toxic materials to be deposited in pristine regions far from emission sources. In cases where they have entered the water table, large water bodies such as the Great Lakes can also become net sources. Polychlorinated Biphenyls (PCBs), Polychlorinated Dibenzo [p] Dioxins and Polychlorinated Dibenzo-Furans (PCDD/Fs) are toxic persistent organic pollutants (POPs) that can be transported over global scales, thereby affecting human health and the ecology. Most PCBs and PCDD/Fs are semi-volatile organic compounds that partition between gas and particle phases in the atmosphere and between the atmosphere and surface soil and water. Since the chemical transformation and removal of these POPs in the gas phase differ from that in the particle phase, the partitioning process is a key factor for simulating their fate in the atmosphere. C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
685
F. Meng et al.
686
We have added modules that simulate PCBs and PCDD/Fs to the Community Multiscale Air Quality (CMAQ) model of the U.S. EPA. This allows us to take advantage of its capability to predict real-time spatially and temporally resolved aerosol concentrations. The added components are the Junge-Pankow adsorption and KOA absorption gas/particle partitioning models to CMAQ, as well as important gas phase chemical reactions involving the PCBs and PCDD/Fs. For gas phase PCBs we also added an atmosphere/water surface exchange model. We compared the performance of the resulting model systems with existing measurements. A total of 22 PCB congeners and the 17 most toxic PCDD/F congeners are included in this model system. Gas phase PCBs concentration
Modeled(pg/m3)
Measured(pg/m3)
PCB18
PCB52 PCB101
Fig. 1 Comparison of modelled gas-phase concentration of PCBs with measurements for January 2000–July 28, 2000
Fig. 2 Modelled surface level PCB gas phase concentrations after model spinup
Figure 1 is the comparison of the predicted results for PCB18, PCB52 and PCB101 with measurements from the Integrated Atmospheric Deposition Network (IADN) sites for the year 2002. The agreement is generally good; the model is in reasonable agreement with the measurement for PCB101 and PCB 52, but overestimates PCB18, the lighter (tri-chloro) congener, which is expected to be predominantly in the gas phase The modelling domain and typical predicted gas phase PCB concentrations are shown in Figure 2. PCB emissions are from isolated continental sources and this result shows significant atmospheric transport on a continental scale. As a result, there is considerable deposition of this material into the great Lakes and related watersheds. Acknowledgments We are pleased to acknowledge the financial support of Ontario Power Generation Inc., Natural Sciences and Engineering Research Council Canada, Canadian Ortech Environmental Inc. and the Province of Ontario.
P4.3 Development of a Quasi-Real-Time Forecasting System over Tokyo Masayuki Takigawa, Masanori Niwano, Hajime Akimoto and Masaaki Takahashi
Abstract We present an evaluation of the distribution of ozone over Kanto region, calculated by using a one-way nested global/regional air quality forecasting (AQF) system. This AQF system consists of the global chemistry-transport model (CTM) part and the regional CTM part. The global CTM part is based on CHASER, and the regional CTM part is based on WRF/Chem. An experimental phase of this model system began operation in July 2006 and has been providing 15-hour forecasts of the distribution of ozone concentrations over Kanto region four times in a day. The time-evolution and horizontal distribution of chemical species calculated by this AQF system were compared to ground–based observations.
The global CTM part is based on CHASER (Sudo et al., 2002). Spectral coefficients are triangularly truncated at wavenumber 42 (T42), equivalent to a horizontal grid spacing of about 2.8. The model has 32 vertical layers. The regional CTM part is based on WRF/Chem (Grell et al., 2005). Two-domains’ calculation has been done in the regional CTM part. The outer domain covers over Japan with 15 km horizontal resolution, and the inner domain covers over Kanto with 5 km resolution. The inner and outer domain in the regional CTM have 31 vertical layers up to 100 hPa. Anthropogenic emissions except automobiles over Japan are taken from JCAP (Japan Clean Air Program) with 1 × 1 km resolution (Murano, private communication), and anthropogenic emissions from automobiles over Japan are taken from EAgrid2000 (East Asia gridded emissions inventory) with 1km × 1km resolution (Kannari et al., 2007). Surface emissions over China, North Korea, and Korea are taken from REAS with 0.5q × 0.5q resolution (Ohara et al., 2007). The lateral boundary of chemical species in the regional CTM part is taken from the global CTM part. The lateral boundary is updated every 3–hourly, and linearly interpolated for each time step. The feedback from the regional CTM part to the global CTM part is not taken into account in the present study, i.e., the one–way nesting calculation has been done between the global and regional CTMs. A 15hour forecast has been produced four times in a day with a 8–10 hour lead time since August 2006. The initial condition of meteorological field for the regional CTM part is taken from Mesoscale Model (MSM) of Japan Meteorological Agency (JMA) for each forecast, and the initial condition of chemical species is taken from the model output driven by the analysis meteorology.
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
697
698
M. Takigawa et al.
To evaluate the model-calculated ozone, the surface ozone mixing ratio was compared to that observed at air quality monitoring stations. In August 2006, 251 stations observed ozone within the inner domain of the regional CTM. For comparison of temporal variation, hourly observed and modelled surface ozone mixing ratios in August 2006 are shown in Figure 1. Observed ozone exceeded 100 ppbv from 3 to 6 August at Hanyuu in Saitama prefecture (3610’ 28’’ N, 13933’ 21’’ E), which is downwind from the Tokyo Metropolitan area. The maximum value in the observation was 162 ppbv at 16Z on 3 August. The model reproduced the ozone maximum on 3 August well. The maximum simulated value was 137 ppbv in the model. The model also successfully captured the decrease from 3 to 7 August, but failed to show the rapid decrease on 8 August. Three typhoons (‘Maria’, ‘Somai’, and ‘Bopha’) existed in this period, and the difficulty of predicting the meteorological field may have led to overestimation of ozone on 8 August. The observed and modelled ozone exceeded 100 ppbv on 11, 13 August, and the model overestimated ozone mixing ratio on 19 August. Modelled ozone mixing ratio was 135 ppbv, whereas the observed ozone mixing ratio was 86 ppbv. Fig. 1 Hourly observed (solid line) and modelled (dashed) surface ozone mixing ratio in August 2006 at Hanyuu in Saitama Prefecture. Units are ppbv
References Grell GA, Peckham SE, Schmitz R., McKeen SA, Frost G, Skamarock WC, Eder B (2005) Fully coupled “online” chemistry within the WRF model, Atmos. Environ., 39, 6957–6975. Kannari A, Tonooka Y, Bada T, Murano K (2007) Development of multiplespecies 1km × 1km resolution hourly basis emissions inventory for Japan, Atmos. Environ., 41, 3428–3439, doi:10.1016/j.atmosenv.2006.12.015. Ohara T, Akimoto H, Kurokawa J, Horii N, Yamaji K, Yan X, Hayasaka T (2007) Asian emission inventory for anthropogenic emission sources during the period 1980–2020, submitted to Atmos. Chem. Phys. Dis. Sudo K, Takahashi M, Kurokawa J, Akimoto H (2002) CHASER: a global chemical model of the troposphere 1. Model description, J. Geophys. Res., 107, doi:10.1029/2001JD001113.
P4.6 Evaluation of an Operational Ensemble Prediction System for Ozone Concentrations over Belgium Using the CTM Chimere Andy Delcloo and Olivier Brasseur
Abstract In the framework of operational air quality forecasts in Belgium, an ensemble prediction system based on the chemical transport model (CTM) Chimere has been implemented. The Chimere model was forced by ECMWF meteorological fields and by the EMEP emission database. The simulation domain covers Western Europe with a spatial resolution of 0.5q. The objective of these ensemble simulations consists of evaluating the impact of uncertainties from emission and meteorological data on the simulated concentrations of pollutants. Such evaluations are important in the operational context, since they contribute to reduce the risk of false alarm and inappropriate broadcast of information to the public. Indeed the forecaster has at its disposal more information to better judge to what extent a change in one or more particular input parameters can influence the modelled pollutant concentrations. A first assessment of the ensemble prediction system has been performed for ozone forecasts during summertime. Currently, the Chimere model is run considering 13 different scenarios in which some variables influencing pollutant dispersion and ozone chemistry (e.g. temperature, wind velocity, cloud cover) are perturbed. The treatment of all these simulations allows defining a confidence interval around the concentrations simulated by the reference (i.e. without change in input parameters) simulation, which contributes also to improve the accuracy of the ozone forecast. Considering physical aspects, the ensemble prediction system contributes to point out – for each specific situation – the most sensitive input parameters that influences ozone concentrations. The observations, used to validate the Chimere model are deduced from an advanced interpolation method (kriging). This “RIO-algorithm”, developed by VITO (Hooyberghs et al., 2006), provides every hour an ozone value which takes into account the representativeness of the location the user is interested in. The ozone maps created with this algorithm have a spatial resolution of 5 km2. Since a CTM is very depended on the meteorological fields, it is interesting for the forecaster to know what will be the outcome of a sudden change in these meteo fields for the next day (D + 1). Therefore we roughly changed (in a first stage) 12 parameters: Cloud cover: (( 0 or 1 ) (CLO, CL1), +50% (Cp5), –50% (Cm5));
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
705
A. Delcloo, O. Brasseur
706
temperature: (+2ºC (Tp2), +4ºC (Tp4), –2ºC (Tm2), –4ºC (Tm4)) and wind speed: (–1 m/s (Wm1), –2 m/s (Wm2), +1 m/s (Wp1), +2 m/s (Wp2)). As an example results are shown for latitude 51 and longitude 4.5 (Figure 1 and Table 1). To validate the results, some general statistics (bias, rmse, correlation) are calculated on this different runs for the forecasted day + 1 (D + 1): Max EPS forecast for D+1 at LAT = 51.0 and LON = 4.5 140
120
Ozone (ppb)
100
80
60
40
20 01/Jun/2006
11/Jun/2006
21/Jun/2006
02/Jul/2006
12/Jul/2006
22/Jul/2006
01/Aug/2006
Fig. 1 Boxplot of maximum daily ozone concentrations (ppb) for the time period 01/06/2006– 31/07/2006. The 10, 25, 50, 75, 90 percentiles and the mean value (black cross) are shown. The red lines corresponds to the observed values with RIO, the green line represents the standard run of Chimere Table 1 Bias, correlation and rmse for the time period 01/06/2006–31/07/2006. STD
CL0
CL1
Cm5
Cp5
Tm2
Tm4
Tp2
Tp4
Wm1
Wm2
Wp1
Wp2 -10.6
Bias
-10.2
-2.3
-27.9
-7.9
-11.4
-9.0
-9.2
-8.8
-6.7
-9.8
-9.5
-10.4
Corr
0.85
0.80
0.65
0.85
0.85
0.86
0.86
0.87
0.87
0.85
0.85
0.85
0.85
Rmse
15.6
13.9
32.8
14.4
16.6
14.5
14.8
14.4
13.6
15.4
15.3
15.8
15.8
The results show that Chimere has a high negative bias for this particular time period. When we reduce the cloudcover, the bias and rmse improves, but the correlation declines. Table 1 shows that in general the changes in wind speed and temperature are less important. Only Tp4 shows significantly better scores for bias, correlation and rmse, compared with the standard run of Chimere.
References Hooyberghs J, Mensink C, Dumont G, Fierens F (2006) Spatial interpolation of ambient ozone concentrations from sparse monitoring points in Belgium, J. Environ. Monit., 8, 1129–1135.
P2.7 Evolution of the Ozone Episodes in Northern Iberia (Cantabric and Pyrenaic regions) Under West European Atlantic Blocking Anticyclones V. Valdenebro, G. Gangoiti, A. Albizuri, L. Alonso, M. Navazo, J.A. García and M.M. Millán
Abstract How main ozone episodes registered in the Basque Country (BC), at northern Iberia (Figure 1), affect the neighbouring areas and which are de mechanisms and pathways to export pollutants to these areas is analyzed. The blocking anticyclones over the British Islands, which are behind most of the ozone episodes in the BC, are related with regional and sub-continental transport of pollutants into this region, as we have documented in previous studies (Gangoiti et al., 2002 2006). The use of a coupled high resolution RAMS-HYPACT modelling system has allowed us to find various mechanisms and pathways for the importing of pollutants into the BC (Figure 3a, b). Now we analyze how these episodes affect the neighboring areas (Cantabric and Pyrenaic regions), which are the mechanisms and pathways to export pollutants from the BC, and how the episodes dissipate. To fulfil these tasks the AirBase ozone data corresponding to the Cantabric and Pirenaic regions are analized and the previous RAMS-HYPACT simulations have been extended in time. 55N 50N
GRID #1
GRID #2
46N
1
44N
GRID #2
45N
42N
40N
GRID #3
BAY OF BISCAY 3 4 6 8 6 7
2
9
11 10
40N
35N
12W 10W 8W
30N
0
20W 15W 10W 5W
0
5E 10E 15E
6W
4W
2W
0
2E
4E
6E
200 400 600 800 1000 1200 1400 1600 1800 2000 2200 2400
HEIGHT (m)
Fig. 1 Topographic map of the three domains used by RAMS-HYPACT, with the selected sources of pollution (left), and the location of ozone monitors (right). The BC is inside grid #3
Our results show that this type of ozone episodes is recorded all along the Cantabrian coast, where they begin in a quasi simultaneous manner and last similarly. Maximum concentrations are registered earlier in the W zone (during the accumulation phase) and later in the SW of France (during the dissipation phase). Pollutants from the BC are transported towards the Atlantic Ocean and coast of Portugal following two main pathways (along the Duero Valley and along the C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
671
672
V. Valdenebro et al.
northern coast of Iberia) at the beginning of the episodes (Figure 3a); then, they can continue their transport along the coast of Portugal towards the African continent. At the final stage of the episodes (Figure 3c), there is a massive transport towards southern France and along the Ebro Valley, towards the Mediterranean Sea.
Fig. 2 Time sequences of the ozone monitors concentrations (Figure 1) during the June 2001 episode
Fig. 3 Main pathways during the episodes: (a) accumulation, (b) peak, and (c) dissipation phase
Acknowledgments The authors wish to thank the Basque Government for their supply of data and the Spanish Ministry of Science and Technology for financing.
References Gangoiti G, Alonso L, Navazo M, Albizuri A, Pérez-Landa G, Matabuena M, Valdenebro V, Maruri M, García JA, Millán MM (2002) Regional transport of pollutants over de Bay of Biscay: analysis of an ozone episode under a blocking anticyclone in west-central Europe, Atmospheric Environment, 36, 1349–1361. Gangoiti G, Albizuri A, Alonso L, Navazo M, Matabuena M, Valdenebro V, García JA, Millán MM (2006) Sub-continental transport mechanisms and pathways during two ozone episodes in northern Spain, Atmospheric Chemistry and Physics, 6, 1469–1484.
P1. Local and urban scale modelling
P1.1 Finite Volume Microscale Air-Flow Modelling Using the Immersed Boundary Method V. Fuka and J. Brechler
Abstract This contribution describes results of computation of a turbulent flow over a square cylinder by 2D large eddy simulation. Solid wall boundary conditions were described by the immersed boundary method. We choose this example as a part of validation of a CFD model we are developing for flows in geometrically complex (namely urban) areas. Square cylinder is here used as a prototype of bluff body, instead of a real 3D building, which will be computed later. For time discretisation of incompressible Navier-Stokes equations, we used the fractional step method. For spatial discretization, the finite volume method was used, but for advective fluxes, we used central high-resolution scheme of Kurganov and Tadmor (2000). The turbulence was treated in the context of ILES (implicit large eddy simulation (Drikakis, 2003)) using the Kurganov-Tadmor method. The complex geometry on Cartesian grid was described by the immersed boundary method (Kim et al., 2001). Due to the computational resources and the present state of our model we performed the calculation in 2D. Although the turbulence is a inherently a 3D phenomenon, according to Bouris and Bergeles (1999) the 2D computation can capture most of important features of the quasi-two-dimensional flow. In addition, in 2D, one can use grid with better resolution. In Figure 1 is a snapshot of the vorticity field at time t = 200. The vortex shedding is clearly visible. Strouhal number of vortex shedding was 0.12, which is slightly less than 0.13–0.14 reported by Bouris and Bergeles (1999). Averaged horizontal velocity profile at the centerline is in Figure 2. Recirculation length was 0.6, which is less than experimental value, but it is consistent with 2D calculations of Bouris and Bergeles (1999).
