Flood Warning, Forecasting and Emergency Response

Flood Warning, Forecasting and Emergency Response

Kevin Sene

Flood Warning, Forecasting and Emergency Response

Kevin Sene United Kingdom

ISBN 978-3-540-77852-3

e-ISBN 978-3-540-77853-0

Library of Congress Control Number: 2008927074 © 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. Printed on acid-free paper 9 8 7 6 5 4 3 2 1 springer.com

Preface

This book provides an introduction to recent developments in the area of flood warning, forecasting and emergency response. The topic spans a wide range of disciplines, including weather forecasting, meteorological, river and coastal detection systems, river and coastal flood forecasting models, flood warning dissemination systems, and emergency response procedures. The text deals mainly with general principles and concepts, but also includes references to a number of manuals, guidelines and papers which provide more detailed information on factors to consider in designing and implementing a flood warning system. Although informal flood warning systems have existed ever since people settled near to rivers and coastlines, improvements to communication and computer systems in recent years have opened up a range of possibilities in many aspects of the flood warning process. These include developments in remote sensing techniques, ensemble forecasting, automated flood warning systems and decision support systems. Some recent research and operational developments in these areas are discussed, although specific brands of equipment (software, instrumentation etc.) are not considered. The topics of performance monitoring, risk based design and prioritisation of investment are also considered in several chapters, with recent developments driven in part by rising public expectations, and by an increasing need for organisations to justify investments in new equipment and procedures. Early warning systems are often described in terms of the detection, warning dissemination, response, recovery and review stages. In many cases, a forecasting component will also be included, and preparedness is essential for an effective emergency response. This structure is also adopted here, although with only a short discussion of the recovery phase, since flood warning and forecasting has a less important role to play once flood levels start to recede, such as estimating when floodwaters will drain, or if any further flooding is imminent. By contrast, the warning aspect is discussed in several locations, including a chapter on the decision criteria used for issuing flood warnings (often called thresholds) and sections on decision support and decision-making under uncertainty. The book is presented in three main sections as follows: ●

Part I – Flood Warning, which discusses the topics of detection, thresholds and dissemination

v

vi ●

●

Preface

Part II – Flood Forecasting, which discusses general principles, specific types of river and coastal forecasting models, and examples of specific applications Part III – Emergency Response, which covers the topic of preparedness, response and review

The types of flooding which are discussed include river flooding, coastal surge, snowmelt, ice-jams, urban drainage, flash flooding, and geotechnical risks, such as Tsunami, dam breaks, and debris flows. The impacts of tropical cyclones, hurricanes and typhoons are also discussed from a flooding perspective, although the meteorological aspects are only considered briefly. Examples of operational systems are also provided from several countries, which in places has led to a need to decide on the most appropriate terminology to use. So, for example, the term catchment is used to describe what in some countries is known as a river basin or watershed, the term cell phone describes mobile or cellular phones, and the term flood defence is used in place of the terms levees or dikes. A glossary provides more detail on the terminology used. Although the book is primarily about real time flood warning, forecasting and emergency response, some of the techniques described have evolved from those used in other applications, such as flood simulation, water resources, hydrometeorology, and ocean modelling, and may be of wider interest. The main difference in flood warning applications is the requirement for rapid decision making, often with incomplete or uncertain information. Supporting tools, such as forecasting models, also need to operate sufficiently quickly and reliably to be of value in the process, again often with less input data than would be available in simulation modeling, although with the option of updating outputs in real time to help to correct for differences between observed and forecast values. There is also often a greater emphasis on the resilience of systems, and on documenting any design, operational and other decisions made during model operation. These differences all add an interesting dimension to this diverse and wide ranging subject.

Acknowledgements

This book has benefited from discussions with many people. Following several years working in fluid mechanics, I joined the Centre for Ecology and Hydrology in Wallingford (formerly the Institute of Hydrology) which provided the opportunity to work on a wide range of research and consultancy projects on flood-related, hydrometeorology, water resources and hydrometry topics in more than twenty countries. The many discussions with colleagues during that time provided a useful grounding for the topics discussed in this book. Subsequently, as part of a large engineering consultancy, I have had the benefit of many meetings, site visits and discussions with operational staff as part of flood warning and forecasting improvement projects and strategies, and on projects to develop best practice guidelines in flood forecasting for the Environment Agency and SEPA. In a rapidly developing field such as flood warning, forecasting and emergency response, much information can also be obtained from internet searches, and many organizations place conference proceedings, reports, manuals, and other useful documents in the public domain. In presenting figures, references and quotations from internet and published sources, both the publisher and myself have attempted to identify and provide citations to the appropriate sources, although we apologise if there have been any unintentional errors. Many people assisted with providing comments on short extracts from the draft text and providing figures, and I hope that I have included their comments accurately. Michael Robbins and Steve Jebson from the Met Office, and Ian Marshall and Hazel Phillips from the Environment Agency, were also very helpful with my requests to use a range of figures and tables in the book. I am also grateful to a number of colleagues for discussing aspects of the text, or providing figures, including Marc Huband, Nick Elderfield, Jayne Lamont, Tom Rouse, and Graham Clark. Also, Yiping Chen for many useful discussions on hydraulic modelling for real time forecasting applications, and Jonathan Wright for his general advice and support. Finally, from Springer, I would like to thank Robert Doe and Nina Bennink for their help and advice throughout the process of writing the book and bringing it to production.

vii

viii

Acknowledgements

Additional organizations that I would like to thank include: ●

● ● ● ● ● ●

● ● ● ●

Environment Agency for Figures 3.5, 4.2, 8.7, 9.3 and Tables 3.1, 3.3, 5.1, 5.5, and 6.1 Federal Emergency Management Agency (FEMA) for Figure 10.1 Her Majesty’s Stationary Office for the text cited in Box 9.2 KNMI, Royal Netherlands Meteorological Institute for Figure 3.1 Met Office for Figures 2.1, 2.3 and 2.4 NOAA/National Weather Service for Figures 7.5 and 7.6 Proudman Oceanographic Laboratory/National Tidal and Sea Level Facility for Figures 7.3 and 7.4 Royal Meteorological Society for Figure 2.2 Scottish Hydraulics Study Group for Table 8.1 STOWA for Figures 10.4 and 10.5 World Meteorological Organisation for Figures 1.2, 1.3, 3.4, 3.6, 7.1, 7.2, 8.2, 8.5 and 11.1

Note that any text/material regarding TCP/WMO does not imply the expression/ endorsement of any opinion whatsoever on the WMO Secretariat concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. Also, I would like to thank the following people for assisting with providing figures, or comments on draft text: ● ● ● ●

● ● ● ● ● ● ● ●

● ●

S. Baig, National Hurricane Centre, for comments on Box 7.3 C. Carron, Environment Agency, for comments on Box 8.2 P. Durrant for Figure 10.3 K. Horsburgh, Proudman Oceanographic Laboratory, for comments on Box 7.2 H. Lewis, North Cornwall District Council, for comments on Box 10.1 J. Nower, Environment Agency, for comments on Box 8.2 T. Peng, World Meteorological Organisation, for comments on Box 7.1 A. Richman, Virtual Environmental Planning, for Figure 9.4 B. Stewart, Bureau of Meteorology, for comments on Boxes 1.1, 7.1 and 8.1 K. Stewart, Urban Drainage and Flood Control District (UDFCD), for Figure 4.4 A. Tyagi, World Meteorological Organisation, for comments on Boxes 1.1 and 8.1 D. Vogelezang and colleagues, Royal Netherlands Meteorological Institute (KNMI), for comments on Box 3.1 and for providing Figure 3.1 H. Vreugdenhil and colleagues, STOWA, for comments on Box 10.2 D. Whitfield, Environment Agency, for comments on Boxes 4.1 and 5.1

Contents

Preface .............................................................................................................

v

Acknowledgements ........................................................................................

vii

1

1 1 8 8 9 13 15

Introduction .............................................................................................. 1.1 The Flood Warning Process .............................................................. 1.2 The Nature of Flood Risk ................................................................. 1.2.1 Flooding in Context .............................................................. 1.2.2 Assessing Flood Risk ............................................................ 1.3 Emergency Response ........................................................................ 1.4 The Role of Flood Forecasting .........................................................

Part I

Flood Warning

2

Detection ................................................................................................... 2.1 Meteorological Conditions ............................................................... 2.1.1 Site Specific Observations .................................................... 2.1.2 Remote Sensing .................................................................... 2.1.3 Weather Forecasting ............................................................. 2.2 River and Coastal Conditions ........................................................... 2.2.1 River/Tidal Level Monitoring ............................................... 2.2.2 River Flow Monitoring ......................................................... 2.2.3 Wave Monitoring .................................................................. 2.3 Instrumentation Networks................................................................. 2.3.1 Telemetry Systems ................................................................ 2.3.2 Network Design ....................................................................

21 21 22 28 33 36 37 39 42 44 44 47

3

Thresholds ................................................................................................ 3.1 Rainfall Thresholds ........................................................................... 3.2 River and Coastal Thresholds ........................................................... 3.2.1 Introduction ........................................................................... 3.2.2 Simple Forecasting Techniques ............................................ 3.3 Performance Monitoring ...................................................................

51 51 56 56 61 67

ix

x

4

Contents

Dissemination ........................................................................................... 4.1 Flood Warning Procedures ............................................................... 4.1.1 Introduction ........................................................................... 4.1.2 Flood Warning Areas ............................................................ 4.1.3 Organisational Issues ............................................................ 4.1.4 Control Rooms ...................................................................... 4.2 Dissemination Techniques ................................................................ 4.2.1 Introduction ........................................................................... 4.2.2 Role of Information Technology .......................................... 4.2.3 Warning Messages ................................................................ 4.3 Design and Implementation ..............................................................

Part II

71 71 71 73 75 77 79 79 81 84 87

Flood Forecasting

5

General Principles.................................................................................... 5.1 Model Design Considerations ........................................................... 5.2 Forecasting Systems ......................................................................... 5.3 Data Assimilation ............................................................................. 5.3.1 Error Prediction ..................................................................... 5.3.2 State and Parameter Updating ............................................... 5.3.3 Other Techniques .................................................................. 5.4 Model Calibration and Performance ................................................. 5.4.1 Basic Concepts ...................................................................... 5.4.2 Model Calibration ................................................................. 5.4.3 Performance Measures .......................................................... 5.5 Model Uncertainty ............................................................................

93 93 97 104 106 107 108 108 108 110 113 114

6

Rivers......................................................................................................... 6.1 Model Design .................................................................................... 6.1.1 Forecasting Requirement ...................................................... 6.1.2 Data Availability ................................................................... 6.1.3 Type of Model....................................................................... 6.2 Rainfall Runoff Models .................................................................... 6.2.1 Introduction ........................................................................... 6.2.2 Process-Based Models .......................................................... 6.2.3 Conceptual Models ............................................................... 6.2.4 Data-Based Methods ............................................................. 6.3 River Channel Models ...................................................................... 6.3.1 Introduction ........................................................................... 6.3.2 Process Based Models........................................................... 6.3.3 Conceptual Models ............................................................... 6.3.4 Data Based Methods .............................................................

123 123 124 126 128 132 132 135 137 139 141 141 142 145 146

7

Coasts ........................................................................................................ 7.1 Model Design Issues ......................................................................... 7.2 Process-Based Models ......................................................................

149 149 156

Contents

8

xi

7.2.1 Astronomical Tide Prediction ............................................... 7.2.2 Surge Forecasting.................................................................. 7.2.3 Wave Forecasting.................................................................. 7.2.4 Shoreline Processes ............................................................... 7.3 Data-Based Methods ......................................................................... 7.3.1 Artificial Neural Networks ................................................... 7.3.2 Other Techniques ..................................................................

156 157 165 167 169 169 171

Selected Applications ............................................................................... 8.1 Integrated Catchment Models ........................................................... 8.1.1 Introduction ........................................................................... 8.1.2 Modelling Approach ............................................................. 8.1.3 Ungauged Inflows ................................................................. 8.2 Flash Flood Forecasting .................................................................... 8.3 Snow and Ice ..................................................................................... 8.3.1 Snowmelt Forecasting ........................................................... 8.3.2 River Ice Forecasting ............................................................ 8.4 Control Structures ............................................................................. 8.4.1 Dams and Reservoirs ............................................................ 8.4.2 River Control Structures ....................................................... 8.4.3 Tidal Barriers ........................................................................ 8.5 Urban Drainage ................................................................................. 8.6 Geotechnical Risks ........................................................................... 8.6.1 Structural Risks ..................................................................... 8.6.2 Earth Movements ..................................................................

175 175 175 177 178 181 185 185 188 190 190 195 198 199 202 203 205

Part III

Emergency Response

9

Preparedness............................................................................................. 9.1 Flood Emergency Planning ............................................................... 9.1.1 General Principles ................................................................. 9.1.2 Risk Assessments .................................................................. 9.1.3 All-Hazard Approaches ........................................................ 9.1.4 Validation and Testing of Plans ............................................ 9.2 Resilience .......................................................................................... 9.2.1 Introduction ........................................................................... 9.2.2 Analysis Techniques ............................................................. 9.3 Role of Information Technology ...................................................... 9.3.1 Introduction ........................................................................... 9.3.2 Geographical Information Systems....................................... 9.3.3 Visualisation and Simulation ................................................

209 209 209 214 217 219 220 220 224 226 226 227 228

10

Response.................................................................................................. 10.1 Flood Event Management ............................................................. 10.1.1 Preparatory Actions ........................................................ 10.1.2 Timelines ........................................................................

231 231 231 234

xii

Contents

10.2 10.3

Decision Support Systems ............................................................ Dealing with Uncertainty ..............................................................

237 244

Review ..................................................................................................... 11.1 Performance Monitoring ............................................................... 11.2 Performance Improvements .......................................................... 11.2.1 Detection ......................................................................... 11.2.2 Thresholds ....................................................................... 11.2.3 Dissemination ................................................................. 11.2.4 Forecasting ...................................................................... 11.2.5 Preparedness ................................................................... 11.2.6 Response ......................................................................... 11.3 Prioritising Investment .................................................................. 11.3.1 Cost Benefit Analysis ..................................................... 11.3.2 Multi Criteria and Risk Based Analysis .........................

249 249 253 254 255 256 257 258 258 260 261 265

Glossary ..........................................................................................................

267

References .......................................................................................................

275

Index ................................................................................................................

299

11

Chapter 1

Introduction

Recent flood events have shown the devastating impact that flooding can have on people and property. Flood warning and forecasting systems can help to reduce the effects of flooding by allowing people to be evacuated from areas at risk, and to move vehicles and personal possessions to safety. With sufficient warning, temporary defences can also be installed, and river and tidal control structures operated to mitigate the effects of flooding. Many countries and local authorities now operate some form of flood warning system, and the underlying technology requires knowledge across a range of technical areas, including rainfall and tidal detection systems, river and coastal flood forecasting models, flood warning dissemination systems, and emergency response procedures. This introductory chapter provides a general overview of the flood warning process, approaches to flood forecasting and emergency response, and the nature of flood risk.

1.1

The Flood Warning Process

Flood warning systems provide a well-established way to help to reduce risk to life, and to allow communities and the emergency services time to prepare for flooding and to protect possessions and property. Actions may also be taken to reduce or prevent flooding; for example, by operating river control structures, and floodfighting activities such as reinforcing flood defences, and installing temporary or demountable barriers. Informal flood warning systems have existed ever since people started to live and work near rivers and coastlines. Heavy rainfall, high river levels, unusual sea states and other cues, such as the sound of running water, all provide useful information on impending flooding, with traditional methods for providing warnings including word of mouth, messengers, and raising flags and storm cones. These approaches still have a valuable role to play, particularly where flooding develops rapidly, and communities must rely on their own resources for the initial response. For example, in remote parts of Australia, farmers may alert others further downstream if river levels are high or flooding has started (Emergency Management Australia 1999) and, following the December 2004 Tsunami, several community leaders were praised for recognising the abnormal sea conditions and issuing an alert in time to prevent major loss of life (e.g. UNESCO 2006). K. Sene, Flood Warning, Forecasting and Emergency Response, © Springer Science + Business Media B.V. 2008

1

2

1 Introduction

The use of a more technological approach started to become widespread with the introduction of telegraph transmission of river levels in the mid to late 19th century in countries such as the USA, France and Italy (e.g. Smith and Ward 1998), followed by telephone and radio telemetry early in the 20th century, and accelerated in the 1950s and 1960s as the computer and electronic industries developed. Developments have included the introduction of operational computer models of the atmosphere (from the 1950s), weather radar and satellite based observations of rainfall (from the 1970s), and automated and internet based methods of warning dissemination (from the 1990s). The widespread ownership of televisions, radios and telephones and, more recently, cell (mobile) phones and computers, has increased the range of methods which can be used for issuing warnings, supplementing traditional door knocking, loud hailer, siren and other techniques. Many countries and local authorities now operate some form of flood warning system, and Box 1.1 summarises estimates by the World Meteorological Organisation for the status of flood warning and forecasting services worldwide. Flood warning is also increasingly considered as part of a multi-hazard response to natural, technological and other risks (e.g. United Nations 2006a). If the performance meets the required levels of accuracy, reliability and lead time, flood warning can also be one of a range or portfolio of non-structural measures which can be used to manage or reduce flood risk in river catchments or along coastlines, together with other measures such as land use planning, and tax and insurance incentives to limit development in flood prone areas. A flood warning system can include rainfall and tidal detection systems, river and coastal flood forecasting models, flood warning dissemination systems, and emergency response procedures. Each link in this chain is important, and the modern emphasis is on a Total Flood Warning System (Emergency Management Australia 1999) or people-centred approach, in which communities provide inputs to the design of flood warning systems, and help with their continuing operation (e.g. Parker 2003; ISDR 2006; Basher 2006; Martini and de Roo 2007). The various components considered in this book (Fig. 1.1) are shown in Table 1.1, although the recovery component is only discussed briefly, since flood warning and forecasting has a lesser role to play at this stage of a flood event (for example, advising on when flood waters will recede). Also, mitigation measures (e.g. land use planning, insurance) are not discussed. Of course, the terminology used varies between countries and organisations, and some aspects may overlap (e.g. Alexander 2002). For example, the US Army Corps of Engineers (1996) identifies the following stages in the flood warning process: Flood-Threat Recognition; Warning Dissemination; Emergency Response; Postflood Recovery; and Continued Plan Management whilst, for tropical cyclone forecasting (Holland 2007) the following ten phases are identified in a typical cyclone season: Pre-Season Check; Routine Monitoring (at least twice daily); Cyclone Information (about 48 hours from estimated landfall); Cyclone Watch or Alert (landfall within 36–48 hours), Cyclone Warning (landfall within at least 24 hours); Imminent-Landfall; Post-Landfall; Impact Assessment; Documentation; and System Review.

FLOOD WARNING Detection

EMERGENCY RESPONSE PREPAREDNESS For example Stakeholder meetings/consultations

FLOOD FORECASTING

Customer satisfaction surveys Public Awareness campaigns Media briefings

Thresholds

School/outreach campaigns Interagency coordination meetings Flood Hazard Mapping

Dissemination

Flood Emergency Plans Table-top and full scale exercises

Response

Business Continuity/Resilience assessments Staff Training Forecasting model improvements

Recovery

Review

System improvements (instrumentation, communications, dissemination etc) Inputs to flood mitigation projects

Fig. 1.1 Illustration of the components of a flood warning, forecasting and emergency response system Table 1.1 Typical components in the flood warning, forecasting and emergency response process Item Component Examples Flood warning

Flood forecasting Emergency response

Detection

Monitoring of meteorological, river and tidal conditions; and meteorological forecasting (e.g. nowcasting, numerical weather prediction) Thresholds The meteorological, river and coastal conditions under which decisions are taken to issue flood warnings (sometimes called triggers, criteria, warning levels or alarms) Dissemination Procedures and techniques for issuing warnings to the public, local authorities, emergency services, and others Rivers, coasts Conceptual, data based and process based models for forecasting future river and coastal conditions Response Emergency works, temporary barriers, flow control, evacuation, rescue, incident management, decision support Recovery Repairs, debris removal, reuniting families, emergency funding arrangements, providing shelter, food, water, medical care, counselling, support to businesses, restoration of services if interrupted Review Review of the performance of all components of the system, and recommendations for improvements Preparedness Emergency planning, public awareness campaigns, training, systems improvements, business continuity/resilience assessments, flood risk mitigation etc.

4

1 Introduction

The resilience of flood warning systems to failure is also an important consideration, and risk based techniques from other technical sectors and types of emergency are gradually being introduced to help to identify potential points of failure, and appropriate risk reduction measures. There is also much debate about the effectiveness of flood warnings (e.g. Drabek 2000; Handmer 2002; Parker 2003) and of computer models and information systems (e.g. Fortune 2006). Clearly, a warning is successful if it initiates action which prevents flooding which might otherwise have occurred in the absence of that warning; for example by triggering the closure of a tidal barrier, or installation of a temporary defence. However, research suggests that success with providing warnings to the public is mixed, although in some countries has improved markedly in recent years through a combination of using flood forecasting models to extend the lead time and accuracy of warnings, a better understanding of how to communicate warnings, and an increased emphasis on community participation and inter-agency collaboration. For example, one recommendation (Emergency Management Australia 1999) is that the flood warning task can be boiled down to providing appropriate responses to the following five questions: ● ● ● ●

●

How high will the flood reach, and when? Where will the water go at the predicted height? Who will be affected by the flooding? What information and advice do the people affected by the flooding need to respond effectively? How can the people affected by the flooding best be given the appropriate information?

A particular issue to consider is that of the requirements for warning lead time, which can range from a few minutes or less for people on a steep sloping river bank to reach higher ground, to many hours or days for some situations, such as raising temporary defences, evacuating large numbers of people, or drawing down a reservoir in advance of flooding. Similarly, the requirements for accuracy, and tolerance to false alarms, will vary between organisations and communities, and can be influenced by education and public awareness exercises. This topic is discussed in more detail in later chapters. One early success story is that of Bangladesh (World Meteorological Organisation 2006b) in which a programme of investment in education, early warning systems, establishing a volunteer network, and emergency planning has led to a significant reduction in the number of casualties from tropical cyclones, storm surges, and tidal and river floods. For example, in 1998, a major storm surge led to about 140 deaths but, in a storm of similar magnitude in 1991, approximately 130,000 people lost their lives. Flood forecasting and warning systems have also led to major reductions in casualties in China in recent years (e.g. Huaimin 2005). Similar improvements can also be cited in many other countries where, due to improvements in flood warning systems, the risk to loss of life from flooding has reduced markedly. Approaches to flood warning, forecasting and emergency response are constantly evolving as technical advances are made, lessons are learned from flood events, and

1.1 The Flood Warning Process

5

ideas are adapted from other technical disciplines. For example, technological developments in recent years have included the introduction of short range rainfall forecasting techniques (nowcasting) which typically combine weather radar observations with the outputs from Numerical Weather Prediction models, and of multimedia systems for issuing warnings. Much social and behavioural research has also been performed into public understanding of, and response to, flood warnings, in some cases building on research in other disciplines, such as health care and emergency response for other natural hazards. Improvements can also be driven by national legislation, rising public expectations, customer satisfaction surveys, performance monitoring, and the introduction of level of service targets (e.g. Andryszewski et al. 2005). Risk based and probabilistic approaches are also increasingly being evaluated and used operationally, building on ideas from meteorological forecasting and elsewhere; for example, in techniques for prioritising investment, and ensemble forecasting. Increasingly, improvements are performed within a framework of targets for flood warning performance at a national level. Chapters 2–4 discuss the topics of detection, threshold setting and dissemination for flood warnings, whilst Chapters 9–11 discuss the preparedness, response and review stages. The remaining chapters (Chapters 5–8) cover flood forecasting for rivers and coastlines.

Box 1.1 The WMO Flood Forecasting Initiative The WMO Flood Forecasting Initiative aims to improve the capacity of meteorological and hydrological services to jointly deliver more timely and accurate flood forecasting and warning products and services for use in emergency preparedness and response. The initiative was launched in April 2003, and the main expected outcomes are (World Meteorological Organisation 2006a): ●

●

●

●

Improved quantitative and qualitative weather forecasting products are available in such a way that these can be directly used for flood forecasting. Medium-range weather forecasting and climate prediction tools can be applied to extend warning times and produce pre-warning information. National Meteorological and Hydrological Services have improved their capacity to cooperate to jointly deliver timely and accurate flood forecasting information. Integrated weather, climate and hydrological forecasting information are available in a relevant format for use by civil organizations responsible for disaster preparedness and mitigation.

Between 2003 and 2006, a series of regional workshops on “Improved Meteorological and Hydrological Forecasting for Floods” was held in West, Central and South Africa, Latin America, Asia, Europe and the Mediterranean (continued)

6

1 Introduction

Box 1.1 (continued) basin countries. These meetings involved hydrologists and meteorologists from about 85 countries and a number of regional and river basin organisations, as well as universities and research institutions. These meetings helped identify the status of flood forecasting and warning in the countries which can be categorised as: (Fig. 1.2)

Fig. 1.2 Overall status of national flood forecasting and warning services (sample-86 countries) (Reproduced from the WMO Strategy and Action plan for the enhancement of cooperation between National Meteorological and Hydrological Services for improved flood forecasting, courtesy of WMO)

●

●

●

Level I – flood forecasting and warning services are limited or not operational, and a significant upgrading and strengthening of the basic data collection and transmission networks is required, together with improvements in the coordination between meteorological and hydrological services and in the dissemination of flood warnings. Level II – the basic infrastructure is in place for flood forecasting and warning services but improvements are needed in data management and flood forecasting modelling, with training in advanced modelling techniques, and some improvements in coordination between meteorological and hydrological services. Level III – well established flood forecasting and warning services using the latest observation and forecasting techniques, and with warnings generally communicated through various media to Government and Civil Protection Agencies, industry and the public. The main requirement identified here was for improved training and staff capacity in some cases.

These workshops were followed by an international conference in Geneva in November 2006 to identify gaps in current procedures and to establish and agree on a framework and action plan to improve national and regional capacities for flood forecasting. The action plan addresses flooding due to flash

1.1 The Flood Warning Process

Box 1.1 (continued) floods, riverine floods, coastal floods, snowmelt floods, ice-jams glacier, lake outburst floods, landslides and mud flows. The review of existing techniques showed a wide range of capabilities, ranging from well developed systems using the latest Numerical Weather Prediction, weather radar, and satellite and modelling technologies, through to some countries lacking the technical or institutional capacity to operate flood forecasting and warning systems (Fig. 1.3).

Fig. 1.3 Main symptoms of insufficient or non-existent national flood forecasting capability (sample-86 countries) (Reproduced from the WMO Strategy and Action Plan for the enhancement of cooperation between National Meteorological and Hydrological Services for improved flood forecasting, courtesy of WMO)

However, for some of the countries with limited capacity (14–16%), hydrological forecasts are provided by a regional transboundary river basin authority, and activities were underway in a further 28–33% of countries to improve and modernise existing monitoring and forecasting systems. More than half of countries surveyed (55%) identified a lack of monitoring equipment (automatic weather stations, weather radars, satellite imagery) as an issue, including some 27% of countries which required a significant upgrading of basic meteorological and hydrological networks and telemetry systems for flood forecasting applications. More general requirements which were identified included the need for improved coordination and cooperation between organisations and countries, guidance materials for a range of subjects including data exchange, warning dissemination, and forecast products, and improved training and capacity building.

7

8

1.2 1.2.1

1 Introduction

The Nature of Flood Risk Flooding in Context

Flooding is a threat to many communities and businesses, and flood risk is increasing in some locations due to development on floodplains, migration to urban areas at risk from flooding, and artificial influences on flow regimes; for example, urban developments can sometimes increase flood risk through changes to runoff characteristics and the drainage paths of floodwater. Climate change may also be increasing the likelihood of flooding in some places through changes in the frequency and severity of storms, patterns of snowfall and snowmelt, and rising sea levels. Estimates by the Centre for Research on the Epidemiology of Disasters (CRED) suggest that, in the period 1974–2003, there were more than 200,000 victims of flooding, with many more people affected for every casualty. In that period, the July 1974 and December 1999 floods in Bangladesh and Venezuela each accounted for about 30,000 deaths, and flood events in India and China accounted for seven of the ten disasters which were identified as affecting more than 100,000,000 people (all figures from Guha-Sapir et al. 2004). During 2007, flooding due to heavy rainfall affected approximately half of all African countries, affecting more than 1,000,000 people with about 400 victims whilst, in India, Bangladesh and Nepal, the death toll from monsoon rains exceeded 2,000 and affected some 30,000,000 people. Compared to other types of natural disaster, floods account for approximately 20–40% of the events which are reported. Floods can also cause extensive damage to property, infrastructure and crops, and can cut across administrative and national boundaries. For example, the 1998 floods in China were estimated to have submerged more than 200,000 km2 of farmland (e.g. Kundzewicz and Jun 2004) whilst, for Hurricane Katrina in August 2005, in addition to causing more than 1,000 deaths, hundreds of thousands of people were evacuated, and billions of dollars of damage was caused to property, businesses and infrastructure, much of this flood related. Other examples include the Midwest floods of 1993 on the Missisippi and Missouri rivers in the USA, which affected more than 15% of the country, damaging or destroying some 50,000 homes, with approximately 54,000 people evacuated (Smith 2004) whilst, in Europe, the flood events of 2002, 2005 and 2006 affected thousands of people in central Europe, and caused more than 100 deaths. The causes of flooding are mainly atmospheric or geotechnical (Table 1.2). Atmospheric hazards include heavy rainfall, causing rivers to flood, sometimes linked to snowmelt and ice-jams in colder climates, and coastal and estuarine flooding due to surge, wave and wind effects, most notably in tropical cyclones, hurricanes and typhoons. Geotechnical factors such as landslides, debris flows and earthquakes can also lead to raised river levels causing inland flooding, and Tsunami waves resulting in coastal flooding. Secondary effects may include overtopping or breaches of river and sea defence structures, debris blockages at bridges and other structures, surcharging of drainage networks in urban areas, and dam failure or overtopping. Due to the short time available for people to react, fast

1.2 The Nature of Flood Risk

9

Table 1.2 Examples of flooding mechanisms Type

Example

Typical types of flooding

Atmospheric

Frontal depressions

Extensive river flooding, coastal surge and wave overtopping, estuary and delta flooding, urban and pluvial (surface water) flooding Fast response/flash flooding and urban and pluvial (surface water) flooding Extreme prolonged rainfall causing a range of river and urban flooding issues Coastal surge and wave overtopping, inland flooding, estuary and delta flooding Extensive river flooding Rapid rises in river levels Fast moving, deep river flows

Thunderstorms Monsoon Tropical cyclones

Geotechnical

Snowmelt Ice jams Glacial lake outburst flows Dam break Defence breach Tsunami Debris flow

Fast moving, deep river flows Extensive inundation of coastal or inland areas Extensive inundation of coastal margins Destructive flows with high mud and rock content

developing floods present a particular risk to life, including flash floods, dam or defence breaches, and some ice-jam and local surge and wave overtopping events. Tropical cyclones, hurricanes and typhoons are all forms of tropical storm, with the term tropical cyclone used in the Indian Ocean, hurricane in the Atlantic and Eastern Pacific Oceans, and typhoon in the Western Pacific. Frontal depressions are most common in mid-latitudes, and can cause prolonged rainfall, as can monsoons which are driven by seasonal variations in temperature between sea and land masses. Thunderstorms can occur at most latitudes, and can cause intense rainfall for periods of typically up to a few hours. Snow and ice related problems affect many high latitude regions on all continents, and high mountain ranges elsewhere. Dam and defence risks are possible anywhere that reservoirs or polders have been constructed, or dams built across lakes, as are breaches in river or coastal flood defences (often known as levees or dikes). Tsunami can affect all ocean basins, but are most prevalent in the Pacific Ocean and in South East Asia (although the December 2004 Tsunami was in the Indian Ocean). Debris flows are a major problem in Central Asia and the Caucasus and in parts of the USA.

1.2.2

Assessing Flood Risk

Flood risk is often expressed as the combination of two factors; probability (or hazard) and consequence (or impact). The probability expresses the likelihood of damaging flood levels or flows being reached, whilst the consequence can be expressed in terms of indicators such as the numbers of properties affected, loss of life, or economic damages.