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
651
V. Fuka, J. Brechler
652
Fig. 1 Instant vorticity field at t = 200
Fig. 2 Averaged velocity profile at the centerline, solid line is the present study, other lines are other numerical results and symbols are experimental values
Acknowledgments This research has been supported by the Grant Agency of the Czech Academy of Sciences, grant no. T400300414 and by the Grant Agency of the Czech Republic, grant no. 205/06/0727.
References Bouris B, Bergeles G (1999) 2D LES of vortex shedding from a square cylinder, J. Wind Engrg. Indus. Aerodynam. 80, 31. Drikakis D (2003) Advances in turbulent flow computations using high-resolution methods, Prog. Aerospace Sci. 39, 405. Kim J, Kim D, Choi H (2001) An immersed-boundary finite-volume method for simulations of flow in complex geometries, J. Comput. Phys. 171, 132. Kurganov A, Tadmor E (2000) New high-resolution central schemes for nonlinear conservation laws and convection-diffusion equations, J. Comput. Phys. 160, 241.
P2.8 High Temporal Resolution Measurements and Numerical Simulation of Ozone Precursors in a Rural Background M. Navazo, N. Durana, L. Alonso, J. Iza, J. A, García, J.L. Ilardia, G. Gangoiti and M. De Blas
Abstract High-resolution numerical modelling results – using RAMS and HYPACT – of an ozone precursors episode, detected on a rural background area in Northern Iberia, are shown. The episode was identified by data analysis of a continuous VOC measurement system. Simulation results show that, when low temporal and spatial resolutions are used, the origin of polluted air masses affecting targets at low heights may be wrongly interpreted. A very complete database of individual non-methane hydrocarbon measurements with high temporal resolution (hourly) in a rural background atmosphere was prepared between January 2003 and December 2005. That kind of database can be used for biogenic NMHC characterization as well as for the identification of the transport and impact of anthropogenic NMHC on rural areas (Durana et al., 2006). The measurement system operated continuously in the centre of the Valderejo Natural Park in northern Iberia. Data coverage was higher than 70% for a total of 59 VOC of both anthropogenic and biogenic origin, with detection limits in the range of pptv. This database was used to describe the behaviour of these compounds, in order to identify the chemical transformations and external impacts arriving to the sampling site, highly representative of a rural background atmosphere.
Fig. 1 Concentration time series of selected aromatic VOCs between August 1 and 7, 2003
A coupled RAMS-HYPACT modelling system, applied over the target region with a temporal and spatial resolution of 1 hour, and 3 × 3 km, respectively (Gangoiti et al., 2006), was used for the duration of the episode, when high VOC levels were detected at night (Figure 1). The Toluene/Benzene ratio was characteristic of an old C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
673
M. Navazo et al.
674
air mass, which could not get to the site unless long-range transport mechanisms were involved (Figures 2 and 3). Fig. 2 Results from the meteorological and dispersion simulation (using RAMS & HYPACT models) for August 5, 2003 at 0300 UTC
Fig. 3 Corresponding N-S (left) and E-W (right) transects to Figure 2
Acknowledgments To the personnel from the Valderejo Natural Park Center, for their friendliness and logistic support; to the Basque Government for his support, and to the Spanish MEC for financing the projects MAMECOVA and TRAMA
References Gangoiti G, Albizuri A, Alonso L, Navazo M, Matabuena M, Valdenebro V, García JA, Millan MM (2006) Sub-Continental transport mechanisms and pathways during two ozone episodes in northern Spain, Atmos. Chem. Phys., 6, 1469–1484. Durana, N, Navazo M, Gómez MC, Alonso L, García JA, Ilardia JL, Gangoiti G, Iza J (2006) Long term hourly measurement of 62 nonmethane hydrocarbons in an urban area: Main results and contribution of non-traffic sources, Atmos. Environ., 40, 2860–2872.
P2.12 High Time and Space Resolution Ozone Modelling in Regional Air Quality Management of a Complex Mountain Area Using Calgrid 2.44 Carlo Trozzi, Silvio Villa and Enzo Piscitello
Abstract In this paper is reported the study of ozone pollution in Autonomous Province of Trento referred to year 2004, with a suite of meteorological (MM5 and Calmet) and photochemical (Calgrid) models in a high detailed space and time resolution over long time period. Trento province occupies a surface of 6.207 km2 (2.9% of national territory); 70% of the surface is located over 1,000 m above sea level. The resident people in Trento province is estimated as many as 450,000. Every year a number of as many as 28 million of tourists involves an additional pressure on the territory. Trento area is a very particular area (a mountain area in Alps) in which the air quality protecttion is a primary goal. CALGRID model (Yamartino et al., 1992) requires input meteorological parameters for every 1 × 1 km cell of the geographical domain. This task is accomplished by Calmet pre-processor (Scire et al., 1992), whose input was the terrain geographical features, land use, various boundary layer specific variables, hourly meteorological parameters coming from the regional network of agrometeorological stations, and upper atmosphere parameters directly modeled with the MM5 model. The non-hydrostatic MM5 model (National Center for Atmospheric Research, 2005) was set up for a coarse domain formed by cells 15 × 15 km. wide with a nest of 5 × 5 km. cells enclosing the entire Calgrid domain. The emission input for Calgrid was based on a highly detailed emission inventory following a bottom-up approach. Emissions were evaluated for 55 point sources, 416 line sources (6 highway sections and 410 extraurban road sections) and 225 area sources (municipality). Line and area sources were assigned to 1 × 1 km grid using proxy variables (mainly derived from land use maps). Emissions were disaggregated on hourly basis using surrogate variables and estimated for 150 Corinair activities. Volatile Organic Compounds emissions were speciated in SAROAD classes for input in chemical mechanism (Carter, 1990). A specific information system was used. A slightly code modified Calgrid v.2.44 model run, with case specific parameters choice and SAPRC97 chemical mechanism, produced interesting results in determining ozone concentrations, in particular for mountainous and high forested terrain, as shown in Figure 1. A subsequent study focused on medium-high populated C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
681
C. Trozzi et al.
682
places like cities in the numerous valleys included in Trento Province and legislative indexes were calculated on that area. The AOT40 index was also calculated, showing a high percentage (75% approx.) of cells exceeding this index.
190 180 170 160 150 140 130 120 110 100 90 80 70 60 50 40 30
Fig. 1 Average ozone concentration from May 1 to August 31, 2004 expressed in µg/m³ as calculated by Calgrid model
Acknowledgments The work was conducted on behalf of Autonomous Province of Trento Environmental Protection Agency.
References Carter WPL (1990) A detailed mechanism for the gas-phase atmospheric reactions of organic compounds. Atmos. Environ., 24A, 481–518. Scire JS, Insley EM, Yamartino RJ, Fernau ME (1995) A User’s Guide for the CALMET Meteorological Model. Yamartino RJ, Scire JS, Hanna SR, Carmichael GR, Chang YS (1992) The CALGRID mesoscale photochemical grid model - I. Model formulation. Atmos. Environ., 26A, 1493–1512. National Center for Atmospheric Research (2005) PSU/NCAR Mesoscale Modeling System. Tutorial Class Notes and User’s Guide, MM5 Modeling System Version 3, January 2005.
P7.3 Intake Fraction for Benzene Traffic Emissions in Helsinki Joana Soares, Ari Karppinen, Leena Kangas, Matti Jantunen and Jaakko Kukkonen
Abstract Benzene (Bz) is well known for its haema and genotoxicity and the carcinogenic effect associated with long time exposure. In urban environment, traffic is an important source for ambient air Bz concentrations. In order to quantify emission-to-intake relationships, intake fraction (iF) was defined as the integrated incremental intake of Bz released from a source (or source category) and summed over all exposed individuals during a given exposure time, per unit of emitted pollutant (Bennet et al., 2000). iF takes into account the dispersion of pollutants, locations and activity of population, and human breathing rates. The calculated iF for Bz is directly applicable to any other inert substance emitted by the traffic, e.g. CO, NOX, so the calculations also provide a ready-to-use tool for health effects studies concerning other pollutants and emission scenarios. This study calculates the spatial distribution of average benzene iF for Helsinki Metropolitan Area (HMA) using the EXPAND model (Kousa et al., 2002). The spatial Bz concentration distributions were obtained by using dispersion models: CAR-FMI (Karppinen et al., 2000) and OSPM (Berkowicz, 2000). A constant breathing rate of 1 m3/day was considered. The EXPAND results for 2000 are shown in Figure 1.
Fig. 1 Spatial distribution of total intake fraction for benzene from mobile sources in 2000
The total iF for HMA is 2.8 × 10-5 with higher values concentrated in residential/ commercial areas, ranging between 10-10 and 10-7, where people spend most of their time. The partial intake fraction (iFi) due to exposure in traffic is 3.1 × 10-6. A study conducted in California (Marshall et al., 2003), shows an iF for Bz in the same order of magnitude: 3.3 × 10-5. Moreover, a total iF of 3.7 × 10-5 was calculated for C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
719
J. Soares et al.
720
Hämeentie Street., in Helsinki, based on street canyon dispersion calculations, and activity and number of inhabitants and workers in this area (Table 1). Table 1 also shows iFi calculated for this sample of the total affected population. Table 1 Intake fraction by inhalation in Hämeentie Street (2000). Groups
Breathing rate (m3/day)
No. of people
Time of exposure (day)
iFi
Inhabitants
1
176
0.56
1.6E-05
1
113
0.37
2.0E-05
1
20,380
1.04E-04
1.1E-06 3.7E-05
Working and costumers In traffic Total iF
It is clearly shown that people travelling through the street canyon are order of magnitude less exposed than the people living and working in that area, due to the short amount of time they are spending in the street. This was also show in the HMA calculations. So if the time spent, e.g., commuting or indoors would be the same, in-traffic exposure would be much more relevant. The iF spatial variation, in particular the detailed calculations for the smallest domain, demonstrates clearly how crucial it is to have access to detailed information on traffic patterns and locations and activities of the people in order to get a reliable estimate on the real burden of pollutants on human exposure and health. Acknowledgments The financial support from CEFIC-LRI is gratefully acknowledge
References Bennet D, Margni M, McKone T, Jolliet O (2002) Intake fractions for multimedis pollutants: a tool for life cycle analysis and comparative risk assessment. Risk Anal. 22, 819–1033. Berkowicz R (2000) A simple model for urban background pollution. Environ. Monit. Assess. 65, 259–267. Karppinen A, Kukkonen J, Elolähde T, Konttinen M, Koskentalo T (2000) A modelling system for predicting urban air pollution: comparison of model predictions with the data of an urban measurement network in Helsinki. Atmos. Environ. 34, 3735–3743. Kousa A, Kukkonen J, Karppinen A, Aarnio P, Koskentalo T (2002) A model for evaluating the population exposure to ambient air pollution in an urban area. Atmos. Environ. 36, 2109–2119. Marshall JD, Teoh SK, Nazaroff WW (2005) Intake fraction of nonreactive vehicle emissions in US urban areas. Atmos Environ 39(7), 1363–1371.
P3.2 Improving Emission Inventory in Lithuania Vidmantas Ulevicius, Vytautas Vebra, Kestutis Senuta and Svetlana Bycenkiene
Abstract It was developed emission inventory and estimation system for raw data about an area and point emission sources in Lithuania: collection, emission estimation, analysis, reporting and public information. Developed emission inventtory and estimation system was used for emission estimation, sectorial and spatial analysis in EMEP grid, reports to LRTAP and NEC preparation in the year 2005. Emission inventory and estimation system’s public information subsystem exposes an emission factors’ browser, a spatial emission maps’ browser and the road transport emission calculator on the website of Institute of Physics. Developed emission inventory and estimation system is described below. Five main subsystems can be distinguished in emission inventory and estimation system according to functionality: a database management system, an emission raw data collection system, an emission estimation and analysis system, an emission data export system and emission data public information system (Figure 1). We use MySQL the database management system for emission data storage, data querying and data analysis platform; we are going to migrate to PostgreSQL database management system with PostGIS extension as more featured geographic data analysis platform. The main tasks of emission raw data collection system are: (1) import emission related data to emission inventory database from various formats; (2) provide interactive forms for data entering. We used the phpMyAdmin, Microsoft Access and MySQL Query Browser for completing these tasks. The main tasks of emission estimation and analysis system are: (1) process raw emission related data and build emission inventory; (2) perform necessary emission inventory analysis and output the results of analysis in form of tables. We have a prepared set of SQL scripts which estimate road transport emission, build emission inventory from raw emission-related data, perform key sources analysis, calculate emission in EMEP grid, perform other sectorial and spatial analysis. The road transport estimation subsystem was developed according to COPERT III methodology (Ntziachristos et al., 2000) by creating the set of SQL queries – this increased emission inventory and estimation system’s effectiveness and performance. There was also developed an emission factors’ table in emission inventory database following the national, EMEP/CORINAIR and some other methodology used by other countries emission estimation. Main task of emission data export system is export emission data stored in emission inventory database for data exchange with air quality modelling systems or other systems. The emission data export system contains data format C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
691
692
V. Ulevicius et al.
and destination specific modules. Currently we have an executive module which exports emission inventory data to the MM5-SMOKE-CMAQ air quality modelling system. The main tasks of emission data public information system are: (1) visually represent emission data on the web site available for public; (2) expose official emission inventory data browser on the web site available for public; (3) expose web-based emission calculators for personal usage on the web site available for public. The emission data public information system is built on Apache with the PHP extension platform. Currently the emission data public information system contains PHP scripts which expose an emission map’s browser, emission factors’ browser and JavaScript based road transport emission calculator on the website of Institute of Physics: http://www.fi.lt/afch.
Fig. 1 Emission inventory and estimation system
Acknowledgments This research was supported by the Agency for International Science and Technology Development Programmes in Lithuania under EUREKA WEBAIR project. The authors gratefully thank for this assistance.
References Ntziachristos L, Samaras Z, ETC/AEM (2000) COPERT III Computer programme to calculate emissions from road transport. Methodology and emission factors (Version 2.1).
P1.4 Inter-Comparison of Gaussian Plume, Street Canyon and CFD Models for Predicting Ambient Concentrations of NOx and NO2 at Urban Road Junctions Richard Hill, Peter Jenkinson and Emma Lutman
Abstract Predictions of Gaussian Plume, Street Canyon and Computational Fluid Dynamics models were compared at five road traffic junctions. CFD may not necessarily reduce uncertainties in air quality assessments, though may be useful for identifying hotspot locations. Technical guidance for the assessment of local air quality under the UK National Air Quality Strategy (NAQS) has identified that previous studies of road-trafficrelated air pollution often may not have considered the effects of junctions adequately (LAQM-TG3, 2003). The influence of road traffic junctions in urban areas is particularly significant, as these areas often have enhanced rates of emission due to traffic congestion and rates of atmospheric dispersion are often reduced, due to the effects of the surrounding buildings on wind flows (Vardoulakis et al., 2003). The combination of these two effects may result in such areas being identified as pollution hotspots through monitoring studies. Predictions of Gaussian Plume (Airviro) and Street Canyon (Aeolius) models were compared with the predictions of a Computational Fluid Dynamics (CFD) model capable of including the complex topographies that occur at urban intersections (Panache). Five case study sites were identified in order to determine the influence of different modelling techniques on the prediction of air concentrations for Local Air Quality Management. The simulations conducted in this project identified that urban buildings affect the predictions of local NOx concentrations significantly. Where streets are flanked by tall buildings these effects have long been recognised and Street Canyon models are typically used for modelling such situations. The modelling assessments for Coventry and Leicester, areas that are similar to the geometries for which Street Canyon models were developed, showed that these models provide the most suitable tools for predicting air concentrations, with similar patterns of dispersion being predicted by the CFD model. Interestingly, realistic predictions were also obtained using the Aeolius model at the Sheffield and Leeds case study sites, areas that have very different topographies to those typical of a street canyon. Model predictions from the CFD code for the Birmingham, Leeds and Sheffield sites showed that particular air concentration hotspots occurred in the wakes of C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
657
R. Hill et al.
658
large buildings and at building faces along the roadside. CFD model runtimes for these simulations and the occasional significant overprediction of concentrations, suggest that the operational use of CFD for NAQS modelling may be prohibitively time-consuming (Table 1). Moreover, CFD simulations may not necessarily reduce the uncertainties in assessments of NO2, particularly when contributions from background sources, the accuracy of emission inventories and the conversion of NOx to NO2 are considered. However, it may be appropriate to consider the types of feature shown by the CFD model to result in local pollution hotspots when developing monitoring strategies in such areas.
Fig. 1 Comparison of model predictions and measurement data (Leeds) Table 1 Statistical comparison of observed background corrected NOx concentrations with those predicted by Airviro (AI), Aeolius (AE) and Panache (CFD). Location Birmingham Coventry Leeds Leicester Sheffield
Fraction within a factor of 2 AI AE CFD 0.50 – 0.63 0.00 1.00 0.63 0.13 0.25 0.25 0.13 0.63 0.38 0.63 0.50 0.13
Normalised mean square error AI AE CFD 1.33 – 1.16 22.95 0.19 0.26 3.02 1.36 1.96 6.21 0.66 16.55 0.97 1.04 11.56
Mean bias AI 0.44 0.04 0.40 0.15 0.52
AE – 0.69 0.59 0.66 0.53
CFD 0.49 0.64 1.07 8.03 7.38
Acknowledgments The authors are grateful to Birmingham, Coventry, Leeds, Leicester and Sheffield City Councils and Defra for funding this work.
References LAQM-TG3 (2003) Part IV of the Environment Act 1995, Technical Guidance. Department for Environment, Food and Rural Affair, London. Vardoulakis S, Fisher BEA, Pericleous K, Gonzalez-Flesca N (2003) Modelling air quality in street canyons: a review. Atmospheric Environment 37(2),155–182.
P2.11 Lake Breezes in Southern Ontario: Observations, Models and Impacts on Air Quality David Flagg, Jeff Brook, David Sills, Paul Makar, Peter Taylor, Geoff Harris, Robert McLaren and Patrick King
Abstract The southwestern Ontario, Canada Border Air Quality Strategy (BAQS-Met) field campaign of summer 2007 investigates the chemical and dynamical influence of lake breeze fronts and urban environments on local and transported pollutant emissions. The presence of both local and long-range transported emissions predisposes southwestern Ontario (ON), Canada to compromised air quality (AQ). Surrounded on three sides by the Great Lakes, frequent lake-breeze fronts (LBFs) complicate this region’s air chemistry and dynamics and generate a unique challenge for air quality modelling. LBFs also contribute to local convection, initiating thunderstorms and potentially enhancing vertical transport of pollutants. The local emissions derive from multiple industrial sites in both the U.S. and Canada, including the Detroit-Windsor metropolitan area. The latter can provoke an urban heat island (UHI) circulation that interacts with LBFs, further complicating the modelling. The Border Air Quality Strategy (BAQS-Met) field study in air quality and meteorology will address the need for improved understanding of the character of this border region (U.S.–Canada). 84°
82°
83° US
43°
122
81° LAKE
LAKE ST.
111
112
LAKE
1.2.5
1.2.7
1.2.6
1.2.4
LEGEND
42° 1.2.3
121
US
Existing surface i ARWIS weather i or U.S. station OME EC supersite met i University supersite
1 2 OFig. 1 BAQS-Met domain
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
679
680
D. Flagg et al.
BAQS-Met is a joint study of Environment Canada (EC), the Ontario Ministry of the Environment (OME), York University, the University of Toronto (UT), University of Western Ontario (UWO) and the U.S. National Weather Service (NWS). The field study consists of a four-week intensive campaign from mid-June through mid-July 2007 and summer-long enhanced meso network (mesonet) measurements from June through August. Both campaigns concentrate on southwestern ON, adjacent to the U.S. border (see Figure 1). BAQS-Met will assess the relative importance of the chemical and dynamical influences of the Great Lakes on regional air quality. Analysis of measurements of LBF-induced vertical motion will contribute to modelling of transport forecasts, air chemistry and thunderstorm initiation. BAQS-Met will also investigate model performance of physical processes over the lakes and its contribution to regional AQ forecasting. Additional studies include modelling the Detroit-Windsor urban boundary layer to examine the role of an urban, coastal environment in tracer transport as well as a review of how threedimensional variational data assimilation (3D-Var) can improve model performance in the boundary layer. The four-week intensive campaign includes an extensive array of measurements from both fixed and mobile surface sites (over land and water) and airborne measurements. So-called ‘supersites’, stations at Bear Creek and Harrow (see Figure 1) each host measurements of gas and particles in addition to surface meteorological variables, including (not all at each site): O3, SO2, CO, NO/NOx, VOC, NH3, Real NO, NO2, NOz, PM1, PM2.5 PM10 and black carbon, with many at up to 1-minute resolution. Deployment of Environment Canada’s Canadian Regional and Urban Investigation System for Environmental Research (CRUISER), based out of Windsor, and a commercial ferry operating on Lake Erie provide mobile surface gas and particle measurements. The Rapid Acquisition SCanning Aerosol Lidar (RASCAL), stationed at Ridgetown, provides lidar measurements. Both OME and UT provide additional mobile gas and particle measurements. Meteorological measurements include a mesonet with 14 stations measuring wind speed and direction, temperature, relative humidity, passive NO2/SO2, O3 and NH3 (in addition to other sites) and, at 10 of the 14 sites, active O3 and PM2.5 measurements. These supplement existing surface stations (EC, NWS, OME), Advanced Road Weather Information System (ARWIS) stations and buoys. Tethersondes at Ridgetown and Windsor and an ozonesonde and VHF wind profiler at Harrow provide vertical profiles, complementing transects from the Canada National Re-search Council (NRC) Twin Otter aircraft, which provides 30 hours of measurement. Flight paths include multi-level transects and spirals over Lake St. Clair, Detroit-Windsor and lake shores in daytime and nocturnal boundary layers as well as single level inter-lake transects. Acknowledgments Funding provided by the Ontario Ministry of the Environment Transboundary Research Program.
P2. Regional and intercontinental modelling
P2.1 Local to Regional Dilution and Transformation Processes of the Emissions from Road Transport Dimiter Syrakov, Kostadin Ganev, Reneta Dimitrova, Angelina Todorova, Maria Prodanova and Nikolai Miloshev
Abstract The objective of the present work is to study in detail the dilution processes and chemical transformations of the generated by road transport from the local scale to the scale of the global models and on deriving some conclusions about the key parameters, which quantify the local dilution and transformation processes impact on larger scale pollution characteristics. It is expected the further development of the current work to give some clues for specification of the “effective emission indices” linking emission inventories to the emissions to be used as input in large scale models. The US EPA Models-3 system (Grell et al., 1994; Byun et al., 1998; Byun and Ching, 1999) was chosen as a modelling tool. The simulations were consecutively carried out in three nested domains. The innermost domain (D3), treated with a resolution of 10 km includes a region with very intensive road transport – the city of London and its “footprint”. The CMAQ “Integrated Process Rate Analysis” utility is used to differentiate the contribution of different dynamic and chemical processes which form the pollution characteristics in the region of interest. Two sets of emission data were used in the present study: (1) the EMEP data was used for all the countries except the UK; (2) for the UK data from the National Atmospheric Emissions Inventory, with a 1 km resolution was used. The biogenic VOC emissions were estimated, using a simple scheme recommended by Lübkert and Schöp (1989). The meteorological background input was taken from US NCEP Global Analyses data. The simulations in D3 were carried out for January and August 2002–2006 for the following emission scenarios: (1) all the emissions (detailed inventory); (2) emissions from the road transport excluded (detailed inventory); (3) all the emissions, averaged over D3; (4) emissions averaged over D3, but emissions (averaged) from the road transport excluded. The combined analysis of these scenarios will make it possible (hopefully) to clarify the role of different dynamic and chemical processes which determine the pollution from road transport pattern and time evolution. Some C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
659
660
D. Syrakov et al.
conclusions about the role of the road traffic emission inventories spatial resolution on the simulated fields and the local to regional scale interaction can also be made. The numerical experiments performed produced a huge volume of information, which have to be carefully analysed and generalized so that some final conclusions, concerning not only clarification of local scale processes of dilution and chemical transformation but also how to account for them in large scale CTMs could be made. Comprehensive survey of the output from all the numerical experiments will be possible only if some integral quantities, characterising the dilution and transformation processes within D3 domain are introduced. The conclusions that can be made at this stage of the studies are: 1. The effect of the road transport emissions is well displayed in both the concentration and process analysis fields. 2. The contributions of different processes have very complex spatial/temporal behavior and variability. 3. Even horizontally/temporally averaged process contributions may be sensitive to emission resolution. Acknowledgments The present work is supported by EC through 6FP projects ACCENT (GOCE-CT-2002-500337) and QUANTIFY (GOGE-003893), and COST Action 728.
References Byun D, Young J, Gipson G, Godowitch J, Binkowski FS, Roselle S, Benjey B, Pleim J, Ching J, Novak J, Coats C, Odman T, Hanna A., Alapaty K, Mathur R, McHenry J, Shankar U, Fine S, Xiu A, Jang C (1998) Description of the Models-3 Community Multiscale Air Quality (CMAQ) Modeling System, 10th Joint Conference on the Applications of Air Pollution Meteorology with the A&WMA, 11–16 January 1998, Phoenix, Arizona. Byun D, Ching J (1999) Science Algorithms of the EPA Models-3 Community Multiscale Air Quality (CMAQ) Modeling System. EPA Report 600/R-99/030, Washington, DC. Grell GA, Dudhia J, Stauffer DR (1994) A Description of the Fifth Generation Penn State/NCAR Mesoscale Model (MM5). NCAR Technical Note, NCAR TN-398-STR, 138. Lübkert B, Schöp W (1989) A model to calculate natural VOC emissions from forests in Europe. Report WP-89-082, IIASA, Laxenburg, Austria.
P2.6 Modelling of Atmospheric Transport of POPs at the European Scale with a 3D Dynamical Model Polair3D-POP Solen Quéguiner and Luc Musson-Genon
Abstract POLAIR3D is an Eulerian 3D atmospheric model designed to handle a wide range of applications. Thus it has been used for passive transport, for impact at European scale, for photochemistry and mercury chemistry. A new version of the model, POLAIR3D-POP, developed to study the atmospheric transport and environmental fate of persistent organic pollutant (POPs) is presented. During the atmospheric transport, POPs can be deposited and re-volatilised to the atmosphere several times before the final destination. A description of the air-surface exchange processes is included in the model to account for this multi-hop transport. The greatest families of pollutants are studied (HAP, dioxins and furans, PCB, lindane and HCB). The results from a model simulation showing the atmospheric transport for the year 2001 at the European scale are presented and evaluated against measurements from EMEP. The Persistent Organic Pollutants (POPs) are organic carbon-based chemical substances. They possess a particular combination of physical and chemical properties such that, once released into the environment, they remain intact for exceptionally long periods of time. They become widely distributed throughout the environment as a result of natural processes involving soil, water and air. They are found at higher concentrations in the food chain and are toxic to both humans and wildlife. The aim of this study is to present a 3D dynamical model, POLAIR3D-POP, describing the atmospheric transport and environmental fate of POPs at the European scale. POLAIR3D-POP is handled basic physical processes by integrating in time the Eq. (1) (Mallet et al., 2007):
wci wt div(ci .V )
div( K .ci ) F i (c) Di Ei
(1)
where i labels a chemical species, c is a vector of chemical concentrations, V is the wind vector, K is the diffusion matrix, F i combines production and loss terms of chemical reactions, Di is the deposition term (dry deposition and wet depositionscavenging), Ei stands for the emissions (surface and volumic emissions).
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
669
S. Quéguiner, L. Genon
670
The change in the POP concentration, ca , in the atmospheric layers with time is described in the model by the Eq. (2):
wc a
wt
1
za
Femis Fexc Oca k air ca
(2)
where za is the thickness of the layer, Femis is the emission, Fexc is the air-surface gas exchange flux, O is the wet deposition and kair is the chemical transformation rate in the air. The soil module POLAIR3D-POP has based on that of the DEHM-POP (Hansen et al., 2004) and that of the MSCE-POP (Gusev et al., 2005). We have considered six types of land use coverage (barren soil, legume/fruit, evergreen forest, deciduous forest, grassland/cereals, and water bodies). The change in the POP concentration in different underlying surfaces, cs , with time is described by the Eq. (3):
wc s
wt
1
zs
F
wet
Fexc Frun _ through k soil c s
(3)
where Frun_through is the amount of chemical running out with the excess water through the bottom soil layer (modelled just for the barren soil). Meteorological data and emissions are used as input to the model simulations. The meteorological data comes from the ECMWF and the emission input are provided by the EMEP database (EMEP, www.msceast.org). We present the model outputs for the atmospheric concentrations, the deposition fluxes and the concentrations in the different types of underlying surface for the HAPs, dioxins and furans, PCBs, lindane and HCB. Measurements from 13 stations are available and the comparison to measurements is presented for the year 2001.