10

1 Introduction

Estimates for the numbers of people at risk from flooding, and affected in individual events, are of course subject to many uncertainties, including the degree to which events are reported, the approach taken to flood risk assessments and, for international comparisons, differences in the datasets and recording methods which are used. However, some studies (e.g. Parker 2000; Smith 2004) suggest that the percentages of people at risk from flooding range from 3% to 5% of the population in the UK and France, to about 12% in the USA, 50% in the Netherlands, and 70–80% in Vietnam and Bangladesh. Estimates are also complicated by transient populations, which can include tourists, hikers, temporary workers, business travellers, and the homeless. Indeed, in some countries, such as the USA, one of the main risks to life from flooding is from people in cars and other vehicles being trapped or swept away by floodwater (e.g. Henson 2001). The link between flood risk and social, political and economic factors, particularly risk to life, is well documented, and can arise from issues such as a lack of public awareness of flooding issues, or controls on floodplain development, limited funds available for flood control and protection (e.g. river and sea defences), low resilience of buildings to flooding (e.g. temporary compared to permanent settlements), and a lack of investment in flood warning, forecasting and emergency response systems. Where these factors are significant, the numbers of people affected by a flood event can be much higher than equivalent events in locations without these problems. Measures of vulnerability to flooding are also increasingly considered in flood risk studies: for example, combining the following factors (e.g. Wade et al. 2005): ● ●

●

Flood hazard (depth, velocity, debris) Area Vulnerability (effectiveness of flood warning, speed of onset of flooding, and type of buildings e.g. low rise/high rise) People Vulnerability (ability to ensure own safety and that of dependents e.g. the elderly, infirm, children)

Of course, vulnerability to flooding can depend on a wide range of physical, environmental, social, economic, political, cultural and institutional factors, and can vary widely between individuals, households and communities; for example, the length of time that people have lived in the floodplain (or if they are visiting the area e.g. tourists), recent experience of flooding, and local institutional capacity to respond to flooding. Some alternative definitions (e.g. World Meteorological Organisation 2006c) express vulnerability in terms of physical, material, constitutional, organisational, motivational and attitudinal conditions or, for tropical cyclones (e.g. Holland 2007), include the availability of existing community level plans and organisational structures, the proportion of cyclone resistant property, the state of protective works (river and coastal defences etc.), and the likely protection from coastal forests and mangroves. When designing a flood warning scheme, a starting point is often to make an assessment of the locations and numbers of people and properties at risk from flooding. Vulnerability studies can also highlight where to target effort in public awareness campaigns, developing flood emergency plans, and in emergency response. Methods for assessing risk include interviews with people who know the

1.2 The Nature of Flood Risk

11

area well, examination of historical flood records (trash mark surveys, aerial and other photographs, newspaper reports, satellite images etc), and hydrodynamic and other modelling techniques. Interviews and historical records can provide useful information, although may give a false impression if any significant changes have occurred since the last major flood event in the level of flood risk or key flooding mechanisms (e.g. construction of flood defences, dredging, urban development). Also, people may not be aware of more serious flooding before they moved to the area. Ground survey and remote sensing techniques can also provide detailed maps of flooding extent, although not necessarily for the peak of the flood, and satellite observations are increasingly being used to monitor flood extents using both optical and microwave frequencies, and to build up databases of flood extent information. Models provide a more formal way of assessing flood risk, and can range from simple correlation and other methods for single locations, through to detailed hydraulic models for river and coastal processes. Some countries (e.g. the USA, Japan and various European countries) have programmes in place to systematically assess flood risk at a national scale through detailed hydraulic modelling of locations with a significant flood risk (Box 1.2).

Box 1.2 Flood risk modelling The national flood risk mapping programmes in many countries use a range of modelling techniques to estimate flood depths, velocities and extents. For rivers, for example, actual or synthetic rainfall events can be fed into a network of rainfall runoff models representing major sub-catchments, whose outputs provide the inputs to a model for the river network and significant features such as floodplains and reservoirs. In areas prone to flooding, the model detail may include all significant controls on river levels and flows, such as bridges, culverts, gates, defences and other features, as well as the main details of the floodplain, using construction and topographic information obtained from conventional survey and remote sensing techniques (e.g. Light Detection and Ranging LIDAR equipment, or Synthetic Aperture Radar SAR equipment). In increasing order of complexity (and, in principle, accuracy), process-based methods for modelling river levels, flows and, in some cases, velocities, on the floodplain can include: ●

●

●

One-dimensional models for the main river channel, with projection of levels onto the floodplain, or separate pathways for main channel and floodplain flows One-dimensional models including floodplain pathways represented via spill units, compartments and/or cells Two dimensional models of the floodplain using ‘bare earth’ digital terrain models based on mass conservation only, or including momentum effects as well (continued)

12

1 Introduction

Box 1.2 (continued) ● Fully two or three dimensional models of the floodplain incorporating features on the floodplain such as buildings, embankments, gulleys etc., and possibly urban drainage networks Hydrodynamic techniques can also be used for modelling inundation of coastal floodplains due to high tidal levels, wave action and surge. Maps may be developed either with or without flood defences, with the no defence case sometimes being used to study the worst case flood extent; for example, if a defence is breached, overtopped or bypassed. Later chapters show several examples of the results from flood risk mapping studies including plan view and virtual reality representations. Having estimated river flows and levels, and possibly depths and velocities, within flooded areas, the resulting flood outlines can then be intersected with information on property locations, and lists generated of properties at risk. The resulting extents can also be related to the gauge heights used for triggering flood warnings (see Chapter 3). These property lists then form the basis for deciding which properties need to receive flood warnings. Vulnerability maps can also be generated to assist in developing emergency plans, although this is performed much less frequently than mapping of flood extent. The resulting flood outlines may also be expressed in terms of probability or return periods, with presentations of maps at 1 in 50, 100, 200 and 1,000 year return periods perhaps the most widely selected. Some sources of uncertainty in flood risk mapping can include the accuracy of input data and high flow rating curves (for river modelling), the various modelling assumptions and parameters, survey data accuracy, local influences around structures, and other factors. Methods for assessing the uncertainty in flood extent estimates are increasingly being explored (e.g. Pappenberger and Beven 2006; Pappenberger et al. 2007), and can potentially feed into decision support and other systems used in preparing for and managing flood events. Probabilistic techniques are also increasingly being used to consider the risk from failures or overtopping at flood defences (e.g. Sayers et al. 2002).

However, whatever the technique used to assess flood risk, one problem is always to assess the extent of mobile and transient populations who may not appear in conventional property and census databases. Examples can include vehicle users, shopping centres, supermarkets, tourists, hikers, outdoor events, and locations such as caravan or mobile home parks, and camp sites. Local visits, and discussions with people who know the area well, may be the best way of determining the extent of this risk, and the options (if any) for providing warnings to these groups, or preventing access in time to minimise the flood risk. Some other problems which can arise with property databases are that they may omit some commercial properties with significant numbers of occupants during working hours, since the correspondence address is at another location (e.g. head

1.3 Emergency Response

13

office), and that some locations with many residents (e.g. apartment blocks) may appear as only a single property. Also, some high-risk locations may not be clearly identified, such as water treatment or industrial works and critical locations such as hospitals, power stations, telecommunications hubs etc. Again, local visits and discussions can help to resolve some of these issues.

1.3

Emergency Response

Emergency response is the process of responding to a flood event, ideally on the basis of a flood warning received. In many countries, there is a separation in responsibilities between the flood warning and forecasting service, and emergency responders such as the police, fire service and local authorities. However, the organisation of a flood warning service can vary widely, with warnings being issued by the meteorological service in some countries, and a range of river management, coastal and local authorities in others. Privately developed systems also operate in some locations, with applications ranging from community based warning systems through to systems operated by owners of major infrastructure such as railways and hydropower schemes. Sometimes warnings may also be restricted to specific types of flooding, such as river flooding or coastal flooding, and exclude other types, such as flooding in urban areas from drainage problems. A major flood event often requires a multi-agency response, involving local authorities, the emergency services, transport operators (road, rail etc.), utility operators (water, electricity, gas, telecommunications), the military, coastguard, medical services, voluntary services, humanitarian aid organizations, and others. The response can include closing transport routes, protection of key installations, such as power stations and water treatment works, reinforcing flood defences, providing rest centers and shelters for people evacuated from properties, and rescue of people and livestock stranded in flood waters. Difficult decisions may also need to be made on issues such as the need to evacuate hospitals and nursing homes (with the evacuation itself presenting risks), precautionary shutdown of power or water supplies, and ordering widespread evacuations of property. During a flood event, individual property owners can also take action to reduce the damage caused by flooding by moving (as appropriate) vehicles, furniture, electrical equipment, personal possessions, valuables, animals and livestock to safer locations, and using sandbags, flood boards and other flood resilience measures to protect their property (if available). For example, in a post event survey of flooding in parts of the Elbe and Danube catchments (Thieken et al. 2007), emergency measures which were reported by residents included: ● ● ● ● ● ●

Put moveable contents upstairs Drive vehicles to a flood-safe place Safeguard documents and valuables Protect the building against inflowing water Switch off gas/electricity Disconnect household appliances/white goods

14 ● ● ● ● ● ●

1 Introduction

Gas/electricity was switched off by public services Protect oil tanks Install water pumps Seal drainage/prevent backwater Safeguard domestic animals/pets Redirect water flow

Businesses can also take actions to reduce damage to stock, equipment and systems and, depending on the time of day, may also be able to advise employees not to come in to work, or to leave early, in order to minimise risk. Flood warnings can also assist river management and coastal authorities with the operation of structures and in other actions to help to reduce or prevent flooding and some examples (Fig. 1.4) include: ●

●

●

●

Flood barriers – installation or operation of temporary or demountable barriers to protect properties and infrastructure from flooding Flood gates – closing gates which at low to medium flows are normally kept open to allow for drainage, access, navigation etc. Flow diversion – diversion of river flows into off-line storage areas to reduce flows further downstream (e.g. washlands, flood retention areas) Pumping – use of high volume pumps to reduce water levels

Fig. 1.4 Examples of river and coastal flood defences and a flood gate for washland drainage

1.4 The Role of Flood Forecasting ●

●

●

●

15

Reservoirs – draw down of reservoir levels in advance of high inflows to provide flood storage to reduce flows further downstream Sandbags – placing sand bags to raise the level of flood defences, fill gaps in defences, or to protect properties Temporary works – emergency repairs to flood defences (levees and dikes) and other locations which might provide a flow route for flood water Tidal barriers – closing barriers or gates to reduce the risk of inland flooding due to surge or high tides

Temporary and demountable barriers are increasingly used for flood prevention, and consist of metal, plastic, rubber and other types of panels, bags or tubes which can be placed at locations where flooding is anticipated, if a flood warning is received in time. Chapters 9–11 describe emergency response in more detail, including the development of flood emergency plans, decision support systems, dealing with uncertainty, and performance monitoring.

1.4

The Role of Flood Forecasting

Although flood warnings can be issued on the basis of observed meteorological, river and coastal conditions alone, the development of flood events can often only be anticipated a short time into the future, and it can be difficult to translate what is observed into estimates of flooding extent. Interpretation can also be complicated by other effects, such as operations at river control structures, storm surges, and inflows from major tributaries. Flood forecasting models can help with these issues, and are increasingly used to improve the lead time and accuracy of warnings provided by a flood warning service. Typically, forecasts are based on observations of river levels and rainfall higher in a catchment (for river flooding), or of tidal levels, wave heights, wind speed and other parameters (for coastal flooding). Rainfall and surge forecasts from atmospheric and oceanographic models may also be used as inputs to further extend the lead time of flood forecasting models. Forecasts may also have wider applications in areas such as river navigation, hydropower generation, water resource management, and pollution incident control. ‘What if’ scenarios can also be performed; for example, using scenarios for future rainfall or snowmelt, or for operational actions such as closing a tidal barrier. Flood forecasts can also be used to automatically trigger the issuing of warnings, or the operation of flow control structures, although the decision to use an automated approach depends on confidence in the model outputs, policy, and other factors, and the vast majority of current systems still require interpretation of outputs by an experienced forecaster. There are many approaches to flood forecasting, ranging from simple empirically based methods to fully integrated catchment or coastal models which, increasingly, incorporate real time hydrodynamic modelling components. Different types of models, or model components, may also be developed for different flooding mechanisms;

16

1 Introduction

Railway

Dam Dam Overtopping

Major road

Wave

Village

Town Defence Town Overtopping

River Out of Bank

Surge, Tide, Wave

Town

Power Station River, tide, surge Farms Caravan Park

River Out of Bank

Chemical Factory Town Surge, Tide, Waves

Fig. 1.5 Illustration of flooding issues which might be included in a regional flood forecasting model

for example, for the situation shown in Fig. 1.5, a range of rainfall runoff, reservoir (dam), river, estuary (delta), and offshore, nearshore and wave overtopping models might be required, optimised to provide forecasts at towns, infrastructure, and transport routes where they are at risk from flooding. One distinguishing feature of forecasting models, compared to off-line simulation models, is the ability to use observed (telemetered) data to modify forecasts as they are generated. Thus, if the forecast at the present time is in error, it can be adjusted to account for the current observed values, and also into the future, based on assumptions about the cause of errors up to the present time (‘time now’), and likely future trends. This real time updating of forecasts (or data assimilation) can significantly improve the accuracy of model outputs, and many techniques have been developed, including error prediction methods and techniques which adjust the internal state of model components, or model parameters. One reason for needing updating techniques is the uncertainty in model outputs, which can arise from many sources. For example, outputs can be affected by errors and uncertainties in measurements of rainfall and levels and flows (for rivers), and wind speed, wind direction, wave height and tidal levels (for coastlines). If meteorological forecasts are used to extend warning lead times, additional uncertainties arise from that component of the system. These issues have long been recognised (e.g. World Meteorological Organisation 1994; Emergency Management Australia 1999; Beven 2008) and it is widely accepted that flood forecasts should be issued with an indication of confidence or uncertainty. The case for probabilistic forecasts in hydrology has been concisely summarised by Krzysztofowicz (2001), which is that:

1.4 The Role of Flood Forecasting ●

●

●

●

17

First, they are scientifically more ‘honest’ than deterministic forecasts: they allow the forecaster to admit the uncertainty and to express the degree of certitude Second, they enable an authority to set risk-based criteria for flood watches, flood warnings, and emergency response; and they enable the forecaster to issue watches and warnings with explicitly stated detection probabilities Third, they appraise the user of the uncertainty; and they provide information necessary for making rational decisions, enabling the user to take risk explicitly into account Fourth, they offer potential for additional economic benefits of forecasts to every rational decision maker and thereby to society as a whole

One widely quoted example of the potential use of uncertainty information in decision making (e.g. Krzysztofowicz 2001) is for the 1997 flooding on the Red River in Grand Forks, North Dakota, which caused flooding to some 5,000 homes, and is described in more detail in Chapter 10. Post event analysis showed that the actual peak was higher than the forecast values, with the question arising that, if the uncertainty in the forecast been known, would sandbagging of levees have been continued to higher levels, avoiding the flooding which occurred? By contrast, an example of deriving economic benefits from ensemble forecasts is in the hydropower industry, where some operators in the USA and Canada gain significant savings from using probabilistic forecasts of seasonal flows (e.g. Howard 2004). In meteorology, probabilistic forecasting techniques have been used since the 1990s, and are nowadays seen as an indispensable tool in weather forecasting. The basis of the method is to adjust the initial conditions for the computer models used to forecast atmospheric and ocean conditions over a range reflecting the uncertainty in current conditions, and possibly model parameters. Stochastic methods, consisting of statistical sampling of inputs or outputs, can also be used. The resulting scenarios, or ensembles, are then used to guide forecasters in the information that they issue to the public and, in some cases (e.g. the Netherlands), the range of estimates is presented in some national television weather forecast bulletins. Similar techniques are also starting to be used in flood forecasting, perturbing the meteorological and other inputs to models (e.g. river flows), and possibly internal model parameters and other factors (e.g. the high flow ends of stage-discharge relationships). The issue of how to communicate the resulting information on uncertainty to decision makers, including the public, is also an active research area (e.g. Todini et al. 2005; National Research Council 2006; Pappenberger and Beven 2006). Chapters 5–8, and 10, describe these topics in more detail. Forecasting models usually also require a computer platform on which to operate, capable of data gathering, the scheduling and control of model runs, alarm handling, and post-processing of model outputs into a form which is useful to forecasters. During a widespread flood event, a purpose-made system may provide the only practicable way of operating and interpreting the output from large numbers of models, particularly if data assimilation is used. Modern systems increasingly make use of spatial techniques for analysing and presenting data and forecasts, and functionality can include map-based indicators (e.g. flashing symbols) of locations

18

1 Introduction

where flooding is expected, overlays of property locations, street maps, aerial photographs, terrain etc., and the facility to ‘drill down’ for additional information and detail at any location. The facility to perform real time inundation mapping during an event is also increasingly available. Although much of the focus nowadays is on automated techniques, simpler techniques still have an important role to play, particularly where budgets are limited, the level of flood risk does not justify investing in a complicated approach, or as a backup to a more sophisticated approach. Low cost community based systems are also widespread, and typically involve nominated members of the community monitoring raingauges, marker boards or river gauges, and issuing warnings by loud speaker, community billboards, and door knocking as appropriate. Information may also be passed to local experts to decide on the appropriate action to take, who may also have access to paper based or computerised forecasting models (e.g. FEMA 2005). Informal systems are also widely used in some countries (e.g. Parker 2003). Another example is the flood forecasting system that was trialled for the two major rivers in Somalia (the Jubba and Shebelli) between 1988 and 1990 (Institute of Hydrology). The system operated on a stand-alone personal computer, with data entered manually based on observations by government workers at some 20 locations along the two rivers, and transmitted verbally to Mogadishu over the government radio network. A low cost approach was used for model development, using a range of simple correlation, flow routing and overtopping models, and a more detailed model for an off-line storage reservoir. Real time updating was included using a simple interactive method which allowed operators to adjust forecasts visually to account for the trend in forecast errors over recent model runs. In operational use, information was received three times per day, and the model runs were used to provide forecasts of future river levels and flows up to 7 days ahead to farmers, irrigation scheme operators, and engineers engaged in river works, together with warnings of high flow conditions. Forecasts were also included in weekly agricultural situation reports. Chapters 5–8 describe forecasting techniques in more detail, including general principles (Chapter 5), river forecasting methods (Chapter 6), coastal forecasting methods (Chapter 7) and a range of applications (Chapter 8), including integrated catchment modeling, and forecasting for flash floods, the effects of snow and ice, control structures, urban flooding, and geotechnical risks such as dam break, defence breach, and tsunami.

Chapter 2

Detection

Most flood warning systems use near real time measurements of meteorological and river or coastal conditions to guide operational decision making. Depending on the application, this may include information on rainfall, wind speeds, sea state, tidal levels, river levels and other parameters, such as snow cover. Remote sensing techniques such as weather radar and satellite may also be used, together with the outputs from Numerical Weather Prediction models and nowcasting techniques. This chapter provides a general introduction to these and other techniques for monitoring meteorological, river and coastal conditions for flood warning applications. Telemetry systems are also discussed, together with approaches to designing telemetry networks for flood warning applications.

2.1

Meteorological Conditions

With only a few exceptions, such as geotechnical risks (see Chapter 8), most flooding problems are linked to atmospheric conditions, and observations or forecasts of rainfall and other parameters often provide the first indication of potential flooding. The main types of meteorological information which are useful in flood warning and forecasting applications include: ●

●

●

Site Specific (or Point) Observations – measurements at a specific location using rain gauges, automatic weather stations etc. Remote Sensing (or Areal) Observations – based on satellite observations, weather radar etc. Computer Model Outputs – from Numerical Weather Prediction (NWP) models, nowcasting techniques, and other approaches

Note that weather forecasting techniques are included in this chapter as a form of detection since, as with site specific and remote sensing techniques, the outputs represent another source of information for the operation of flood warning and forecasting systems. When considering these approaches, there are various trade-offs in terms of the spatial resolution, accuracy and lead times of each technique. For example, site K. Sene, Flood Warning, Forecasting and Emergency Response, © Springer Science + Business Media B.V. 2008

21

22

2 Detection

specific observations provide an indication of actual conditions at certain locations in a catchment or coastal reach, but may be unrepresentative of the overall conditions which lead to flooding. By contrast, remotely sensed data provide an overall picture of the distribution of the parameter being observed (e.g. rainfall, snow cover), but require some assumptions or a model to translate observations to conditions at the ground or sea surface. This introduces an additional source of uncertainty, and measurements are sometimes of too coarse a resolution to be useful. Weather forecasting techniques provide additional lead times, and usually also provide detailed spatial information for the parameters being forecast (rainfall, wind, soil moisture etc.), but obviously rely on the outputs from computer models, which again can introduce an additional source of uncertainty. Chapters 5–8 discuss some of these various trade-offs between lead time and accuracy, whilst other factors which need to be considered include the likely reliability during flood events, and the choice of an appropriate degree of instrumentation in relation to the level of flood risk; a topic which is discussed further in Section 2.3 and Chapter 11 when considering techniques for prioritising investment in flood warning schemes. For river flooding applications, rainfall is often a key parameter, although other meteorological parameters which may be required include observations or estimates for air temperature, wind speed, net and solar radiation, soil moisture, snow cover, river ice cover and ice jam locations, and reservoir and lake evaporation. For coastal flooding applications, information on atmospheric pressure, and wind speed and direction, is often a key input to surge and wave forecasting models and, for tropical cyclones (and hurricanes and typhoons), information on storm size, intensity, track and speed is also important. Table 2.1 summarises these and some other requirements for the various threshold based and forecasting techniques described in later chapters. The remainder of this section discusses the technological background to these various techniques.

2.1.1

Site Specific Observations

The techniques for observing meteorological parameters at specific locations are well established (e.g. World Meteorological Organisation 1994b, 2000; Strangeways 2007). For flood warning and forecasting applications, near real time measurements are usually required, and methods suitable for telemetry include tipping bucket raingauges, cup or ultrasonic anemometers (wind speed), wind vanes (wind direction), radiometers (solar, net and reflected radiation), hygrometers (humidity) and neutron or capacitance probes (soil moisture). Instruments can be installed on land, or on moored buoys, ships and other offshore locations. These and other devices can of course also be used if values are transmitted manually; for example, by voice, email, fax or telegraph. Evaporation can be measured directly using evaporation pans, or, less often, by turbulence monitoring techniques which integrate water vapour transport through a

2.1 Meteorological Conditions

23

Table 2.1 Some common requirements for meteorological data and forecasts in flood warning and forecasting applications Parameter

Category

Examples of techniques

Rainfall

Site specific observations Remote sensing

Raingauges, disdrometers Satellite, weather radar, microwave links Numerical Weather Prediction models, nowcasting Capacitance probes, neutron probes, lysimeters Satellite (e.g. Synthetic Aperture Radar) Numerical Weather Prediction models, nowcasting Ablation stakes, snow pillows, snow cores Satellite based optical or infrared channels Numerical Weather Prediction models, nowcasting Automatic Weather Stations Satellite based infrared sensors (temperature) Numerical Weather Prediction models, nowcasting Satellite, weather radar Storm scale/mesoscale Numerical Weather Prediction models, nowcasting Solar radiation, net radiation (or individual components), soil heat flux Evaporation pan, turbulence measuring devices (also indirect methods using wind speed, radiation and humidity) Numerical Weather Prediction models, nowcasting

Weather forecasting Soil moisture

Site specific observations Remote sensing Weather forecasting

Snow cover

Site specific observations Remote sensing Weather forecasting

Atmospheric conditions (air temperature, humidity, wind speed etc.)

Site specific observations Remote sensing Weather forecasting

Storm scale information

Remote sensing Weather forecasting

Radiation

Site specific observations

Evaporation

Site specific observations

Weather forecasting

fixed pathway between two sensors. Alternatively, methods such as the Penman equation are widely used for estimating open water evaporation from wind speed, temperature, humidity and (possibly) net radiation measurements, with the Penman Monteith approach used for estimating evapotranspiration from grass, vegetation etc. Often the various sensors can be combined into an automatic weather station, which may monitor some or all of the following parameters:

24

2 Detection

Fig. 2.1 Examples of inland and offshore automatic weather stations (Kevin Sene and © Crown Copyright 2007, the Met Office)

● ● ● ● ● ●

Rainfall Air temperature Humidity Wind speed and direction Solar and net radiation Soil or water temperature

Figure 2.1 shows two examples of automatic weather stations, consisting of a temporary installation above a tropical lake in Southeast Asia for a study into long term trends in lake evaporation (Sene et al. 1991) and a buoy mounted instrument being inspected off the coast of the UK. Of the various meteorological parameters which could be monitored, for flood warning and forecasting applications, perhaps the two of most interest are rainfall and snowmelt, and these are described in more detail below. 2.1.1.1

Rainfall

For measuring rainfall, tipping bucket raingauges are probably the most widely used method for flood warning and forecasting applications, and record rainfall when the depth reaches a sufficient amount (or weight) to cause a bucket mechanism to tip. Typical bucket sizes are equivalent to rainfall depths in the range 0.1–2.0 mm, with the choice of tip size often based on the maximum rainfall intensities

2.1 Meteorological Conditions

25

anticipated at the site. Each tip is recorded, together with the time of the tip, and can be reported by telemetry directly, or accumulated to fixed time intervals before transmission. Weighing raingauges, by contrast, use springs, vibrating wires or balance weights to record the weight, and hence depth, of rainfall, whilst drop-counting gauges (e.g. Stow et al. 1998) use optical techniques or electrodes to record individual drops of a fixed size released through a constriction. Depth type gauges accumulate rainfall and use an electrode (e.g. Oi and Opadevi 2006) or float mechanism, linked to a recording device, to record the depth of rainfall and hence the incremental changes in given time intervals. Disdrometers, which use a laser or ultrasound beam to detect falling rainfall, are a newer technique for recording rainfall. These devices work on the principle of detecting the passage of raindrops through a beam of light (e.g. Nemeth 2006) or ultrasound, with appropriate signal processing to estimate rainfall amounts. In principle this approach requires less maintenance than traditional raingauges since there is no capture of rainfall. Factors to consider include processing for a range of droplet sizes (and fall velocities), for different types of precipitation (rainfall, snow, hail etc.), and for wind driven effects as rain passes through the beam. Low cost (micro) vertically pointing precipitation profilers are also another recent development for single site measurements of rainfall. Manually operated (storage) raingauges can also be useful for providing rainfall information to assist with post event evaluations of flooding, and more generally to improve understanding of the rainfall distribution in a region or catchment when developing rainfall runoff forecasting models. Measurements are typically made on a daily or monthly basis. Best practice in the installation and use of raingauges, and the strengths and limitations of different designs, is well documented (e.g. World Meteorological Organisation 1994b, 2000) but some specific problems which can arise in high wind and rainfall conditions include: ● ● ● ● ●

Splashing – both into and out of the gauge Exposure – sheltering by obstacles such as trees or buildings Wind influences – from the airflow over the gauge and at the site Snowfall – blocking by snowfall (raingauges only) Flooding – submergence of the gauge if it is installed in a flood prone site

For flood warning or forecasting applications, the recording interval to use will depend on the capability of the equipment and associated electronics, but should ideally be sufficiently frequent to resolve the key features of events. Typically a 5 minute, 15 minute or hourly value is used. More generally, for all types of site specific measurement, the question arises of how representative the measurements are of overall catchment conditions, and of appropriate techniques to use for estimating catchment average rainfall for input to rainfall runoff forecasting models. Box 2.1 describes some techniques for estimating catchment average rainfall.

26

2 Detection

Box 2.1 Catchment rainfall estimation Raingauges give estimates of rainfall at a single point whereas, for many flood warning and forecasting applications, area averaged values are required; for example, for catchment average rainfall. For small catchments, or reasonably uniform rainfall, a single raingauge may be representative. However, usually a number of gauges within the catchment, and possibly from nearby catchments, will be used in the averaging process, and some techniques include: ●

●

●

●

●

Arithmetic mean – which simply takes the average value for all selected raingauges, giving equal weight to all gauges without considering their spacing or the rainfall distribution in the catchment. Thiessen polygons – in which polygons are derived by joining the mid points of the lines between adjacent raingauges, with the weights based on the proportion of the catchment area attributed to each gauge within the catchment, divided by the catchment area. Isohyetal method – which derives lines of equal rainfall based on the observed values, from which a catchment average value can be derived. Surface fitting methods – which include a range of automated techniques, such as multiquadratic, inverse distance, triangular planes (TIN) and polynomial methods. Geostatistical techniques – such as Kriging which also interpolate values but using functions for the dependence of values on distance between gauges for all combinations of gauges. Methods such as co-Kriging also bring in auxiliary variables such as elevation or aspect.

Additional factors such as topography, aspect, runoff coefficient, and soil type can also be brought into some of these weighting schemes. The methods are presented in approximately increasing order of complexity and accuracy and there have been numerous studies into the merits of the various approaches (e.g. Creutin and Obled 1982; Goovaerts 2000). In particular, elevation and rain shadow effects can be significant, as illustrated in Fig. 2.2 for average annual rainfall estimates for the Lesotho Highlands, whose peaks rise to approximately 3,500 m in places. Analyses of weather radar data and computer modelling can also assist with understanding storm characteristics, such as typical storm scales, preferred directions of travel, local topographic influences etc., and in developing appropriate catchment averaging schemes. Sometimes it is found that, where there are no anticipated major spatial variations in flood generating rainfall (e.g. frontal events in low lying areas), the simpler fixed weight methods can provide reasonable results. However, where spatial and topographic variations (continued)

2.1 Meteorological Conditions

27

Box 2.1 (continued)

Fig. 2.2 Annual rainfall distribution along a transect through the Lesotho Highlands (Royal Meteorological Society/Sene et al. 1998)

are significant, more complex methods may be required if real time systems and software are available to support this type of analysis. 2.1.1.2

Snow Cover

Information on snow depth, water equivalent and snow extent can be required for input to the snowmelt forecasting component of flood forecasting models, and also for more empirical techniques for estimating the consequences of snowmelt. The challenge in snow monitoring is that the depth and type of snow cover can vary significantly over small distances compared to typical catchment scales, so that only a limited sample of values can usually be obtained. Satellite monitoring can assist in assessing snow coverage, whilst observation techniques for estimating depth include snow (or ablation) stakes and snow cores. Traditional depth measuring techniques rely on an observer sending values by telephone, email, radio etc., but automated techniques have also been developed. For example, radio isotype methods can be used in which the water equivalent of snow is estimated from absorption of gamma radiation in the vertical or horizontal plane and, in principle, can provide real time estimates of water equivalent snowfall (e.g. World Meteorological Organisation 2000). Tipping bucket raingauges can also be used to measure the water equivalent snow depth if they are fitted with heaters, although may under-record the true amount of snowfall, and show a lag between

28

2 Detection

snowfall and the recorded values. Measurements of air temperature may be used to help in interpreting the readings from raingauges when snowfall is thought to be a factor. Perhaps the most extensive ground based monitoring of snow depth and extent is by the SNOTEL (SNOwTELemetry) monitoring network in the USA (Schaefer and Paetzold 2000). Observations were started in the mid-1970s and the network consists of more than 700 automatic sensors installed in mid and high elevation areas of the western United States and Alaska to record snowpack, precipitation and temperature, typically at hourly intervals. For the snowpack component, a typical installation includes one or more snow pillows, consisting of a flat circular container filled with non-freezing fluid, and a pressure sensor to record the changes in hydrostatic pressure due to the weight of the snow layer. A downward looking sensor may also be used to monitor snow depth. Some sites also include automatic weather stations, soil moisture and soil temperature sensors. Data transmission is by meteorburst telemetry (see Section 2.3) and power is from battery packs and solar panels. A snow pillow network is also used in Norway, supplemented by satellite and manual observations, to monitor snow conditions for flood forecasting and other applications (Rohr and Husebye 2005).

2.1.2

Remote Sensing

2.1.2.1

Weather Radar

Ground based weather radars use electromagnetic waves to detect precipitation in the form of raindrops, snowflakes and hail (often called hydrometeors). A typical installation consists of a radome, a tower, and buildings housing the computer and generator equipment needed to operate the device. Figure 2.3 shows a typical radar installation from the United Kingdom. There are many books, review papers and guidelines describing the principles of weather radar operation (e.g. Collier 1996; World Meteorological Organisation 2000; Cluckie and Rico-Ramirez 2004; Meischner 2005) and only a few details are provided here. For most types of radar, the beam is rotated about a vertical axis and the type and quantity of precipitation is inferred from the power of the back-scattered energy, whilst the location (distance) is inferred from the time of travel of the signal (the beam is pulsed to provide an interval in which to detect the returned signal). Rainfall intensity is estimated using relationships between drop size density and the power of the received signal. Many different hardware options are available including: ●

Wavelength – attenuation by rainfall reduces with increasing wavelength, but longer wavelength radars require a larger dish and are usually more expensive. In order of increasing wavelength are X-band (3 cm), C-band (5.5 cm) and S-band (10 cm) radars, where typical wavelengths are shown in brackets. Most weather radars use C or S bands, although X-band radars have been used operationally in some applications.

2.1 Meteorological Conditions

29

Fig. 2.3 View of the internal workings of a weather radar, and the Chenies Radar in the UK (© Crown Copyright 2007, the Met Office)

●

●

●

Dual polarisation – use of horizontal and vertical polarisation to help with identifying hydrometeor shapes and hence types (e.g. with larger drops showing more deformation between planes). Multiple beams – to detect vertical variations in reflectivity to assist in correcting radar outputs for gradient and other effects, including the option of multiple level scans. Doppler – to detect the direction of motion of hydrometeors to help in filtering out ground clutter and estimating wind speed and direction.

The power of the reflected signal decreases with the range of the precipitation from the radar due to attenuation by droplets, dust and other factors, and the spread of the beam. The beam may also be transmitted at a positive angle to the horizontal, and may overshoot precipitation at lower levels, including orographic growth of rainfall in hilly regions and evaporation and wind drift/dispersion at low levels. The use of a slight negative beam angle is also used for some weather radars in mountainous regions. However, the beam will eventually either overshoot rainfall due to the curvature of the earth, or encounter terrain causing anomalous reflections. The accuracy of a weather radar therefore decreases with range, so a regional or national radar network is often designed to achieve an acceptable coverage at a reasonable cost, perhaps focused on areas with the highest rainfall or flood related risks. For flood warning and forecasting applications, another option is to use a denser network of low cost short range radars to infill gaps in the main radar network in areas of interest such as major population centres, or as the main component

30

2 Detection

of the radar network; for example, for the Local Area Weather Radar network in Denmark (Pedersen et al. 2007). In addition to considerations of range and beam angle, the outputs from a weather radar may in addition be affected by meteorological factors (e.g. Collier 1996) such as: ● ●

●

●

●

Bright band – due to the beam intersecting the melting layer increasing reflectivity Anaprop – caused by distortion of the beam in strong temperature or humidity gradients, causing the beam to intersect the ground surface causing false returns Ground clutter – intersection of the beam with hills, mountains and other obstacles (e.g. buildings, masts) Hail – causing an increase in the strength of the reflected signal compared to rainfall Drop size distribution – assumptions about typical distributions may be less valid in certain conditions, such as drizzle

Many of these factors can be reduced by post processing of the received signal, including linking into other sources of information such as the outputs from Numerical Weather Prediction models. Areas of research include making use of real time information on rainfall from vertically pointing radars, microwave communication links and disdrometer installations, and using Digital Terrain Models to help in identifying sources of ground clutter. The signals from individual radars are also often combined to produce a so-called composite or mosaic image. Measurements are usually presented on a gridded basis, after being transformed from the original polar coordinates. Many radar systems have a range of sophisticated visualisation and analysis software, for example allowing rainfall estimates to be accumulated at catchment level, sequences of radar images to be animated, and values to be sent to other systems (e.g. flood forecasting systems). Examples of composite images at a continental scale include the outputs from the NEXRAD system of radars in the USA, and the OPERA project, which combines more than 150 radar outputs for countries across Europe (e.g. Harrison et al. 2006). If the raingauge network is of sufficient density and quality, radar estimates of rainfall may also be adjusted to take account of raingauge measurements of rainfall, in an attempt to correct for low level and other effects missed by the radar. The methods used include multi-quadratic, Bayesian and other techniques (e.g. Moore et al. 2004; Todini 2001). In applying these techniques, of course, there are also uncertainties in the accuracy of the raingauge measurements, and in particular how representative they are of the spatial distribution of rainfall. In addition to using raingauges, the radar outputs can also be improved using outputs from Numerical Weather Prediction models, satellite imagery, wind profiles and other sources of information (e.g. lightning detectors), and this type of nowcasting product is described later. The resolution at which radar data is provided depends on the type of signal processing algorithms used, and the distance from the radar. In the UK, for example, values are available at grid lengths of 1, 2 and 5 km, and 5 or 15 minute time intervals, depending on the distance from the radar (the corresponding ranges are up to 50, 100 and 250 km). Figure 2.4 illustrates the appearance of images at these scales for a heavy rainfall event in the south of England.

2.1 Meteorological Conditions

31

Fig. 2.4 Illustration of weather radar images at a resolution of 1, 2 and 5 km (© Crown Copyright 2007, the Met Office)

In the figure, the distance between the towns of Oxford and Watford is approximately 60 km. The figure shows that the degree to which a weather radar can resolve the spatial distribution of rainfall depends on the grid resolution, which in turn depends on the distance of the catchment from the nearest radar. Signal quality may also be influenced by topographic and other effects, particularly in mountainous regions.

32

2 Detection

Radar coverage maps, typically estimated using digital terrain models, and line of sight calculations, can give an indication of the likely resolution and coverage of radar for a given location or catchment.

2.1.2.2

Satellite

Satellite observations offer the potential to provide estimates of rainfall and other parameters (e.g. sea state) for input to river and coastal flood forecasting models, and are routinely assimilated into Numerical Weather Prediction and nowcasting models. Images of cloud cover are also widely used by flood warning services, although the quantitative use of information, such as estimates of rainfall rate, is much less widespread, in part due to the relatively coarse resolution of measurements compared to the scale of some smaller catchments. Some examples of geostationary satellite systems which have been used in flood forecasting applications include: ●

●

Meteosat – operated by Eumetsat and the European Space Agency for weather forecasting including infrared, visible and radiation budget sensors Geostationary Operational Environmental Satellites (GOES) – operated by the National Oceanic and Atmospheric Administration (NOAA) for weather forecasting with multiple sensors

Geostationary satellites maintain a fixed position relative to the earth’s surface, whilst polar orbiting satellites are at lower altitudes (hence with better resolutions) although may only pass overhead a given location every few days. The sensors used vary depending on how long ago the satellite was built, and the primary applications. However, in general, most meteorological and oceanographic satellites are able to monitor cloud cover, the radiation budget (radiation, reflected energy), surface and cloud top temperatures, snow cover and other parameters. The GOES and Meteosat systems form part of the World Meteorological Organisation (WMO) World Weather Watch (WWW) Global Observing System (GOS) programme (World Meteorological Organisation 2003) which also includes a number of other satellites launched as part of national programmes for environmental and meteorological monitoring (e.g. GMS-Japan, METEOR and GOMS – Russian Federation; FY-1 and FY-2 – China). For flood forecasting applications, one technique of interest is the estimation of rainfall intensities from cloud top temperatures, with cooler temperatures indicating cloud tops at greater altitudes, and therefore possibly with greater depths. Observations of cloud temperature relative to surrounding regions, and of cloud morphology, can also be used to help to discriminate between clouds with similar cloud top temperatures but different rainfall producing potential, such as cirrus and cumulonimbus clouds (e.g. Golding 2000; World Meteorological Organisation 2000; Grimes et al. 2003). Methods may use single images (cloud indexing methods), or sequences of images (life history methods) to assess cloud development and movement. These techniques have also been used to estimate rainfall for input to rainfall runoff forecasting models (e.g. Grimes and Diop 2003)

2.1 Meteorological Conditions

33

Passive and active microwave measurements also show potential in estimating rainfall intensity and the soil moisture at land surfaces (e.g. Crow et al. 2004; Love 2006), although the algorithms which are used need to interpret the signals from different types of land surface including open areas, forest, water, ice, snow and urban areas. Rainfall rates at the surface may also be inferred from the radiation received from sources such as liquid water droplets or suspended ice particles. For active systems, similar principles are used to ground based weather radar, and are actively being developed as part of NASA’s Tropical Rainfall Measuring Mission (TRMM) and the planned international Global Precipitation Measurement Mission. Satellites can also be used for monitoring snow cover, and the formation and break up of ice in rivers and lakes, to provide advance warning of likely flooding problems. 2.1.2.3

Other Techniques

Some other remote sensing techniques which have been considered for rainfall detection in flood forecasting applications include: ● ●

Microwave techniques – use of horizontally transmitted beams to detect rainfall Lightning detection – inferring rainfall amounts from lightning activity

Microwave techniques estimate the path averaged rainfall rate from the attenuation in the signal, and could potentially make use of the extensive transmitter networks used by cell phone operators (e.g. Leijnse et al. 2008). For example, as part of the MANTISSA project (Rahimi et al. 2003; Holt et al. 2005), experiments were performed using dual frequency microwave links with path lengths from 9 to 23 km for a catchment in the northwest of England, and results compared with raingauge and weather radar estimates of rainfall. Uncertainties can arise from unknowns such as the drop shape, temperature and size distributions. Lightning detection methods (e.g. Price et al. 2007) aim to provide forecasts from the short term up to a few hours ahead for heavy rainfall linked to thunderstorms. Lightning activity can be monitored remotely at a global level using space and ground based observations, and in principle can be used to track the progression of thunderstorms. Historical rainfall-lightning relationships can be established from past records, and used together with satellite observations and Numerical Weather Prediction models to estimate rainfall intensity in real time. Lightning data is also assimilated into some forms of nowcasting model as described in the next section.

2.1.3

Weather Forecasting

The topic of weather forecasting covers a wide range of numerical, empirical, observational and other techniques, and in operational forecasting the final decisions on the forecasts to issue are often taken based on a combination of these approaches. For flood forecasting and warning applications, the following two approaches are of particular interest:

34 ●

●

2 Detection

Numerical Weather Prediction models – primarily for rainfall forecasts (Quantitative Precipitation Forecasts) for input to rainfall runoff models, and wind fields for input to coastal models, but possibly also for a range of other surface variables which may be calculated (e.g. soil moisture, air temperature). Typical maximum useful lead times can be 3–5 days or more for deterministic forecasts, and up to 10–15 days for ensemble forecasts, although can be considerably less for events such as thunderstorms. Nowcasting systems – which provide short term forecasts based on a combination of weather radar, satellite and other observations and, increasingly, the outputs from Numerical Weather Prediction models.

The distinction between these two approaches is not clear cut, since Numerical Weather Prediction models also make extensive use of observed data from the sea, ground, air and space to initialise model runs via a process called data assimilation. A simple definition here is that a nowcast is a short term forecast, based primarily on weather radar, typically for times of up to 3–6 hours ahead. Seasonal forecasting systems, combining statistical and other modelling approaches, are also increasingly being used, and have been used in forecasting snowmelt, for example (see Chapter 8).

2.1.3.1

Numerical Weather Prediction

Numerical Weather Prediction models form the basis of the forecasting service offered by many meteorological services, and solve approximations to the equations describing mass, momentum and energy transfer in the atmosphere (e.g. World Meteorological Organisation 2000). The equations may be solved over a global domain, or domains limited by horizontal extent. The boundary conditions for the limited area models are then derived from the larger scale models (i.e. the models are nested). The equations are usually solved on a layered grid, with typical horizontal scales of 10–100 km for global scale operational models, and 1–10 km for local models, and up to 100 layers representing vertical development in the atmosphere. Local models may be called local area, mesoscale or storm-scale models, depending on the type and spatial extent of modelling approach adopted. Sub-models may be included for a range of processes, including cloud development and decay, energy and water transfer at the ocean and land surfaces, and interactions with topography and other obstacles. As noted earlier, models are initialised using a process called data assimilation which can be a major undertaking, using measurements taken from raingauges, weather stations, weather radar, lightning detectors, aircraft, ships, wave buoys, radiosondes, satellites, and other sources. Models typically run on a 1, 6 or 12 hourly timestep, and the data assimilation component can often take a significant proportion of the time between model runs.

2.1 Meteorological Conditions

35

Model outputs can include the wind field, rainfall, potential temperatures, specific humidity, surface pressure, evapotranspiration, snow depth, surface and soil temperatures, soil moisture, cloud water and ice, and other variables. Other outputs can include the convective cloud base and cloud top elevations, sea surface roughness, vertical velocities, and other parameters. Results are usually processed further into specific ‘products’ which vary from country to country but may include a general outlook, synoptic charts, surge forecasts, daily forecasts, strong wind warnings, heavy rainfall warnings, flash alerts, and other forms of output tailored to meet each user’s requirements. Other types of output which may be useful in hydrological applications include estimates for soil moisture conditions and snow cover. Due to the intrinsic uncertainties in both the models, and the data assimilation process, it is now standard practice in many meteorological services to use an ensemble forecasting approach, in which the initial model state is perturbed and multiple realisations of model runs are performed to provide an indication of the uncertainty in the forecasts. With current computing power, typically of the order of 10–100 ensemble runs are performed at each time step. In some countries (e.g. the Netherlands), the ensemble outputs may be presented as part of national weather forecasts on television in the form of an estimated range of values for parameters such as air temperature or rainfall, or as probabilities of occurrence. Multi-model techniques are also used, in which the outputs from several models are displayed in a common format to see the variability between different formulations (e.g. Garcia Moya et al. 2006; Rotach et al. 2007). Probabilistic and ensemble forecasts are also increasingly being introduced into flood forecasting applications, and are discussed further in Chapters 5 and 8. For river flood forecasting applications, a key requirement is often to translate the meteorological model outputs to a scale more appropriate to hydrological modelling. Both statistical and dynamical techniques are used (e.g. Rebora et al. 2006; Schaake et al. 2005). Statistical techniques can include multi-fractal cascades, nonlinear autoregressive models, and processes based on the superposition of rainfall cells at different scales (cluster models). Dynamical techniques can include nesting of higher resolution atmospheric models for the catchment or region of interest within models with a coarser resolution but wider spatial extent (e.g. Environment Agency 2007). For large catchments, upscaling may also be required to help to preserve hydrological spatial characteristics over large distances, particularly where there are significant topographic or climatic variations. 2.1.3.2

Nowcasting

The term Nowcasting covers a range of techniques which use spatial extrapolation of current observations of rainfall from weather radar, sometimes guided by or combined with the outputs from Numerical Weather Prediction models. For short

36

2 Detection

lead times, these techniques can perform better than Numerical Weather Prediction models, and their relative simplicity allows more frequent model runs (e.g. every few minutes) and higher model grid resolutions (e.g. 1–10 km). The maximum lead times provided can be several hours, although values of 3–6 hours are often quoted. Nowcasting methods often use the assumption that, if the speed, size and direction of travel of a storm is known at the present time, then the future development can be estimated by extrapolation, at least at short time scales. Methods range from simple extrapolation of current conditions, neglecting possible growth or decay, to techniques using a wide range of sources of information to help to estimate the future evolution of a rainfall event or thunderstorm or tropical cyclone (e.g. Franklin et al. 2003). A more sophisticated approach is to use the outputs from Numerical Weather Prediction models. Forecasts can then be generated by extrapolating the motion of areas of rainfall, using the wind fields and other forecast outputs from these models to guide this response, including allowance for the development and dissipation of rainfall (e.g. Golding 2000; Wilson 2004). Sub-models may be included to forecast the development of convective cells (thunderstorms) using conceptual (life-cycle) models and probabilistic techniques. As with Numerical Weather Prediction models, ensemble and probabilistic approaches are increasingly being used in Nowcasting, with research also considering how seamless ensembles can be generated covering a range of timescales, from nowcasting through to Numerical Weather Prediction and seasonal forecasting. For example, the Short Term Ensemble Prediction System STEPS (e.g. Bowler et al. 2006) recognises the inherent uncertainty in forecasts over a wide range of scales, including the fact that smaller scales are shorter lived and less predictable, and blends extrapolation, stochastic noise and Numerical Weather Prediction model outputs on a hierarchy of scales. The system generates 50 member ensembles of rain rate and accumulation at a 2 km grid, 5 minute resolution to provide forecasts at lead times of typically up to 6 hours ahead.

2.2

River and Coastal Conditions

Near real time measurements of river and tidal levels, wave conditions, and river flows are important in many flood warning and forecasting applications. There is much in common between the techniques used for river and coastal monitoring, although river gauges may be affected by debris and sediment loads, whilst tidal gauges may experience a harsher environment in terms of salinity and wind and wave loading. Various types of instruments are also deployed in the open oceans although are not described here, including free drifting floats, ocean gliders, and ship-borne measurements. For river monitoring, depending on the nature of the catchment, information may also be required on levels in reservoirs and off-line storage reservoirs, on flow depths on floodplains, and for ice conditions, pump settings or flows, borehole levels,

2.2 River and Coastal Conditions

37

and other parameters. For both river and coastal monitoring, additional information may also be required on gate settings (e.g. at reservoirs, or tidal barriers), and the condition of river and sea defences and other key assets, particularly if there is a suspected risk of breaching or overtopping. Some monitoring techniques for these applications are discussed briefly in Chapter 8. The techniques used for river and coastal monitoring are well established (e.g. World Meteorological Organization 1980, 1998; Hershey 1999; Intergovernmental Oceanographic Commission 1994) and only a few key points are presented here, together with some recently developed techniques. It is convenient to categorise techniques as follows: ●

●

● ●

River/tidal level monitoring – shaft encoder (float), pressure transducer, bubbler gauges, downward looking devices, sensor networks, satellite altimetry River flow monitoring – ultrasonic and electromagnetic devices, gauging structures, particle imaging velocimetry Wave monitoring – recording of wave heights, periods etc. Position monitoring – applicable to gates, ice monitoring, flood defences etc.

Position monitoring devices are not described in detail but include shaft encoders, ice motion detectors (e.g. doppler radar, or instruments linked by wire to plates anchored in the ice), and strain gauges. Fixed or panning CCTV and webcams for visible light, low light or infrared are also increasingly used for monitoring locations prone to blockages, ice formation, or other problems, and for estimating parameters such as wave overtopping rates at sea defences. For flood forecasting and warning applications, unless manual observations are used, all devices require a means of translating movement into an electrical signal for data logging, and onward transmission by telemetry. Instruments should also be installed with electronics above the highest likely flood levels. Only instruments suitable for telemetry are described in the following sections.

2.2.1

River/Tidal Level Monitoring

Level monitoring devices record water levels using a range of techniques including: ●

●

Float recorders – a float contained in a stilling well, installed either in a downpipe within the water body, or set into the ground and connected by a horizontal pipe to the river, reservoir or sea. The float moves up and down with water levels, causing the pulley from which it is suspended to rotate, and the rotation is detected electronically by a shaft encoder. Pressure transducers – are typically submerged at the end of a downpipe which acts as a protective conduit for the wire connecting the device to the data logger. The pressure which is recorded depends on the depth of water above the sensor. Pressure sensors have also been used in urban areas to detect flooding on roads, for example.

38 ●

●

●

●

2 Detection

Bubbler gauges – typically release bubbles of inert gas (e.g. nitrogen) or air from an orifice and are supplied from a gas canister or compressor. The pressure required to displace water from the submerged orifice depends on the water pressure and hence the depth of water above the device. Such devices also include a pressure sensor, and are sometimes called pneumatic gauges. Downwards looking devices – include radar, ultrasound and acoustic devices suspended above the water surface by a purpose-made frame, or from existing structures (e.g. bridges, piers), which estimate levels based on the time of travel of transmitted and reflected signals. Although acoustic devices can be operated in the open, usually they are contained within a narrow sounding tube contained within a stilling well (e.g. for tidal monitoring), whilst radar and ultrasound devices do not normally require a protective conduit for the beam. Sensor networks – are a newer technology (pervasive or grid computing) consisting of networks of small low power pressure sensor devices with integral microprocessors and transmitters (e.g. radio or microwave), programmed to collaborate to form networks which can reconfigure automatically if any one sensor fails, and are much cheaper to install and operate than conventional instruments (e.g. Hughes et al. 2006). Satellite altimetry – is widely used for monitoring ocean levels, and shows potential for monitoring of river levels, particularly on large rivers (e.g. Beneviste and Berry 2004; Xu et al. 2004; Zakharova et al. 2005).

For river applications, the time interval for measurements can be set at a value based on the expected rate of rise and fall of river flow hydrographs or reservoir levels, and ideally would provide several values on the rising limb in a flood warning application (although this may not always be practicable in a fast responding river). Figure 2.5 shows a float in stilling well and a pressure transducer installation. The examples are for a river float in stilling well device, and a reservoir pressure transducer installation with radio mast and a staff gauge for manual observations. Each method has its own strengths and limitations. Devices installed below the water surface face the risk of damage by debris during a flood event, or blockage of the equipment by sediment or ice. In some countries, heaters or other forms of protection may be required to ensure operation in ice conditions. For tidal applications, and to a lesser extent reservoirs and lakes, the gauge output may also be affected by wave action. Individual sensor types of course have their own limitations, and may require corrections for drift, temperature effects, density effects and other factors. In rivers, downward looking devices may return ambiguous signals when there is significant debris floating on the water surface (trees etc.). For all types of device, data recording and transmission may be affected by floodwater if the data logger and telemetry electronics are installed at too low a level or in a location prone to erosion or impact by debris. Also, datum values need to be established and regularly checked so that water level measurements are consistent over time and can be related to national datum values. Tidal gauges may also incorporate a datum probe or switch which operates at a known sea level so that datum offsets or errors in the tidal record can be identified.

2.2 River and Coastal Conditions

39

Fig. 2.5 Examples of water level recording devices

Instruments may also need to be able to record over considerable ranges in levels; for example, ranges of 10 m or more are not uncommon in some tidal and river monitoring locations. Also, if water levels can fall below the height of the sensor (e.g. in low flow periods), then the instrument needs to be able to cope with dry conditions, and possibly high air temperatures, and blowing sand. However, in tidal applications, sometimes a gauge is designed to become exposed to the air at low tides to allow the datum or instrument to be checked at regular intervals.

2.2.2

River Flow Monitoring

For flood warning applications, measurements of levels may sometimes be all that is required. Indeed, compact self contained units are available commercially combining a pressure transducer or float recorder with a solar power or battery supply and a direct connection or telemetry link to a warning device (e.g. a bell, siren or cell phone) which is triggered if one or more preset levels is exceeded (see Chapter 4). However, for many river monitoring applications, an estimate of flow is required and, unless a purpose made flow monitoring gauge is installed (see later), values must be obtained by calibration of a stage-discharge relationship or

40

2 Detection

rating curve established based on concurrent measurements of levels and flows or discharge (i.e. spot gaugings). Most hydrometric services have programmes in place for regular gaugings, and often also for high flow measurements during flood events. The technique for measuring flows typically involves taking a series of measurements of depths and velocities across the river, using one or more velocity measurements at each location, and integrating values to estimate the flow. Velocity measuring devices include propeller meters (current meters), and Acoustic Doppler Current Profilers (ADCP). Except at low flows, when wading can be used, current meters must normally be suspended from a bridge, boat or cableway, whilst ADCP devices can be floated across the river surface from the river banks or using towed, manned or radio-controlled floats and boats. On a large river, a single measurement of flow can be a time consuming operation, with potential health and safety and access issues during a flood event, particularly for measurements taken at night and in fast flowing, debris laden water, although remotely controlled winch systems (e.g. Park et al. 2006) are a possible way of helping to avoid these risks. Other less widely used methods include dilution gauging, in which the change in concentration of a tracer such as salt water or dye is recorded along a river reach, and slope-area methods, in which the change in water surface elevation is measured along a reach and the flow estimated from hydraulic formulae. The procedures for establishing a stage discharge relationship are well established (e.g. International Standards Organisation 1996, 1998) and Fig. 2.6 shows a simple example. Stage discharge relationships are often represented by power law equations with parameters obtained from a least squares fit regression to observed values. Look up tables and polynomial functions are also used, although to a lesser extent.

Stage (metres)

10

1

0.1 0.1

1

10 Discharge (cumecs)

Fig. 2.6 A simple example of a stage discharge relationship

100

1000

2.2 River and Coastal Conditions

41

The interpretation of stage discharge relationships can be complicated by a number of factors including: ● Weed growth – seasonal or intermittent growth of weeds causing channel constrictions ● Backwater influences – influences from downstream of the site such as from gate operations, high flows in tributaries etc. ● Tidal influences – tidal influences affecting water levels at the gauge, perhaps only for exceptional tides ● Channel profile changes – changes in the channel bed profile at or near the site due to erosion, scour, sedimentation, dredging etc., and changes in the river cross section as levels rise (e.g. flows going out of bank onto a floodplain or bypassing the station) ● Ice cover – formation of ice constricting river flows to varying degrees at certain times of the year, causing backwater effects for ice downstream, throttling flows for ice cover next to the instrument, and with a range of effects for ice cover upstream ● Hysteresis – differences in flow values for rising river levels and falling river levels These and other factors can cause curves to change with river depth, season and over time, leading to multiple equations valid only for given periods or seasons. Also, errors in the high flow end of stage-discharge relationships are perhaps one of the main sources of uncertainty in flood forecasting models. However, the stage discharge approach is probably the most widely used method internationally for estimating flows in rivers. Some techniques for extending curves include hydraulic modeling (1D, 2D or 3D), slope-area methods, based on peak water levels estimated from photographs, maximum level recorders, or from trash left after flood events, and velocity-area methods, based on direct survey of the river cross section area, and extrapolation or estimates of velocity. Given the uncertainties in estimating river flows from levels alone, various techniques have also been developed to provide a more direct measurement of flows, although are often significantly more expensive in terms of initial capital costs. The main types of device include: ●

●

Ultrasonic devices – which measure flows at one or more depths by the travel times of ultrasound waves between senders and receivers set at angles (typically in the range 30–60 degrees) to the main river flow. The average velocity at each depth can be estimated from the difference in time of travel between pulses with an upstream and downstream component, and can be integrated to provide an estimate of overall flow. Electromagnetic devices – which record the electromotive force induced by water flowing over a coil buried in the river bed, which is notionally proportional to the water velocity. The signal is detected by electrodes in the river banks, and additional probes may be required to allow corrections to be made from other electromagnetic sources near to the station.

42 ●

2 Detection

Gauging structures – purpose made structures designed or adapted for flow measurements, including various types of weirs, flumes and other structures, with estimates of flow obtained from levels recorded at one or more prescribed locations in the structure and a corresponding theoretical formula.

For all three techniques, it is usually necessary to make a number of spot gauging measurements when the instrument is first installed to check or establish the calibration. Also, for gauging stations in particular, but also for ultrasonic and electromagnetic devices, occasional spot gaugings are often made during routine operation to check for any drift in the calibration. All three types of device can be affected by the problems of ice, algae, weed growth, sediment and damage from debris, although electromagnetic instruments are less affected by weed growth. In addition, for gauging structures, the usual assumption is that the depth at the structure is controlled at the structure, and is independent of levels downstream. However, under high flow conditions, structures may drown out, so that the theoretical relationship no longer applies. One approach to estimating flows at structures where this occurs is to install a second level recorder downstream and to use theoretical or modelled relationships to estimate flows in these conditions. To provide better sensitivity to changes in depth, some structures also include changes in channel cross section, with additional channels becoming effective at higher flows, and only a narrow channel in operation at low flows. V notch weirs also achieve a similar effect. In flood warning and forecasting applications, another option is sometimes to calibrate an existing structure, built for other applications (e.g. navigation, irrigation), if it provides a stable control on water levels. One newer technique which shows some promise is the use of automated computer analysis of video camera images of existing tracers (e.g. foam, flotsam) on the water surface (e.g. Creutin et al. 2003). The technique, Particle Image Velocimetry, gives an estimate of surface flow velocities, which can be related to overall flows either using standard formulae or previous current meter measurements. The method relies upon suitable tracers being present on the water surface, and can be affected by shadows and reflections; however, it offers the promise of being able to estimate flows at a low cost from remote locations. Similar trials have also been performed using hand held or bridge mounted radar devices.

2.2.3

Wave Monitoring

In flood warning applications, estimates of wave height and direction are useful to assist in deciding whether to issue coastal flood warnings, and for input to coastal flood forecasting models. Typical wave periods might only be a few seconds, so the information is usually recorded over intervals of a few minutes or more and expressed in terms of the spectral properties of the wave field, from which key statistics such as significant and maximum wave heights, dominant and average wave period and direction, wave spread, mean water level, wind speed and direction, and wave spectra, can be estimated. The output across a number of wave monitoring locations provides a spatial picture of wave distributions.

2.2 River and Coastal Conditions

43

The main techniques for monitoring waves include (Massel 1996; World Meteorological Organisation 1998): ● ● ●

Measurements from below the sea surface Measurements at the sea surface Measurements from above the sea surface

For sub-sea devices, the signal can be transmitted by cable to the shore, or to a nearby buoy to be transmitted by radio or satellite. Pressure transducers are the most widely used method, with the pressure at the instrument varying with wave height. The resulting spectrum is then corrected for hydrodynamic attenuation with depth. However, depth corrections start to be of a similar magnitude to typical wave pressure signals for water depths much beyond 10–15 m, and also tend to filter out higher frequency signals, limiting the depths at which pressure transducers can be used (e.g. World Meteorological Organisation 1998). Vertically pointing echo sounders can also be used, although the signal may be affected by bubbles from breaking waves. Measurements at the sea surface are typically made from wave buoys, in which the vertical acceleration is measured using an accelerometer mounted on a gyroscopically stabilised platform, although solid state techniques are increasingly used. Wave heights can then be inferred from the acceleration terms. Motion can also be monitored in the two horizontal planes (roll and pitch) to provide spectral estimates of wave direction. Some devices also use Global Positioning Systems (GPS), and solid state inertial motion sensors which provide combined values for surge, sway, heave, roll, pitch, yaw and heading. Telemetry can consist of radio or satellite links. Lightships can also be used, in which the accelerometer output is combined with pressure sensor outputs to detect horizontal motions. A ship provides a more stable platform, although is less sensitive to smaller waves, whilst a buoy needs to be carefully installed so that the mooring does not influence the motion significantly. These methods are more appropriate for deep water, and shallow water techniques include capacitance probes and resistance probes, which can be mounted on structures such as piers or platforms. These devices consist of a series of sensors along a board (wave staff), where the signal depends on the depth of wave immersion, although can be affected by breaking waves. Devices which use ultrasonic or electromagnetic velocity meters can also be used to measure the two horizontal components of wave orbital velocity which, in conjunction with a pressure recorder or capacitance or resistance probe, can provide useful directional information. Downward looking devices of the types described earlier, such as laser, infrared, and acoustic range finding devices, can also be used to monitor waves if a suitable platform is available, although can be affected by reflections and other influences from the structure. For model calibration, satellite based estimates of long term wave state can be derived using synthetic aperture radar and other spaceborne instruments, although for polar orbiting satellites observations at a given location are only made once every orbit. Shore based high frequency radar also provides a method for monitoring wave states and sea surface currents over large areas.

44

2 Detection

2.3

Instrumentation Networks

Flood warning and forecasting systems usually rely on a network of meteorological, river and/or coastal instruments. Individual types of instrumentation may also be combined; for example, an automatic weather station may be installed on a wave buoy, or a raingauge at a river gauging station. Monitoring networks can also serve a range of purposes in addition to flood warning and forecasting, such as water resources monitoring, marine forecasting, and climate change monitoring, requiring a compromise between these different applications. For example, a water resources gauge may be installed close to a river confluence to monitor the entire runoff from a catchment but, at high flows, suffer from backwater influences from the main river, possibly making it unsuitable for use in a flood forecasting application. For new sites, issues of site permissions, power supply, access for installation and maintenance, and other factors may lead to gauges being installed in locations that are not ideal. The requirements for telemetry connections may also influence the locations at which gauges are installed. The following sections discuss some options available for telemetry of real time information, and give a brief introduction to the design of networks for flood warning applications. Chapter 11 also discusses some of the economic considerations in network design, and in choosing an appropriate solution tailored to the level of flood risk.