References Gusev A, Mantseva E, Shatalov V, Strukov B (2005) Regional Multicompartment Model MSCE-POP. EMEP/MSC-E, Technical Report 5/2005. Hansen KM, Christensen JH, Brandt J, Frohn LM, Geels C (2004) Modelling atmospheric transport of J-hexachlorocyclohexane in the Northern Hemisphere with a 3D dynamical model: DEHM-POP, Atmos. Chem. Phys., 4, 1125–1137. Mallet V, Quélo D, Sportisse B, Ahmed de Biasi M, Debry E, Korsakissok I, Wu L, Roustan Y, Sartelet K, Tombette M, Foudhil H (2007) Technical Note: The air quality modeling system Polyphemus. Atmos. Chem. Phys., 7, 5479–5487
P2.10 Modelling the Impact of Best Available Techniques for Industrial Emissions Control in Air Quality: Setting Up Inventories and Establishing Projections R. Rodriguez, P. Maceira, J.A. Souto, J. Casares, A. Sáez and M. Costoya
Abstract Strategies for industrial emissions control mainly depend mainly on the best available techniques, BATs, their economical feasibility, and their impact over air quality. A bottom-up comprehensive methodology for industrial emissions inventories and projections is presented and applied to the estimation of emissions for different scenarios in Galicia (NW Spain). Starting from a 2001 emissions scenario, the application of different BATs was considered, and emissions projections were obtained for 2010. Finally, SMOKE (CEP, 2003) was applied to integrate the emissions inventories in air quality modelling. From the results obtained, a strong SO2 emissions reduction will be achieved, and a slight reduction of NOx emissions in the utilities sector is expected. Changes (either increments or reductions) on other pollutants emissions are feasible. An emissions inventory is a strategic tool for environmental management. In this work, a bottom-up comprehensive methodology (based in Source Classification Codes; U.S. EPA, 2004) for establishing industrial emissions inventories, is presented. The proposed methodology is applied to the estimation of industrial emissions for different scenarios in Galicia (NW Spain) using SMOKE (CEP, 2003), which is also used for a subsequent air quality evaluation. The process for establishing emissions inventories and the projections derived from the scenarios proposed is divided in three stages: (a) data acquisition and classification of emissions calculation and estimation methods; (b) structure of the emissions inventory, for multiple applications; and (c) emissions projections, based both in industrial growth rates and technological changes. This methodology was tested in the industrial emissions inventory of Galicia (NW of Spain) (Casares et al., 2005) covering 370 major industrial plants, selected for their potential atmospheric emissions from around 3,000 installations. Annual emissions inventories for 1999, 2000 and 2001 years were obtained, covering as pollutants: SO2, NOx, CO, PM10, CO2, CH4, N2O, NMVOC, PAH, benzene, Cl-HCl, F-HF, NH3, PFCs (CF4 and C2F6) and heavy metals (As, Cd, Cr, Cu, Hg, Ni, Pb, Zn, and their compounds). Finally, the inventory for 2001 was
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
677
R. Rodriguez et al.
678
adopted as a reference, and different emissions projections for 2010 year were obtained. Figure 1 shows results of SO2 emissions by sector for 2001 and 2010 (projection); based on existing regulations (EU Directive on large coal combustion plants), and economic growth rates expected (Mantzos et al., 2003). 0,00 0,0 4600 ,0 00 ,0 00 00 0,00 0,0 3550 ,0 00 ,0 00
kg/y ear
4 0,00 0,0 00
3 0,00 0,0 00
20 01
20 10
2 0,00 0,0 00
1 0,00 0,0 00
Oth ers
Wa ste trea teme nt
Po we r pl ants
Co gen era tio n p la nts
D iv ers e ma ch in ery
Tran sp ort e qu ip men t
M etal s
Me tal pro du cts
Lea the r
C era mi c, co nc rete …
Pl as tic s an d ru bb ers
Pe troc he mi cal in du stry
Pri nti ng
Ch emi ca l i nd us try
Fu rni ture
Pu lp an d p ap er in du stry
Tex ti l fa cto rie s
Wo od pro ce ssi ng
Min in g
Fo od pro ces si ng
0
Fig. 1 SO2: 2001 emissions and 2010 emissions projection classified in 20 industrial sectors at Galicia
A slight reduction of NOx emissions is expected. Other pollutants’ emissions change in different directions (positive and negative impacts) depending on the industrial sector considered. Acknowledgments This work was financially supported by the R&D Spanish Programme (REN2002-02988/CLI) and R&D Galician Programme (PGIDT03PXIC20901PN).
References Carolina Environmental Program (CEP) (2003) SMOKE v.2.0 User Manual. The University of North Carolina at Chapel Hill, NC, USA. Casares JJ, Rodríguez R, Maceira P, Souto JA, Ramos S, Costoya M, Sáez A (2005) Inventory, analysis and projection of industrial air pollution emissions in Galicia. University of Santiago de Compostela, Spain (in Spanish). Mantzos L, Capros P, Kouvaritakis N, Zeka-Paschou M (2003) European Energy and Transport. Trends to 2030. European Commission. U.S. Environmental Protection Agency (2004) AP-42, Fifth Edition. EPA Publications, Washington, DC, USA.
P7.6 New Approaches on Prediction of Maximum Individual Exposure from Airborne Hazardous Releases John G. Bartzis, Athanasios Sfetsos and Spyros Andronopoulos
Abstract One of the key problems in coping with deliberate or accidental atmospheric releases is the ability to reliably predict the individual exposure during the event. Due to the stochastic nature of turbulence, the instantaneous wind field at the time of the release is practically unknown. Therefore for consequence assessment and countermeasures application, it is more realistic to rely on maximum expected dosage rather than actual one. Recently Bartzis et al. (2007), have inaugurated an approach relating maximum dosage as a function of the exposure time, concentration mean and variance and the turbulence integral time scale. Such approaches broaden the capability of the prediction models such as CFD models to estimate maximum individual exposure at any time interval. In the present work a further insight is given to this methodology and an alternative correlation is proposed based on theoretical considerations. The methodology to utilize such correlation types is further justified. Recently Bartzis et al. (2007) have inaugurated an approach relating the parameter
C max 'W C
to the fluctuating intensity I and the C max 'W C
§ 'W 1 1.5 I ¨¨ © TL
'W ratio as follows: TL · ¸¸ ¹
0. 3
(1)
It is reminded that the correlation (1) has been calibrated with 1s concentration signals representing the approximation of the instantaneous concentration statistical properties. It is suggested to be applied for such cases and exposure time duration 'W t 1s . Although such restrictions do not pose any serious problem for most atomspheric applications there is a need to remove such limitations and widen the applicability of those types of approaches. The theoretical background of the present approach can be summarized in the following notations: – The instantaneous concentration pdf can be approximated by a Gamma distribution – The approximation introduced by Venkatram (1979) for time average concentration variance is nearly valid C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
725
J.G. Bartzis et al.
726
The problem closure has been obtained by assessing the FLADIS T16 field stationary data (Bartzis, 2007) and applying best fit analysis. The new obtained correlation has as follows:
C max 'W C
1 4.57 I [(1 0.37
'W 1.45 )] TL
(2)
The FLADIS T17 Experiment is similar to T16 Experiment but with different meteorological conditions (Nielsen et al., 1994). Both correlations (1) and (2) are compared against the T17 experimental data. In Figures 1 and 2 the Factor of two (FAC2) and the Index of Agreement (IA) are plotted respectively. Both methods are able to predict maximum exposure within a factor of 2, but the new model is more accurate as expected on the basis of IA. 1
Index of Agreement
Factor of 2
105
100
0.99
0.98
0.97
95 New Model
Old Model
Fig. 1 FAC2 comparison all sensors
New Model
Old Model
Fig. 2 IA comparison all sensors
Conclusions (1) An new model to estimate maximum individual dosage has been derived applicable to any exposure time based on fluctuating intensity I and the ǻIJ / TL ratio. (2) The methodology inaugurated by Bartzis et al. (2007) has been further refined taking into consideration the theoretical pdf of the instantaneous concentration.
References Bartzis JG, Sfetsos A, Andronopoulos S (2007) On the individual exposure from airborne hazardous releases. The effect of atmospheric turbulence. Journal of Hazardous Materials, accepted for publication. Venkatram A (1979) The expected deviation of observed concentrations from predicted ensemble means, Atmospheric Environment, 13, 1547–1549. Nielsen M, Bengtsson R, Jones C, Nyren K, Ott S, Ride D (1994) Design of the FLADIS field experiments with dispersion of liquified ammonia, Risø–R– 755(EN), Risø National Laboratory, Roskilde, Denmark.
P2.9 Nonlinearity in Source-Receptor Relationship for Sulfur and Nitrate in East Asia Woo-Sub Roh, Seung-Bum Kim and Tae-Young Lee
Abstract Source-receptor (S-R) relationships for sulfur and nitrate are being actively sought in East Asia where great amounts of pollutants are being emitted into the air (e.g., Park et al., 2004). Nonlinearity in the advection and chemical processes is suggested to cause problems when model is used for the derivation of S-R relationships (Bartnicki, 1999). We have examined the effects of nonlinearity in the derivation of S-R relationship for sulfur and nitrate for East Asian region using a comprehensive model. Nonlinearity problem is investigated by examining the effects of varying emission rate in a source region on the deposition of pollutant in receptor regions as in the following steps (East Asia is divided into five source/receptor regions for this study): 1. Calculation of air quality, dry and wet deposition with full emission sources. 2. Same as step (1), except that emission rate of a particular species in a particular source region is reduced to 0%, 25%, 50% and 75% of full emission. Emission reduction is considered for SOx, NOx, NH3, VOC + CO in a separate manner. 3. Calculation of the difference in the deposition amount in a receptor region between full emission and reduced emission experiments. These calculations are carried out for March and July 2002. Air quality and pollutant deposition are calculated using an Eulerian, comprehensive acid deposition model (CADM) (Lee et al., 1998). Horizontal grid size is 60 km and vertical grids are stretched with a stretch ratio of 1.15. Meteorological fields are prepared using CSU RAMS with four-dimensional data assimilation (Pielke et al., 1992) (version 4.4). One-hourly fields are produced and then supplied to CADM. The emission rates of SOx, NOx, NH3, CO, and VOCs for China, South Korea and Japan are from the Long-range Transboundary Air Pollutants in Northeast Asia (LTP) project (Park et al., 2004). Air quality and pollutant deposition are calculated using CADM for March and July 2002. Calculated monthly total deposition amounts for July are shown in Figure 1. Dry deposition of sulfur reflects the distribution of emission. But dry deposition of nitrate shows a smoother pattern than that for sulfur. Wet deposition also shows significant difference between sulfur and nitrate. Wet deposition over Japan is significant for nitrate, while it is similar to back ground level for sulfur. Total deposition amount reflects more the distribution of wet deposition, due to
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
675
676
W. Roh et al.
large amount of precipitation over East Asia during July. Effects of nonlinearity are being investigated.
Fig. 1 Monthly total deposition amount for July for sulfur (left panels) and nitrate (right panels). Top and bottom panels are for dry and wet deposition amounts, respectively (unit: mg/m2)
Acknowledgments This study has been supported by the Global Environment Research Center/National Institute of Environmental Research through the project “Development of next-generation model for prediction of long-range transport of air pollutants in Northeast Asia.”
References Bartnicki J (1999) Computing source-receptor matrices with the EMEP Eulerian Acid Deposition Model. EMEP/MSC-W note 5/99. Park I-S, Kim J-C, Lee D-W (eds.) (2004) Annual report for the 4th year’s Joint Research on Long-range Transboundary Air Pollutants in Northeast Asia. Secretariat of Working Group for LTP project, 2004, NIER, Korea, 392 pp. Lee T-Y, Kim S-B, Lee S-M, Park S-U, Kim D-S, Shin H-C (1998) Numerical simulation of air quality and acid deposition for episodic cases in eastern Asia. Korean. Journal of Atmospheric Sciences 1(2), 126–144.
P6. Interactions between air quality and climate change
P6.1 On the Effective Indices for Emissions from Road Transport Kostadin Ganev, Dimiter Syrakov and Zahari Zlatev
Abstract The emissions from the road traffic are in a way different from the ship and airplane emissions: (i) the road network can be pretty dense in some cells of the large scale model grid; (ii) the emissions are continuous with time; (iii) the road traffic sources are close to earth’s surface. That is why the concept of deriving effective emission indices from the interaction of an instantaneous plume with the ambient air is perhaps not so convenient in the case of road transport emissions. On the other hand, the vertical turbulent transport is a very important process near earth’s surface, which means that it is relatively easy to parameterize the vertical structure of the pollution fields and so relegate the considerations to a twodimensional problem within a layer where the emissions heterogeneity can be important for the nonlinear chemical reactions. Due to the limited volume the vertical parameterization can not be discussed in details. It will be mentioned only that it is based on the heavy particles dry deposition parameterization in the surface layer, suggested by Ganev and Yordanov (2005) and for the case of N admixtures results in the following system of equations:
wci Lci Ai Bij c j J (ci ci ) Wij c j wt
Ei ,
i 1,..., N ,
(1)
where E i ( x, y, t ) accounts for large scale pollution source, ci ( x, y, t ) and ci ( x, y, t ) are the large scale pollution content and the respective concentration averaged in the layer, ci ( x, y, t ) is the concentration above the layer, ci ( x, y, t ) (h z 0 )ci ( x, y, t ) , Ai (c1 , c 2 ,..., c N ) – the term describing sources and sinks of the i th admixture, due to chemical transformations, {Bij } and {Wij } are diagonal matrixes describing the large scale absorption by earth’s surface and gravity deposition, J is a parameter describing the pollution exchange between the near surface layer and the upper atmosphere, L is the operator describing the horizontal transport. It is assumed that the mesoscale effects on large scale characteristics are small enough, which after some additional assumptions leads to the following formulation of the small disturbances problem in a domain D :
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
713
K. Ganev et al.
714
wGci LGc j D ij Gc j Bij Gc j JGci 0 , D ij wt Gci (t
0) W .(GE i GBij c j ) , Gci
1 wAi h z 0 wc j
(2)
0 at D boundaries,
i 1,..., N (3)
From mass-balance considerations in Eq. (2) and after additional simplifications it ml becomes clear that the pollution outflows/inflows from cell D of the large scale ml model (the mesoscale emission disturbances) ei for a time step W can be calculated by: eiml
1 0.5W (D ij Bij ) GC jml (W ) , GC i ml
³³ Gci ( x, y,W ) dD
i 1,..., N , (4)
D ml
which actually makes the calculation of effective emissions for each time step possible. The functionals GCi ml can be calculated when the problem (2, 3) is solved for the time period >0,W @ , but there is also another way to obtain them. As the problem (2, 3) is linear, the technique of functions of influence can be applied (Marchuk, 1982; Ganev, 2004) and GCi ml can be also expressed in the form:
GCi ml W ³³ c( i ) k ( x, y,0).GE k GBkj c j dD ml *
(5)
D
ml *
The advantage is that the solutions c( i ) k of the adjoin equations in this case can be factorized – one of the multipliers accounts for the chemical transformations and the other, which accounts for the horizontal transport can be analytically obtained in an explicit form. Acknowledgments The present work is supported by EC through 6FP projects ACCENT (GOCE-CT-2002-500337) and QUANTIFY (GOGE-003893), and COST Action 728.
References Ganev K (2004) Functions of influence and air pollution models sensitivity, Compt. Rend. Acad. Bulg. Sci., 57, 10, 23–28. Ganev Ʉ, Yordanov D (2005) Parameterization of dry deposition processes in the surface layer for admixtures with gravity deposition. Int. J. Environ. Pollut., 25, 1–4, 60–70. Marchuk GI (1982) Mathematical Modeling in Environmental Problems (in Russian), Nauka, Moscow.
P5. Aerosols in the atmosphere
P5.1 Quantifying Source Contribution to Ambient Particulate Matter in Austria with Chemical Mass Balance Receptor Modeling A. Caseiro, H. Bauer, I. Marr, C. Pio, H. Puxbaum and V. Simeonov
Abstract In this work, we apply the CMB model to a set of samples collected in Vienna. Those samples were chemically characterised for a wide range of chemical species as were samples from aerosol sources. The set of samples represent periods in which the threshold value of the PM10 level was exceeded, thus providing an insight over the causes of such episodes. Particulate matter concentrations at street level have been a raising concern for many years already, in particular due to its health effects (WHO, 2006). Such situation has been the background for ambient air European legislation regarding particulate matter (Council Directive 1999/30/EC). Yet, many European cities are not in agreement (EEA, 2006). A broad understanding of the identity, the sources and their intensity, the atomspheric interactions and the sinks of particulate matter are thus required. Chemical mass balance (CMB) is based on the principle of mass conservation, so that the mass of aerosol in a given location is a linear combination of the mass emitted by each source. The use of CMB requires the knowledge of the chemical composition of the aerosol at a given location and at a given time along with the chemical composition of the different sources that contribute to it. It is then possible, using a multi-linear regression, to calculate the composition of each source to the ambient aerosol (Gordon, 1988). Vienna is the capital city of Austria with about 1.8 million inhabitants. Four sampling sites were selected for this study: Schafberg, SCH (Residential area, north-west city fringe), Rinnböckstrasse, RIN (City centre, near city-highway), Kendlerstrasse, KEN (City centre), Lobau, LOB (Background area (Danube meadows) at the south-east of the city). Chemical species accounting for a wide range of chemical classes were measured using a suite of eight analytical techniques. The 12 aerosol sources accounted for in this study are road dust, diesel exhaust, gas combustion, biomass combustion, vegetative detritus, cooking, HULIS (humic-like substances, secondary organic aerosol), ammonium sulphate, ammonium nitrate, sea salt, break wear and tyre wear.