2.3.1

Telemetry Systems

For telemetry of real time data, the following options are widely used in flood warning and forecasting applications (e.g. World Meteorological Organisation 1994b): ●

●

●

●

●

●

Telephone lines (PSTN) – connections via land-lines using the public switched telephone network. Each instrument has a unique telephone number which can be dialled to retrieve data or check the condition of the instrument. Mobile telephone (GSM, GRPS) – similar to PSTN lines but using cell phone technology. Radio – Ultra High Frequency (UHF) or Very High Frequency (VHF) communication links. Satellite – transmission from the instrument to an orbiting or geostationary satellite for relay to a ground station. Meteorburst – use of naturally occurring ionisation in the atmosphere left by meteor trails to reflect radio waves between a base station and outstation. Meteor impacts are sufficiently frequent that reasonable data transfer rates can be achieved. Internet – broadband, Ethernet and wireless connections.

Each approach has advantages and limitations and Table 2.2 provides some examples of these considerations. Other considerations can include power consumption, licence requirements, and purchase and installation costs. Some generic examples of use of these techniques include:

2.3 Instrumentation Networks

45

Table 2.2 Examples of strengths and possible limitations in telemetry methods Method

Strengths

Possible limitations

Telephone, broadband, local wireless ethernet

Uses an existing network

May incur connection and usage charges Requires a reliable public network Land lines and exchanges can be damaged by flooding and high winds etc. if not designed to avoid these problems May incur connection and usage charges Possible data drop-outs in heavy rainfall Networks can be affected by power cuts during flood events User needs to establish and maintain the network (equipment, permissions, licences etc.)

Simple to set up and operate

Cell phone

Uses an existing network Simple to set up and operate

Radio

Probably no connection charges other than radio licence fees once the network is established User retains full control of the network

Satellite

Instruments can be installed anywhere visible to the satellite No requirement to establish a network

Meteorburst

No requirement to establish a network Signals can be transmitted over long ranges

●

●

Line of sight required for transmission possibly requiring repeater stations, or limiting the range for coastal applications May be affected by interference May incur data transmission charges Possibly no suitable satellite visible Transmission may be restricted to the time of overpass (for orbiting satellites) or transmission time slots determined by the operator Relatively high power transmitter required Possible delays whilst waiting for suitable transmission conditions

WHYCOS – a World Meteorological Organisation initiative to install river and climate monitoring stations on the main rivers worldwide, which is being developed as a series of regional projects. Instruments typically use Data Collection Platforms (DCPs) transmitting via the Meteosat satellite system and other systems (World Meteorological Organisation 2005). ALERT (Automated Local Evaluation in Real Time) – a set of radio based communication protocols, sensing technologies and data formats which is widely used in the USA and elsewhere for locally operated flood warning systems incorporating raingauges, river level and other sensors (NOAA/NWS 1997).

There are also many national and regional examples of applications of these techniques; for example, in the United Kingdom, the public switched telephone network

46

2 Detection

(PSTN) is used almost exclusively for data links to river and raingauge instrumentation whilst, in the USA, the Meteorburst system is used for transmitting data from the snow monitoring SNOTEL network described in Section 2.1. Manual systems, in which levels are relayed by telephone or radio, are still widely used in some countries. For the limitations which are listed, many potential solutions have been devised by suppliers, and can work well in some situations. Also, careful design can eliminate some problems. Networks consisting of more than one type of telemetry link are also an option if no single method is appropriate, or if backup transmission routes are required at each instrument in case of failure of any one method (e.g. radio backed up by cell phone). Interfaces may also be required to locally operated systems, such as the SCADA (Supervisory Control and Data Acquisition) systems which are sometimes used at reservoirs, hydropower schemes and other control structures. The connections to individual instruments typically consist of a data logger, to keep a record of values which can be downloaded at each visit, and a modem, to translate the signal into a form suitable for transmission by telemetry. The logger and data link may allow for multiple sensors, as with an automatic weather station, for example. Additional channels may also be used for sensors internal to the instrument or the logger/modem housing to monitor the status and environmental conditions of the instrument; for example, battery or solar panel condition, air temperature, and humidity, and sometimes a GPS unit for time and location information (e.g. for satellite telemetry). Telemetry connections can be bi-directional or one way only. Simplex connections are links in which the instrument sends packets of data at predefined times, or when a critical threshold is exceeded, whilst duplex connections allow downloading of data on demand. A duplex system provides the flexibility to increase the sampling (polling) rate of instruments when required (e.g. as a flood starts to develop) and also allows the operational status of the instrument to be checked remotely. By contrast, simplex systems are cheaper to install and operate, although with the risk of communications clashing between instruments if they transmit at the same time. Overall control of a telemetry network is typically from one or more central computers which will periodically poll, or update, values from the network. Modern data gathering systems typically include a wide range of functionality including: ● ●

● ● ● ● ●

●

Interfaces to a range of data sources and systems Map based displays of instrument locations and data values, and spatial data (e.g. weather radar data) Summary presentations of instrument status and data returned Data validation facilities (possibly) Report and graph generation facilities Alarm Handling options with transmission of alerts by email, SMS, fax etc. A wide range of options for onward transfer of data (e.g. to a flood forecasting model, automated dialling system, or permanent database system) Database options for short term on-line storage of data, and longer term off-line storage of data

2.3 Instrumentation Networks

47

Alarms can include rainfall depth duration values, river level thresholds, tidal level thresholds, and other types of threshold (see Chapter 3). Some systems may also be programmable, so that simple flood forecasting models such as level to level correlations can be operated on the telemetry system as a back up to the main forecasting system. Also, multicriteria alarms and rules might be included (e.g. IF X > Y AND Z > A THEN…). In control rooms, large wall mounted ‘mimic panels’ can help with providing an overview of current system status against a backdrop of key information, such as reservoir locations, towns and catchment boundaries, although are increasingly being replaced by computer displays. A hydrometeorological database is usually either an integral component of the system, or may be operated alongside the system as a long term repository for the near real time data. Many such systems are available commercially or have been developed by national hydrological and meteorological services, and the functionality might typically include: ● ●

● ●

●

●

Database summary options including key metadata for individual stations Statistical reporting functionality (e.g. hydrological year books, extreme value statistics) Data validation tools for checking, correcting and infilling erroneous data values A wide range of map based, reporting and graphical options for display and printing of data A range of data conversion options (e.g. from river levels to flows, or hourly values to daily values) Possibly a range of data analysis options (e.g. for stage discharge relationships, flood frequency analysis)

For database and telemetry systems, most modern systems provide options to facilitate the exchange of spatial and time series data through agreed data formats (e.g. XML) and metadata standards.

2.3.2

Network Design

The topic of network design for river flood warning and forecasting applications is covered in a number of guidelines, manuals and papers (for example, World Meteorological Organisation 1994a, 1998; USACE 1996; NOAA/NWS 1997; Environment Agency 2002, 2004; Sene et al. 2006), although recommendations can be specific to local meteorological conditions (desert, mountain, tropical etc.), the types of flooding mechanisms experienced, and other catchment and coastline specific factors. Some general issues to consider in network design include: ● ●

The accuracy, reliability and lead time requirements for flood warnings The likely performance of any new or existing instrumentation under flooding conditions

48 ●

●

2 Detection

The reliability of any existing or proposed telemetry links under flooding conditions The level of flood risk at the location, or locations, for which flood warnings are required

Here, flooding conditions can include high river or tidal levels, and the associated high winds and heavy rainfall which often accompany flood events, and an assessment of likely performance under these conditions usually forms part of the design study (e.g. is the instrument range sufficient to monitor all likely conditions, and are the electronics above likely maximum flooding thresholds). Backup power units and lightning conductors may also be needed and, in cold climates, heaters may be needed to ensure operation in snow or ice. The lead time requirement for flood warning can also influence network design. For example, for river monitoring sites, to assess local conditions, ideally an instrument would be installed at or near the location for which flood warnings are required. However, typical rates of rise of river levels in flood events may be so fast that the flood warning threshold level would have to be set to a low value to achieve a useful lead time, causing too many false alarms. Some ways of extending the lead time would therefore be to install a gauge further upstream, or to develop a forecasting model to the original proposed gauge location. Both methods introduce some uncertainty into the flood warning process, and both approaches might be used to help to reduce that uncertainty, possibly also using data assimilation and a probabilistic approach for the forecasting component, as described in later chapters. Similarly, for coastal locations, a tide gauge may be available at or near the location of interest, but if, for example, several hours of advance warning are needed to evacuate properties or to operate a tidal barrier, then locations further afield would need to be considered (or additional instruments installed), probably combined with use of surge forecasting models. Offshore monitoring also provides early warning of deep swell and Tsunami events not linked to local storms. For locations with complex wave and surge patterns (e.g. some harbours, and coastal reaches), on site monitoring is often the only way to resolve these effects. For raingauges, the flood warning or forecasting requirement may be simply to give an idea of rainfall in the general area, or to provide estimates of catchment rainfall or rainfall distribution for lumped or distributed rainfall runoff models. If the raingauge density is insufficient, then additional raingauges might be installed, including gauges in nearby river catchments. Existing alternatives, such as weather radar, might also be considered, if the coverage and accuracy is sufficient in the locations of interest. Given that major operational decisions may be taken based on the data provided, the issue of reliability (or resilience) is also important, and often one or more backup instruments may be identified in case of failure of any one instrument during a flood event. For a river monitoring gauge, that might be a gauge further upstream, whilst for a tide gauge another gauge might be selected from the same coastal reach. Backup instruments might also be installed at the same site or nearby locations, particularly in high risk locations (e.g. city centres).

2.3 Instrumentation Networks

49

A number of techniques can assist with network design including: ●

●

●

●

Digital terrain models – for radio path or line of sight studies, for estimating catchment characteristics (area, slopes, elevations etc.), and for viewing potential instrument locations against a backdrop of topography, flood risk locations, and other factors Hydraulic and hydrological analyses – to study the likely response of the catchment or coastal reach at the proposed instrument locations (e.g. rate of rise of levels for typical events, typical depth-duration values for rainfall, times of travel of flood or surge waves from distant locations, possible backwater and confluence influences etc.) Meteorological analyses – using historical raingauge data, weather radar data and possibly Numerical Weather Prediction model outputs to help in developing an understanding of flood generating conditions, with the likely scale, speed, and direction of storms all being important factors in deciding on appropriate raingauge locations Temporary gauges – installation of temporary gauges, maybe without telemetry, to investigate river or coastal characteristics at potential sites, and to check site security and feasibility (e.g. risk of vandalism, objections from nearby residents etc.)

More generally, it is often worthwhile considering other current or planned applications of the data; for example, for other purposes (e.g. water resources monitoring, ocean climate monitoring, port and harbour operations), or for providing flood warnings to additional locations. For example, considerable cost savings can sometimes be realised by considering opportunities to share data between departments or organisations, or by finding alternative nearby site locations which would serve more than one purpose. Another consideration, particularly for flood forecasting applications, is the level of uncertainty which can be tolerated. For example, for river forecasting models, it is often not economically feasible to place raingauges and river gauges in all major subcatchments, with the result that some inflows to the model (lateral inflows) will need to be estimated, introducing a source of uncertainty into the process. Also, it might be desirable to install more raingauges to obtain a better idea of rainfall distribution in and around the catchment, and to upgrade gauging stations so that they are better able to record accurate values at high flows. These various trade-offs and compromises are all part of the process of network design, and in part are one of the motivations for the increasing interest in using probabilistic and ensemble techniques to help to quantify the uncertainty (see Chapters 5–8 for examples).

Chapter 3

Thresholds

Flood warning thresholds define the meteorological, river and coastal conditions at which decisions are taken to issue flood warnings, whilst flooding thresholds are the values at which flooding occurs. Normally, a flood warning threshold will be set to achieve an acceptable lead time before the flooding threshold is reached, or may be time based (as with tropical cyclones, for example). Alternative names for flood warning thresholds include triggers, criteria, warning levels, critical conditions, alert levels and alarms, and sometimes a range of values will be required as warnings are escalated from advisories (or watches, or pre-warnings) through to full warnings. Threshold values may be set based upon experience or analysis of historical data, or using conceptual, data based or process based modelling studies. Values may be fixed (static) for all flood events, or dynamic, varying depending on how each event unfolds. This chapter describes a range of techniques ranging from simple fixed flood warning thresholds through to probabilistic approaches, together with several examples of approaches to performance monitoring of thresholds.

3.1

Rainfall Thresholds

Observations or forecasts of heavy rainfall often provide the first indication of likely river flooding. Some typical uses of rainfall thresholds are for the initial mobilisation of staff (e.g. opening an incident room), and moving to an increased frequency of monitoring river conditions and operation of flood forecasting models. Rainfall values can be obtained from observations (e.g. raingauges, weather radar, satellite) or forecasts (e.g. nowcasts, Numerical Weather Prediction models), with observed values usually providing higher accuracy, but with a shorter lead time before the onset of flooding. Best practice is to calibrate methods directly to the type of input data (or forecasts) to be used operationally, to account for any systematic or other differences between rainfall measurement and estimation techniques. Rainfall amounts can also be used directly to initiate flood warnings although, due to the various uncertainties in how rainfall translates into river flows (see later), this approach is used much less widely, with a greater risk of a high false alarm rate compared to warnings based on river levels. K. Sene, Flood Warning, Forecasting and Emergency Response, © Springer Science + Business Media B.V. 2008

51

52

3 Thresholds

For some types of rainfall inputs, such as weather radar observations, or rainfall forecasts, rainfall values will usually be available on a gridded basis, so that the criteria for raising or displaying an alarm might apply to a single grid square, or to the average value across a region or in a river catchment. Information on rainfall amounts and accumulations can also be presented spatially; for example, as maps of rainfall amounts with overlays of catchment boundaries, rivers, topography and flood risk locations, and in terms of probabilities of exceedance (if using ensemble rainfall forecasts). Spatial estimates for rainfall distribution can also be derived for raingauges, if required, using the techniques described in Chapter 2. Rainfall threshold (or alarm) criteria are often expressed in terms of the quantity (depth) of rainfall in a given period (duration) which has the potential to cause flooding. A range of depth-duration values may be used; for example, an alarm might be raised if rainfall is forecast to exceed 25 mm in any 3 hour period, or 40 mm in any 6 hour period. Alternatively, thresholds may be expressed in rainfall frequency terms calculated from a statistical analysis of historical records, such as the 1% or 10% exceedance probability, or the 1 in 100 year or 1 in 10 year return period. Values can be tested by analysis of long term historical rainfall records; for example, by comparing the number of alarms which would have been raised compared to the number of actual flooding events (or near misses), and estimating the number of false alarms which would have occurred (see Section 3.3). Of course, rainfall values alone do not provide a full indication of flooding potential, since the catchment characteristics (topography, land use etc.), current catchment state (e.g. soil moisture, snow cover) and other factors (e.g. reservoir levels) may also influence the magnitude and timing of flooding. Rainfall thresholds are therefore often combined with indicators of catchment response and the current catchment state. Table 3.1 illustrates a simple approach of this type, and is adapted from one of several examples in Environment Agency (2002). Here, the codes and locations are for three hypothetical Flood Warning Areas, and the Forecast Rainfall values could be either best estimates, or worst-case scenarios. The depth/duration pairs (e.g. 20/6) are in units of mm of rainfall, and hours, and Table 3.1 Illustration of rainfall alarm criteria (Adapted from Environment Agency 2002; © Environment Agency copyright and/or database right 2008. All rights reserved) Flood Watch Criteria based on SMD

Forecast Rainfall (mm) Code and location

6 12 18 hours hours hours

24 hours

SMD (mm) <5

FW021 Bridgetown

3.8

5.8

8.8

10.8

39.0

FW022 Southford

3.0

5.0

8.0

10.0

50.3

FW023 Northtown

3.0

5.0

8.0

10.0

50.8

20/6 25/12 24/6 30/12 24/6 30/12

5–20

21–40

>40

25/6 30/12 28/6 35/12 28/6 35/12

30/6 40/12 32/6 40/12 32/6 40/12

30/6 45/18 35/6 45/18 35/6 45/18

3.1 Rainfall Thresholds

53

catchment conditions are expressed in terms of the soil moisture deficit (SMD), which is the depth of rainfall which would be required to bring the catchment to saturated conditions (i.e. the amount of water required to bring the soil to field capacity). Some other possible indicators for catchment conditions (e.g. World Meteorological Organisation 1994; USACE 1996) include recent rainfall, current river flows, Catchment Wetness Index, Base Flow Index, Antecedent Precipitation Index, and borehole levels. Where, as is often the case, direct observations are not available, values are often computed from the soil moisture accounting component of rainfall runoff models (see Chapter 6) or as a secondary (diagnostic) output from the land-atmosphere component of Numerical Weather Prediction models (e.g. Cox et al. 1999). Satellite based methods also show potential for remote sensing of soil moisture. Another approach to setting thresholds is to use a catchment rainfall runoff and flow routing model to explore the rainfall amounts required to achieve flooding for a range of durations and catchment initial conditions. One approach is to first derive a typical storm profile from historical data, describing the variation in rainfall during the course of an event. These values are then scaled by magnitude and duration, and the resulting synthetic storms used as input to the catchment model. For each duration, the depth required to reach flooding thresholds is noted, perhaps for a range of catchment conditions, and the resulting table of values can then be used as the basis for estimating the rainfall thresholds for that location. Other factors, such as reservoir drawdown at the start of an event, or the depth of water in off-line storage areas, might also be considered in setting thresholds. Some possible criteria for flooding thresholds include bank full flows, peak river levels exceeding a threshold at which flooding commences, or flood flows of a given probability (return period). The latter method is often used for ungauged catchments and in ensemble forecasting approaches (see Chapter 5). For example, these types of method form the basis of the Flash Flood Guidance concept (FFG) developed by the National Weather Service in the USA (Sweeney 1992). Flash Flood Guidance is defined as the amount of rainfall of a given duration over a small basin needed to create minor flooding (bank full) conditions at the outlet of the basin. The approach has been used operationally since the 1970s and was integrated into a system called the Flash Flood Guidance System in 1992, and has more recently been considered for providing early alerts for debris flows (NOAA-USGS 2005). Chapter 8.2 provides some examples of international initiatives using this approach. In the original version of the method, threshold values of runoff were estimated based on the outputs from a lumped rainfall runoff modelling approach. More recent developments (National Weather Service 2003) have included improvements to the method for areas of the country where rainfall intensity and land characteristics have more influence on flash flooding than soil moisture (e.g. some desert regions), and the introduction of a distributed (grid based) hydrological modelling approach for estimating thresholds and for real time soil moisture accounting. Operationally, estimates of rainfall depths and durations (e.g. 1, 3, 6, 12 and 24 hours) are compared with the threshold values appropriate to the estimated soil moisture conditions. The resulting exceedance over threshold values can then be mapped.

54

3 Thresholds

Spatial analysis tools are also available to examine rainfall accumulations, rainfall intensities and guidance values at point or catchment scale. Some other developments in the area of rainfall threshold based approaches (e.g. Martina et al. 2006; Georgakakos 2005, 2006; Reed et al. 2006; Collier 2007; Fouchier et al. 2007) have considered or implemented systems which include various permutations of the following techniques: ●

●

●

●

●

Bayesian techniques requiring optimisation of a utility function combining the perception of stakeholders, historical losses, and perhaps the losses from false alarms Alternative soil accounting approaches using a variety of conceptual and processbased catchment models Artificial neural network methods which improve forecasting skill by ‘learning’ from meteorological and streamflow response Thresholds based on the return periods/recurrence intervals of model flows based on long term simulations using historical or synthetic rainfall data Development of indicators of flash flood potential which can be searched in real time including soil moisture, channel constriction/debris risks, storm depthduration, direction and velocity

Ensemble approaches are also increasingly being used, in which the probability of rainfall is displayed on graphs, maps and tables and compared to probability based thresholds. Risk-based approaches, combining probability and consequence, can also be used, and Box 3.1 provides an example of an operational system in the Netherlands which uses ensemble forecasts of rainfall to provide rainfall alarms to assist with water management operations in polder regions. Chapter 5 provides further information on ensemble forecasting techniques and Chapter 10 gives more background on risk-based and cost loss approaches to decision making. In addition to the use of rainfall thresholds, various other meteorological indicators have been considered for use in providing early warning of flooding, with an emphasis on probabilistic techniques. One of the earliest methods was a combined deterministic/stochastic approach which was developed for application in the Mediterranean areas of France (Obled and Datin 1997; Bernard 2004). Observations and forecasts of rainfall and other parameters at lead times of 2–3 days or more are linked to an archive of rainfall and other parameters for past events; for example, geopotential or temperatures. The technique can also be used at shorter lead times, using stochastic modelling to link observations up to time now with likely future scenarios (again based on an historical archive), with the option of conditioning forecasts on nowcasts and likely limits on daily rainfall for the catchment for the type of storm being observed. Another approach is to use operational mesoscale and other Numerical Weather Prediction models to monitor parameters which are thought to be good precursors of flooding, including: ● Potential vorticity – as an indicator of atmospheric stability ● Convective Available Potential Energy (CAPE) – an indicator of the energy available for a storm to develop

3.1 Rainfall Thresholds

55

Box 3.1 The KNMI precipitation alert system The Netherlands is a low lying country with extensive areas of reclaimed land, known as polders, and more than half of the population lives in areas below sea level. The main flood risk in the polder areas arises from heavy rainfall falling directly on the polders, often combined with high river levels or sea levels (due to surge, wind and wave action), which may limit the ability to remove excess water. A network of pumping and drainage systems is used to manage water levels in polders, and rainfall observations and forecasts play an important role in optimising these operations. Rainfall alarms are used to mobilise staff, alert third parties, obtain emergency pumping equipment (if required), and to help in deciding when to start pumping operations. Since 2003, as part of a collaborative project with the Union of Water Boards, the Royal Netherlands Meteorological Institute (KNMI) has been issuing probabilistic rainfall alarms to selected Water Boards based on ensemble forecasts and observed rainfall data. For lead times up to 36 hours ahead, deterministic forecasts are used from the national HIRLAM Numerical Weather Prediction model whilst, at longer lead times, 50 member ensemble outputs are obtained from the European Centre for Medium-Range Weather Forecasts (ECMWF) Values for rainfall depth and duration are estimated for each polder based on weather radar observations of rainfall in the previous 5 days, and rainfall forecasts for the next 9 days (Fig. 3.1). Critical rainfall depth-duration values are defined by each Water Board on the basis of historical rainfall records and flooding histories and can depend on soil type, the proportion of urban areas, storage capacity, pumping capacity, the time of year, and other more subjective

Fig. 3.1 Example of a 9 day probabilistic rainfall forecast (KNMI)

(continued)

56

3 Thresholds

Box 3.1 (continued) factors. The probabilities at which alarms are issued are calculated from a risk assessment which compares the cost of mitigating actions (pumping etc.) with the estimated losses (damages) if no action is taken. The so-called cost loss ratio gives an indication of the appropriate probability thresholds to use for each polder, which can be refined based on experience. Thus the criteria for issuing warnings are defined in terms of a risk profile for each polder consisting of a range of depth, duration and probability combinations (e.g. 50 mm in 72 hours with a 33% exceedance probability). Over a number of events, consistent use of these profiles should help to minimise the economic impacts from flooding. The criteria are checked automatically on an hourly basis and, when individual values are exceeded, the relevant Water Board managers receive an email alert or text message, and the alert is also published on a secure website. A user group meets at regular intervals to share experience on use of the system, and in particular to discuss approaches to the setting and verification of appropriate risk profiles, since this is a new and developing application for ensemble forecasting. Planned developments include use of radar rainfall nowcasts for shorter term rainfall forecasts, and use of wind and tide forecasts, since wind effects can add to the flood risk in larger polders via wave and local surge effects. Reference: Kok C J, Vogelezang D H P, Wichers Schreur B, Holleman I Description and use of the automated warning system for the Dutch water boards, KNMI.

● ●

Precipitable water – or precipitable water vapour Intensity and direction of the low level flow

These indicators can be calculated from wind, moisture, pressure and other fields (e.g. Collier et al. 2005; Environment Agency 2007). These developments are often linked to the general trend to develop higher resolution (storm scale) models better able to forecast the development of convective and other events. Lightning activity has also been considered as a possible indicator of the likelihood of heavy rainfall and flash flooding (e.g. Price et al. 2007).

3.2 3.2.1

River and Coastal Thresholds Introduction

River and tidal thresholds (or triggers) are a key component in many flood warning systems, and define the levels and possibly other variables (e.g. wind speed and direction) at which the decision to issue a flood warning should be taken, or other actions initiated

3.2 River and Coastal Thresholds

57

(e.g. mobilisation of staff, more frequent monitoring). They are sometimes called Action Thresholds. Some systems may also automatically issue a warning at these levels without any human intervention (e.g. using email, sirens or pagers) although there are many issues to consider in taking an automated approach of this type; for example, the likely false alarm rate, and the possibility of missed warnings (see Chapter 4). Other types of thresholds can include parameters such as ice motion and river flows. Time based criteria may also be used in some situations; for example, the time before landfall for hurricanes, typhoons and tropical cyclones (see Chapter 9 for further discussion of this topic). Observations are normally made by telemetry but, where this is not available, or extra safeguards are required, observers and patrols may be deployed on site, or other methods such as CCTV or webcams used. Community representatives may also monitor conditions in some flood warning schemes. On site observations can be particularly useful where site specific flood risks can occur, such as defences breaching, waves overtopping at sea defences, or bridges being blocked by debris, and as additional backup for high risk locations such as town centres. Threshold values are normally defined based on a combination of experience, analysis of historical data, and possibly detailed hydraulic and other modelling of river or coastal response. Values are usually chosen to achieve the required warning lead time, without causing an unacceptable number of false alarms and, for instruments at the location of flooding, are set in relation to the flooding threshold, as illustrated in Fig. 3.2 for the case of a river level threshold. Here, the flooding threshold is the gauge reading at which flooding impacts begin (and for which a warning is required), such as property flooding, or flooding of roads, or overtopping of flood defences (levees), and is sometimes called a Result Threshold. In practice, the actual warning lead time will be less than the potential

River Level Flooding Threshold

Flood Warning Threshold

Warning Lead Time

Time

Fig. 3.2 Illustration of a flood warning threshold for an at site gauge

58

3 Thresholds

value indicated in the figure since factors such as decision times, and flood warning dissemination times, must be accounted for, as described in Chapters 4, 5 and 10. Also, it is advisable to include some allowance (contingency) in the setting of values to allow for uncertainty in data, models and event specific factors. The terminology and approaches used vary between organisations and countries, but some typical types of threshold (or trigger) include: ●

●

●

At site or local values – where the flood warning is issued based on values at or near the location for which the flood warning is required Upstream or remote values – where the flood warning is issued based on values at a site further upstream in the river network, or further offshore or around the coast in the case of coastal triggers, to provide additional lead time at the site of interest Forecast values – where the flood warning is issued based on the output from a river or coastal forecasting model for the site or other location of interest

For each type of threshold, there is a trade off between the accuracy, reliability and timeliness which can be achieved; for example, if a threshold is lowered, this normally increases lead time, but may also increase false alarm rates (e.g. USACE 1996), whilst forecast values may be set at a higher threshold (e.g. a flooding or result threshold) than for at site values due to the additional lead time available from model outputs. To provide additional lead time and resilience, a site may have more than one type of threshold, with warnings being issued on the basis of exceedance of any one value, or other permutations. Values may also be nominated as the primary, secondary (or backup) or failsafe threshold, with the choice depending on the relative performance of each type of threshold. Within each category of threshold, there may also be a range of values for different operational and warning conditions. For example, a site might have standby or alarm values which are set at a low level for early warning of possible events, and mobilisation of staff, and a range of flood warning values to escalate the severity of the warning as river or sea levels rise. Also, as flood hazard mapping techniques improve, and flood warning dissemination systems become more sophisticated, it is increasingly becoming possible to target warnings to smaller areas, or even to individual properties, with the advantages of reducing the number of false alarms experienced by property owners, and allowing for a more phased approach to warning and evacuation of properties. If this approach is used, then each zone or sub-area will have its own warning threshold level, both for the At-Site gauge (if available) and for any Remote gauges or Forecast values. Of course, the terminology and formats used for flood warning procedures, and the criteria for escalating and downgrading alerts, differ widely between organisations but the general principle of escalation of warnings, followed by confirmation that the threat has passed, is widespread. Figure 3.3 provides an example of this general approach for a river flood warning application. In this hypothetical example, the Flood Warning Area at Newtown is divided into four sub-areas or zones, identified by codes FW001 to FW004. The corresponding warning threshold levels are shown for the At-Site gauge, and the Remote gauge would also have its own set of values (not shown). If a forecasting model output is available, that too would have a set of values based on a consideration of flooding thresholds,

3.2 River and Coastal Thresholds

59

FW004 FW003 FW002 FW001

R

IV

ER

Remote Gauge

At Site Gauge

FW001 – Riverside Paths

FW002 – Riverfront Apartments FW003 – Town Centre FW004 – Power Station

To Bridgeham Fig. 3.3 Illustration of at site and upstream thresholds (not to scale)

model lead times, and other factors. Table 3.2 provides a simplified illustration of how these values might be implemented into a set of operational Flood Warning Procedures for the At-Site gauge (sometimes called an Action Table or Flood Intelligence Card), although it is important to note that the details of warning messages and operational responsibilities differ widely between countries and organisations. Values are expressed in terms of gauge readings, but absolute values might also be included, relative to a national datum level. Other thresholds (sometimes called Information Thresholds) might also be included to indicate other useful information, such as the highest level recorded at the site, and peak levels for historical flood events. A similar table would also be produced for the Remote Gauge, with a separate set of values. As illustrated, the gauge at Newtown might also be a Remote gauge for another Flood Warning Area further downstream, and an example is included for the town of Bridgeham. Following the initial standby alarm, a series of flood warnings is issued as the event escalates, and operational instructions are also issued where this requires direct contact with other organisations or the public. As noted in Chapter 4, this allows an audit trail of actions and decisions to be maintained during the event, including any departures from the agreed approach as each warning level is exceeded (for example, based on other information which may be available, such

60

3 Thresholds

Table 3.2 Illustration of flood warning thresholds for the example in Fig. 3.3 Observed level (m) Forecast level (m) Action required 3.2

>3.8

3.5

>4.0

3.8

>4.2

4.0

>4.4

4.2

>4.5

3.1

<2.5

STANDBY ALARM Issue FLOOD WATCH: WW001 – Newtown and Bridgeham Issue OPERATIONAL INSTRUCTION: OP001 – Loud Hailer Patrol crew standby for Newtown Record confirmation of receipt of FLOOD WATCH: WW001 from Newtown Town Council, Power Station at Newtown, Bridgeham City Council Issue FLOOD WARNING: FW001 – riverside paths at Newtown Issue FLOOD WARNING: FW002 – riverside apartments at Newtown Issue OPERATIONAL INSTRUCTION: OP002 – start patrols along flood defences in Newtown Town Centre, assign liaison officer to Newtown Police Emergency Command Centre Issue FLOOD WARNING: FW003 – Newtown Town Centre Issue OPERATIONAL INSTRUCTION: OP003 – Loud Hailer Patrol in Newtown Town Centre Issue FLOOD WARNING FW004 – power station at Newtown Issue FLOOD WARNING FW011 – riverside properties in Bridgeham Issue ALL CLEAR FW001 – Riverside Paths at Newtown FW002 – Riverside Apartments at Newtown FW003 – Newtown City Centre FW004 – Power Station at Newtown FW011 – Riverside Properties at Bridgeham

as information from operational staff observing river levels on site). Finally, an All Clear is issued when the river levels have dropped back to standby levels, and the forecast indicates that levels will continue to drop. In this example, a forecasting model output is also available, and the Forecast Thresholds are included in the procedures. The extent to which the Forecast Thresholds are formally integrated will depend on organisational policy, the confidence in model outputs, and the expertise and background of duty officers. Some examples of the way that the forecast outputs could be used include: 1. Issue the warning either if the observed value is exceeded, or if the forecast value is exceeded 2. Issue the warning if both the observed and forecast values are exceeded 3. Consider issuing the warning if the observed value is exceeded, using the forecast outputs to take the final decision 4. Generate warnings to individual properties or groups of properties from real time forecasts of the inundation extent

3.2 River and Coastal Thresholds

61

Other possibilities might also be envisaged. A maximum forecast lead time (horizon) or minimum observed level might also be specified, with forecasts at longer lead times or for lower levels given less weight or not considered at all due to the increase in uncertainty with increasing lead time. The first approach takes advantage of the potential additional lead time from forecasts, but raises the prospect of more false alarms if the model overestimates levels. The second approach helps to guard against the risk of a model providing erroneous outputs, but would possibly lead to less lead time in warnings, and could result in missed warnings if the forecasting model underestimates levels. The third approach relies mainly on the forecasting model outputs, with the observed value acting as an initial alert level above which issuing a warning should be considered. Another possibility is to introduce the concept of soft and hard limits, or a contingency for uncertainty, in which there is a range of levels in which the duty officer can provide an input to the decision making process, but once the hard limit is reached a warning must be issued. This again increases the number of decision criteria, although some forecasting and telemetry systems can help in automating application of this approach. The fourth approach requires a forecasting model which is able to estimate flood inundation extents in real time, such as a one-dimensional or two-dimensional hydraulic model. The resulting extents can then be intersected with maps of property locations, and lists of property addresses and contact details generated for a range of forecast lead times. In principle, these can then be used to automatically generate warnings to individual properties (e.g. by telephone or cell phone) once the forecast has been approved. However, this is a new and developing area, and the issue of confidence in the model outputs again needs to be carefully considered before implementing this approach. Another factor to consider in deciding on the degree of automation is the worst-case scenario of a widespread flood event. For a small number of locations, it may be practicable for a duty officer to inspect every output (observed levels, and forecast values, if available) and take a decision based on experience and judgment. However, in a major event, many hundreds of threshold levels may be exceeded during the course of the event, and duty staff will have less time to consider the accuracy or appropriateness of each value, except possibly in high risk locations, where major decisions need to be taken (e.g. on evacuating population centres, or closing down key utilities). Similar considerations apply for very fast responding catchments or coastal reaches.