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
711
712
A. Caseiro et al.
CMB is based on the assumption that the mass collected on a filter is a linear combination of the mass emitted by the different sources and that the chemical build-up of the aerosol remains constant, both during the aerosol transportation and sampling. It states that one can identify the contribution of various classes of sources by measuring the concentrations of many chemical species in ambient air samples because their composition pattern is sufficiently different. Exceedences of the 50 µg/m3 PM10 value only occurred in the cold period. From those days, a set of eleven periods was chosen to model the source contributions. The difference between the total calculated mass and the measured mass was, in average, 18% for SCH, 15% for RIN and KEN, and 13% for LOB. Secondary inorganic aerosol was a general strong source in exceedence periods. In average terms, it is by far the predominant source of PM10 in exceedence episodes, being more than two times more important than the second most contributive source in the background sites (SCH and LOB), and about one and a half time more important in the city centre sites (RIN and KEN). Road dust was the second most important source of PM10 in exceedence days. Particularly, this source was very strong for the episodes Feb1, Mar2 and Apr1. These episodes were coincident with the period subsequent to the melting of the snow, when the road gravel material, Dolomite limestone, happens to be triturated by the road traffic. Biomass burning was also a major source. The main contributor to this class is residential wood burning, which occurs mainly in the colder periods, contributing to about 15–24% of the PM10. Diesel exhaust, secondary organic aerosol and vegetative detritus were the last important sources. The former was quite constant for all the exceedence episodes and the latter less present in city centre sites. Gas combustion, cooking, salts, break wear and tyre wear were not key contributors to PM10.
References Council Directive 1999/30/EC: Limit values for sulphur dioxide, nitrogen dioxide and oxides of nitrogen, particulate matter and lead in ambient air. EEA (2006) Air pollution at street level in European cities, EEA Technical report No 1/2006, European Environment Agency, ISBN 92-9167-815-5, ISSN 17252237, © EEA, Copenhagen 2006. Gordon GE (1988) Receptor models. Environmental Science and Technology, 22, 1132–1142. Health risks of particulate matter from long-range transboundary air pollution, Joint WHO/Convention Task Force on the Health Aspects of Air Pollution, E88189 © World Health Organization Regional Office for Europe, Copenhagen 2006.
P2.5 Regional Transport of Tropospheric Ozone: A Case Study in the Northwest Coast of Iberian Peninsula Santiago Saavedra, María R. Méndez, José A. Souto, José L. Bermúdez, Manuel Vellón and Miguel Costoya
Abstract Relevant tropospheric ozone levels are frequently reached in the NW coast of the Iberian Peninsula (Galicia) during spring and summertime under high pressure conditions (Logan, 1998). In this study, the origin and associated phenomena to tropospheric ozone episodes in rural areas at that region are considered. Most of them are produced by regional ozone transport from Southern and Eastern regions of the Iberian Peninsula. In addition, analysis and simulation of a typical episode (12–22 September 2003) are presented. Tropospheric ozone episodes in rural areas of Western Europe have been reported in the past (Logan, 1998). These phenomena appear in Galicia, an Atlantic region with complex topography and strong sea influence. Therefore, a systematic analysis of ozone episodes from 2002 to 2006 was done, considering both field measurements and modelling results. For a typical episode, mesoscale modelling with PSU/NCAR MM5 (Grell et al., 1995) was applied in order to get a better understanding about the origin of O3 peaks. Setting an ozone hourly ground level concentration (glc) threshold of 150 Pg/m3, (close to first legal threshold, 180 Pg/m3) 26 episodes were identified. Then, an analysis was done considering: (a) field measurements in the region, and its surroundings; and (b) EURAD meteorological and air quality modelling (Memmesheimer et al., 2001). Analysis of regional conditions during 12–22 September 2003 typical episode shows a synoptic pattern dominated by Central-Europe anticyclone. Weak SE circulation was caused by this synoptic situation in the NW of the Iberian Peninsula, changing to southerly flux on 15th. Then, a typical summertime low pressure gradient was established at Central Peninsula: at this point, maximum ozone glc is achieved, with hourly averages above 135 Pg/m3 (reaching up to 190 Pg/m3). Period from 15 to 17 September 2003 was selected to simulate meso-E meteorology and back-trajectories using MM5. Figure 1a shows back-trajectories, covering 12-hour backwards, and showing the transport of air masses from N of Portugal, as in EURAD operational forecast. Figure 1b shows a W-E wind profile over the region, where sea breeze regime (due to the weak pressure gradient coupled with warm land temperatures) causes recirculation of air masses from the coast to inland. C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
667
668
S. Saavedra et al.
Although in other cases (Alonso et al., 2000) aged pollutants layers from the coastal O3 can be created, in this episode recirculation contributes to the entrance of O3-polluted air from the South, from upper to lower levels, increasing the effect of O3 external contribution.
Fig. 1 MM5 simulations results: (a) 12-hour back-trajectories that reach the N of Galicia on 15/September/03 at 16 UTC. Odd and even indexes show 2 and 1,000-m height, respectively; (b) simulated flows showing a convective cell produced by NW sea breeze, opposite to SE synoptic wind, on 16/September/03 at 16 UTC
Acknowledgments This work was financially supported by Endesa Generación, S.A. and the R&D Spanish Programme (CTQ2006-15481/PPQ). Air quality data were provided by Environmental Departments of As Pontes Power Plant (Endesa company), Xunta de Galicia and Junta de Castilla-León (Spain), and Ministerio do Ambiente of Portugal. EURAD operational forecasts and technical supports by MeteoGalicia and CESGA are acknowledged.
References Alonso L, Gangoiti G, Navazo M, Millán M, Mantilla E (2000) Transport of tropospheric ozone over the bay of Biscay and the eastern coast of Spain, Journal of Applied Meteorology, 39 (4), 475–486. Grell GA, Dudhia J, Stauffer DR (1995) A description of the Fifth-Generation Penn State/NCAR Mesoscale Model (MM5), NCAR/TN-398 + STR, Boulder, USA. Logan JA (1998) Trends in the vertical distribution of ozone: an analysis of ozonesonde data, Journal of Geophysical Research, 99 (D12), 25553–25586. Memmesheimer M, Jakobs HJ, Piekorz G, Ebel A, Kerschgens MJ, Friese E, Feldmann H, Geiß H (2001) Air quality modeling with the EURAD model. In proceedings of the 7th International Conference on Harmonisation within Atmospheric Dispersion Modelling for Regulatory Purposes, Belgirate.
P2.15 Saharan Dust over Italy: Simulations with Regional Air Quality Model BOLCHEM Mihaela Mircea, Massimo D’Isidoro, Alberto Maurizi, Francesco Tampieri, Maria Cristina Facchini, Stefano Decesari and Sandro Fuzzi
Abstract Saharan dust is transported over Mediterranean area, dust reaching often different regions of Italy. To the scope of predicting the advection of dust and its physical and chemical properties over Italy, a dust emission scheme has been implemented in the air quality model BOLCHEM, which solves simultaneously the chemical and meteorological equations. This study demonstrates the ability of BOLCHEM to predict the dust events over Italy and evaluates the impact of differrent parameterizations used in the dust production. The dust aerosols, besides of changing climate through the scattering and absorption of solar and thermal radiation, also affect the environment by fertilizing marine and terrestrial ecosystems that in turn influence the carbon cycle. Moreover, the dust particles contribute substantially to the total aerosol mass usually employed in the developing of the environmental policy regulations, therefore, a reliable forecast of dust events is mandatory over Italy, often affected by Saharan dust transport. This study describes the dust model implemented in the air quality model BOLCHEM and shows its ability to forecast dust events over Italy. The study also investigates the dependency of dust production on threshold friction velocity and number of dust size bins. The air quality model BOLCHEM (D’Isidoro et al., 2005; Mircea et al., 2007) comprises a meteorological model, an algorithm for airborne transport and diffusion of pollutants and two photochemical mechanisms. The meteorology is coupled online with the chemistry. The dust model implemented in BOLCHEM was developped by Tegen et al. (2002) and is based on the soil-derived dust emission scheme designed by Marticorena and Bergametti (1995). The horizontal and vertical dust fluxes are calculated based on the location of the preferential dust sources, soil texture, surface roughness, vegetation cover, soil moisture content and surface wind velocity. The ratio between the vertical and the horizontal dust fluxes varies with the type of soil and the size of the particle mobilized. The size distribution of the mobilized dust depends on both the surface properties (soil texture) and the surface wind speed. The threshold friction velocities used to initiate the dust emissions are computed as a function of particle size following Marticorena and Bergametti (1995), but assuming constant roughness within the model grid cells (0.001 cm). Moreover, the simulation shown here was carried out with a threshold friction velocity lowered by a factor of 0.75 since lower thresholds velocities improve model results compared to observations at global level. The comparison of the dust C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
687
688
M. Mircea et al.
event occurred on 16 July 2003, simulated by BOLCHEM and seen by the AQUA/MODIS satellite/sensor, shows that the model is able to predict well both the extent and the timing of the dust event over Italy. In both images, it can be noted that the plume of dust over the Mediterranean comes from north-west and north of Africa and goes straightforward to the center and north of Italy with only a little veil over Sicily and Messina Strait. These results substantiate that the model uses reliable surface land/soil information, meteorological conditions and transport scheme.
Fig. 1 Saharan Dust over Italy: July 16, 2003 at 13 UTC. Model simulation (left) and satellite image from Aqua/Modis (right)
Dust emissions calculated with the box dust model shows that the production of dust depends more on threshold friction velocity than on number of dust size bins. The effect of threshold friction velocity is similar for accumulation and coarse mode while the number of size bins impacts only on accumulation mode. These results will be further used in the calibration of the dust concentrations calculated by the model. Acknowledgments This work was conducted in the frame of ACCENT and GEMS EC projects, Italian MIUR project AEROCLOUDS, and were also supported by the Italian Ministry of Environment through the Program Italy-USA Cooperation on Science and Technology of Climate Change.
References D’Isidoro M, Fuzzi S, Maurizi A, Monforti F, Mircea M, Tampieri F, Zanini G, Villani MG (2005) Development and Preliminary Results of a Limited Area Atmosphere-Chemistry Model: BOLCHEM, First ACCENT Symposium, Urbino 12–16 September 2005. Marticorena B, Bergametti G (1995) Modeling the atmospheric dust cycle: 1. Design of a soil-derived dust emission scheme, J. Geophys. Res., 16415–16430. Mircea M, d’Isidoro M, Maurizi A, Vitali L, Monforti F, Zanini G, Tampieri F, (2007) A comprehensive performance evaluation of the air quality model BOLCHEM over Italy, submitted to Atmos. Environ. Tegen I, Harrison SP, Kohfeld K, Colin Prentice I, Coe M, Heinmann M (2002) Impact of vegetation and preferential source areas on global dust aerosol: results from a model study, J. Geophys. Res., 107, D21, doi:10.1029/2001JD000963.
P1.2 Simplified Models for Integrated Air Quality Management in Urban Areas B. Sivertsen, A. Dudek and C. Guerreiro
Abstract The Norwegian Institute for Air Research, NILU has been requested by the World Bank to support the Hanoi Urban Transport and Development Project (HUTDP) with the Urban Air Quality Management Subcomponent primarily executed by Hanoi’s Department of Natural Resources, Environment and Housing (DONREH). As part of the evaluation NILU has performed air quality modeling in order to assess the importance of air pollution from mobile sources in Hanoi. For this purpose, NILU has combined two models; the NILU developed air quality modelling planning system AirQUIS and the Simple Interactive Model for Better Air Quality (SIM-Air) developed by the World Bank. AirQUIS was used to simulate the computation of an emission inventory for key pollutant and estimate the impact of the sources on air quality. SIM-Air will be used to assess and evaluate health effect impact, and allowed for various policies measures, economic and technical options to be evaluated for their environmental and health impacts and cost effectiveness. This model uses simple mathematical and financial tools to evaluate different air quality management options. All the management options are linked to cost and health impacts based on percentage change in the considered option.
1. Introduction NILU has developed the AirQUIS GIS based air quality management and dissemination system to perform integrated assessment and planning for improving air quality (http://www.nilu.no/airquis/). A comprehensive management system such as AirQUIS requires large specific datasets. There is therefore a need to develop simple interactive decision support tools to assist local authorities to assist local authorities to carry out screening processes and take appropriate decisions and actions for air quality management, especially in developing countries where available data and resources are limited.
C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
653
654
B. Sivertsen et al.
2. System Approach The concept developed by NILU can be summarised in three stages: estimation of top-down emission inventory, calculation of ambient concentrations using dispersion modelling and evaluation of health effects, management options and related costs. For estimating emissions for a specific area, NILU uses a similar integrated approach as in the Simple Interactive Model for Better Air Quality (SIM-AIR) presented by the World Bank (Shah and Saikawa, 2005). Where detailed emission inventory is not available, estimates of emissions are performed as a top down approach. In NILU’s approach, the emissions are estimated for a defined gridded domain, with user defined resolution, covering the area being studied. Input data for the modelling system include: (1) population data, (2) meteorological data, (3) emission data, (4) emission factors for source categories, (5) dose response functions, and (6) cost estimates.
3. Results and Discussion The models generated concentration distributions of NOx, PM10 and SO2 over the city of Hanoi (Sivertsen and Dudek, 2006). As a first estimate/surrogate for the total population exposure the estimated concentrations have been used together with the population distributions to estimate the person-weighted concentrations. The relative contribution from each of the vehicle categories has been estimated for NOx, PM10 and SO2. NOx exposure due to traffic emissions is caused by truck emissions (36%), motorcycles (22%), petrol driven cars (20%), diesel cars (12%) and about 8% due to buses. A similar estimate for the PM10 contributions indicated that the total population exposure in Hanoi is due in 23% to traffic sources; 15% to industrial sources and 62% to other undetermined sources (including “background”). This simple integrated decision support tool meets the requirement of a first screening of air pollution problems in defined areas and requires less computing power than existing advanced air quality management tools including complex dispersion models. Simple decision support systems of this kind may also help stakeholders air quality management planning or action plans and easier assessment of different options.
References Shah J, Saikawa E (2005) Interactive Database for Emission Analyses (IDEAHanoi) Version 1. (Developed in the East Asia Region of the World Bank.) Sivertsen B, Dudek A (2006) Support for the Review of Air Quality Management Sub-component for the Hanoi Urban Transport and Development Project. Modelling air pollution in Hanoi. Kjeller (NILU OR 83/2006)
P7.4 Source Apportionment of Particulate Matter in the U.S. and Associations with In Vitro and In Vivo Lung Inflammatory Markers Rachelle M. Duvall, Gary A. Norris, Janet M. Burke, John K. McGee, M. Ian Gilmour and Robert B. Devlin
Abstract Associations are well established between particulate matter (PM) and increased human mortality and morbidity. Fine particulate matter (particle diameter < 2.5 Pm) is most strongly linked to adverse health impacts. The toxicity of PM may depend on the PM source and composition which will vary by location. While a number of epidemiological studies have shown that certain PM sources are associated with specific health outcomes, the underlying mechanisms are still unclear. To investigate these mechanisms, continuous weekly PM2.5 samples were collected for four consecutive weeks (24 hours a day for seven days) in six cities across the U.S. as part of the Multiple Air Pollutant Study (MAPS). Sample composites were constructed for each site and particles were extracted in water. Samples were analyzed for trace metals (via Inductively Coupled Plasma – Optical Emission Spectroscopy), ions (via Ion Chromatography), and elemental carbon (via thermal methods). Sources contributing to the PM2.5 samples were identified using the EPA Chemical Mass Balance (CMB8.2) model. Both in vitro and in vivo experiments were conducted to measure a variety of toxicological outcomes. For the in vitro analysis, PM extracts were applied to cultured human lung epithelial cells and the production of different lung inflammation/injury markers (Table 1) was measured by real-time reverse transcriptase polymerase chain reaction (RT-PCR). For the in vivo analysis, particle extracts were instilled into mouse lungs at different doses (25 and 100 Pg). Indicators of lung injury and inflammation (Table 1) were measured in bronchoalveolar lavage fluid and plasma by enzyme-linked immunosorbent assay (ELISA). The relationship between the toxicological measures and PM2.5 sources was evaluated using linear regression. A few of these plots are displayed in Figure 1. For the in vitro health markers, mobile sources and secondary sulfate (from coal combustion) were related to increased IL-8 production (r2 = 0.39 and r2 = 0.79, respectively). Combustion sources and soil were associated with increases in COX2 (r2 = 0.38 and r2 = 0.48), and secondary sulfate was associated with increased HO1 (r2 = 0.51). For the in vivo health markers, wood combustion was associated with increased MIP-2 production (r2 = 0.95), whereas mobile sources were associated with increased IL1-E and TNF-D (r2 = 0.94 and r2 = 0.99, respectively). These findings confirm that PM2.5 sources are associated with specific health outcomes. C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
721
R.M. Duvall et al.
722 Table 1 In vitro and in vivo health markers analyzed. In vitro markers Interleukin-8 (IL-8) Cycolooxygenase-2 (COX-2) Heme oxygenase-1 (HO-1)
In vivo markers Macrophage inhibitory protein (MP-2) Interleukin 1-ȕ (IL1- ȕ) Tumer necrosis factor alpha (TNF-Į)
Legend: Ŷ Salt Lake City Ÿ Seattle x Phoenix ż Sterling Forest Ɣ South Bronx * Hunter College
(Sterling Forest data removed )
5.00
y = 2.77x + 0.35 R2 = 0.79
0.00 0.00
1.00
30.00
HO-1
IL-8
10.00
2.00
3.00
4.00
20.00 y = 7.58x + 0.46 R2 = 0.51
10.00 0.00 0.00
3
Secondary Sulfate (P g/m )
1.00
2.00
3.00
4.00
Secondary Sulfate (Pg/m3)
800.0
500.0
400.0
y = 329x + 255 R2 = 0.94
200.0 0.0 0.00
TNF-D
IL1- E
600.0
0.50
1.00
1.50
2.00
250.0
0.0 0.00
Mobile Sources (Pg/m 3)
y = 185x + 161 R2 = 0.99 0.50
1.00
1.50
2.00
Mobile Sources (Pg/m 3)
Fig. 1 PM sources compared to in vivo and in vitro health markers
Disclaimer Although this work was reviewed by EPA and approved for publication, it may not necessarily reflect official Agency policy.
P2.4 SPECIATE – EPA’s Database of Speciated Emission Profiles J. David Mobley, Lee L. Beck, Golam Sarwar, Adam Reff and Marc Houyoux
Abstract SPECIATE is the U.S. Environmental Protection Agency’s (EPA) repository of total organic compound (TOC) and particulate matter (PM) speciation profiles for emissions from air pollution sources. The profiles are key inputs to air quality modeling and source-receptor modeling applications. This paper addresses Version 4.0 of the SPECIATE Database. The SPECIATE Database is an important EPA product which serves as the repository for source category-specific emission speciation profiles. The profiles contain weight fractions of species of both volatile organic compounds (VOC) and Particulate Matter (PM). The weight fractions of VOC species are grouped into reactivity classes to support air quality modeling for ozone. The profiles of PM species weight fractions are specific to particle size ranges and are being used to support air quality modeling for PM and visibility. The Database has also supported air toxic assessments and is essential for source-receptor modeling applications. The Database was first computerized in 1988. Although accessibility to the Database has been sustained through the Clearing House for Inventories and Emission Factors (CHIEF) website, updates to SPECIATE have languished since the mid-1990s due to decreasing budgets. The US National Research Council in its report on Research Priorities for Airborne Particulate Matter (NRC, 2004), the Clean Air Act Advisory Committee in its report of the Air Quality Management Working Group (CAAAC, 2005), NARSTO in its Emission Inventory Assessment (NARSTO, 2005), and other groups have recommended that the Database be extensively updated and maintained in a dynamic manner. Given the importance of SPECIATE to the process of air quality management, a team was organized to undertake an update of the Database. The scope of the team’s project was to: (1) update the Database with profiles from the literature and EPA source test data sets; (2) link the new profiles to Source Classification Codes (SCCs) in the National Emissions Inventory (NEI); (3) assign any new species to reactivity classes; and (4) update the air quality models to use the new information. The final report, “SPECIATE 4.0 – Speciation Database Development Documentation (US EPA, 2006)” summarizes the development and provides guidance on use of the Database. The Database is posted on the CHIEF Website. The final version of the Database has been integrated into the Emissions Modeling Platform for subsequent research and regulatory modeling applications. The ability to speciate the emissions inventory with the new SPECIATE composite profiles will C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
665
666
J. Mobley et al.
bring substantial benefits to the fields of air quality modeling and source apportionment Results from these improved analyses will enable the development of more effective control strategies for sources of these species. Further, estimates of uncertainty in the results of air quality and receptor models will be improved by providing the most representative and up-to-date information to characterize emissions from the myriad point, area, mobile, and biogenic sources that contribute to ambient pollutant concentrations. The initiative to update SPECIATE has produced: x 2,856 PM profiles, 1,258 of which are new profiles x 1,215 gas profiles, 648 of which are new profiles x 1902 unique species, 1012 of which are new species
SPECIATE 4.0 represents a significant enhancement of the data available to characterize emissions by species and source category. Air quality modeling and source-receptor modeling applications have benefited from using these enhanced speciation profiles. Additional efforts are needed to capture new data from current testing based on data submitted via the protocol for database expansion. The user community can support the Database development by supplying electronic data with full references. Acknowledgments The authors acknowledge Ying Hsu, Randy Strait, and Frank Divita of E.H. Pechan & Associates, Inc. for their support to the SPECIATE project.
References NRC (2004) Research Priorities for Airborne Particulate Matter: IV. Continuing Research Progress, National Research Council, National Academies Press, Washington, DC. CAAAC (2005) Recommendations to the Clean Air Act Advisory Committee, Air Quality Management Working Group, January 2005. NARSTO (2005) Improving Emission Inventories for Effective Air Quality Management Across North America – A NARSTO Assessment. NARSTO-05001, September 2005. US EPA (2006) SPECIATE 4.0 – Speciation Database Development Documentation. EPA/600/R-06/161, US Environmental Protection Agency, Research Triangle Park, NC, November 2006.
P7.7 The Detroit Exposure and Aerosol Research Study Ron Williams, Alan Vette, Janet Burke, Gary Norris, Karen Wesson, Madeleine Strum, Tyler Fox, Rachelle Duvall and Timothy Watkins
Abstract The Detroit Exposure and Aerosol Research Study (DEARS) was designed to assess the impacts of local industrial and mobile sources on human exposures to air pollutants in and around Detroit, Michigan. Daily integrated measurements were made of personal exposure, and residential indoor and outdoor concentrations in six neighborhoods throughout the Detroit area. Concurrent data were collected for comparison at a central community ambient monitoring location and a regional background site. These data collected in DEARS can be used to evaluate local air quality and explore the application of air quality models to assess human exposure in an urban area.
Keywords Air quality, human exposure, modeling, particulate matter, toxic air pollutants, Detroit The Detroit Exposure and Aerosol Research Study (DEARS) will provide needed information on defining the factors that impact individual exposures to various sources of particulate matter (PM) and toxic air pollutants and will contribute to the scientific information that is needed to inform decisions on standards to protect air quality. The study included three years of field data measurements (summer 2004 through winter 2007). Approximately 120 adult participants living in detached single-family residences were randomly selected and enrolled from six neighborhood census areas. These neighborhoods were selected because they each represented a variation of potential industrial and regional source influence, housing stock, and proximity to automotive emission sources. Selected participants and/or their residences were involved in five days of summertime monitoring and five days of wintertime monitoring each year. Data collected included various PM size fractions and select pollutant concentrations such as volatile organic compounds, carbonyls, metals, and criteria pollutant gases (see Table 1). Data from the DEARS will be used to conduct analyses to improve the understanding of relationships between sources, air quality, and human exposures and to evaluate the uncertainty of using community-based monitoring as surrogates for true human exposures. Air quality modeling is an integral part of DEARS. Both the regional-scale, photochemical model, CMAQ, and the local-scale, dispersion model, AERMOD will be used. The modeling domain for CMAQ will be centered on the Detroit area and will utilize a 12 km horizontal grid resolution, while the AERMOD receptor domain will be much smaller, spanning an area of 36 by 48 km, with receptors C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
727
R. Williams et al.
728 Table 1 DEARS measurement parameters. Parameter PM2.5 (mass, elements) PMcoarse (mass, elements) EC-OC (PM2.5) EC (PM2.5) Nitrate Gases (O3, NO2, SO2) Aldehydes VOCs SVOCs PAHs Air exchange rate
Personal X – – X – X X X – – –
Indoor X X X X X – X X X X X
Outdoor X X X X X X (NO2 only) X X X X –
Ambient X X X X X X X X X X –
placed at 1 km intervals. The predicted concentrations from CMAQ and AERMOD will be combined where appropriate, using a one-atmosphere “hybrid” approach (Isakov et al., 2007). This hybrid approach allows the preservation of the granular nature of the dispersion model while properly treating the chemistry and transport offered by the photochemical model. In addition, the most recent version of CONCEPT (Consolidated Community Emissions Processing Tool) will be used to produce link-based mobile emissions for PM and air toxics to provide a more refined allocation of mobile emissions for this project and to improve the ability to analyze the local-scale impact of mobile emissions on the urban air quality. Figure 3 shows the conceptual approach for DEARS air quality modeling.
Fig. 3 Conceputal Approach for DEARS Air Quality Modeling
Disclaimer and Acknowledgments Although this work was reviewed by EPA and approved for publication, it may not necessarily reflect official Agency policy. The research was funded and conducted through contract 4D- 5653-NAEX (University of Michigan), 68-D-00-012 (RTI International), and 68-D-00-206 (Alion Corp).
References Isakov V, Irwin J, Ching J (2007) Using CMAQ for exposure modeling and the importance of sub-grid variability for exposure estimates. Journal of Applied Meteorology and Climatology, Bostan, MA, 46(9): 1354–1371, (2007).
P2.3 The Role of Sea-Salt Emissions in Air Quality Models Raúl Arasa, Maria R. Soler and Sara Ortega
Abstract In this study, we show our investigations about the effect on atmospheric aerosol concentrations caused by sea-salt emissions generated in open oceans and in surf zones. We use the CMAQ/MM5/MECA air quality modelling system, taking and do not taking into account sea-salt emissions in order to find that they produce several differences in particulate concentrations.
1. Introduction Atmospheric particles, in particular sea-salt ones, play an important role in climate and atmospheric chemistry. Several studies show the importance of sea-salt emissions coming from both open oceans and the surf zone due to their interactions with other atmospheric species. The aim of this study is to investigate the effect of both emission sources (open ocean and surf zones) on atmospheric particles concentration in Catalonia, located in the northeast part of Spain. In order to achieve this goal, we simulated a summer 2003 period using: an emission model created by the authors known as MECA, MM5 meteorological model, and CMAQ 4.5.1 photochemical model with its sea-salt module AERO4 enabled. Different simulations were performed taking and not taking into account sea-salt emissions. Preliminary results indicate that the addition of sea-salt module alters particle concentrations causing relative changes in total PM10 and PM2.5 concentrations.
2. Modelling System and Set-up PSU/NCAR mesoscale model MM5 v3.7 was used to generate meteorological fields, the inputs of the air pollution modelling system. Meteorological simulations were performed for four two-way nested domains with 27, 9 and 3 km resolutions. The coarsest domain covers Spain, part of France and part of Italy. An inner domain of 30 x 30 cells (9 km) covers Catalonia, while two other 3 km resolution domains (the smallest ones) cover two areas whose interest lie in their high pollution level measurements. Characteristics of these simulations are described in Soler et al. (2007). The chemical transport model used in this study is the U.S. EPA models-3/CMAQ v4.5.1. In order to investigate the sea-salt effect on atmospheric C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
663
664
R. Arasa et al.
particles concentration, we did two simulations: with and without the sea-salt module AERO4 enabled. MECA, developed by the authors (Ortega et al., 2006), was the emission model used. This model was applied over domains number two, three and four; while for the biggest domain the emission inventory was quantified by the top-down approach using EMEP emissions. These emissions include the most important primary air pollutants from vegetation, on-road traffic, industries, fossil fuel consumption, and domestic-commercial solvent effects.
3. Results and Discussion The simulated period, 10–14 June 2003, was characterized by an anticyclonic situation favouring the development of mesoscale winds such as diurnal and nocturnal sea breezes. AERO4 effect on modelled PM2.5 and PM10 concentrations was evaluated by calculating the concentration relative differences between enabling and disabling the AERO4 module. In PM2.5 case, results suggest that the inclusion of sea-salt emissions decrease PM2.5 concentrations during all daily periods over sea areas. This reduction reaches values up to 60%. On the other hand, in this particular case we do not detect any clear tendency during nighttime over land and shorelines. The relative changes, positive or negative, are in any case comparatively small: between –20% and 20%. During day time there is some tendency to PM2.5 concentrations increase. Eventually, the increment could be high but located over inland in small delimited points. In PM10 case, the most important effect controlling the concentration relative differences is the wind speed. Wind velocities higher than 4 ms-1 increase sea-salt concentrations, especially on coast areas where the sea breeze could be intense and the emissions coming from the surf zone are remarkable. This PM10 result was not unexpected because the parameterization used to take into account sea-salt emissions depends on particle size and increases exponentially as a wind velocity function. Acknowledgments This project was funded by the Spain Government through CGL2006-12474-C03-02 grant.
References Ortega S, Alarcón M, Soler MR, Pino D, Grasa J (2006) Cálculo de emisiones relevantes en la modelización fotoquímica mesoscalar, Medioambiente en Iberoamerica: visión desde la Física y la Química en los albores del siglo XXI 1, 171–178. Soler MR, Bravo M, Ortega S (2007) The use of meteorological and dispersion models in stratified boundary layers, Develop. Environ. Sci. 6, 199–208.
P4.7 The Use of MM5-CMAQ-EMIMO Modelling System (OPANA V4) for Air Quality Impact Assessment: Applications for Combined Cycle Power Plants and Refineries (Spain) R. San José, J.L. Pérez, J.L. Morant and R.M. González
Abstract Since 2000 the EPA Models-3 Community Multiscale Air Quality Modelling System (CMAQ) has become in one of the state-of-the-art air quality tools to perform a full analysis of the air concentrations in a determined domain in space and time. The CMAQ modelling tool incorporates several chemical mechanisms, several numerical solvers, several boundary layer parameterizations, etc. and the user has to select the best option according to their experience and knowledge to perform any air quality simulations in historical and/or forecasting mode. The MM5 meteorological mesoscale model and/or the new generation of mesoscale meteorological models based on WRF model can be used as input for CMAQ. Additionally, our laboratory has developed along the last ten years a sophisticated emission model which provides with the corresponding spatial and temporal resolution, the emission data required by CMAQ to perform the full simulations. This system – MM5-CMAQ-EMIMO (OPANA V4) – can be used to forecast air concentrations in space and time or to simulate different periods of time in the past. In order to perform an air quality impact assessment of an industrial plant such as a combined cycle power plant or a refinery, we should run our air quality modelling system for a period of time in the past according to the EU Air Quality Directives (and also US EPA Regulations for non-EU applications). The EU Directives related to the limits in air concentrations are concerned with a number of exceedances of a specific concentration along one year. There are limits for 8-hour, days and year. In this contribution we show the methodology to be used to fulfil the EU Directives for different applications – combined cycle power plants and refineries – in Spain and some results. The use of the system for one-year period to adapt the results to the actual legislation is presented.