3.2.2

Simple Forecasting Techniques

Simple forecasting techniques provide an alternative or supplement to observed thresholds, and typically use information from remote sites to estimate conditions at the site of interest, or information on the rate of rise at the site itself. This category of methods is sometimes called a ‘Simple Triggers’ approach (e.g. Graham and Johnson 2007), and may introduce dynamic thresholds, which vary between events, rather than fixed values.

62

3 Thresholds

The distinction between these methods, and forecasting models of the type described in Chapters 5–8, is not clear cut, and threshold based techniques, as described in the previous section, can also be viewed as a simple type of forecasting, based on assumptions about typical rates of rise and travel times for flood events. Here, the distinction is taken to be: ● ●

●

The methods do not attempt to model physical processes. The methods can easily be applied using non-computerised approaches such as graphs, look up tables and charts (although can be computerised if required). Where a computerised approach is needed, the calculations are simple enough to perform on a telemetry system, if required, rather than a dedicated flood forecasting system.

The advantage of using paper based techniques is that the methods can be applied by staff without computer skills, and are quick and cheap to implement. Also, where more sophisticated techniques are available, the methods can be used as a backup in case of system failure (e.g. due to power cuts), and as a cross check on the plausibility of the outputs from more advanced techniques. For example, an observer on site might relay observations of river levels by telephone or hand held radio for a duty officer to use, even if both the telemetry and forecasting systems have failed. Similarly, if the methods are implemented on a telemetry system, then this can provide additional backup and resilience to the flood forecasting system, and modern telemetry systems are often capable of running simple types of model. Simple forecasting techniques include correlations (single and multiple regressions), multicriteria approaches (e.g. look-up tables, carpet plots, nomograms), transformation matrices, rate of rise triggers, and time of travel (isochrone) maps.

3.2.2.1

Correlations

Correlations are widely used in flood warning applications, and relate parameters at the location where an estimate is required to real time observations or forecasts at one or more remote locations. For rivers, correlations are usually performed in terms of river levels or flows, and can be calculated using either peak values for a number of representative historical events, or values for the full flow range. Peak to peak correlations are sometimes called Crest Stage forecasts, and Fig. 3.4 shows two examples (World Meteorological Organisation 1994). The first figure shows a peak level to level correlation between two river stations, whilst the second shows a set of correlations in which the choice of relationship depends on current runoff conditions. Other secondary variables may be used including snow cover, air temperature, soil moisture, and time of year. Whole hydrograph correlations are usually estimated assuming a typical lag time between upstream and downstream stations. An optimum value can also be estimated by repeating the calculation for a range of assumed lag times and finding the value which gives the smallest correlation coefficient. Correlations are normally performed in

3.2 River and Coastal Thresholds

63

Fig. 3.4 Examples of level to level and flow to flow correlations (Reproduced from the WMO Guide to Hydrological Practices – Data Acquisition and Processing, Analysis, Forecasting and Other Applications, courtesy of WMO)

terms of levels since this avoids the need for a rating equation or stage discharge relationship at each site to convert levels to flows (see Chapter 2). However, levels at an individual gauge can be affected by other influences, such as backwater effects from tributaries, gate operations or tidal influences, which can introduce additional uncertainty into the relationship. One of the risks in this approach is also that these effects may only become apparent in large flood events, beyond the range of calibration. Flow based correlations help to avoid this problem, but as noted introduce the uncertainties arising from stage discharge relationships (where used), and can only be applied between sites for which flow estimates are available. Correlations can also be affected by inflows, storage and losses between the two stations being used (e.g. tributaries, or spills onto floodplains, or lakes and reservoirs), and by the extent and motion of storm events over the catchment. One option in this situation is to use a multiple regression approach to include other records in the relationship, such as gauges further upstream, and on tributaries or floodplains (e.g. Torfs 2004), or lake or reservoir levels. Another consideration is whether a correlation derived from peak values should be applied over the full flow range to derive a forecast hydrograph. This approach is often used operationally and sometimes works well, although it is important to note that lag times and wave speeds vary over the flow range, so that the shape of the rising limb of the hydrograph can be considerably in error, which is important if the errors are close to flood warning threshold values. Correlations can also be used in tidal applications, with examples including single or multiple regressions for the following situations, often including a time delay factor between the observed or forecast point and the location for which the estimate is required: ●

●

Estuaries – relating levels at a point in the estuary to tidal conditions and/or river levels upstream Tide Gauges (observations) – relating levels at a tide gauge to conditions at gauges further around the coast; for example, to estimate the likely development of surge events

64 ●

3 Thresholds

Tide Gauges (forecasts) – relating offshore forecasts of level from a surge model at one or more node points to conditions at the shoreline

Additional parameters which may be included are wind speed and direction, and possibly wave heights. Relationships are often expressed in the form of look up tables advising on when to issue warnings for various combinations of actual or forecast tide levels, wind speed, wind direction, maximum surge, and maximum wave heights. Some other ways of presenting information include equations, nomograms or carpet plots. For example, Fig. 3.5 shows an example of a flood warning plotting chart for an estuary location in North West England where water levels are influenced by both fluvial flows and tidal levels (although in this case the graph represents the theory behind the model and has now been replaced by automated forecasting techniques, including use of tidal forecasts). The chart gives estimates of levels at Lancaster Quay based on forecasts for tidal levels at Fleetwood Dock and river flows at Caton. For both rivers and coastlines, hydrodynamic and other numerical models can also be used to guide the development of correlations and other types of threshold, such as time based thresholds. Multiple model runs can be performed for a wide range of scenarios to explore the key influences on flood response at the location(s) of interest, possibly deriving a range of relationships for different initial conditions

Fig. 3.5 Example of a flood warning plotting chart for a coastal location (Environment Agency 2004, © Environment Agency copyright and/or database right 2008. All rights reserved)

3.2 River and Coastal Thresholds

65

for snow cover, reservoir storage and other factors, and examining response for levels beyond the range of the historical data. In some cases, a considerable amount of exploratory work may be required to determine the appropriate combinations of variables to use in real time. This approach is often used for generating operational look up charts and tables to assist with the operation of tidal barriers, for example. As another example, transformation matrices derived from detailed off-line scenario modelling for wave transformation and overtopping are used operationally in coastal flood warning procedures for some locations around the coast of England and Wales (Environment Agency 2004a). Models of this type can also be implemented in real time, if the model run time is short enough, and the model stability and convergence is acceptable, and Chapters 6 and 7 discuss this topic in more detail.

3.2.2.2

Time of Travel Maps

For river catchments, time of travel or isochrone maps (e.g. World Meteorological Organisation 1994) can be a useful aid in flood warning applications, and show estimated or typical travel times from the onset, centroid or peak of a rainfall event, to the peak of flows being observed at various points in the river network, or travel times between locations in the network. Values can be presented in the form of tables, graphs or as shaded or contour maps of equal travel times. Figure 3.6 shows an example of a time of travel map based on times to the lowermost point in the catchment (World Meteorological Organisation 1994).

Fig. 3.6 Example of an isochrone map (Reproduced from the WMO Guide to Hydrological Practices – Data Acquisition and Processing, Analysis, Forecasting and Other Applications, courtesy of WMO)

66

3 Thresholds

In this case, the lines of equal time (isochrones) were estimated assuming an average velocity of flow in the river channels, whilst some other methods for estimating the time of travel include: ● Analysis of historical rainfall and river level and flow data ● Area based methods using empirical overland flow models to estimate velocities ● Unit hydrograph rainfall runoff modelling techniques ● Catchment models combining rainfall runoff, flow routing and/or hydrodynamic modelling techniques Values can also be estimated using Geographical Information Systems, with other options for presenting information including overlays of catchment boundaries, gauging stations, flood risk locations etc., and annotations giving information on catchment response times and peak flows for historical flood events. If the estimates are based on historical data, then a range of representative events needs to be selected, and mean or median values for lag times derived across all events, or individual values derived for different types of event (e.g. snowmelt, floodplain flows etc.). For modelling based approaches, assumptions also need to be made about the magnitude, distribution, speed and direction of the storm events which are used. For all types of analysis, it may be advisable to also consider the scenarios likely to give the smallest lag times in a catchment to indicate the worst case for flood warning applications. Although mainly used for river flood warning, this technique can also be useful in coastal applications to give an idea of the timing of the peak in astronomical tides around a coastline, and the typical movement of surge peaks.

3.2.2.3

Persistence Methods

Another simple forecasting technique is to assume that some aspects of the current observed response will persist into the future. Some approaches which have been applied operationally include: ●

●

Rate of rise methods – which extrapolate the rate of rise of a hydrograph or tidal levels as levels rise towards threshold values. Typical or fastest likely rate of rise values can be estimated from historical data and/or hydrodynamic modelling results for a number of events and then used operationally to forecast the likely time of crossing of thresholds. Rate of rise values can also be estimated dynamically during an event, in which case parameters which can be varied to optimise the performance of the model include the averaging time over which the rate of rise is calculated, and the required lead time. An optimisation can then be performed to maximise the success rate of warnings, and minimise the number of false alarms (e.g. Graham and Johnson 2007). Similar techniques can also be applied to reservoirs, in which different categories of warning are issued based on the current water surface elevation, and the rate of rise of levels. Constant offset – in which a constant value is applied to observed data to compensate for event specific factors. For example, Tissot et al. (2005) compare a simple persistence based method with a range of more sophisticated modelling

3.3 Performance Monitoring

67

techniques in which the differences between observed tidal levels, and the estimated tidal harmonics, are assumed to persist for the duration of the forecast.

3.3

Performance Monitoring

Threshold based approaches to flood warning are widely used and may be progressively improved using experience gained over successive flood events. For example, for a river gauging station, if post event analysis shows that a warning was issued too late at a site, then the warning threshold might be lowered to allow more lead time, although possibly at the expense of an increase in false alarm rates. Similarly, if a threshold is resulting in too many false alarms, then after careful analysis the value might be increased, provided that this does not increase the risk of missing actual events. Alternatively, more accurate approaches might be investigated, such as development of a flood forecasting model. As part of the development of a flood warning service, it is usual to review the performance of flood warning thresholds on a regular basis, and after each major flood event, and when other changes occur which may influence performance (e.g. flood defence construction work, dredging, instrument replacements, changes in forecasting models etc.). Flooding thresholds should also be regularly reviewed, although it is less likely that they will need adjusting. However, some examples of when this might be required include when a flood defence is raised or repaired changing the level at which it is likely to be overtopped, and if additional properties require adding to the warning system. As with all components of an operational flood warning system, any changes to thresholds and alarms should be fully tested and documented before implementation, and discussed with key stakeholders who may be affected by those changes. A wide range of methods can be used for monitoring the performance of thresholds. Similar techniques can also be used at the design stage of a flood warning scheme to explore how values would have performed based on the historical data available to date. In practice, the values used are often a compromise between the need for an adequate warning lead time, to avoid missing flood events, and to minimise false alarms rates. Values for warning lead times can be estimated by examining historical records to determine the time difference between crossing of the warning and flooding thresholds. Note that this time is not the same as the lead time provided to recipients of flood warnings which, as noted earlier, may include additional time delays; for example the time taken for decision making, or in issuing a warning. Estimates of these actual lead times are more difficult to obtain, but methods which are used include post event surveys of people who were flooded, and examination of the records (logs) maintained during the event by flood warning duty officers, the emergency services and others (e.g. for the times of phone conversations, and for the estimated time of onset of flooding). One simple way to present information on lead times is as a histogram showing the lead time performance across a number of events. Values might also be tabulated by gauge, Flood Warning Area, or catchment or coastal reach, as illustrated in Table 3.3 for a single gauge across a number of flood events (adapted from an example in Environment Agency 2002).

68

3 Thresholds

Table 3.3 Example of a lead time summary for several flood warning areas (Adapted from Environment Agency 2002; © Environment Agency copyright and/or database right 2008. All rights reserved)

Flood warning area FW001 – riverside paths at Newtown FW002 – riverside apartments at Newtown FW003 – Newtown City Centre FW004 – power station at Newtown FW005 – riverside properties at Bridgeham

After start <2 of flood hours

2–4 hours

4–6 hours

6+ hours

Modal value (hours)

Target (hours)

0

0

7

4

0

2–4

2

1

3

2

0

0

<2

2

0

0

2

0

0

2–4

3

0

0

0

1

0

4–6

4

2

0

1

0

0

After

2

Table 3.4 Simple 2 x 2 contingency table for flood warning threshold evaluation Flooding threshold exceeded Flood warning threshold exceeded

Yes No

Yes

No

A C

B D

In this hypothetical example, which is based on Fig. 3.4, the warning lead times for FW001, FW003 and FW004 were satisfactory for all events, but were late for one event at FW002, and below the target value for three events. Also, for FW005, on two occasions the warning threshold was not reached until after flooding started at Bridgeham. This might indicate the need to adjust the value for FW002, and possibly for a new approach for FW005. An alternative way of examining performance is using a contingency table approach as shown in Table 3.4 for the case of a river or tidal level gauge. The example uses information on the crossing of flood warning and flooding thresholds but, as described earlier, could also be extended to include an evaluation of the dissemination component of the system, based on information obtained from post event surveys and incident logs (e.g. was a warning received? was your property flooded?). Based on this table, a number of parameters (categorical statistics) can be defined, including the following three performance statistics which are widely used in flood warning and forecasting verification studies: ● ● ●

Probability of Detection (POD) = A/(A + C) False Alarm Ratio (FAR) = B/(A + B) Critical Success Index (CSI) = A/(A + B + C)

The CSI parameter is sometimes called a ‘Threat Score’. These statistics can be accumulated across a number of flood events at a single site, or a number of sites,

3.3 Performance Monitoring

69

Table 3.5 Example of a verification analysis for raingauge data Raingauge 1

Raingauge 2

Year

Flood warning issued

10 mm in 3 hours

15 mm in 6 20 mm in hours 12 hours

10 mm in 3 hours

15 mm in 6 hours

20 mm in 12 hours

1998 1999 2000 2001 2002 2003 2004 2005 2006 2007

1 2 2 1 1 2 1 2 1 1

5 8 8 3 6 6 6 8 6 4

5 5 5 4 5 3 5 6 2 5

11 10 8

15 6 10

11 4 7

2 5 3 4 2 5 3 5 2 3

and give an indication of overall performance, and are discussed further in Chapter 5. Histograms or cumulative frequency plots can also be produced as a guide to the thresholds which provide the best compromise between probability of detection and the number of false alarms. Similar techniques can also be used for rainfall threshold values; for example, Table 3.5 shows a hypothetical analysis to assist with verification, or setting, of rainfall thresholds for two raingauges, based on analysis of 10 years of historical rainfall data. The Flood Warning column indicates the number of flood warnings issued each year at this frequent flooding location, which of course may not always indicate that flooding actually occurred. For Raingauge 1, the analysis indicates that the False Alarm Ratio is high for the shorter duration thresholds, but is of the order 50–75% for the 12 hour duration, which may be acceptable for applications such as providing an initial alert to duty officers of the need to start monitoring rainfall and river levels more closely. For the newer gauge (Raingauge 2), false alarm rates are approximately twice those of the other gauge, so the depth-duration thresholds could possibly be adjusted, or perhaps the gauge is not representative of rainfall in the catchment. Additional checks would also be required to confirm that the successful alarms are linked to the same rainfall events which led to the flood warnings. For evaluation of flood warning performance, it can also be useful to introduce the concept of a ‘near miss’ (e.g. Environment Agency 2004b), in which levels are within some defined tolerance of the threshold value. For example, if flood warnings are being provided for an area behind a flood defence, then the tolerance might be set equal to (or to some factor of) the design freeboard, on the basis that any level within that tolerance is a cause for concern. Some additional examples of approaches to verification are presented in Chapter 5 for the case of evaluation of flood forecasting model outputs, and in some cases these methods might also be used for verification of warning thresholds. It is also worth noting that many of the ideas used in flood warning verification have been developed from other fields, in particular meteorology, for which the science of

70

3 Thresholds

forecast verification is perhaps better established (e.g. Stanski et al. 1989; Jolliffe and Stephenson 2003). One outcome of performance monitoring may be that a decision is taken either that a more sophisticated approach is needed (e.g. a flood forecasting model, or additional instrumentation), or that the level of service hoped for cannot be achieved in practice with current budgets or technology. For reporting at organisational, regional or national level, it may also be useful to aggregate performance statistics across large numbers of Flood Warning Areas, which in turn can be used as a basis for deciding on future investment and other requirements to improve performance. Chapter 11 discusses some of these issues in terms of the economic performance of an overall flood warning, forecasting and emergency response system.

Chapter 4

Dissemination

Although organisational structures can differ widely between countries, a regional or national flood warning service typically has a wide range of responsibilities, which can include monitoring meteorological, river and coastal conditions, development and operation of flood forecasting models, and dissemination of flood warnings to the emergency services, local authorities and the public. Other responsibilities may include operation of control structures to mitigate flooding impacts, assisting with or coordinating the emergency response (evacuation, sandbags etc.), and contributing to post event assessments. These various activities may be performed within an overall framework of flood warning targets and performance monitoring, so that the lessons learned from each flood event guide future investments and technological improvements. This chapter discusses some of these organisational and procedural aspects to providing a flood warning service, and gives an overview of techniques for disseminating flood warnings and for implementing a flood warning system. Later chapters describe how the flood warning service fits into the wider emergency response to a flood event, which can potentially involve participants from many different organisations.

4.1 4.1.1

Flood Warning Procedures Introduction

Flood warning procedures define the actions that flood warning staff should take as a flood event develops. Some reasons for establishing clearly defined procedures include: ●

●

●

During the pressure of a major flood event, there may be little time available for analysis and discussion regarding whether to issue individual flood warnings Given that floods can occur any time of day or night, less experienced staff may be on duty, and need clear guidance on the actions to take If procedures are not available, or do not cover all likely eventualities, vital actions may be overlooked

K. Sene, Flood Warning, Forecasting and Emergency Response, © Springer Science + Business Media B.V. 2008

71

72 ●

4 Dissemination

Increasingly, there a need for organisations to maintain an audit trail of actions during a flood event, and to perform detailed performance analyses and post event reviews after the event

The format of Flood Warning Procedures varies widely between organisations and can range from short documents or charts through to detailed manuals and computerised decision support systems. Procedures can cover just a single location, through to large numbers of flood warning areas in a region, and Table 4.1 illustrates some of the topics which may be covered.

Table 4.1 Some typical items covered in flood warning procedures Items

Description

Detection

Procedures for routine monitoring and telemetry of meteorological, river and coastal conditions against predefined alert criteria (thresholds), and for other forms of monitoring such as CCTV or webcams and control gate settings (Chapter 2) Actions to take when monitoring a potential flood incident (e.g. opening the incident room, calling in additional staff) Summaries of the threshold values and other criteria under which flood warnings should be issued and other actions taken (Chapter 3) Details on who flood warnings should be provided to, and by what means, including operating instructions The operation of flood forecasting models and interpretation of the model outputs (Chapters 5–8) Guidance on operation of key systems, equipment and other facilities, and backup plans in case of failure Requirements for recording information on warnings issued, actions taken, maintenance of communication logs, reporting etc., and external liaison with the public, media and other organizations both during and following the event Other actions which may be useful to perform during and after a flood event if possible to help with post event analysis and reporting (e.g. aerial photography, high flow calibration of equipment, surveys of flood extent and properties affected etc.) Guidance on safe working near water, and dealing with emergencies For failure of any key aspect of the flood warning system, including the need to relocate the incident room if there is a threat of flooding affecting access, escape or equipment Names, addresses, phone numbers for key individuals from various organisations, including representatives for vulnerable people or communities at risk from flooding (hospitals, care homes, the elderly etc.)

Pre-warning activities

Action or flood intelligence tables

Dissemination Flood forecasting Systems Reporting

Flood Event Recording

Health and safety Contingency plans

Contacts

4.1 Flood Warning Procedures

73

Different sets of procedures may also be available for different types of flood event; for example, river flooding and coastal flooding. Threshold values for individual sites are often a key component of a flood warning system and are described in more detail in Chapter 3. They summarise the conditions under which flooding may occur, and the meteorological, river or coastal conditions for issuing warnings or for operational response (if applicable). Some examples of operational response can include: ● ● ● ● ● ●

●

Vehicle or foot patrols of flood defences at risk from breach or overtopping Initiation of sandbagging, raising of temporary defences and barriers etc. Visual observations of gauge board levels at rivers, reservoirs, coastal reaches Operation of sluice gates, tidal gates, and other flow control structures On site (visual) verification of closure of canal gates, flap gates etc. On site (visual) checks for blockages by debris at culverts, bridges and other structures, and clearance of blockages (if possible) Operation of pumps, flow diversion structures etc.

Depending on the organisational structure, there may also be a requirement for staff to coordinate the on site dissemination of warnings by loud hailer, door knocking, portable sirens etc., although in some countries that task may be performed by local authorities, the police, community representatives, or other groups. Some additional information which may be included in procedures includes photographs of the sites described, safe access routes under normal conditions and for various flooding scenarios, descriptions of flooding mechanisms at specific locations, and detailed instructions on the operation of structures. A single Flood Warning Manual can cover many different sites, each with its own set of Action Tables, and so can be a lengthy document. In developing procedures, if resources are available, it is also useful to test them regularly using table top or full scale response exercises. As described in Chapter 9, a table top exercise attempts to mimic the decision making processes and pressures which occur during a real flood event, and may make use of computer generated visualisations, simulated television news reports, and other items to add to the realism. The coordinator will introduce a range of scenarios and complications during the course of the exercise following a timeline for the event.

4.1.2

Flood Warning Areas

A major task in developing flood warning procedures is often to define the districts or properties for which warnings will be provided. Locations can be identified from consultations, street maps, and site visits, making use of flood risk maps as well if these are available (see Chapter 1). However, if a map based approach is used, then an additional step is to convert these results into operationally useful units for providing flood warnings. For example, the flood outlines derived from modelling studies may cut through individual properties or groups of properties (e.g. an industrial site or hospital

74

4 Dissemination

Road 1 in 100 year (1%) flood outline

Sub Area C Road 1 in 50 year (2%) flood outline Town boundary

Sub Area A Sub Area B

Fig. 4.1 Example of defining the extent of a Flood Warning Area from flood risk mapping outputs

grounds), or indicate that access and escape routes to some properties might be cut by flood water, even though the properties themselves are not at risk. Also, there may be advantages in extending areas so that warnings go to identifiable groups of people, or people outside the area which might be flooded, if they are in a position to help those at risk (e.g. the elderly, hard of hearing etc.). Issues like these often need to be discussed with other participants in the flood warning process and with community representatives, and the flooding outlines may need to be adjusted based on those discussions. Also, as described in Chapter 3, with the increasing sophistication of warning dissemination techniques, another option is to subdivide areas based on the probability of flooding, so that flood warnings are progressively provided to more and more people as levels rise. This subdivision can be both linear (along the river or coastal reach) and lateral (moving laterally away from the river or coastline). Each sub area would then have its own set of threshold values to allow warnings to be extended to more properties as river levels rise. For example, Fig. 4.1 shows a simple example in which the flood risk outlines, estimated from hydraulic modelling, are used as a guide to the development of a Flood Warning Area for a hypothetical town called Newtown. The following three zones or sub areas are established in the Flood Warning Area: ● ● ●

Sub Area A – Riverside paths, sports ground and road at Newtown Sub Area B – Town Centre and Southside District at Newtown Sub Area C – Northside District at Newtown

In this example, the sub areas are extended in some places outside the main flood risk zones to cover complete communities for which access may be affected, or for which there may be a history of flooding not represented in the hydraulic model. Also, for Sub Area B, rather than escalate warnings to a small number of properties in the 1–2% risk band, all properties to the south of the river are combined into one sub area. There are of course many issues to consider in defining the extent of warning areas, including whether property owners in areas not at direct risk would welcome information on potential flooding (or not), and national policy on this issue. A possible extension of this method is to derive estimates of the probability of flooding in real time, where the probabilistic component arises from the uncertainties in observations and forecasts, and other unknowns. The resulting probabilistic flood outlines can then be combined with consequence to give a measure of risk

4.1 Flood Warning Procedures

75

(probability x consequence), opening the way to a risk-based approach to issuing flood warnings. This is a new and developing area which is discussed further in Chapters 5 and 10, and users with differing risk tolerances might wish to be warned at different levels; for example, utility operators might require warnings at a low probability so that staff can be mobilised and contingency planning started. By contrast, some property owners might wish to avoid false alarms and so only require warnings when the likelihood of flooding is better defined.

4.1.3

Organisational Issues

The organisation of a flood warning service varies widely between countries and, depending on the scale of the overall system, duties might include some or all of the following activities: ●

●

●

●

●

●

Detection – design, installation and operation of rainfall, river level, reservoir, tidal level, wind, wave, and other monitoring equipment (e.g. for snow cover, soil moisture) Design – design of flood warning schemes, including contributing to decisions on who should receive flood warnings, setting flood warning thresholds, deciding how warnings should be disseminated, and under what circumstances Dissemination – monitoring measurements and forecasts against thresholds, and issuing warnings following agreed procedures, and public awareness activities Operational – taking actions to mitigate flooding, such as patrols, channel clearance, operation of river control structures, or installing temporary barriers Management – general management activities including defining staff rotas, procurement, performance monitoring and reporting, research and development etc. Forecasting – development and operation of flood forecasting models to provide estimates of river levels, river flows, tide levels, wave overtopping etc.

Of course, some of these tasks might be unnecessary for a small-scale community based system, where the primary needs are for detection and the dissemination of warnings. However, for a regional or national flood warning service, most of these tasks will usually be necessary, although some might be shared with other organisations. For example Box 4.1 provides an introduction to the flood warning service operated by the Environment Agency for England and Wales. A common example of shared responsibilities is a separation between the meteorological service, which provides weather forecasts, and the organisation responsible for operating flood forecasting models and issuing flood warnings (although in some countries these functions are combined). Another example is a split in the responsibilities for issuing warnings; for example, responsibility may only extend to issuing warnings to other organisations such as local authorities or the police, or may extend to issuing warnings directly to the public. Other approaches can also be found, as in the United States for example, where, in addition to providing warnings at a national scale, the National Weather Service works in partnership with many smaller scale ALERT, IFLOWS and other local flood warning systems (NOAA/ National Weather Service 1997).

76

4 Dissemination

Box 4.1 The flood warning service in England and Wales (Environment Agency) Recent estimates suggest that in England and Wales approximately five million people in two million properties are at risk from flooding from a 1% (1 in 100 years) event, including almost 400,000 businesses. More than 70% of properties currently receive a flood warning service, with a target lead time for warnings of at least 2 hours, where this is technically feasible. The flood warning service in England and Wales is operated by the Environment Agency, whose responsibilities include installation and operation of raingauges, river gauges, tide gauges, and other instrumentation, the development and operation of river and coastal flood forecasting models, the implementation of flood warning schemes, monitoring weather radar outputs, and issuing flood warnings to local authorities, the emergency services and the public. The Environment Agency also has many wider responsibilities; for example, in flood defence, and water resources. Flood warnings (Fig. 4.2) are issued by more than 20 local offices supported by flood forecasts provided by eight regional offices, and meteorological forecasts (rainfall, wind, surge etc.) from the UK Meteorological Office. Some limited real time information from other organisations, such as water companies and canal operators, is also available and used in the flood forecasting and warning process.

Fig. 4.2 Flood Warning codes in England and Wales (Environment Agency, © Environment Agency copyright and/or database right 2008. All rights reserved)

(continued)

4.1 Flood Warning Procedures

77

Box 4.1 (continued) Flood warnings are issued using an automated dissemination system called Floodline Warnings Direct. This allows alerts to be issued by email, Internet, text messages, fax, telephone and cell phone, and includes text-tospeech conversion software for phone based methods. All messages are available in a range of languages. The system has the capability to target warnings to small groups of properties and individuals, and is supplemented by a range of community-based methods which, depending on location, can include sirens, flood wardens, loud hailers and door knocking. Television, radio, and Internet based approaches are also used, including the Environment Agency’s web based Floodline service (http://www.environment-agency. gov.uk/subjects/flood/floodwarning/) Flood Warning Areas are defined using a community based approach and, in some high risk locations, are divided into zones so that the number of properties warned can be increased as the extent of flood inundation increases. Individually worded messages may also be issued to key contacts in the emergency services, local authorities, utilities, and communities. Public awareness activities, media training, and community and inter-agency liaison all have a high profile between and during flood events, with local approaches guided by targets and procedures established within the national Flood Awareness Campaign. Several methods are used to provide early warning of potential flooding, including heavy rainfall, daily rainfall, flash warning, and surge tide forecasts from the UK Meteorological Office, and catchment-based alarm values set on weather radar and raingauge observations and forecasts. River and coastal conditions are monitored against a range of pre-defined threshold levels and, if these are exceeded, a warning will normally be issued. Flood forecasting models assist throughout this process and, where the model outputs are known to be reliable, are formally integrated into flood warning procedures. Operational threshold levels are also used to initiate operational response, such as establishing foot patrols in high risk areas, raising temporary defences (barriers, sandbags etc.), and operating river and coastal control structures to help to mitigate or avoid flooding.

4.1.4

Control Rooms

For a flood warning authority, monitoring of rainfall, river and coastal conditions is usually performed from one or more control rooms. Typically these are equipped with computers to monitor rainfall, river and tidal conditions, and the outputs from forecasting models, together with telephone, fax, and automated communications systems for disseminating warnings and liaison with other organisations.

78

4 Dissemination Wall Maps

Whiteboards

Warning Status

Meteorology

Telemetry

Incident Manager

Meeting Area / Table

Flood Forecasting Operations Manager

Communications

Dissemination

Fig. 4.3 Example layout for a Flood Incident Room

The control centre could just be a single room in an office, or a separate building dedicated to providing flood warnings. The location should be outside any possible area which may flood, or for which access may be impeded by flood waters, with an alternate location available in case of problems. Some other equipment which may be available includes: ● ●

●

●

●

●

●

Maps – large scale maps on walls or on chart tables Mimic boards – large wall mounted displays showing flood risk areas, gauges, and other features (e.g. control structures, gates) Whiteboards – wall mounted boards for drawing sketches etc., possibly including electronic whiteboards to transmit images to other offices Media kit – equipment to assist staff with providing television and radio briefings to the media Television/radio – to keep up to date with news reports and how the event is being reported Briefing area – an area for staff briefings and for visitors from government, the media etc. to catch up with and observe operations Hot desks – networked workstation areas for temporary visitors to work

Figure 4.3 shows one possible example for the layout for the incident room for a small regional centre. In the figure, the numbers of computers and telephones shown are illustrative only and other devices such as printers and fax machines are not shown; also most key systems are likely to have complete backups in case of failure. A separate location for media briefings is also usually advisable (Holland 2007).