Keywords Industrial plants, air quality, forecasting In this contribution we show the methodology used for applying the MM5-CMAQEMIMO air quality modeling system for performing air quality impact assessment according to the EU legislation. We will show the results for different applications for combined cycle power plants and incinerators in Spain. The calibration and validation process in all the studies is an essential step to carry out the fuell air C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
707
708
R.S. José et al.
quality impact assessment. Figure 1 shows one year ozone data obtained in different air quality monitoring stations in Madrid Community during 2005 and averaged and compared with ozone modeled data obtained in every grid cell where the monitoring stations are found. This plot is made with 365 days × 24 hours = 8,760 data for the full year 2005.
Fig. 1 Correlation coefficient between one year modeled ozone data (MM5-CMAQ-EMIMO) and measured data averaged over several stations in Madrid Community Air Quality Monitoring network
The correlation coefficient is 0.796 and it represents the accuracy of the results of the study. The methodology used is based on the so-called ON-OFF approach which means that a full MM5-CMAQ-EMIMO modeling system is run over three different domains with 405 × 405 km, 88 × 105 km and 24 × 24 km with 9, 3 and 1 km spatial resolution for the full year under the so-called OFF scenario which includes all biogenic and anthropogenic emissions present during 2005 in the area of the study and a full run of a so-called ON scenario which is exactly the same than OFF scenario but adding the maximum expected emissions of the proposed combined cycle power plant or incinerator. The differences ON-OFF represent the impact of the expected future industrial plant – this procedure is valid for any industrial plant not only for power plants or incinerators. The percentiles present in the EU Legislation for the different pollutants are calculated based in the full one year run. The quality of the simulations are based on the correlation coefficient and several other statistical tools currently available for this analysis.
P4. Model assesment and verification
P4.1 Tropospheric Ozone and Biogenic Emissions in the Czech Republic K. Zemankova and J. Brechler
Abstract Terms of formation, amount, spatial distribution of tropospheric ozone and the contribution of biogenic sources of VOC to ozone precursors were studied using numerical model for summer photochemical smog simulation (SMOG model). Biogenic emissions of VOC were estimated using semi-empirical model proposed by Guenther et al. (1995). North-eastern part of the Czech Republic (Hruby Jesenik area) was selected as a model domain and comparison of two different model runs with measured data from this area is presented. Only antropogenic emissions of VOC were taken into account in the first run of the model whereas biogenic emissions of VOC were added into model inputs in the second run.
Tropospheric ozone is a significant element of atmospheric pollution, especially in the summer period when photochemical reactions, which lead to ozone formation, are remarkably enhanced by intensive sunlight and high temperatures. Main ozone precursors such as nitrogen oxides (NOx) and volatile organic compounds (VOC) are released to the atmosphere not only from antropogenic sources, but they are emitted in considerable amounts from biogenic sources as well. Lagrangian puff model SMOG was used for evaluation of a difference in tropospheric ozone concentration while its formation has been simulated with and without the contribution of VOC from natural sources. Short decription of the SMOG model can be found in Section 2.1, for further details see Bednar et al. (2001). Emissions of biogenic VOC were estimated on the basis of semi-empirical model suggested by Guenther et al. (1995) and its principal ideas are described in Section 2.2. Model results from both model runs are compared with measured data in Section 3. SMOG model is a chemical transport model developed at the Department of Meteorology and Environment Protection, Charles University in Prague. It is a lagrangian puff model where a continuous plume of pollution is divided into several separate puffs preserving the original intensity of emission fluxes from individual sources. Thanks to this approximation SMOG model is able to model non-stationary situations under varying meteorological conditions. Each puff has its own trajectory according to meteorological preprocessor – model ETA. A dispersion of puffs into all three dimensions with the normal distribution as well as chemical interaction of individual puffs with each other is expected. C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
693
K. Zemankova, J. Brechler
694
Isoprene and monoterpens are considered to be the most prominent VOC released from natural sources (Simpson et al., 1995). Emission flux F (ȝg m-2 h-1) of each chemical compound is according to Guenther et al. (1995) calculated as:
F
(1)
H D J
where H is an ecosystem dependent emission factor (ȝg C m-2 h-1 at photosynthetically active radiation (PAR) flux of 1,000 ȝmol m-2 s-1 and leaf temperature of 303.15 K), D is foliar density (kg dry matter m-2) and Ȗ is a dimensionless correction factor which accounts for the influence of PAR and leaf temperature in case of isoprene and for the influence of leaf temperature only in case of monoterpene. Comparison of daily mean concentrations of tropospheric ozone calculated by the SMOG model for June 2000 (dashed line- with antropogenic emissions of VOC only, dotted line- with inclusion of biogenic emissions) with measured data (solid line) are shown below. Two model grid points were selected to correspond with monitoring stations Cervenohorske sedlo and Jesenik. Cervenohorske sedlo
Jesenik
concentration /Pg.m-3
160
220
140
120
170
100
120 80
70 60
20
40 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 30
days
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 30
days
Acknowledgments Authors would like to acknowledge Czech Hydrometeorological Institute, Prague and Ekotoxa, Opava who for providing measured values and land cover data.
References Bednar J, Brechler J, Halenka T, Kopacek J (2001) Modeling of summer photochemical smog in the Prague Region. Phys. Chem. Earth (B), 26, 129–136. Guenther A, Hewitt N, Erickson D, Fall R, Geron Ch, Graedel T, Harley P, Klinger L, Lerdau M, McKay WA, Pierce T, Scholes B, Steinbrecher R, Tallamraju R, Taylor J, Zimmerman P (1995) Global model of natural organic compound emissions. J. Geophys. Res., 100, 8873–8892. Simpson D, Guenther A, Hewit CN, Steinbrecher R (1995) Biogenic emissions in Europe. 1. Estimates and uncertainties. J. Geophys. Res.-Atmos., 100(D11), 22875–22890.
P4.8 Verification of Ship Plumes Modelling and Their Impacts on Air Quality and Climate Change in QUANTIFY EC 6FP Project Tomas Halenka, Peter Huszar and Michal Belda
Abstract The impact of emission from transportation on climate change is being quantified in EC FP6 Integrated Project QUANTIFY. In Activity 2 the analysis of the dilution and transformation of the emission from microscale at exhausts and plumes till mesoscale distribution will be provided from all modes of transportation. In this contribution the mesoscale simulations of ship emission impact on atomspheric pollution are studied with emphasis to compare the simulation with reality analyzed by means of flight measurement during the field campaign. In framework of the project the modeling studies are supposed to support the field campaign as well. The sensitivity of the impact on air quality and composition is analyzed as well with respect to ship emissions. Here the couple of non-hydrostatic model MM5 (PSU/NCAR) and Eulerian model CAMx (ENVIRON International Corporation, 2006) is used to support the measurement campaign. This couple with double nesting enables very high resolution both in meteorological conditions and chemistry in the region of interest, with outer domain of resolution 36 × 36 km, inner one with resolution 12 × 12 km covering the Channel. Meteorological fields generated by MM5 drive CAMx transport and dry/wet deposition. There are problems with the emission inventories available, emissions from EMEP 50 × 50 km database are interpolated and represent average ship emissions in the Channel, other emissions are combination of EMEP and UAEI (United Kingdom Atmospheric Emission Inventory). In our setting CB-IV chemistry mechanism is used (Gery et al., 1989). To see the impact of ships emission on the chemical composition at the surface we present here sensitivity test on outer domain simulation. In Figure1 ship corridors are well visible in ozone concentration field and the simulations clearly identify impact of ships emission on the chemical composition at the surface. The performance of the couple was tested first on pre-project campaign data (provided by H. Schlager). This pre-project campaign off-line test shows reasonable comparison between simulation and flight measurement. There are results for O3 displayed in Figure 2, where very good agreement can be seen in ship corridor low level flight, out of it the disagreement is due to the limited extent of CAMx coverage, it covers rather boundary layer processes. Another limitation is valid as the model is not working with actual emission data of individual ships but involving some average emission in model grid. Thus, resulting concentrations represent rather the average value across the grid and the comparison of ship corridor C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
709
710
T. Halenka et al.
background in the flight measurement is more appropriate. Individual peaks of NO concentration from the individual ship stacks captured by flight measurement cannot be resolved by the model (not shown), better comparison is provided when more complex chemical processes on longer time scale undergone.
Fig. 1 Pre-campaign simulation of chemical fields (ozone concentration, ppm). Left – simulation with ship emission included, right – no ship emission
Fig. 2 Comparison of pre-project campaign simulation to flight measurement for O3
Acknowledgments This work is supported in framework of EC FP6 Integrated project QUANTIFY (GOCE 003893) as well as under local support of the grant of Programme Informacni spolecnost, No. 1ET400300414 and Research Plan of MSMT under No. MSM 0021620860.
References ENVIRON International Corporation (2006) CAMx Users’ Guide, version 4.40 Gery MW, Whitten GZ, Killus JP, Dodge MC (1989) A Photochemical kinetics mechanism for urban and regional scale computer modeling. J. Geophys. Res., 94, 925–956.
P7.2 What Activity-Based Analysis and Personal Sampling Can Do for Assessments of Exposure to Air Pollutants? Doina Olaru and Jennifer Powell
Abstract This paper gives an example of the benefits of personal sampling (PS) and activity-based analysis (AA) for exposure assessment. Air quality and exposure studies traditionally assume the same exposure for people living in the same area and neglect individual mobility within the urban space and the the time spent indoors. This results in underestimation of true personal exposure. Combining PS with AA overcomes this limitation, as it tracks individuals through their daily routines and regards exposure at the contact point with pollutant. Two small longitudinal studies confirm significant differences in exposure profiles, despite similar activity spaces of the subjects. Substantial epidemiological evidence suggests that fixed monitoring stations measurements cannot assess the population exposure for several reasons: – Spatial and temporal resolution are too rough to identify peak exposures, more relevant from a health perspective (Michaels and Kleinman, 2000). – The real population exposure occurs as a result of conducting various activities in spatially sparse locations, and the fixed station measurement does not allow this detailed spatial analysis (Violante et al., 2006). – Population spends most of their time (more than 80%) indoors (Myers and Maynard, 2005). As exposure represents the bridge between air quality and human health risks in health studies (Curtis et al., 2006), it important to pinpoint places (microenvironments) with higher pollution concentration and how individuals spend their time there. The best way to estimate personal exposure is to consider the microenvironmental activity patterns and personal behaviour (Kotzias, 2005; Sørensen et al., 2003; Han and Naeher, 2006). This paper shows the value of spatial and temporal variation accounts in activity patterns and exposure concentration for exposure assessment. It reports on two GIS-based longitudinal studies conducted in Melbourne (2004 – 18 weeks winter and summer) and Perth (2006 – five weeks winter) for NO2 exposure of three female subjects, members of the same family (daughter, mother, and grandmother), living in the same house. Samplers were deployed in seven fixed locations (home – bedroom/bathroom/living, kitchen; work/school; in the car) and they were also worn by the subjects on a badge. The location of their daily activities and the routes were geocoded for spatial analysis and mapping of activity spaces and exposure C. Borrego and A.I. Miranda (eds.), Air Pollution Modeling and Its Application XIX. © Springer Science + Business Media B.V. 2008
717
718
D. Olaru, J. Powell
profiles. The methodology allowed us to gain considerable insights into the magnitude and variation of exposure, showing that assessment of exposure based on ambient concentrations is 20–30% lower than the personal exposure. This is consistent with previous findings presented by Violante et al. (2006) and Kaura et al. (2005). Indoor activities are significant contributors to exposure to NO2 and exposure concentration during walking had similar levels to those recorded in the car. The intra-individual variation of personal exposure depended on the day of the week and on the season, being higher on winter and on weekend days. The activity spaces of the subjects are much alike, as the young girl and the two women carry out activities close to their homes. However, the three female subjects are at different life-cycle stages, which influence their activity routines. This aspect, superimposed on the spatial variation in pollution found within the activity spaces, lead to variable exposure profiles. Even higher discrepancies arise when considering the intensity of activity. The two longitudinal case studies demonstrate how activity-based methodology combined with personal sampling is able to capture the spatial, temporal, and behavioural variation of exposure, identifying individuals potentially at greater risk, not only because of the concentrations of pollutants, but also due to the exposure durations and individual susceptibility to adverse effects. Acknowledgments The authors thank their colleagues Kate Boast and Paul Selleck from CSIRO – Marine and Atmospheric Research for analysing the samplers.
References Curtis L, Rea W, Smith-Willis P, Fenyves E, Pan Y (2006) Adverse health effects of outdoor air pollutants, Environment International 32, 815–830. Han X, Naeher LP (2006) A review of traffic-related air pollution exposure assessment studies in the developing world, Environment International 32, 106– 120. Kaura S, Nieuwenhuijsenb MJ, Colvile RN (2005) Pedestrian exposure to air pollution along a major road in Central London, UK, Atmospheric Environment 39, 7307–7320. Kotzias D (2005) Indoor air and human exposure assessment – needs and approaches, Experimental and Toxicologic Pathology 57, 5–7. Michaels RA, Kleinman MT (2000) Incidence and apparent health significance of brief airborne particle excursions, Aerosol Science and Technology 32, 92–105. Myers I, Maynard RL (2005) Polluted air – outdoors and indoors, Occupational Medicine 55, 432–438. Sørensen M, Autrup H, Møller P, Hertel O, Jensen SS, Vinzents P, Knudsen LE, Loft S (2003) Linking exposure to environmental pollutants with biological effects, Mutation Research 544, 255–271. Violante FS, Barbieri A, Curti S, Sanguinetti G, Graziosi F, Mattioli S (2006) Urban atmospheric pollution: personal exposure versus fixed monitoring station measurements, Chemosphere 64, 1722–1729.
Author Index A Aarnio P., 634 Ainslie B., 145, 146, 147, 148, 151, 591, 601 Akimoto H., 109, 136, 139, 699 Aksoyo÷lu S., 101 Alaviippola B., 634 Albergel A., 28 Albizuri A., 673 Alexandersen S., 200, 204 Alfarra M., 101 Alonso L., 673, 675 Alpert P., 360, 364 Andronopoulos S., 726 Anfossi D., 28, 82 Anlauf K., 163, 541 Arasa R., 665 Arnold D., 663 Astitha M., 507, 508 Astrup P., 200 Aulinger A., 298 B Baklanov A., 3, 4, 5, 6, 7, 10, 186, 201, 204, 640 Baldasano J., 54 Bartnicki J., 685 Bartzis J., 726, 727 Batchvarova E., 18, 22, 23 Bauer H., 712 Beck L., 667 Bedogni M., 387, 434 Belda M., 579, 710 Belfiore G., 28 Bergan T., 685 Bermúdez J., 669 Bewersdorff I., 298 Blas M., 675 Blot R., 516 Borrego C., 27, 191, 463 Bouchet V., 165, 436, 437, 472, 541, 542 Bowers J., 63 Božnar M., 697
Brandt J., 570 Brasseur O., 706 Brauer M., 591, 596, 597 Brechler J., 653, 695 Brook J., 163, 681 Builtjes P., 191, 280, 289, 528 Bullock O., 173 Burke J., 642, 722, 728 Buzzelli M., 591 Bycenkiene S., 693 C Carmichael G., 483, 485 Carnevale C., 428, 429, 431, 716 Carvalho A., 191, 198, 369 Casadei S., 387 Casares J., 679 Caseiro A., 712 Castelli S., 28, 81, 82 Chai T., 483, 487, 488 Chang J., 63 Chaxel E., 46, 47, 48 Chemel C., 46, 50 Cho S., 163, 170 Chollet J., 46, 47, 48 Christensen J., 570, 571 Christensen K., 200 Cimorelli A., 618, 625 Constantinescu E., 483, 491, 495 Costoya M., 669, 679 Cousineau S., 436, 541 Crevier L., 472 D D’Isidoro M., 423 Daescu D., 483 Davidson P., 227 Decesari S., 689 Delcloo A., 706 Denby B., 280, 283 Dennis R., 330, 351 Deutsch F., 254, 255, 454, 550 Devlin R., 722 Dharmavaram S., 445 Dimitrova R., 661 729
Author Index
730
D'Isidoro M., 422, 689, 703 Dore A., 127, 128, 133 Draxler R., 227, 228 Dudek A., 655, 656 Duhamel A., 436 Dumont G., 454 Durana N., 675 Duvall R., 722, 728 E Elbern H., 289, 295, 493 Elolähde T., 634 Emeis S., 724 F Facchini M., 689 Fierens F., 454 Finzi G., 428 Fisher B., 378, 488 Flagg D., 681 Forkel R., 724 Fox T., 728 Frohn L., 570, 571 Fuka V., 653 Fuzzi S., 689 G Gadgil A., 265 Galperin M., 701 Ganci F., 28 Ganev K., 661, 714 Gangoiti G., 673, 675 Ganor E., 360, 364 García J., 673, 675 Garcia V., 341 Geels C., 570 Gégo E., 341, 396, 416, 418 Genikhovich E., 182, 183, 187, 701 Genon L., 671 George B., 642 Gilliam R., 236, 237, 567 Gilliland A., 324, 325, 396, 414, 417, 418, 561 Gilmour I., 722 Godowitch J., 414, 418 Goldberg R., 607 Gonçalves M., 54 Gong S., 163, 436, 476
Gong W., 74, 163, 165, 229, 436, 437, 476, 477, 508, 541, 542 González R., 37, 708 Gracheva I., 182 Grašiþ B., 697 Griesser E., 724 Grsic Z., 691 Gryning S., 18, 19, 20, 22 Guerreiro C., 655 Guerrero P., 54 H Haakenstad H., 685 Halenka T., 579, 710 Hanna S., 63, 445 Hansen K., 562, 570 Harris G., 681 Hayden K., 541 He Z., 101 Hedegaard G., 570, 571, 572, 574, 576 Henderson S., 591, 595, 596, 597, 601 Hill R., 659 Hinneburg D., 90 Hogrefe C., 341, 396, 401, 403, 414, 417, 561, 607, 610, 614 Horálek J., 280, 281 Horii N., 136 Houyoux M., 608, 667 Hurley P., 30, 209, 211, 212, 213, 214, 215 Huszar P., 579, 710 I Ilardia J., 675 In H., 118 Irwin J., 396 Isakov V., 616, 618, 619, 642, 647, 729 Iza J., 675 J Janssen L., 254, 593 Janssen S., 454 Jantunen M., 720 Jenkinson P., 659
Author Index
731
K
M
Kaasik M., 333, 334, 532, 536, 538 Kallaur A., 472 Kallos G., 507, 508 Kaminski J., 6, 145, 149, 473 Kangas L., 720 Karppinen A., 307, 634, 636, 720 Kasibhatla P., 396 Katsafados P., 507 Kauhaniemi M., 634, 637 Keller J., 101 Kerschbaumer A., 72 Khajehnajafi S., 445 Kim S., 677 Kim Y., 118, 430, 653 King P., 681 Kinney P., 607 Kishcha P., 360, 361, 362, 364, 365 Kitada T., 244, 246 Knowlton K., 607 Kondragunta S., 227 Kordova L., 360 Korsholm U., 3, 6, 10, 11, 12 Koskentalo T., 634 Koslan K., 445 Kousa A., 634, 635, 636 Kryza M., 127 Ku J., 396 Kukkonen J., 154, 307, 333, 532, 634, 720 Kulmala M., 532, 533 Kurata G., 244 Kurokawa J., 136
Macdonald A., 541 Maceira P., 679 Makar P., 163, 165, 170, 436, 437, 476, 477, 541, 542, 681 Marin R., 324 Marr I., 712 Marshall J., 591, 720 Martilli A., 47, 145 Martins A., 657 Martins V., 191 Mast M., 724 Mathur R., 227, 232, 236, 498, 500 Matthias V., 298, 299 Maurizi A., 422, 689, 703 Mavromatidis E., 507 McConnell J., 145 McGee J., 722 McLaren R., 681 Meagher J., 227 Ménard S., 472, 541 Méndez M., 669 Meng F., 687 Mensink C., 254, 454, 550 Menut L., 369, 371, 487 Mihele C., 163 Mikkelsen T., 200, 201, 206 Millán M., 673 Miloshev N., 661 Miranda A., 191, 193 Mircea M., 422, 423, 689, 703 Mlakar P., 697 Mobley J., 218, 667 Monforti F., 422, 703 Monteiro A., 191, 194, 463, 465, 470 Moran M., 163, 170, 436, 437, 438, 440, 441, 442, 472, 541, 543 Morant J., 37, 708 Mortensen S., 200 Moussafir J., 28, 29
L Larson T., 591, 598, 601 Leaitch R., 541 Lee K., 118, 120, 229 Lee T., 677 Leithead A., 541 Li S., 163 Liggio J., 163, 170 Linkosalo T., 154, 156 Luecken D., 625, 626 Lutman E., 659 Lutz M., 72 Lynn B., 607
N Nagashima T., 136 Napelenok S., 324, 325, 327 Narayan J., 436 Navazo M., 673, 675 Neary L., 145
Author Index
732
Nethery E., 591 Nickovic S., 360, 361, 508 Nicolau J., 369 Nielsen S., 32, 200, 727 Niwano M., 109, 114, 699 Nollet V., 406 Nolte C., 561 Norris G., 722, 728 O Ohara T., 136, 137 Olaru D., 718 Ortega S., 665, 666 Ortega X., 663 Otte T., 236, 499 Özkaynak H., 616 P Pabla B., 541 Pace T., 218 Pavlovic R., 436 Pedersen T., 200 Pérez J., 37, 361, 708 Piazzola J., 516, 517 Pierce T., 218, 476, 499 Pinder R., 324, 417, 551 Pio C., 712 Pirovano G., 387 Piscitello E., 683 Pisoni E., 428, 716 Pleim J., 236, 237, 352 Porter P., 341, 396 Pouliot G., 218, 219, 498 Powell J., 718 Prank M., 333, 532 Prévôt A., 101 Princevac M., 315 Prodanova M., 661 Puxbaum H., 712 Q Qian W., 315, 581 Quante M., 298 Quéguiner S., 671 R Rajkovic B., 691 Ranta H., 154, 157
Rao S., 341, 344, 396, 414, 611 Reff A., 667 Reisin T., 81 Renner E., 90, 524 Riikonen K., 634 Rodriguez R., 679 Roh W., 677 Roselle S., 352, 498, 499 Rosenzweig C., 607 Røsting B., 685 Roy B., 218, 219, 222 Russo M., 657 Ruuskanen T., 340, 532, 534 S Saadi J., 218 Saavedra S., 669 Sáez A., 679 Salbu B., 685 Saltbones J, 685 Saltbones J., 701 Samaali M., 436, 442, 541 San José R., 37, 38, 708 Sandu A., 483, 487, 491, 495 Santos R., 657, 658 Sarwar G., 351, 352, 498, 667 Sassi M., 163, 436, 477, 541 Sauntry D., 163, 541 Schaap M., 191, 194, 195, 280, 282, 289, 290, 526, 551 Schäfer K., 724 Schayes G., 182, 183, 187 Schere K., 119, 173, 227, 325, 342, 352, 416, 499, 563, 608, 617, 626 Schröder W., 90 Segers A., 280, 289 Seibert P., 663 Senuta K., 693 Sfetsos A., 726 Sghirlanzoni G., 387, 388, 390 Siddans R., 289, 291 Siljamo P., 154, 307 Sills D., 681 Simeonov V., 712 Sistla G., 396 Sivertsen B., 655, 656 Sloan J., 101, 687 Soares J., 634, 720
Author Index
733
Sofiev M., 154, 155, 182, 186, 307, 308, 309, 313, 333, 334, 532, 534, 701 Sofyan A., 244, 245, 246, 247, 251 Sohn M., 265, 266, 267, 270, 272, 273 Soja A., 218, 221 Soler M., 665 Sørensen J., 200, 201, 203, 204, 206, 718 Souto J., 669, 679 Sreedharan P., 265, 273, 276 Stahl C., 625 Stendel M., 570, 571, 572 Stern R., 72 Steyn D., 145, 146, 147, 148, 151, 152, 591 Strapp J., 541 Strawbridge K., 163, 170 Stroud C., 163, 170, 436, 477, 541 Strum M., 728 Su J., 591, 598, 601 Sugata S., 136 Suppan P., 724 Sutton M., 127, 130 Sykes I., 64, 445, 447 Syrakov D., 661, 714 Szykman J., 218
Tjemkes S., 289 Todorova A., 661 Tomé M., 657 Tong D., 625, 629 Trozzi C., 683
T
W
Takahashi M., 109, 699 Takigawa M., 109, 699 Talbot D., 472 Tampieri F., 422, 689, 703 Tang S., 127, 128, 130 Tang Y., 483, 485 Tanimoto H., 136, 137 Taylor P., 398, 681 Tchepel O., 463 Tedeshi G., 516 Terrenoire E., 406, 407 Teshiba M., 109 Theobald M., 127, 134 Timmermans R., 289, 291 Tinarelli G., 28, 29, 81, 82, 697
Watkins T., 642, 728 Wayland R., 227 Wesson K., 728 White J., 63 Wiens B., 163 Williams R., 39, 528, 642, 728 Witlox H., 445, 447, 451 Wolke R., 90, 524 Wong D., 236
U Ulevicius V., 693 Uno I., 136, 137, 139 V Valdenebro V., 673 Vankerkom J., 254 Vargas A., 663 Vautard R., 289, 295, 369, 370, 389, 407 Vebra V., 693 Vellón M., 669 Venkatram A., 315, 316 Vergeiner J., 724 Vette A., 642, 728 Vieno M., 127, 128 Villa S., 683 Vitali L., 422, 703 Vogel B., 6, 351, 353, 354 Volta M., 422, 428, 716 Vujadinovic M., 691
Y Yamaji K., 136, 138, 141 Yang R., 687 Young J., 236
734
Z Zanini G., 422, 423, 703 Zanoni A., 387, 390 Zemankova K., 695 Zhan T., 315 Zhang B., 687
Author Index
Zhang J., 102, 163, 445, 476, 541, 545, 594 Zheng Q., 436 Zlatev Z., 714
Subject Index A AERMOD, 251, 253, 355, 360, 656, 658, 663, 687, 769 aerosol composition, 114, 137, 139 aerosol concentration, 155, 160, 268, 371, 372, 556, 560, 563 aerosol concentrations, 139 aerosol feedbacks, 39, 46 air quality, 40, 42, 44, 49 Air Quality Forecast, 264, 270, 512 AirQUIS, 696 analytical solution, 218, 220 AOD, 327, 335 AOT, 154–160, 330 Asian dust, 52, 155 atmospheric transport model, 163, 169, 170, 578 B BEIS3, 519, 539, 544, 608, 648 C
CMAQ, 51, 73, 80, 90, 91, 154, 155, 172, 187, 209, 265, 336, 362, 363, 424, 477, 537, 538, 539, 540, 601, 639, 646, 663, 666, 687, 703, 729, 736, 749, 769 coastal area, 176, 288, 452, 556 Complex Geometries, 695 Complex Terrain, 739 CORINAIR, 92, 292, 471, 567, 570, 735, 758 critical load, 421 D Data assimilation, 301, 318, 325, 336, 345, 353, 362, 371, 387, 527, 733 Dispersion Modelling, 71, 88, 676, 712, 739 DMS, 44 DREAM, 400 E
CAFE, 291, 294, 297, 328, 470 Calgrid, 108, 110, 725 Calmet, 725 CAMx, 81, 137–138, 143, 427, 428, 430, 433–435, 461, 547, 548, 550 CBL, 249 CFD model, 64, 70, 73, 81, 694, 701 chemical transport model, 42, 110 Chemical Transport Model, 39, 46 Chernobyl accident, 47, 705 Chimere,449, 747, 748 Chlorine Chemistry, 491 City-Delta, 334 Clean Air Act, 256, 648, 651, 654, 709, 710 Climate Change, 51, 466, 554, 601, 608, 732, 746
elemental carbon, 108, 112, 470, 477, 637, 763 EMEP, 81, 92, 128, 163, 232, 292, 325, 339, 371, 372, 410, 422, 435, 447, 471, 475, 566, 570, 611, 622, 703, 708, 713, 720, 735, 747, 751 emission reduction, 74, 80, 92, 114, 138, 189, 428, 468, 471, 473, 615, 645, 668, 673, 757, 758 ensemble Kalman filter, 318, 327, 523, 537 EURAD, 711 F Fluent, 699 frame, 118, 123, 427, 468, 470, 474, 732, 765 735
Subject Index
736
G
O
Great Lakes, 216, 276, 729 grid resolution, 88, 97, 167, 265, 273, 275, 372, 413, 620, 666, 769
one-way nesting, 74, 219, 231 organic carbon, 137, 173, 232, 269, 422, 461, 713 Ozone concentrations, 90, 379, 465, 612
H heterogeneous chemistry, 46, 231, 477, 516, 589 HIRLAM, 39, 191, 238, 346, 372, 374, 727 I Iberian Peninsula, 91, 236, 404, 711 Inverse modelling, 369 L Lagrangian Particle Model, 64 Long-Term Simulation, 172 low wind speed, 59, 64, 141 M Mediterranean region, 400 mercury deposition, 209 mercury emissions, 536 microscale urban flow, 124 mineral dust, 402, 407, 554 Minerve, 739 mixing height, 63, 640, 641, 705, 765, 766 MM5, 73, 99, 137, 155, 229, 231, 272, 283, 289, 337, 343, 392, 397, 429, 458, 505, 513, 602, 607, 618, 621, 647, 660, 666, 672, 704, 711, 725, 736, 765 MODIS/TERRA, 154 multi-objective analysis, 757 MUSCAT, 43, 52, 126, 564
P PAH, 337, 721 particle formation, 128, 131, 293, 550, 572 particulate sulfate, 525, 547 personal exposure, 632, 643, 686, 759, 769 photochemical simulation, 379 plume dispersion, 123, 132 point sources, 168, 200, 248, 292, 373, 515, 539, 582, 621, 634, 639, 640, 648, 725 PREV’AIR, 410 public health, 380, 631, 633, 647, 651, 656, 682 R RAMS, 117, 173, 179, 557, 621, 715 receptor modelling, 435 regional climate scenarios, 602 remote sensing data, 39 S Saharan Dust, 132, 400, 731 sensitivity analysis, 162, 369, 594, 615 SILAM, 191, 218, 345, 371, 572 SKIRON, 408, 548, 550 SORGAM, 110 SPM, 281 surface roughness, 171
N
T
nested grid, 225, 659 nitrate concentration, 594
TKE, 120, 146
Subject Index
U urban area, 55, 76, 83, 124, 199, 394, 427, 666, 674, 681, 718, 762, 769 urban flow, 124 urban heat island, 82, 723
737