4.2 Dissemination Techniques

79

Since floods can occur at any time of day or night, a rota of key staff will usually be established, equipped with laptop computers, cell phones, radios or pagers so that they can easily be contacted when away from the office. The incident room may be permanently staffed, or limitations may be placed on how far duty officers can travel from the incident centre when on call. At the handover between shifts, a briefing may be held for incoming staff on the current situation, and a package of key manuals, equipment, situation reports and other information formally handed over. During normal operations, when no flooding is occurring or anticipated, the duty officer might only monitor weather, river or coastal conditions daily, or a few times a day, and perhaps have other duties unrelated to flood warning. For a full time flood warning service, routine day to day duties can include review and improvements to existing flood warning schemes, issuing routine river and coastal situation reports and bulletins, development of new flood warning schemes, training, post event reporting, reporting against organisational or national targets, planning and liaison with other flood response organisations (local authorities, emergency services etc.), public awareness campaigns (newspaper, television, radio, meetings, leaflets etc.), commissioning public satisfaction surveys, installation of monitoring equipment, system improvements (telemetry, forecasting, dissemination etc.) and other activities. When flooding conditions appear possible, the frequency of monitoring (and forecasting model runs, if available) will typically be increased, and additional staff put on standby or called in to the incident room. The number of staff required for a fully operational incident room can be high, and can include representatives from local authorities, the police, and other organisations. The organisational structure differs considerably between countries, but could include a general incident manager, a manager for operational staff deployed on site, monitoring and forecasting specialists, communications experts, a press or public relations officer (or technical staff trained in media relations), and other specialists from within and outside the flood warning team with detailed knowledge of the catchments or coastal reaches at risk. As described in Chapter 9, other organisations, such as the emergency services and local authorities, may establish their own command and control centres. Ideally one centre will be designated to lead the response, with representatives from all other key services and functions present, with clearly documented procedures describing the division of responsibilities between different organisations.

4.2 4.2.1

Dissemination Techniques Introduction

Flood warnings may need to be issued to the public, emergency services, local authorities and others with an interest in when and where flooding is likely to occur, or who are involved in the emergency response.

80

4 Dissemination

Again, terminology differs, but dissemination techniques can broadly be separated into indirect methods, community based methods, and direct methods, with some alternative classification schemes including use of the terms General and Specific (Emergency Management Australia 1999), individual, community and broadcast (e.g. Andryszewski et al. 2005a, 2005b) and Pull or Proactive Mode, and Push or Reactive Mode (Martini and De Roo 2007). Direct methods provide a warning targeted at specific individuals or organisations, and have the advantage in some cases of confirmation that the recipient has received the warning, which is particularly important for the representatives of communities, local authorities, emergency services and other key responders. Indirect warnings by contrast provide a more general warning and can potentially reach large numbers of people, whilst community based methods fall in between these two extremes. Some examples of these techniques include: ●

●

●

Indirect – television, radio, Internet, teletext, telephone help line, RSS, newspapers Community – sirens, fixed, mobile or helicopter loud hailers, megaphones and public address systems, bells, storm cones, flags, cascade systems, motorcycles, billboards/signs (electronic/manual), road barriers, flood wardens Direct – telephone, cell phone (voice, text), door knocking, fax, telex, pagers, two way radio, email, leaflet drops

For some methods, the distinction between these approaches is blurred. For example, sirens may be installed at one or more strategic locations to provide complete coverage of an area, but can be operated either locally, or indirectly from a control centre over a telemetry network. Similarly, some cell phone networks have the capability both for direct communications, and to broadcast emergency messages to all phones within range of the nearest network tower. Where loud hailers or hand operated sirens are used, these are typically operated either by people patrolling the streets on foot, or from vehicles, often following a pre-planned, timed route covering all areas for which a warning is required. Fixed installations may also be used in locations where there is a regular flood risk. Cascade systems (or telephone trees) may also be used, in which contacts are initially with one small group of key people by telephone or in-person, who in turn each warn a second tier of people, and so on until all intended recipients have received the warning. All methods have their own advantages and potential drawbacks, and many organisations use at least two alternate approaches, both in case of failure of any one method, and because research has shown that people are more likely to respond if they receive information in varying ways and from more than one source, including any existing informal networks (e.g. Parker 2003; Andryszewski et al. 2005a, 2005b). Table 4.2 illustrates some potential issues with approaches to issuing direct warnings (although note that all of the methods shown can work well in many situations, and much depends on the institutional and cultural setting, and the resilience built into the design of dissemination procedures).

4.2 Dissemination Techniques

81

Table 4.2 Some issues to consider with a selection of direct warning methods Issue

Examples of methods affected

Transmission/telemetry network can be affected Telephone/cell phone, sirens (remotely by power failure controlled), two way radio (if using repeater stations), Internet High workload for staff (control centre) Telephone/cell phone (if manually operated) High workload for staff (on site) Loud hailer, door knocking, leaflet drops High record keeping requirements between Telephone/cell phone, internet (email) flood events Possibly a significant time delay between Telephone/cell phone (if manually operated), issuing a warning, and all recipients loud hailer, door knocking, leaflet drops receiving the warning Recipient of the warning may not be the Telephone/cell phone (automated messaging), decision maker (e.g. on evacuation) loud hailer, sirens, leaflet drops, internet (email) Relies on recipient having device with them, Cell phone, two way radio and switched on Message will probably not be received by Telephone, loud hailer, door knocking, siren, recipients away from the property leaflet drop (e.g. at work) Message is not voice or text based and relies Sirens, storm cones, flags on recipients understanding the meaning and that it applies to a flood event No direct confirmation that a warning has been Telephone/cell phone (automated messaging), received and understood loud hailer, siren, bell ringing, leaflet drop, others Transient populations cannot easily be added to Telephone/cell phone (automated messaging), the list of recipients (e.g. people in vehicles) internet (email) Less effective in rural areas with widespread Loud hailer, siren, bell ringing, door knocking, properties others Reception may be affected by high ambient Loud hailer, siren, bell ringing noise (e.g. factories), high winds, window insulation etc, and/or be an issue for the hard of hearing Dissemination may be affected by flood waters Loud hailer, door knocking, leaflet drop interrupting access. Possible health and safety issues for staff and Loud hailer, door knocking, leaflet drop volunteers

4.2.2

Role of Information Technology

Developments in computer and communication technology in recent years have led to a range of new approaches for issuing warnings which complement existing techniques. They also provide the opportunity to issue direct or community warnings to much larger numbers of people than has been feasible in the past using manually based methods alone.

82

4 Dissemination

Table 4.3 Examples of Internet based flood warning systems Country

Operator

Name or link

Australia Bangladesh

Bureau of Meteorology Flood Forecasting and Warning Centre Finnish Environment Institute Ministry of Ecology and Sustainable Development Rhineland Palatinate Flood Warning Centres Japan Meteorological Agency Environment Agency, SEPA NOAA/National Weather Service

http://www.bom.gov.au/hydro/flood/ http://www.ffwc.gov.bd/

Finland France Germany Japan United Kingdom USA

http://www.environment.fi/ http://www.vigicrues.ecologie.gouv.fr/ http://www.hochwasserzentralen.de/ http://www.jma.go.jp/en/warn/index.html Floodline http://www.nws.noaa.gov/

Perhaps the most widely used indirect approach is the Internet, with information on web addresses provided via television, radio, newspaper and other public awareness campaigns. Table 4.3 lists examples of Internet based flood warning systems from several countries. Some typical functionality can include information on the time of issue of the warning, text, graphical and map based representations of the areas at risk, contacts for more information, advisory information on acting upon the warning, and search facilities by location, river, town etc. For example, Fig. 4.4 shows a display used by the Urban Drainage and Flood Control District in Colorado, which combines information on river levels, likely impacts, historical flood heights, and site specific issues, and provides links to maps, additional information on rainfall, and a range of tabulated outputs. Increasingly, multimedia dissemination systems are also being used to allow warnings to be targeted more precisely at specific groups of people (see Box 4.1, for example). Some typical functionality for this type of system might include: ●

●

●

●

●

●

Information stored on people and properties at risk from flooding, including preferred methods of contact, and alternate contact methods Information stored on flood warning codes and the messages to use (fax, voice, other) The facility to link flood warning codes to specific groups of people or properties, including map based displays and definition of groups Automated dialling of phone numbers and sending of emails etc., perhaps with computer generation of text and voice messages, including multiple languages Automated logging of warnings issued, including time of issue and time of receipt, including call back facilities in case of no response Automated generation of summary statistics

4.2 Dissemination Techniques

83

Fig. 4.4 Greyscale example of water level templet from the Information Services and Flood Warning Program, Urban Drainage and Flood Control District, Denver, Colorado

In some systems, summary information can also be generated for the overall numbers of people or properties warned, for statistics on the average time delays experienced between issuing and receiving warnings, and other performance measures, such as the percentage of people who acknowledged messages. Systems may be opt-in, with people identified as being at risk of flooding choosing to be included, or opt-out, with the default being to include people unless they confirm otherwise. A particular problem for any warning system is that of so-called transient populations, such as road users, pedestrians on riverside or coastal paths, business travellers, hikers, and people in campsites. Options for automated transmission of location dependent warnings include using the traffic alert systems available with digital radio, targeting voice and text messages to cell phones within a given range of a transmitter, and remotely activated electronic warning signs to warn of potential flooding of roads and river and coastal footpaths. One other approach to dissemination, which is sometimes used if false alarms are not a major issue, or floods develop very rapidly, is to link the detection and dissemination components of the system directly, without human intervention. For example, river level detectors might be linked to an alarm bell or siren to alert gate operators to the need to take action. Some other examples include: ●

A community based heavy rainfall warning system in Central America, in which rainwater piped into a container is detected using three electrodes set at different

84

●

●

4 Dissemination

depths, which trigger audible alarms, and send automated messages by cell phone or landline to selected community representatives (Oi and Opavedi 2006). Pressure transducers installed at low level in road kerbs with a direct link to nearby electronic road signs to warn road users of potential flooding, as used in parts of Texas, for example. Also, automated road barriers in some parts of the USA (e.g. canyons). An alert system in Nepal for Glacial Lake Outburst Floods which used the outputs from river level sensors to trigger warnings in turn from a series of sirens further downstream, connected by radio telemetry.

However, the decision to use a fully automated approach will depend on a range of factors, including system reliability, the consequences of failure, and tolerance to false alarms, and most operational systems rely on expert inputs from duty officers at some point in the warning process. Recent developments have also considered how to provide warnings to rural or dispersed communities, with particular emphasis on a low cost, sustainable approach, as illustrated in Box 4.2. More generally, research on flood warning technologies increasingly aims to improve the targeting of warnings, allowing more effective use of staff and other resources, and avoiding the unnecessary of evacuation of properties. Forecasting models can play a useful role here; for example, as described in Chapters 5 and 10, systems and models are now available which, when combined with property databases and digital terrain models, are capable of mapping likely flood inundation extents at each time step in a model run, and generating lists and maps of properties likely to flood at specific times into the future. When coupled to automated dissemination systems (such as voice messaging systems), warnings can in principle be issued to individual properties, together with estimates of the likely depth, start time and duration of flooding. However, a cautionary note is that much relies on the model accuracy, so manual intervention is still likely to be required at some point in the process to provide a check on model outputs.

4.2.3

Warning Messages

The content and wording of warning messages again varies widely between countries and much research has been done on the most effective ways to issue warnings to the public and non-specialists, and on how warnings are perceived (see Chapter 11 for examples). Some general principles are to provide a clear and accurate description, in familiar (non technical) language, and ideally contrasting the severity of the current situation to recent events which people may remember or can relate to. For example, in some countries, colour coded marker boards are used on river banks and buildings illustrating the flood levels likely to be reached for different stages of warning.

4.2 Dissemination Techniques

Box 4.2 Examples of international developments in dissemination technologies Recent years have seen the development of a number of techniques which combine the latest communication technologies with a low cost, sustainable approach. These methods are often aimed particularly at rural communities, and include the option to broadcast warning messages for many types of natural hazards. Some examples include: ●

●

●

RANET – the RANET (Radio and Internet Technologies for the Communication of Hydro-Meteorological and Climate Information for Rural Development) project is an international initiative supported by National Meteorological and Hydrological Services, Non Governmental Organisations, and others. It aims to make early warnings about natural hazards and other climate and weather related information available to rural populations and communities, and operates in Africa and parts of Asia and the Pacific. Activities also include identification of appropriate dissemination technologies, training, and capacity building. Information is sent via uplink to satellites for broadcasting every hour and can be received by computer, digital radio, and cell phone, with other techniques under development. Additional information is also transmitted on topics such as general health, agriculture and basic education by a range of information providers (Sponberg 2006). Village knowledge centres (Tamil Nadu, India) – are focal points for information on various issues of interest to rural and coastal communities, such as fish movements, wave heights, weather forecasts, health, education, agriculture and other community activities. The centres provide access to computer, radio, telephone and Internet equipment and include public address systems. Information kiosks are also widely available or planned in large numbers of villages in India (UNESCO 2006). The ISLAND project – was a collaborative exercise between organisations in Vietnam, Cambodia, Laos and several European countries to explore communication needs for hazard-related information (floods, pollution, epidemics, forest fire etc.), particularly in rural communities. Initial consultations showed that sustainability could be improved if the scope was widened beyond crisis forecasting and management to provide other information of daily interest to rural communities, such as weather forecasts and market data (e.g. for crops and livestock), together with longer term information on hazards (e.g. how long fields were likely to remain flooded). The project also explored new ways of presenting information such as community electronic billboards and multimedia information sent directly to cell phones (Morel and Despres 2006).

85

86

4 Dissemination

The warning should also include the time of issue and the location and expected time and duration of flooding, recommended actions, and the time for the next update. Locations are better expressed in terms of places where people live or work (communities etc.), rather than in terms of river or coastal reaches or monitoring locations. Messages should also be from a single authority, or intermediaries (e.g. community representatives) or, if not, following an agreed plan for different organisations to issue different components of the message. The following items show a suggested generic format for a warning bulletin based on a comparative study of methods used in several European countries (Martini and De Roo 2007): ● ● ● ●

●

● ●

●

●

●

Header – a short title describing the event and/or its development Date and time – of the bulletin’s delivery and its time of validity Name – of the bulletin’s provider (the organisation) Core message – short and clear description of the current situation, and its forecasted development Data – observed and forecasted data; comparison with past and historical events; flood warning level if available; time and level of the forecasted peak Uncertainty – level of forecasting uncertainty together with explanations Local/personal advice – where appropriate; feared impacts on public life (transport, communication networks, …..) and advice to face them; refer to seasonal activities if appropriate (holidays, sports events, ….) Information – about further broadcast/information technologies to ease telephone service General permanent information – flood warning level scale if available; emergency or other useful contact points for more information; links to other information providing systems (Internet addresses, telephone numbers,…); where to find general information about flood risk in the area Date and time – of the next information/forecast

Where lead times are sufficiently long, experience suggests that warnings are more effective if a staged approach is used, allowing people to prepare for the possibility of flooding before needing to take action. For example (World Meteorological Organisation 2006), a generic set of warnings for natural hazards such as typhoons, hurricanes and floods is: ●

●

An advisory informs people within a designated area of probable weather or hydrological conditions that could lead to hazardous situations, but they are not yet severe enough to move to the next stage of alert. People should take note of an advisory and be aware of any change in conditions. A watch alerts the public of the possibility of a particular hazard and provides as much information as is available on its intensity and direction. Such forecasts are issued well in advance of a weather event such as a cyclone, when conditions are suitable for development of severe conditions. When a watch is announced, people should take steps to prepare to protect their lives and property. Depending on the circumstances they may need to prepare for evacuation.

4.3 Design and Implementation ●

87

A warning is a forecast of a particular hazard or imminent danger issued when extreme conditions have developed and are occurring or have been detected. It is time to take appropriate action.

In some situations, the initial stages of warnings may only go to emergency services and governmental organizations. Different countries use different approaches; for example, Box 4.1 illustrates the four stage warning system used in the UK which is defined as follows: ● ● ●

●

Flood Watch – flooding of low lying land and roads is expected. Flood Warning – flooding of homes and businesses is expected. Act now! Severe Flood Warning – act now! Severe flooding is expected with extreme danger to life and property. All Clear – no further flooding is expected. Water levels will start to go down.

Some conditions which might lead to issuing a Severe Flood Warning include high risk to life (due to depth, velocity etc.), large numbers of properties likely to be flooded, severe disruption to infrastructure and the ability of responders to act, and the risk of flood defences failing or overtopping. More generally, there is an increasing trend to provide estimates of uncertainty with flood warnings and forecasts, and Chapters 5, 10 and 11 discuss this topic, together with a more detailed discussion on how effective warnings can contribute to reducing risks during a flood event.

4.3

Design and Implementation

The design of a flood warning system can require many stages, including consultation with community representatives, local authorities and the emergency services, installation of monitoring equipment, development of forecasting and dissemination systems, writing procedures, and a range of management, training and other activities. Several guides have been published on the main steps involved both for flood warning systems, and other types of early warning system, and some examples are summarised in Table 4.4. Some typical areas to consider in designing and implementing a flood warning system can include: ●

●

●

● ●

User requirements – from consultations, consideration of flood warning performance targets (if any), and other criteria Risk assessment – formal assessments of the locations at risk from flooding based on historical flood events, modelling and consultations Detection – assessment of the availability, quality and reliability of existing real time data on rainfall, rivers, tides etc. (as appropriate), and installation of new sites if required Thresholds – definition of the criteria under which warnings will be issued Flood forecasting – review of any existing flood forecasting models and development of new models (if required)

88

4 Dissemination

Table 4.4 Examples of guidelines on designing and implementing various aspects of flood warning and other early warning systems Country

Organisation

Australia

Emergency Management Australia Environment Agency

England & Wales

USA

Generic

Generic Generic

●

● ●

●

●

●

Document(s)

Reference

Australian Emergency Emergency Management Australia (1999) Management Manuals; Flood Warning volume A series of best practice Tilford et al. (2007), guidelines on river, Environment Agency estuary and coastal (2002, 2004) flood forecasting NOAA/National NOAA/NWS Automated local flood warning systems handWeather Service book (1997) (see also USACE 1996) World Meteorological Guide to hydrological World Meteorological Organisation practices (forecasting Organisation (1994) sections) World Meteorological Global guide to tropical Holland (2007) Organisation cyclone forecasting ISDR Developing early warning ISDR (2006) systems: a checklist

Dissemination – decisions on the most appropriate methods to use, preparation of appropriate messages, implementation of databases etc. (as required) Procedures – development of flood warning procedures Preparedness – liaison with other organisations involved in flood response, and development of flood emergency plans (Chapter 9) Resilience – assessment of all aspects of the proposed system for possible points of failure, particularly during extreme weather and flood events (Chapters 2, 3, 5, 9, 11) Communication – design and implementation of public awareness campaigns (Chapter 11) Performance Monitoring – developing procedures for ongoing monitoring and evaluation of the scheme (Chapter 11)

An important first step is usually to decide on the main aims of the flood warning scheme or system and any performance requirements or targets. Consultees can include members of the public, or their representatives, the emergency services, local authorities and others. Approaches to consultation can include workshops, town meetings, site visits, household surveys, telephone surveys and questionnaires. Requirements may be expressed in terms of lead time, accuracy, ways of receiving information, and other choices. This approach, particularly if it is driven by members of the communities involved, also raises awareness of the proposed scheme and builds a sense of ownership (Emergency Management Australia 1999).

4.3 Design and Implementation

89

Another key question is the likely budget available, and the economic case for the scheme, and this point is discussed in Chapter 11, together with a range of techniques for prioritising investment in schemes (cost benefit, multi-criteria methods etc.). Some aspects of the design may also be guided by organisational or national targets and standards, again as discussed in Chapter 11. Later chapters also discuss the design of flood forecasting systems (Chapters 5–8), techniques for examining resilience (Chapter 9), and performance monitoring (Chapter 11).

Chapter 5

General Principles

Flood forecasting models are an important component in many flood warning and emergency response systems. Models can assist by providing advance warning of the likely timing and magnitude of flooding, and in helping to understand the complexities of a flood event as it develops. Models outputs may also be used in decision support systems for flood event management and the operation of flow control structures. The techniques used for flood forecasting have many similarities to the methods used for simulation modelling of river and coastal processes. However, the design may be constrained by the availability of real time data, and computer systems on which to operate the model, although there is the advantage that model outputs can be updated to help to account for differences with observed values; a process which is often called data assimilation. Ensemble and probabilistic techniques are also increasingly being developed to provide information on model uncertainty to users of model outputs, and to allow a more risk-based approach to decision making. This chapter provides a general introduction to these issues and to the topic of flood forecasting model calibration and performance monitoring.

5.1

Model Design Considerations

In designing a flood forecasting model, some important factors to consider include: ● ● ● ● ●

The forecasting requirement The real time data available to support operation of the model The forecasting system on which the model will be operated The required model performance The time, budget and skills available for model implementation

The overall design will often be a compromise based on these various considerations. Section 5.2 discusses forecasting systems, whilst the requirements for data depend on the particular application, and examples are provided in Chapters 6–8 for a range of forecasting applications. The issues of time, budget and skills, and the level of flood risk, often form part of a wider decision making process on the economic justification for a flood warning system, and are discussed in Chapter 11. K. Sene, Flood Warning, Forecasting and Emergency Response, © Springer Science + Business Media B.V. 2008

93

94

5 General Principles

The forecasting requirement depends on the needs of users of the model outputs, and Table 5.1 lists some examples of applications which might be considered. For a given situation, a model may help in addressing some or all of these requirements. Although it is difficult to generalise, in moving down Table 5.1, the types of model required will generally be more complex, and take longer to develop. For example, for a river flow model, to estimate the flood peak at a single location, a simple rainfall runoff model may in some cases be sufficient whilst, to estimate flow depths and velocities at property and street locations, a real time hydrodynamic model of the floodplain flow would ideally be required. Similarly, for coastal forecasting, to estimate inundation extent behind a sea defence, offshore-nearshore wave transformation, wave overtopping, surge and floodplain models might be required. In particular, models to assist with control structure operations, and decision support applications, can sometimes require considerable exploratory work to implement. Of course, exceptions can always be found and some of the choices available are discussed in Chapters 6–8 when considering the strengths and limitations of alternative modelling approaches for river and coastal forecasting applications. Another aspect of the forecasting requirement is to specify where forecasts are required. These locations are often known as Forecasting Points, and can include:

Table 5.1 Some typical applications of flood forecasting models (Adapted from Environment Agency 2002; © Environment Agency copyright and/or database right 2008. All rights reserved) Requirement

Typical applications

To provide additional warning lead time

Providing the emergency services and public with advanced warning of the likely times for the onset of flooding or the peak of the flood event As above; also providing estimates for peak values, perhaps also linking to maps of likely flooding extent based on off-line simulation modelling or past experience As above; also considering when to issue the message that the flood risk has reduced or passed As above, but providing real time updates to the likely spatial extent of flooding, together with depths and velocities at general or specific locations (districts, roads, houses etc.) Optimising response to mitigate flooding (e.g. at dams, river control structures, tidal barriers etc.) and possibly to reduce penalty payments and opportunity losses Providing advice on the implications of various problems arising during an event; for example, failure of a flood defence, blockages by debris, pumps failing etc. Also exploring different options for response (e.g. controlled diversion of flows)

To estimate peak levels or flows

To estimate flooding duration

To model flooding depths, extent and, possibly, velocities

To provide information to assist with operation of control structures

To provide information to assist with event-specific factors

5.1 Model Design Considerations ● ●

●

95

Locations where there is a flood risk Locations where real time information is available to evaluate or update model outputs Locations where forecasts are required to assist with operations of structures

Some alternative names for Forecasting Points include Forecast Points, or Flood Forecast Points or, for coastal applications, Coastal Cells or Units. One option available for the development of a forecasting model is to focus the modelling effort on achieving acceptable model performance at these locations, but to use a simpler approach elsewhere provided that this does not affect performance at those points. This approach can result in considerable savings in the cost and time for model development and/or reductions in model run times. For example, for a hydrodynamic river model, detailed survey data may only be required at and around the Forecasting Points, rather than for the whole catchment whilst, for a nearshore hydrodynamic coastal model, the grid resolution can be tailored to the areas which have the most influence on levels, surge and wave action at the Forecasting Points. The performance requirements for the model can also be defined in terms of the requirements at Forecasting Points; for example, the required accuracy of forecasts of peak river levels, or the lead time provided for surge forecasts. Section 5.4 discusses some other possible model performance and calibration criteria. However, these requirements may not always be achievable with the data, models and budget available, and this needs to be factored into the overall design, and potential users of model outputs warned of any such limitations. Chapter 11 discusses a range of approaches to prioritising the development of forecasting models and other components in the flood warning process. In many cases, the forecast lead time is one of the key design criteria, and may dictate the overall design of the model. For example, for a river forecasting model, if the catchment response time is less than the required lead time, then rainfall forecasts will usually be required as inputs to the model. Additional tasks required in this case could include an investigation of rainfall forecast accuracy (and whether this is suitable for use with the model), establishing a real time data feed of those forecasts to the model and (ideally) calibration of the model to a historical archive of forecast values, if this is practicable. Also, in a real time application, it is important to distinguish between the potential lead time provided by the forecasting model, and the likely lead time for flood warnings. The warning lead time can be considerably shorter than the potential lead time due to time delays in the system including: ●

●

●

●

Polling time – the time delay between a parameter (e.g. rainfall) being measured, and being available for transfer to the forecasting system Forecasting system run frequency – the time delay whilst waiting for the forecasting system to initiate the next model run Pre-processing time – the time taken to prepare and validate the data for input to the model, and to perform any preparatory analyses (e.g. infilling of missing values) Model run time – the time taken to initialise model states (if required), run the model(s) in the forecasting environment, including any probabilistic simulations, and perform any real time updating required

96

5 General Principles Peak rainfall data received

Warning Decision to issue warning received by all property owners Postprocessing completed

Model Run completed

Pre-processing completed

Flooding threshold exceeded Flood peak reached

Rainfall Event

Warning lead time Forecast lead time Catchment response time

Fig. 5.1 Illustration of the time delays in issuing a warning for a single rainfall runoff model

●

●

●

Post-processing time – the time taken to collate and process the model outputs into reports, graphs, maps etc., and to raise alarms (if required) Decision time – the time taken for the forecaster to review the outputs and decide whether to approve the forecast, and for the recipients of that forecast to decide whether to issue a warning (including time for discussions etc.) Dissemination time – the time taken between a flood warning being issued, and all of the intended recipients receiving it

Figure 5.1 illustrates these various potential time delays for the idealised case of a rainfall runoff model providing flow forecasts for a single Forecasting Point following a short duration, intense rainfall event (see USACE 1994; Environment Agency 2002; Carsell et al. 2004 for similar examples). In the figure, the relative magnitudes of the various time delays are illustrative only and are exaggerated in places; also the catchment response time is assumed to apply from the mid-point of the rainfall event, although various other definitions are used (for example, starting from the centroid of the rainfall event). However, for this particular example, the warning lead time available would be considerably less than the catchment response time or the forecast lead time. This lead time could be extended by using rainfall forecasts although, as described in Chapters 2 and 6, with the accuracy likely to decrease with increasing rainfall forecast lead time. In practice, for a simple model like this, operating on a dedicated forecasting system, the overall time delay due to these factors can be small compared to the physical response times of the river or coastal reach. However, if many models are being operated in parallel (e.g. as on a regional forecasting system), or if more complex model types are being used (e.g. hydrodynamic models), or probabilistic or ensemble techniques are being used, these time delays can be significant, and may in some cases lead to the decision to use an alternative modelling approach. For example, hydrodynamic models of flood defence systems developed for design

5.2 Forecasting Systems

97

studies can sometimes take many hours to run, unless they have been optimised for real time use, in which case improvements in run time of one or two orders of magnitude or more are sometimes possible (e.g. Chen et al. 2005). Chapters 6–8 discuss some run time issues associated with various types of river and coastal forecasting techniques, and Chapter 9 discusses some other sources of time delays in the emergency response process. However, for the forecasting component alone, some possible options for obtaining solutions more quickly include (e.g. Chen et al. 2005; Environment Agency 2007): ●

●

●

●

●

Model emulators – simpler models, such as transfer functions, which can emulate the behaviour of more complex models Restructuring of models – so that the more computationally intensive components of the model can be run on demand, or at a less frequent time step (e.g. a hydraulic model for a specific flood risk area) Filtering or clustering of ensembles – use of only a subset of values if ensemble techniques are used (although with many issues to consider regarding the representativeness of the sampling technique) Computer processing – using faster processors, or structuring the model so that computing effort can be shared between more than one processor Model rationalisation – improvement of the underlying model to improve run times, convergence and stability

The issue of whether the decision time can be eliminated by automatically issuing warnings based on forecasts is an interesting question, and ties in with various topics which are discussed in Chapters 4 and 11; for example, the uncertainty in the model outputs, the time available for an emergency response (e.g. for flash flood forecasting), performance targets, and tolerance to false alarms. Many (but not all) modellers take the view that some human intervention and interpretation is essential in the overall process, where there is time to do this. Also, based on estimates of model uncertainty, or ensemble forecasts, or experience, the forecaster may choose to wait for additional model runs before deciding whether to issue the forecast, in the expectation that by then the model uncertainty will have reduced (see Chapter 10).

5.2

Forecasting Systems

A flood forecasting system provides the operating environment within which flood forecasting models can be operated, and is sometimes called the system environment. Table 5.2 shows some of the key functionality which is typically available in modern forecasting systems and Box 5.1 provides an example of an operational system. The precise options available will depend on the system developer or vendor. Systems typically run continuously all year around (24/7) and are required to meet specified standards for availability, reliability and downtime. For additional resilience, many systems offer multiple redundancy in computer hardware, software, and data transfer routes in case of failure of any one component.

98

5 General Principles

Table 5.2 Some typical functionality in modern flood forecasting systems Item

Function

Description

Pre-processing

Data gathering

Polling of instruments directly, or receiving data from a separate telemetry system (see data interfacing) Interfacing to a range of real time data feeds and forecast products from various sources (meteorological, river, coastal) Real time validation using a range of time series, statistical, spatial and other validation methods Transformation of input data into the values required by the modelling system (e.g. catchment rainfall estimates), including infilling missing values by interpolation and other methods Scheduling and control of model runs, and error handling Application of real time updating and data assimilation algorithms Automatic fall-back to alternative options in case of failure of one or more components (models, data inputs etc.) Processing of model outputs into reports, maps, graphs, web-pages etc. Intersection of inundation extents (if computed) with street and property maps etc. to generate information on areas at risk at each time step Raising alarms when thresholds are forecast to be exceeded, using map based displays, email, pager, text messaging etc. Automated calculation and reporting of information on model performance and system availability Maintenance of a record of data inputs, model run control settings, model forecast outputs, operator identities etc. The facility to replay model runs for post event analysis, operator training and emergency response exercises Map based, graphical and other displays of input data, forecast outputs, alarms etc., including overlays of aerial and satellite photography For running scenarios defined during the design phase or in real time (e.g. for future rainfall, defence breaches, gate operations etc.) Interactive tools for off-line configuration of models, data inputs, output settings, alarms etc. Off-line tools for calibration of models

Data interfacing

Data validation Data transformation

Model runs

Model run control Data assimilation Data hierarchy

Post processing

Model outputs Inundation mapping

Alarm handling

Performance monitoring Audit trail

Replay

User interface

Model outputs

What if functionality

System configuration Model calibration

Although models can be operated without a forecasting system, this can rapidly become complicated if there are multiple Forecasting Points to consider, or forecast lead times are short, or real time updating is required, or forecasting duty officers have other tasks to perform. Organisations are also increasingly required to provide

5.2 Forecasting Systems

99

Box 5.1 The National Flood Forecasting System in England and Wales The National Flood Forecasting System (NFFS) is the operational flood forecasting system used by the Environment Agency in England and Wales. The system can run a wide range of river and coastal forecasting models, and provides forecasts to numerous Forecasting Points on rivers and coastal reaches across England and Wales. Operationally, the system gathers real time data from a wide range of sources, including regional telemetry systems (rainfall, river, reservoir, tide data etc.), and Met Office weather radar data and weather forecasts (rainfall, surge, wind speed and direction etc.). An extensive range of data validation, aggregation and transformation tools is available. The range of model types and options includes: ● ● ● ● ● ● ● ● ●

Conceptual rainfall runoff models Transfer function rainfall runoff models Flow routing models One-dimensional hydrodynamic models Snowmelt models Reservoir models Coastal wave transformation models Specialised models related to structure operations Data assimilation tools

Hydrodynamic models are optimised for run times, stability and convergence before loading onto the system, and several thousand kilometres of river network is represented in this way, allowing complex backwater, confluence, control structure and tidal effects to be modelled in real time. Real time inundation mapping is also being evaluated for some high risk locations. A range of simpler methods, such as correlations and look-up tables, is also included as a backup, and sometimes as the primary model type where a more expensive approach is not justified. Models are typically operated at least daily to monitor for potential flood risk. If flooding seems likely they are run on a more frequent basis to keep forecasters up to date on when and where floods are expected. The latest model outputs are normally available to inspect via the map based user interface, and a variety of graphical and other reporting formats. The system includes an extensive range of forecasting performance monitoring tools, including contingency table, statistical, and skill-score approaches. The system includes the facility to run ‘what if’ scenarios to explore the effect of different rainfall forecasts, control structure operations, and event specific factors (e.g. defence breaches), and includes data hierarchies in case of instrument or telemetry failure. There is also the facility to raise alarms when rainfall and level thresholds are exceeded but, for operational reasons, these are normally raised on the regional telemetry systems. (continued)

100

5 General Principles

Box 5.1 (continued) Ensemble and probabilistic flood forecasting techniques are actively under development. The software and database systems are hosted on two server clusters more than 100 km apart to provide resilience in case of local problems (flooding, power cuts etc.), with automated switching from duty to standby servers, or parallel running at both locations. The system operates ‘around the clock’ without user intervention, and stand-alone versions are available to allow testing and integration of new models, and user training. A web server also allows outputs to be viewed. A novel feature of the system is that it is open source, in the sense that any model complying with the open XML-based published interface can be ‘plugged’ into an existing network of models, provided that it is physically sensible to do so. The development of NFFS (e.g. Werner and Van Djik 2005) was a major undertaking and included migrating existing models from a range of older legacy systems. The system became fully operational in 2006 and passed its first major test in June and July 2007, when flood events affected more than 20% of the catchment area in England and Wales.

an audit trail of decisions made during a flood event, including use of the outputs from forecasting model runs, and a forecasting system can help to provide this functionality. Automation of model runs can also free up skilled staff to spend more time on interpretation and discussion of data and model outputs, rather than routine data processing and analysis. Some possible exceptions are situations where only a small number of models needs to be considered, the data entry requirements are modest, or model runs are only required at irregular intervals (for example, for some types of coastal or groundwater flooding). Figure 5.2 illustrates a typical configuration for a forecasting system operating both catchment and coastal flood forecasting models. In this example, real time data flows are received from a network of raingauges, river gauging stations, automatic weather stations, weather radars, and tide gauges. Rainfall and surge forecasts and composite weather radar data are also received from a meteorological service or department. Satellite, reservoir and snowcover/snowmelt information might also be included, although this is not shown on the figure. The data inputs are handled by a separate telemetry system, which feeds data to the forecasting system and to an off-line hydrometric database system (not shown). Two independent forecasting systems are shown; the operational (Duty) system and a backup (Standby) system, which operates in parallel and can switch automatically to being the live system in case of problems. Normally, the hardware for these independent systems would be located at different sites, both out of the floodplain, and ideally separated by a sufficient distance that both would not be adversely affected at the same time by widespread catastrophic events (fire, flood, earthquake,

5.2 Forecasting Systems

101

Numerical Weather Prediction

Weather Radar

Rainfall Forecasts Composite Radar Rainfall

Raingauges

Weather Stations River Gauges

Tide Gauges

Telemetry System

Surge Forecasts

Forecasting System (Duty)

Pre - Processor

Pre - Processor

Model Run Control & Database

Model Run Control & Database

Post - Processor

Post - Processor

Alarm Handling

Alarm Handling

Forecasting System (Standby)

Flood Warning Dissemination System

Fig. 5.2 Illustration of a possible configuration for a flood forecasting system

major power failure etc.). The telemetry system may also have backup facilities, although this is not shown here. Many other configurations can be used, of course; for example, the forecasting system might also manage the polling (collection) of data, avoiding the need for a separate telemetry system, although introducing the risk of interruptions in data collection if there are problems with the forecasting system. Also, the forecast outputs might be fed back to the telemetry system to allow both forecasts and observed data to be displayed there, and for the alarm (threshold) handling on forecasts to be performed in parallel with that for observed data, providing greater consistency and avoiding duplication of functionality, although at the expense of some loss of system redundancy. Various other combinations can also be envisaged, depending on the relative roles and responsibilities of the various organisations involved in the data collection and forecasting process. Internationally, many types of forecasting system have been developed for individual forecasting services, or are available commercially, and the system which is used can be a key factor in choice of modelling approach since usually only a specific range of model types can be configured on a particular system. There may also be model run time, licencing and other issues to consider. Increasingly, though, forecasting systems use a toolkit approach, with several choices of model, which can be configured as appropriate for each modelling problem, perhaps using an openly published interface which allows any type of model to be used which conforms to this interface (e.g. Fortune 2006). Table 5.3 presents several examples of

102

5 General Principles

Table 5.3 Examples of national or regional scale flood forecasting systems Country or region

Abbreviation

System

Reference

Australia

AIFS

Australian Integrated Forecast System - flood forecasting component National Flood Forecasting and Response System Central America Flash Flood Guidance system National Information System for Flood Control and Drought Relief Watershed Simulation and Forecasting System National Flood Forecasting Service Del Plàta Basin Hydrological Warning System (Argentina, Bolivia, Brazil, Paraguay, Uruguay) National Flood Forecasting System, FEWS Scotland

Elliott et al. (2005)

Bangladesh Central America

CAFFG

China

NISFCDR

Finland

WSFS

Norway Del Plata river basin

United Kingdom

NFFS/FEWS

USA

AHPS

Advanced Hydrological Prediction Service

Paudyal (2002) http:// www.ffwc.gov.bd/ Sperfslage et al. (2005) Huaimin (2005)

Vehviläinen et al. (2005) Røhr and Husebye (2005) Goniadzki (2006)

Werner and van Djik (2005), Cranston et al. (2007) http://www.nws.noaa. gov/oh/ahps/

systems operated at a national or regional scale, and further examples can be found in Chapters 6–8 and in the references to those chapters. Later sections and chapters describe the process of selecting and calibrating an appropriate model, or network of models, for use on a forecasting system. Once those stages have been completed, the next step is usually to configure the models for real time use. The approach to configuration varies between systems, but might include creating a database or hypertext file (e.g. XML) describing how models link together, and how data flows through the chain of models. Alternatively, some systems offer the option of setting up the configuration using a graphical user interface and Fig. 5.3 shows an example of how an interactive configuration editor might appear for a simple catchment forecasting model. The symbols used are for illustration only, but the principle of abstracting the chain of models into a network through which data can flow, is common to many systems. In particular, the representation in the system need not be to the same scale and layout as the physical situation. In this example, the evaporation function could consist simply of a typical seasonal profile, or be estimated using real time data from an automatic weather station. The flow routing component might consist of a single ‘black box’ model, or be represented by a series of nodes at cross sections along the river reach (e.g. a hydrodynamic model). In the latter case, the inflow components could be connected to the appropriate nodes so as to better represent the timing and attenuation of the hydrograph at locations further downstream, with additonal components to represent lateral (ungauged)

5.2 Forecasting Systems

103

1

1 2

A 2 C

B

A

C

B

D

Flow Routing Model

River Gauging Station

Evaporation Function

Town Raingauge

D

Rainfall Runoff Model

Fig. 5.3 Example of model configuration in a flood forecasting system

inflows. The river gauging station locations might also be selected as real time updating points, with the choice of updating procedure selected via the user interface. In addition to the appearance of graphs, maps, tables and other forms of output, some key considerations in configuring a model onto a forecasting system typically include: ●

●

●

●

The choice of time step (or model run frequency) – this is unlikely to be less than the polling or transfer intervals for real time data or rainfall and surge forecasts, but can be multiples of those values. Ideally, a value will be chosen that is sufficient to resolve the details of any event, particularly for the rising limb of a hydrograph, as flooding thresholds are approached. The approach to model initialisation – how will rainfall runoff, flow routing, hydrodynamic, reservoir, coastal and other models (as required) be initialised for routine operation and when starting after a gap or interruption in operations, and what choices will be made for saving model states between runs (if required)? Integration of the system into operational procedures – how will the forecast outputs be used in the process of issuing flood warnings, and will the system be operated continuously, or on demand as a flood event develops? Are there certain times of the day at which model outputs will be required for inspection, and are there benefits in operating the model less frequently (e.g. once per day) when flooding is unlikely and, if so, what is the process for switching from normal operation to a raised state of alert (e.g. more frequent model runs)? Data storage requirements – data volumes can increase rapidly when both input data and forecast runs are archived, particularly if a probabilistic or ensemble approach is being used. Recent values may need to be kept in a ‘rolling barrel’ store which after each run stores the oldest values off-line, then overwrites them with the latest values.

Many systems also allow a data hierarchy to be defined in case of failure of one or more input data streams, such as an instrument or a telemetry link, or the overall telemetry system. A hierarchy or choice of models can also be defined to run in

104

5 General Principles

parallel on some systems; for example, if a hydrodynamic model run fails to converge, then the outputs from a simpler backup hydrological routing or correlation model might be used instead. One common example of a data hierarchy is for raingauges and, as a simple illustration, the following set of rules might be one possibility for the rainfall runoff models shown in Fig. 5.3: ● ● ● ● ●

If both raingauges are operating, use raingauges 1 and 2. If raingauge 1 fails, use raingauge 2. If raingauge 2 fails, use raingauge 1. If raingauges 1 and 2 fail, use weather radar data (if available). If the weather radar data feed fails, use a standard storm profile.

This sequence could be configured to run automatically on the system, with alerts provided to the user that replacement data streams have been used in model runs. Also, to account for possible calibration and other differences in each data stream, ideally the model should be calibrated and optimised for each combination of inputs, with a hierarchy of parameter sets also available, although this may not be practicable in many cases. More complicated scenarios, using other more distant raingauges, and combinations of gauges, could also be envisaged. A related option provided by some systems is the functionality to perform what-if scenarios during a flood event; for example, to explore the impact on flood magnitude, timing and extent for situations such as operating a river control structure, or eventspecific problems occurring such as a bridge being blocked by debris, or a breach occurring in a flood defence. One commonly used example is a set of scenarios for future rainfall and some options could include: ● ● ● ●

The forecast rainfall follows a pre-defined profile. The forecast rainfall matches values from a significant historical flood event. The forecast rainfall continues at the current intensity. Rainfall stops.

Normally, a set of pre-defined scenarios will be calibrated and configured ready for use during an event although, in some cases (e.g. gate operations, blockages), the forecasting system may provide users with direct access to model parameters and initial conditions via the user interface (e.g. dialog boxes) so that these can be changed manually. Ensemble and probabilistic techniques might also be used as described in Section 5.5.

5.3

Data Assimilation

Although there are many complicating factors in real time modelling, one advantage compared to off-line simulations is that observations of river or coastal conditions are usually available to compare with model outputs. If the observations are reliable, then the forecasts can be updated to help to take account of the differences between observed and forecast values.

5.3 Data Assimilation

105

Many different techniques have been developed for forecast updating, and these are often separated into the following three main categories (e.g. Reed 1984; World Meteorological Organisation 1994; Moore 1999), although the distinctions between methods can sometimes be blurred: ●

●

●

Error prediction – in which forecast outputs are adjusted directly, based on models calibrated to the time series of differences between observed and forecast values State updating – in which the initial conditions of the model are adjusted to achieve a better match between the observed and forecast values Parameter updating – in which the parameters in the model are adjusted to achieve a better match between observed and forecast values

State updating is often called data assimilation by meteorologists and coastal modellers, whilst the terms ‘real time updating’ or ‘real time adaptation’ are often used by hydrologists to describe all three approaches. Error prediction is sometimes also called error correction, output updating or real time adjustment. Although data assimilation can significantly improve the accuracy of flood forecasts, and is often recommended as best practice, a few issues to consider include (e.g. Environment Agency 2002): ●

●

●

Updating does not remove the need to have a well calibrated model, able to represent response for a wide range of types of event. In particular, some forms of updating algorithm can struggle with correcting errors in the timing of peaks. The quality of the updated forecast will depend on the quality of the input data, and erroneous data can degrade, rather than improve, the accuracy of forecasts. Usually, it is advisable to validate data inputs either automatically or manually before they are used for data assimilation. For real time control applications, the use of updating needs to be factored into the system design from the start, since otherwise unwanted feedback effects can develop; for example, control gates ‘hunting’ for optimum settings.

If data quality remains an issue after validation, then one solution might be to restrict updating to ranges in which values are known to be reliable; for example, if the high flow end of a stage discharge relationship is suspect, or levels start to become insensitive to flows (e.g. for some floodplain flows), then updating could be performed only up to a certain threshold value. At the simplest level, one approach to updating a forecast is to inspect the observed and forecast values on a graph, and to adjust the forecast ‘by eye’ to compensate for any differences. Figure 5.4 illustrates this process for a river flow hydrograph. In this example, the forecast is below the observed values throughout the period up to the start of the forecast period and, for the earlier event, the forecast peak was later than the observed peak. There are also minor differences in hydrograph shape or volume. The adjusted forecast attempts to visually compensate for these errors by applying both timing and magnitude corrections. The human eye is remarkably good at distinguishing between errors in timing and magnitude, and in deciding on the appropriate adjustments to make. However,

106

5 General Principles

Flow

Observed Flow Adjusted Forecast Original Forecast

Time ‘Time Now’

Fig. 5.4 Illustration of a visual approach to forecast updating

although systems have been developed which allow a forecaster to make adjustments of this type by editing or blending graphical outputs on a computer display, this approach can be impractical to apply if there are many Forecasting Points to consider, or there are many other demands on a forecaster’s time. In practice, therefore, the majority of updating procedures operate automatically without user intervention, although the original and updated forecasts are often displayed together, with the option to accept or reject the adjusted forecast.

5.3.1

Error Prediction

One distinguishing feature of error prediction methods is that they are usually independent from the forecasting model, and can be applied as part of the post-processing of model outputs. State and parameter updating techniques, by contrast, tend to be specific to the type or ‘brand’ of model, although with some exceptions. Error prediction methods to some extent mimic the visual adjustments described earlier, although may use a sophisticated range of time series analysis techniques. The basis for the approach is that, although the sequence of errors (residuals) is unknown before the start of the event, the natural persistence in catchment and coastal processes will tend to cause the outputs to be consistently higher or lower than observed values for at least part of the event, with timing errors also showing persistence over time. A time series model can then be fitted either to historical datasets, resulting in a fixed set of parameter values, or can be dynamically fitted during an event as it progresses. The forecast values from ‘time now’ are then adjusted based on the output from the time series model. The effects of these adjustments vary between approaches but tend to force the forecast values to match observed values at ‘time

5.3 Data Assimilation

107

now’, with the magnitude of the adjustments decaying into the future as the information content of the observed values reduces. Some examples of automated approaches to error prediction include autoregressive techniques (e.g. AR, ARMA), transfer function, and neural network techniques (e.g. Reed 1984; Rungo et al. 1989; Serban and Askew 1991; Moore 1999; Goswami et al. 2005).

5.3.2

State and Parameter Updating

State updating techniques aim to adjust the initial conditions of the model to obtain a better match between observed and forecast values at the start of the forecast period (‘time now’), and possibly over a number of time steps in the period leading up to that point (e.g. Wohling et al. 2006). The approaches used vary between different categories of model and can be very model specific. For conceptual models (see Chapter 6), state updating typically involves updating the internal stores in the model (for a rainfall runoff model) or, in the case of a reservoir, the initial level for the simulation. For example, the adjustment may be made using a gain factor which redistributes water between selected stores in the model (e.g. Moore 1999). By contrast, process-based models such as hydrodynamic river models (see Chapter 6) and coastal surge models (see Chapter 7) usually represent the catchment or coastal response using a grid-based approach. In this case, the updating problem is to distribute the forecast errors across the whole domain of grid nodes based on what is usually a relatively small number of observation points (river gauges, tide gauges etc.), whilst preserving conservation of mass and momentum, and without producing unwanted transient effects from the input of potentially large corrections at the grid squares closest to the points of observation. For coastal models, many different approaches have been considered, including empirical, sequential and variational methods, sometimes involving minimising of a cost function which depends on functions of the differences between modelled and observed values. A major international initiative (the Global Ocean Data Assimilation Experiment: GODAE) is also exploring new and improved techniques for data assimilation in ocean forecasting models (http://www.godae.org/), including the use of satellite altimetry to widen the extent of data available for assimilation. Perhaps the simplest state updating technique of all is simply substitution of observed values for forecast values; a technique which has been used for both river and coastal models with some success, although at the risk of introducing oscillations and other unwanted behaviour into the model forecasts. Techniques which adjust the input data to the model, such as inflows or rainfall values (e.g. Serban and Askew 1991) are also sometimes considered as a form of state updating. For example, for hydrodynamic models of rivers, one approach is to distribute the error into the tributary and main river inflow components (e.g. Rungo et al. 1989). For the future, low cost sensor networks, of the type described in Chapter 2, may also provide the possibility of using many more updating points than is possible at present, due to their small (unobtrusive) size and low cost of installation.

108

5 General Principles

By contrast, parameter updating techniques seek to update the model parameters, rather than the state variables. For process-based models, and some types of conceptual models, parameter updating is little used, and might at first sight seem inappropriate, since the basis of these modelling approaches is to use parameter values which have some physical meaning. However, due to measurement, calibration and other uncertainties in the modelling process, model parameters can never be known precisely, and parameter updating therefore seeks to compensate for this lack of knowledge by adjusting parameters within defined bounds. An example of use of this approach is in adjusting the roughness coefficient which is often used to parameterise friction losses in a river channel in hydrodynamic models.

5.3.3

Other Techniques

For data-based models, such as transfer functions (see Chapters 6 and 7), the distinction between approaches is perhaps less clear cut, and methods which have been used for data assimilation include Recursive Least Squares, Adaptive Gain and Instrumental Variable techniques, and various forms of Kalman filter, including extended Kalman filter and ensemble Kalman filter approaches (e.g. Young and Tomlin 2000; Beven 2001; Romanowicz et al. 2006). Kalman filter techniques have also been applied to other types of model; for example, to operational coastal surge forecasting models in the Netherlands (Verlaan et al. 2005), and to both rainfall runoff and hydrodynamic modelling applications in the form of the ensemble Kalman filter (e.g. El Serafy and Mynett 2004; Weerts and El Serafy 2005; Butts et al. 2005).

5.4 5.4.1

Model Calibration and Performance Basic Concepts

The principles of model calibration for catchment and coastal models are well established and are discussed in many books and technical papers (see Chapters 6 and 7 for examples). However, for flood forecasting applications, some additional factors need to be considered, and these are briefly described in this section. These include the overall structure of the model (if the model configuration can be varied), and the optimisation criteria to be used in the calibration. Once a model is operational, the performance will also need to be monitored, both to advise users of the likely accuracy, and to guide future improvements to the model and of the real time sources of data used in its operation. As with off-line models, it is useful to know how well the model can represent the timing and magnitude of peak levels, flows, surge, wave, wind and other outputs, and the time history of values during the event (e.g. the shape of a river flow

5.4 Model Calibration and Performance

109

Level

A

Flooding Threshold

B

Flood Warning Threshold C

Flows subsequently observed

Time Fig. 5.5 Illustration of forecasting issues related to threshold levels

hydrograph). However if the model is used in a flood warning application, then the performance in the time leading up to the crossing of flood warning threshold levels may also be of interest, together with other criteria, such as the success rate of issuing warnings, and the number of false alarms. A related issue is that, due to model errors, some forecasts may not reach critical threshold values, although the flows which are subsequently observed pass those thresholds. Figure 5.5 shows three different examples of forecasts for a single hypothetical event (for example, from three different modelling approaches). In this example, Forecast A correctly indicates that the flooding threshold will be exceeded, although at a slightly later time than actually occurred, whilst Forecasts B and C do not. Also, Forecasts A and B indicate that the flood warning threshold will be exceeded, but Forecast C does not. Another scenario which can occur, of course, is that the actual flows do not reach critical thresholds, but the forecast predicts that those thresholds will be exceeded, in which case a false alarm would be triggered by the forecast output. These examples consider the case of single forecasts, issued at a given time (‘time now’). The start time of a forecast is often called the Forecast Origin, whilst the maximum lead time which the forecast can reliably provide is called the forecast lead time or forecast horizon. For river flow forecasting, the maximum useful forecast lead time is of the order of the catchment response time, if observed rainfall values are being used, but can be extended by using forecasts of catchment rainfall. The model performance at each lead time can be assessed, leading to the concept of fixed lead time forecasts, as illustrated in Fig. 5.6 for the case of a river flow forecast. The figure shows seven forecasts at successive time intervals, with the Forecast Origins shown as circles. In this example, the Forecast Origins coincide with the actual flow values (as might be the case for some forms of real time updating), although this need not necessarily be the case. As an example, the figure also shows the values three time steps ahead for each forecast which, for hourly model runs,

110

5 General Principles

Flow

Time

Fig. 5.6 Illustration of Fixed Lead Time forecasts (circles are fixed origins, triangles are 3-hour ahead forecasts)

would be called the 3-hour ahead forecasts. These values can be joined by a smooth (interpolated) curve to create the so-called 3-hour fixed lead time forecast. In this example, at this lead time, the forecast flows are both later, and lower than, the values which were subsequently observed (and which are shown by the dashed line). Fixed lead time forecasts can be constructed for a range of lead times, and calibration criteria and performance measures can be developed which make use of these values. However, if data assimilation techniques are used, then the values obtained may be significantly improved at shorter lead times, so any performance values quoted should note whether or not updating was in operation. An additional consideration is whether a probabilistic or ensemble approach is used, since additional calibration and performance measures may be required as discussed later.

5.4.2

Model Calibration

The objective in model calibration is to develop a model which best meets the requirements of the modelling study. In flood forecasting applications, examples might include the forecast lead time requirement, and the ability to estimate the timing and magnitude of flood peaks, the times of crossing of thresholds, or levels at a river control structure. Chapters 6 and 7 give some examples of how the requirement can influence the design of river and coastal flood forecasting models. The approach to calibration depends partly on the choice of model, and can include calibration in off-line/simulation mode, assuming perfect foresight of input time series data (e.g. rainfall), or calibration in real time mode, only making use of data up to the time at which each evaluation is performed. In simulation mode, the model is optimised against a number of historical flood events, possibly also including values for the intervening periods to assess the full range of performance. In real time mode, the focus of calibration is typically on the performance of fixed lead time forecasts.

5.4 Model Calibration and Performance

111

Simulation mode is widely used for calibrating process-based and conceptual models, with any real time updating component calibrated at a later stage. Databased models, by contrast, are often calibrated in real time mode, with the data assimilation aspects sometimes integral to the overall model formulation. Also, as discussed in Chapters 6 and 7, many data-based models are event based, in the sense that they are calibrated only to specific flood events whilst, for some process-based and conceptual models, it is essential to calibrate over much longer periods of data. Flood forecasting models can also consist of networks of interconnected models; for example, integrated catchment models consisting of rainfall runoff, flow routing and other model types (e.g. reservoir models), or coastal models comprising offshore, nearshore and foreshore components. For each model in the network, the steps in model calibration can include: ●

●

● ●

Model identification – choice of an appropriate structure for the model (if there is a choice) Choice of calibration criteria – decisions on which criteria to use in model calibration, and the relative importance of each choice Model optimisation – optimisation using best fit, trial and error and other approaches Model validation – tests of model performance using additional datasets, not used in the original calibration

Model Identification techniques can include trial and error for different configurations of model (e.g. alternative choices of stores in a conceptual rainfall runoff model), or automated searches of a wide range of configurations (e.g. for some types of transfer function model). There are many approaches to model calibration and validation, including automated optimisation techniques, such as hill climbing, genetic algorithm, Monte Carlo, and simulated annealing approaches (e.g. Beven 2001; Anderson and Bates 2001). Optimisation criteria can include the timing and magnitude of peak values, measures of the overall shape of the hydrograph (bias, mean absolute error, root mean square error, Nash Sutcliffe efficiency etc.), and threshold crossing measures (e.g. the mean timing error in threshold crossings). Multi objective or multicriteria techniques can also be used. For some of the more physically based model types, some model parameters may also be fixed, or restricted to certain ranges, depending on catchment, river or coastal characteristics. Values can also be calculated on a fixed lead time basis, giving an indication of how model performance changes with increasing lead time. For example, plots can be produced of how mean square error, or Nash Sutcliffe Efficiency, decreases with increasing lead time, and a cut-off value identified beyond which performance drops below an acceptable level. Additional threshold based criteria can also be defined using a contingency table approach, as illustrated in Table 5.4. Based on this table, the following parameters can be defined: ● ● ●

Probability of Detection (POD) = A/(A + C) False Alarm Ratio (FAR) = B/(A + B) Critical Success Index (CSI) = A/(A + B + C)

112

5 General Principles Table 5.4 Simple 2 × 2 contingency table for flood forecasting model evaluation Threshold crossed (observed) Threshold crossed (forecast)

Yes No

Yes

No

A C

B D

Each type of criterion has its own strengths and limitations; for example, measures based on peak values obviously overlook many other aspects of model performance, whilst some whole hydrograph measures (e.g. mean square error) can be sensitive to small timing errors or outliers, and may not consider how performance varies with magnitude (unless they are reported only above certain thresholds), or the sign of errors (e.g. bias). Threshold based measures are closely linked to the operational flood warning requirement, but are based on individual threshold values, although these methods can be extended to sample over a range of possible choices of threshold values (e.g. Environment Agency 2004). Some trial and error may be required to achieve a reasonable compromise across a number of flood events and criteria, together with adoption of some best practice principles for model calibration for flood forecasting, which include (e.g. Environment Agency 2002): ●

●

●

●

●

●

●

●

Data validation – use validated, quality controlled data, particularly for values during flood events (e.g. assessing the high flow performance of stage-discharge relationships). Data sources – calibrate models to the same sources of data that will be used in real time to avoid bias and other problems (e.g. if using radar rainfall data in real time, calibrate to historical radar not raingauge data). Type of event – choose calibration datasets for events of the type(s) which the model is required to represent (e.g. frontal events, thunderstorms, snowmelt). Data currency – use datasets representative of current conditions (e.g. since flood defences were constructed, instruments installed, channels dredged etc.). Run frequency – set the model run frequency (if possible) to adequately resolve the type of flood events being modelled, particularly in the time leading up to flood warning thresholds. Data assimilation – use real time updating where the model type supports this, and the data quality is good enough. Model initialisation – for types of models where this is important, focus on providing realistic initial conditions. Model validation – validate the model outputs against a number of flood events not used in the original model calibration.

Also, for combinations of models, the performance of the overall network should be assessed as well as for individual models within the network. As part of the model documentation, flooding mechanisms and other factors not represented in the model should be highlighted to operational staff, together with the likely uncertainty in model outputs, and the acceptable limits of model performance (maximum flows, lead times etc.).

5.4 Model Calibration and Performance

5.4.3

113

Performance Measures

Once a flood forecasting model has been developed, and integrated into the operational forecasting environment, a period of pre-operational testing will often follow, during which the performance and reliability of the model is assessed, before using it in the flood warning process. This monitoring will then continue as part of the routine assessment of the model, and in particular to help to identify areas for improvements to the calibration and the quality of the input data. Also, any sudden deterioration in performance can be identified; for example, due to changes in the accuracy of input data (e.g. stage-discharge relationships), or in catchment or coastal characteristics (e.g. recently constructed flood defences). Usually, model performance will need to be evaluated following each major flood event, both for post-event reporting and to detect any event-related problems which need to feed into the model development programme. Modern flood forecasting systems often have the facility to automatically calculate a wide range of model performance measures, including some of the real time measures described in the previous sections (e.g. contingency tables, and fixed lead time performance). The types of methods used, and the scale of the assessment which is possible will, in part, depend on the availability of observed data to use in the assessment, and may be limited for some types of model (e.g. process-based models). Also, some types of information which it would be desirable to measure in real time (e.g. inundation extent and depths) are difficult to capture other than by post event survey, and may not be practicable to assess for every event. For assessment of individual flood events, many of the calibration criteria discussed in the previous section can provide useful information on model performance; for example (Environment Agency 2002, 2004; Werner and Self 2005): ●

●

●

●

●

Graphs of model outputs at fixed lead times, compared to the (subsequently) observed values Tables summarising errors in the timing and magnitude of peaks, and event-based measures of performance, such as the R2 coefficient or root mean square error Threshold based measures such as Probability of Detection, False Alarm Rate, Critical Success Index, Over-Prediction Ratio, and timing errors in crossing of thresholds expressed in absolute, mean square or probability of detection terms Lead time based measures such as the bias, average or median error in lead times, or the distribution of errors, and the first forecast of threshold time Map based comparisons of inundation extent (if applicable)

The extent of validation will depend on the application, and in some cases several measures will be evaluated to provide an overall view of the different aspects of model performance, and, if relevant, may extend to analyses of performance with respect to specific aspects of model performance; for example, closures at a flow control structure, or flow diversions to an off-line storage reservoir. Values can also be normalised to facilitate comparisons of trend and variability between different locations. Also, since real time updating can have a significant influence on performance, it is important to indicate whether or not updating was used in the model runs (if applicable).

114

5 General Principles

For evaluating long term performance over a number of events, or when considering the performance of a number of models in a single event (e.g. for post event reporting), it is useful to consider methods for aggregating outputs to allow an overall picture of performance to be obtained. In this situation, it can also be useful to introduce some more operationally focussed performance measures; for example, the success rate for issuing flood warnings, or for property owners acting upon warnings received. However, measures of this type depend on factors beyond just the model performance, and are discussed in more detail in Chapter 11. Some examples of aggregation methods for evaluating forecasting model performance include: ● ● ● ●

Histograms of performance measures across a number of events Plots of performance measures against lead time Tabulated values of performance measures by river reach and frequency Maps of performance statistics to highlight spatial trends

Ideally, values will be presented in non-dimensional form to facilitate comparisons between models, locations and events. For example, for performance monitoring of the Storm Tide Forecasting Service (STFS) operated by the Meteorological Office in the UK (see Chapter 7), some performance statistics which are used include contingency tables based on crossing of alert levels, histograms of surge forecast lead times for alerts raised relative to a threshold, and tables summarising the estimated mean, maximum, root mean square and standard deviation of errors (in metres) across a number of events. For evaluating the performance of probabilistic and ensemble forecasts, options available for meteorological forecasts (e.g. Jolliffe and Stephenson 2003) include the Brier Skill Score (including continuous versions), Ranked Probability Score, Relative Operating Characteristic, Reliability and Sharpness, and these types of measure can also be adapted and extended for flood forecasting applications (e.g. Laio and Tamea 2007). The general aim is usually to assess aspects of the ensemble such as how close the median is to the deterministic value, how well the probability of an event occurring is represented (over the long term), and the confidence which can be placed in the probabilistic estimates, and the value or utility of the forecast (e.g. expressed in terms of cost-loss functions).

5.5

Model Uncertainty

As noted in Chapter 1, uncertainties in the flood forecasting process can arise from many sources, and it is widely agreed that flood forecasts should be issued with an indication of confidence or uncertainty (e.g. Krzysztofowicz 2001). Information on uncertainty can also help with deciding where to focus effort on future model development and data improvement programmes. Some sources of uncertainty in flood forecasting models can include (e.g. Butts et al. 2005):

5.5 Model Uncertainty ● ●

● ●

115

Random or systematic errors in the model inputs (boundary or initial conditions) Random or systematic errors in the observed data used to measure simulation accuracy Uncertainties due to sub-optimal parameter values Uncertainties due to incomplete or biased model structure

Table 5.5 provides examples of some additional sources of uncertainty in river and coastal forecasting models.

Table 5.5 Some of the main sources of uncertainty in river flood forecasting models (Environment Agency 2007; © Environment Agency copyright and/or database right 2008. All rights reserved) Component River models Catchment averaging procedures (for raingauges)

Choice of model type and structure

Model calibration

Operational

Real time updating procedures

Typical sources of uncertainty Representation of physical processes (topography, elevation etc.) Type of rainfall event (convective, frontal, orographic etc.) Rain gauge density and distribution Instrumental problems at one or more of the rain gauges used Lumped, semi-distributed, distributed rainfall inputs Representation of catchment runoff processes River channel and floodplain representation Under/over parameterisation (parsimony) Flood defence loading/fragility (if represented) Gate operations Representation of ungauged inflows Representation of abstractions/discharges Representation of groundwater influences Effectiveness of optimisation routines Choice of optimisation criteria Availability of sufficient high flow events for calibration Skill of person calibrating the model Changes in catchment/channel characteristics since model was calibrated Use of different input data streams from those used in the original model calibration (e.g. radar rainfall or forecasts instead of raingauges) Events outside the range of the model calibration Model stability problems Representation of initial/antecedent conditions Representation of snowmelt (if applicable) Instrument/telemetry downtime problems (rainfall) Appropriateness for the type of model used Sophistication of calibration software Quality of the high flow data used both for calibration and in real time Event specific problems (backwater, bypassing, debris etc.) Instrument/telemetry downtime problems (flows) (continued)

116

5 General Principles

Table 5.5 (continued) Component Coastal models Model boundary conditions

Choice of model type and structure

Calibration Operational

Data assimilation

Typical sources of uncertainty Magnitude and timing of changes in wind direction and storm track Subgrid scale/secondary depressions Peak values for astronomical tides Grid resolution – inadequate representation of local bathy metric and topographic features that cause changes in local water levels Coupling of offshore and nearshore models Availability of sufficient extreme events for model verification Influence of mobile/shingle beaches Changes in characteristics since model was calibrated Events outside the range of the model calibration Instrument/telemetry downtime problems Values for trigger levels Extent of improvement and limitations on data quality

The uncertainty in weather radar data, satellite data, meteorological forecasts, and other inputs could also be considered where applicable. There are many techniques for estimating the uncertainty in model outputs in real time and these include: ●

●

●

●

Sensitivity tests – simple tests of alternative data inputs, model structures, parameter values etc. Multi-model approaches – in which the outputs from several models are compared to examine the range of estimates (e.g. Georgakakos et al. 2004). Probabilistic approaches – in which multiple forecast realisations are produced, typically based on stochastic sampling from probability distribution functions for the model parameters, input data, and/or boundary conditions. Ensemble approaches – which use a collection of forecasts obtained by perturbing the input data and/or model parameters for a model over plausible ranges (and can include multi-model ensembles).

Some data assimilation techniques, such as the Kalman Filter, also automatically provide an estimate of uncertainty as part of the assimilation process (see Section 5.3). One general distinction between approaches is whether the likely uncertainty in input data, parameters etc. is defined beforehand based on previous forecasts, or assessed dynamically, during a flood event, based on current forecasts and observed data. Sensitivity tests can include ‘what if’ scenarios, of the type described in Section 5.2, whilst Rotach et al. (2007) provide an example of a real time multi-model approach in which the outputs from a wide range of atmospheric models, and conceptual and process-based catchment flood forecasting models, can be compared in a common format, with colour coding on maps and tables to show where threshold levels have been exceeded for specific locations.

5.5 Model Uncertainty

117

Probabilistic and ensemble approaches are an active area for research in flood forecasting, with several operational systems under development worldwide. Probabilistic approaches generate multiple scenarios for model outputs by random sampling from probability distributions for the model parameters, initial conditions, boundary conditions, or input data, and many hundreds or thousands of scenarios might be generated offline, for subsequent analysis, or in real time. For example, Pappenberger et al. (2004) describe Monte Carlo sampling of rating curve and roughness coefficient uncertainty in a hydraulic model, whilst Pappenberger et al. (2007) describe studies into the influence of uncertainty on flood inundation extent using multiple combinations of effective model parameters for a 2D flood inundation model. By contrast, ensemble techniques employ a more focussed sampling technique (sometimes on account of constraints on model run times) to derive a smaller number of realisations (typically of the order 10–100) which span the likely range of outcomes (Box 5.2).

Box 5.2 Some general principles of ensemble flood forecasting Figure 5.7 shows a simple example of an ensemble approach for a catchment flood forecasting problem. In this example, five possible tracks are shown for an idealised storm approaching the catchment. The resulting catchment rainfall estimates for each realisation could then be propagated through a rainfall runoff and flow routing model to estimate the uncertainty in flows in the lower parts of the catchment. The figure also shows some other sources of uncertainty, including uncertainty in catchment antecedent conditions, stage discharge relationships and river channel survey data. The ensemble rainfall inputs might be combined with models to represent the uncertainties in these components as well, so that these various sources of uncertainty are also included in the flow estimates. Ensemble outputs can be presented in a wide range of formats, including graphical, map-based and tabulated outputs. The raw outputs can also be used as input to a decision support system, as discussed further in Chapter 10. Figure 5.8 shows some idealised examples of graphical methods for presenting information on uncertainty in a flood hydrograph. The figure shows the following four types of display together with the deterministic forecast (single line): • Spaghetti plot – the raw ensemble outputs (of which 6 are shown for illustration) • Plume – flow values within defined probability bounds (which often include additional sets of bounds with appropriate colour coding) • Whisker plot – giving the median, 25 and 75 percentile values (say) and the maximum and minimum values • Stacked histogram or bar chart – for a pre-defined set of values (e.g. 25%, 50%, 75% and 100%) (continued)

118

5 General Principles

Box 5.2 (continued) Level

Antecedent conditions Rating River flows

Flow

Level Rating Flow

Storm River Gauging Station Town Raingauge

Fig. 5.7 Illustration of some sources of uncertainty for a catchment flood forecasting problem Flow

Flow

Time

Time

Flow

Flow

Time

Time

Fig. 5.8 Illustration of graphical presentations of ensemble outputs for a flow hydrograph

(continued)

5.5 Model Uncertainty

119

Box 5.2 (continued) Note that the probability distributions shown are idealised, and are not directly comparable between the different graphs; for example, spaghetti plots usually show a complex array of overlapping possible scenarios. Many other ways of presenting probabilistic information are also possible, including tabulated, map based and other formats, such as persistence tables, and strike probability plots for hurricane landfalls. Experience from other areas (e.g. meteorology) suggests that forecast products which include measures of uncertainty should be developed in close consultation with end users to help choose the most appropriate form of presentation, and to provide advice on how the information can be interpreted.

Some current research themes in ensemble forecasting include: ●

●

●

●

Downscaling – of the outputs from Numerical Weather Prediction models to the scales of interest for hydrological modelling using both statistical and dynamic techniques (e.g. Rebora et al. 2006), and (if applicable) blending ensembles generated at different time scales into a seamless ensemble. Also, generation of higher resolution and shorter lead time ensembles (e.g. local area and storm scale NWP models, and probabilistic nowcasting techniques) Computational efficiency – in producing multiple ensemble flood forecasting model outputs in a time which is operationally useful, with potential solutions including emulators, parallel processing, simplification and rationalisation of models, and filtering or clustering of ensembles Decision support – how to use stochastic and ensemble forecasts in making decisions on issuing flood warnings, operating control structures etc., and to communicate information on uncertainty in flood forecast products to the public and emergency response organizations (see Chapters 8 and 10) Hindcasting – reanalysis or derivation of long term ensemble simulations of the outputs from Numerical Weather Prediction models in their present day form (resolution, parameterizations etc.) to provide test datasets for use in developing ensemble flood forecasting techniques

Two major research programmes which are considering these and other topics are: ●

●

Hydrological Ensemble Prediction Experiment (HEPEX) – an international collaboration in ensemble flood forecasting techniques involving scientists from the National Weather Service in the USA, the European Centre for Medium-Range Weather Forecasts (ECMWF) and the European Joint Research Centre, Canada, Italy, Brazil, Bangladesh and elsewhere (Schaake et al. 2005) COST-731 – uncertainty in advanced meteo-hydrological forecast systems – a European initiative to examine meteorological and hydrological techniques for ensemble flood forecasting, including the use of forecasts in decision making (e.g. flood warning), and involving meteorological and hydrological services from more than ten European countries (European Cooperation in Science and Technical Research)

120

5 General Principles

Table 5.6 also presents several examples of research and operational studies in flood forecasting applications, whilst later chapters provide additional examples for coastal applications, and of using probabilistic forecasts to assist in optimising decision making for flood control and emergency response (Box 5.3). Table 5.6 Examples of research and operational applications of probabilistic and ensemble flood forecasting techniques System Extended Streamflow Prediction System, USA

European Flood Alert System (EFAS) Bureau of Meteorology, Australia

Environment Agency, UK Lower Severn, UK

Blue River, USA; Welland and Glen catchment, England

Probabilistic or ensemble component

Reference

Statistical sampling based on long Schaake et al. (2005) term hydrological records; and development of short term ensembles for a range of sources of uncertainty ECMWF ensemble rainfall forecasts Thielen et al. (2004) Ensemble Quantitative Precipitation Forecasts, stochastic nowcasting techniques, and multi-model assessments Ensemble Surge Forecasts, stochastic wave modeling techniques Data assimilation and propagation of uncertainty in a network of transfer function models Ensemble Kalman filtering

Elliott et al. (2005)

Tozer et al. (2007) Romanowicz et al. (2006)

Butts et al. (2005)

Box 5.3 Decision making using probabilistic forecasts Probabilistic and ensemble flood forecasts provide a range of possible future flow scenarios and, for flood warning applications, it is interesting to consider how this information can be used to help in making decisions about whether or not to issue a warning. The development of techniques for defining probabilistic thresholds is an active research area and, in some cases, builds on ideas which are already well established in ensemble forecasting in meteorology (e.g. Jolliffe and Stephenson 2003). Perhaps the simplest approach, and one which is widely used, is to use a qualitative (“eyeball”) assessment of the spread of forecasts to provide an indication of uncertainty, and whether any of the ensemble members exceed thresholds of interest. Outputs can be viewed in a range of formats, including spaghetti plots, plumes, whisker plots and histograms. For example, when ensemble rainfall forecasts are used as inputs to a rainfall runoff model, the uncertainty in runoff estimates would be expected to be greater for some types of events, such as convective rainfall events, and with increasing lead time. Over a number of events, a forecaster could build (continued)

5.5 Model Uncertainty

121

Box 5.3 (continued) up experience in how the uncertainty varies between locations and the type and magnitude of event, and of the confidence to attach to forecasts in different situations and from different models. Other criteria, such as the clustering of ensembles, have also been shown to be useful in some studies. A logical next step is to interpret the distribution of forecasts in terms of key flooding and other thresholds, ideally leading to a single binary (yes/no) decision. As for deterministic (single valued) forecasts, a flooding threshold can be defined for the parameter of interest (e.g. rainfall, levels, flows) but, in addition, an exceedance probability needs to be defined for the proportion of ensemble members which exceed that value. Action could then be taken (e.g. issuing a warning) when the appropriate percentage of ensemble members exceeds the critical value. Additional criteria could also be introduced; for example, some studies have shown that false alarm rates can be reduced if the persistence in threshold exceedance between two or more model runs is also considered. The interpretation of ensembles in terms of thresholds also allows other forms of output to be made available to the forecaster, such as colour coded maps showing the locations at which flows (or other parameters) exceed threshold values. To estimate appropriate values for probabilistic thresholds, perhaps the simplest approach is to assess forecasting performance over a number of events by trial and error. The optimum value can be deduced in terms of statistical measures such as Probability of Detection, average lead time, and the number of false alarms (see Section 5.4). More generally, several studies have suggested (e.g. Roulin 2007) that techniques developed in meteorology for assessing the economic value of forecasts (e.g. Richardson 2003) might also be adapted for flood forecasting applications. In this approach, the value of a forecast can be expressed as the economic benefit, over a number of events, from having the forecast available, compared to the situation of only having climatological information available. The economic value (or reduction in mean expense) typically depends on the performance of the forecast (expressed in terms of Probability of Detection, and False Alarm Rate), the probability threshold which is selected, and the average costs and losses (or risk profile) for the recipient(s) of the forecast, and is often expressed as a ratio to the equivalent value for a perfect forecast. Probability thresholds can then be selected to provide the maximum economic value for each individual user (or group of users), introducing the concept that the value of a forecast depends not only on its performance, but also on the risk tolerance and economic circumstances of the user. Here, the costs are those incurred in taking mitigating action, whilst the losses are those due to flooding in the absence of a warning. For example, for a temporary or demountable flood defence barrier, assuming that appropriate action is always taken when a forecast is issued, a cost (e.g. in staff and transportation costs) is incurred each time that the barrier is installed, including occasions when the forecast provides a false alarm. (continued)

122

5 General Principles

Box 5.3 (continued) However, losses (e.g. damage to property and vehicles) are only incurred when flooding occurs but is not forecast. Chapter 10 provides some further discussion of this topic, and illustrates how costs and losses can be summarised in an expense table, similar to the contingency table shown in Table 5.4. Some extensions of the approach are to consider other factors, such as only partial mitigation of losses, a partial response to warnings (e.g. only some people take action), and variations in costs and losses with forecast lead time. Utility functions can also be used to bring in other factors which cannot easily be expressed in monetary terms, such as the differing risk profiles of recipients, and tolerance to false alarms. For the more complex approaches, thresholds need to be computed dynamically for each event, and the calculations are probably best performed within a decision support system. Cost-loss analyses of this type show potential for flood forecasting applications, particularly where flooding is frequent, or decisions need to be taken on a regular basis, and are already used in some of the real time control and decision support systems which are discussed in Chapter 8 for water supply reservoirs and hydropower schemes. However, as with many other types of flood-related analysis, the issue of extreme (rare) events needs particular consideration in order to provide statistically meaningful samples, and techniques developed in other areas, such as for design flood estimation (e.g. regional pooling groups and continuous simulation), might provide one route to further development of these techniques for application to low probability, high impact events. The social and behavioural aspects of response to extreme events also need to be considered (for example, tolerance to false alarms, and modifications to response when faced with large or catastrophic losses).

Chapter 6

Rivers

Flood forecasting models for rivers can range from simple empirical relationships to complex integrated catchment models. Forecasts may be based simply on river level or flow observations at locations upstream of the site of interest, or use observations and possibly forecasts of rainfall to gain additional lead time. This chapter begins with a discussion of the main factors which can influence the design of a river flood forecasting model, including the forecasting requirement and the availability of real time data, and then describes two main categories of model; rainfall runoff models, and river channel (flow routing) models. Examples are provided for a range of process-based and conceptual modelling techniques, and for data-based approaches such as transfer function and artificial neural network models.

6.1

Model Design

River forecasting models aim to estimate river conditions at or near sites of interest in a river basin (or catchment), such as locations at which there is a flooding history, or where studies suggest that there may be a significant flood risk. Forecasts may also be required at specific structures, such as reservoirs or flow control gates, to assist with real time control of river flows to mitigate flooding. The modelling approaches used in flood forecasting applications have many similarities with the techniques used for catchment simulation, and the factors which need to be considered during the initial catchment conceptualisation include the catchment response (response times, lakes, reservoirs etc.), local influences on river levels (tidal, backwater, confluences, structures etc.), any artificial controls on flows (dams, river gates, tidal barriers, abstractions/discharges etc.), and other factors, such as snowmelt and groundwater influences. Chapter 8 discusses modelling approaches for several of these possible catchment forecasting problems. For real time applications, the following factors also need to be considered when considering the most appropriate approach to use:

K. Sene, Flood Warning, Forecasting and Emergency Response, © Springer Science + Business Media B.V. 2008

123

124 ● ●

● ●

●

6 Rivers

Forecasting requirement – the intended use of the forecasts Data availability – the availability of data and other information in real time and for model calibration Forecasting system – the system(s) on which models will be operated Performance requirements – for run time, accuracy, lead time and other variables Type of model – process-based, conceptual or data-based

Model design is therefore often a compromise, and the overall approach may also be constrained by additional factors such as costs and system limitations, as discussed in Chapter 11. Chapter 5 discusses some of the factors to consider regarding forecasting systems and model performance requirements, whilst the following sections present a brief summary of the remaining topics.

6.1.1

Forecasting Requirement

The forecasting requirement is often one of the main criteria in selection of an appropriate modelling approach. Some factors to consider can include the level of flood risk in the catchment, the specific locations at which forecasts are required, and the intended operational use of the forecasts. As noted in Chapter 5, the locations at which forecasts are required are often called Forecasting Points, and the model (or models) used will typically be optimised to achieve the required performance at those locations. For example, if the forecasting model is based on an existing simulation model, it may be desirable to remove some of the model complexity away from these key locations to make the model run faster, or to ensure model stability under all combinations of flow conditions. This is a commonly used approach with real time hydrodynamic models, for example. For river flood forecasting applications, some examples of possible locations for Forecasting Points include: ● ● ● ● ●

Flood warning areas (to assist with issuing warnings) High risk locations (power stations, hospitals etc.) River control structures (to assist with structure operations) River gauging stations (for real time evaluation and updating of forecasts) Reservoirs and dams (to assist with operations to reduce flood risk)

For a given location, there may be more than one Forecasting Point; for example, a single Flood Warning Area could include points at several anticipated overtopping locations in a flood defence system. Where Forecasting Points are separated by some distance, another consideration is whether to develop individual models appropriate to each point, or an integrated catchment model covering the whole catchment (or, at least, the catchment above the furthest point downstream). The integrated solution may be more complex and expensive to develop initially, but may reduce telemetry requirements, and may

6.1 Model Design

125

allow solutions to be developed for intermediate Forecasting Points where the risk and benefits are too low to justify development of a forecasting model specifically for that location, such as small numbers of properties or farmland. Chapter 8 provides more background on integrated catchment modelling techniques. The intended use of forecasts can also strongly influence the types of model which are selected. For example, it can be useful to view the problem from the point of view of potential users of the forecasts, such as the emergency services, local authorities, and the public, and Table 6.1 provides some examples of how user requirements can translate into a modelling requirement. Here, the term hydrograph refers to the variation in river levels or flows over time. These examples, although simplified, illustrate that the complexity of the modelling approach can potentially vary considerably between applications. Models may also be required for purposes other than flood forecasting; for example, for forecasting water resource availability, or water depths and velocities for navigation, and again Chapter 8 discusses this topic in more detail. In some of these applications, the main interest is in forecasts of river levels and flows in the time leading up to crossing of a threshold value which, as described in Chapter 3, can be at levels some way below the flood peak. This requirement can have implications for the choice of model; for example, for hydrodynamic models, if thresholds are set at river levels for which flows remain in channel, then there may be no particular requirement to model the details of floodplain flows or flood

Table 6.1 Illustration of how user requirements can translate to a modelling requirement (Adapted from Environment Agency 2002, © Environment Agency copyright and/or database right 2008. All rights reserved) Question

Possible minimum modelling requirement

Will flooding occur? When will the flood be at its worst? When will flooding begin? What depths (and, possibly, velocities) will be reached at this street, road, railway etc.? How long will the flooding last?

Expected value for the peak level Magnitude and timing of the peak Time at which a threshold level is reached The peak level reached and/or the flow on the floodplain

When will flood levels drop?

Which properties will be flooded? Where will flood defences be overtopped first (or where should sandbags be placed?) When should this control gate be operated? Does the reservoir level need to be reduced?

Times of crossing thresholds on the rising and falling hydrograph Time of dropping below a threshold; possibly also a floodplain and/or reservoir/storage drainage model Flows and volumes on the floodplain and location of any overtopping The peak level reached at one or more locations in a flood defence system, and possibly defence breach risk modelling Can vary widely, from very simple models to decision support systems incorporating optimisation algorithms

126

6 Rivers

defence overtopping in the neighbourhood of Forecasting Points, which can simplify the overall model development (although it is usually useful to know the likely timing and magnitude of peak levels at telemetry sites). If there are no limitations on budget, then one option is to produce the best model that is technically feasible, including floodplain flows, river control structures, and other factors, as appropriate. However, even the best models make some assumptions, and may not include all flooding mechanisms (e.g. urban drainage), so it is important to be realistic about what the model can (and cannot) achieve in discussions with potential users of the forecast outputs. Also, sometimes a simple model, such as a regression relationship, may meet the requirement, representing a considerable saving in development time and costs compared to a more sophisticated approach. In many cases, the choice of modelling approach is also influenced by the level of flood risk at the selected Forecasting Points. For example, if only a few properties are at risk from occasional flooding, there may be less justification for developing a complex model than for a major city, with thousands of properties at risk. These economic aspects of model selection are discussed in Chapter 11 as part of a wider discussion of the costs and benefits of flood warning.

6.1.2

Data Availability

In addition to the availability of historical calibration data, and survey and other static data, a key consideration in designing river forecasting models is the availability of near real time data for model operation and real time updating. Whilst historical records are useful for exploring catchment rainfall distributions and flow response, and in model calibration, ultimately the model needs to be configured to operate using the real time data feeds which are available. As discussed in Chapter 2, these can include rainfall observations from raingauges, weather radar and satellite, and rainfall forecasts from nowcasting and Numerical Weather Prediction models, and observations of river level and flows, but may also include information on control structure settings, pumping operations, reservoir levels, evaporation, catchment conditions, and other parameters. Also, as described in Chapter 10, the forecasting model may form part of a wider decision support system incorporating information on flood defence condition, locations of temporary defences (barrier, sandbags etc.), and the current flooding situation (floodplain depths etc.). The requirement for forecast lead time is an important consideration in the choice of modelling approach, since this may influence whether a river channel (flow routing) model relying simply on information from locations further upstream is sufficient, or whether a rainfall runoff model, representing the relationship between rainfall and flow, is required. Generally there is a trade-off between increasing lead time and decreasing forecast accuracy (e.g. Reed 1984; Environment Agency 2002), as illustrated in Table 6.2.

6.1 Model Design

127

Table 6.2 Illustration of the trade-off between forecast lead time and accuracy Approach in order of decreasing lead time Typical modelling approach Rainfall forecasts (e.g. Numerical Weather Prediction) Rainfall observations (e.g. raingauge, weather radar) River flow forecasts upstream of the Forecasting Point River flow observations upstream of the Forecasting Point

Rainfall Runoff Models Rainfall Runoff Models Flow Routing Models Flow Routing Models

Various attempts have been made to link the choice of model to catchment response; for example, Reed (1984) proposed the following criteria based on the characteristic response time of the catchment (TP): TP ≤ 3 hours 3 ≤ TP ≤ 9 hours TP ≥ 9 hours

Rainfall runoff modelling plus quantitative rainfall forecasts Rainfall runoff modelling Flow routing

However, this was only proposed as a rule of thumb, and the many assumptions in this approach were noted. Later approaches have included additional factors such as a distinction between overland and river channel flows (e.g. Lettenmaier and Wood 1993), and allowances for the various time delays in the data processing, model run, and dissemination chain which are discussed in Chapters 5 and 9 (e.g. Tilford et al. 2007). Map based presentations of catchment response can also help in deciding on an appropriate choice of models; for example, by plotting lines of equal response times based on analyses of historical data, or modelling studies, together with the locations of key Forecasting Points. Figure 6.1 illustrates one example (Box 6.1) and Chapter 3 provides a further example, and more sophisticated contoured or grid based analyses are easily performed using Geographical Information Systems. When estimating response times using historical rainfall and flow data, care should be taken to choose events of a similar magnitude and type to those for which the model is required, if this information is available. For example, response times can vary significantly depending on catchment antecedent conditions, and whether river flows are in-bank or extend onto the floodplain. For a given forecasting problem, some issues which may influence the choice of real time data inputs used, and the overall modelling approach, include: ●

●

Spatial density – Chapter 2 includes a short discussion on network densities for raingauges and river gauges for various applications. For example, for flood forecasting applications, it is often desirable to have a river gauge at or near each of the key Forecasting Points of interest. For rainfall runoff models, another consideration is whether to use process-based (distributed) models when the main sources of input data are also gridded (e.g. weather radar data, Numerical Weather Prediction model outputs), or to aggregate the gridded outputs to catchment scale, and use a conceptual (lumped) or data-based approach (see later). Data quality – The accuracy and consistency of model outputs will depend on the quality of data inputs (‘rubbish in-rubbish out’), so difficult choices may need to be made on whether to include instruments which are sited conveniently

128

●

●

●

●

6 Rivers

for inclusion in the model, but have poor or unreliable outputs. In particular, performance during heavy rainfall and high flow conditions will need to be considered, particularly for river gauges; for example, the risk of flows bypassing the gauge, backwater influences, and the accuracy of the high flow end of stagedischarge relationships (see Chapter 2). Data reliability – Since a primary requirement of a forecasting model is to operate during flood events, the reliability of real time data feeds during heavy rainfall is of particular importance; for example, is a monitoring station likely to be flooded, or telemetry links interrupted? A fall-back hierarchy of data inputs may need to be considered, and the model calibrated or tested for each of these scenarios (see Chapter 5). Model initialisation requirements – Certain types of model may have particular requirements for real time information to initialise the model state. Examples include reservoir models, for which information on reservoir levels and gate settings (if applicable) is usually required, hydrodynamic models, and some types of rainfall runoff model, which may need long runs of historical data to reinitialise after a break in operations. Data assimilation – If real time updating is used, data quality issues are of particular importance for any river gauging stations which are used as updating locations. If data quality is a concern, then other options may need to be considered, such as not using updating at those locations, or only updating within specified flow ranges (see Chapter 5). Ensemble forecasting – If a probabilistic or ensemble approach is to be used for rainfall or other inputs (see Chapter 5), then multiple model runs may need to be performed at each forecasting time step, with possible issues of model run times and post processing of model outputs.

6.1.3

Type of Model

In choosing an appropriate modelling solution, there are many types of model which could potentially be used, and one widely used classification scheme is as follows: ●

●

●

Process or physically based models – which model the spatial variations in catchment or river response in detail, typically using physically based equations or functions, on a regular or irregular grid (sometimes called distributed models) Conceptual models – which, although to a certain extent physically based, conceptualise the overall catchment or river response, whilst still representing the main features of the response Data based models – which use systems analysis concepts, such as transfer functions and artificial neural networks, to capture the main features of river response (and are sometimes called data driven, metric or black box models)

In some applications, the choice of approach may be decided by external factors, such as the time and budget available, local policy, the model calibration software

6.1 Model Design

129

Box 6.1 Model selection example Figure 6.1 illustrates a simple example of the model selection problem for a small catchment. There are two towns along the main river with a significant flood risk, and a Forecasting Point is required in each location. There are two raingauges in the catchment, and four river gauging stations, all of which are on telemetry. However, there is no weather radar data available of suitable accuracy for rainfall estimates.

1 Reservoir 1 hr

A

3 hrs

2 C

D

B

2 hrs

5 hrs

2 hrs

7 hrs

River Gauging Station Town Raingauge

Fig. 6.1 Example of a simple catchment modelling problem

A wide range of modelling solutions could be proposed, ranging from a simple correlation based approach, to a full hydrodynamic modelling solution. Configuration 1 Rainfall Runoff 1

Rain 1

Gauge D

Configuration 2 Rainfall Runoff 1

Rain 1

Rain 2

Gauge A

Rainfall Runoff 2

Gauge B

Rainfall Runoff 3

Gauge C

Routing 1

Gauge D

Fig. 6.2 Some possible model configuration options

(continued)

130

6 Rivers

Box 6.1 (continued) Figure 6.2 shows two intermediate modelling solutions; Configuration 1, which uses a single rainfall runoff model to Gauge D, with flows at sites further upstream estimated by scaling on catchment area or correlations (not shown), and Configuration 2, which uses all four river gauges in the catchment, with a hydrological or hydrodynamic flow routing model extending from Gauges A, B and C to Gauge D. In Configuration 2, the use of a hydrodynamic model would allow forecasts to be extracted at internal node points in the model if required; for example at the uppermost Forecasting Point whilst, if a hydrological routing approach is used, the model could extend from Gauge A to Gauge D, with the inflows from Gauges B and C included at appropriate node points. As is often the case, the models are configured around the locations of telemetry gauges, rather than to natural features in the catchments such as confluences. This also allows the forecast model outputs to be updated based on real time data from those locations. Configuration 1 ignores possible differences in catchment response between the various subcatchments, whilst Configuration 2 includes some representation of this effect, and of confluence influences at the uppermost Forecasting Point. Of course, several other configurations could be envisaged, with one obvious choice being to estimate the rainfall in individual tributary subcatchments using a catchment averaging approach based on both raingauges, and possibly additional raingauges outside the catchment, and another option being to include a sub model for the influence of the reservoir.

available, the ability to run models in real time, and the availability of real time data for model operation. Some types of model may also be better suited to certain types of forecasting problem than others, and Chapter 8 presents examples of a variety of forecasting approaches for flash floods, snowmelt, river ice, reservoirs, control structures, urban drainage and geotechnical risks such as dam break and flood defence breaches. The skills and modelling preferences of individuals can also be a factor and, whilst it might be anticipated that the more complex types of model would have better performance, this is not necessarily always the case in real time applications (e.g. Beven 2001; Arduino et al. 2005). For example, for rainfall runoff models, Table 6.3 provides some examples of the advantages which are sometimes stated for each general type of model when used for flood forecasting applications. A similar table could also be produced for flow routing models. Here, a parsimonious model is one which is no more complex than necessary to predict the observations sufficiently accurately to be useful, and links to the concept of equifinality, which is that there may be many models of a catchment (e.g. parameter sets for an individual type of model) that are acceptably consistent with the observations available (e.g. Beven 2001).

6.1 Model Design

131

Table 6.3 Some potential advantages of different rainfall runoff modelling approaches Type

Description

Process based models

Well suited to operate with spatially distributed inputs (weather radar, Numerical Weather Prediction model outputs, multiple inflow locations etc.) Can represent variations in runoff with both storm direction and distribution over a catchment Parameter values are often physically based and can be related to catchment topography, soil types, channel characteristics etc., including (possibly) the potential to represent events outside the range of calibration data, and for ungauged catchments Fewer parameters to specify or calibrate than in the process based approach Fast and stable for real time operation Easier to implement real time state updating than for process based models Parsimonious, run times are fast, and models are tolerant to data loss Can be optimised directly for the lead times of interest The model fitting or data assimilation approach automatically provides a measure of uncertainty for some types of model

Conceptual models

Data based models

Later sections also discuss some other factors to consider in the choice of approach, including model run times, model performance outside the range of calibration, model stability, the representation of variations in runoff within a catchment, and of catchment initial conditions, and the number of parameters which need to be estimated or calibrated. Another approach to the question of model selection is to use more than one model for a given forecasting problem and to compare the outputs to see if they agree on likely future flows and, in particular, on the possibility of flooding thresholds being exceeded (e.g. Rotach et al. 2007). Some additional factors to consider in choosing the optimum modelling approach include: ●

●

Is the availability and quality of real time data sufficient to develop a useful model? Are the catchment and flooding processes understood well enough to develop a suitable model?

For example, it is often the case that the near real time information available is not as complete or reliable as would ideally be required, or that there are uncertainties in how the catchment responds to rainfall, or the precise conditions which cause flooding to occur. If rain gauge information on rainfall is insufficient, then one option is to use inputs from other techniques with an improved spatial coverage, such as weather radar data, Numerical Weather Prediction or nowcasting model outputs, and satellite data. Best practice would then be to investigate the performance of these inputs using historical records for the catchment (or nearby catchments)

132

6 Rivers

and, if satisfied with the performance, to calibrate the model directly to the inputs which will be used in real time. If the river gauge coverage is incomplete, one option might be to delay implementation of the model until sufficient real time data is available. Temporary monitoring equipment, such as water level recorders, might also be installed, and exploratory modelling studies performed to better understand the catchment response. An interim model could also be developed, providing a basic framework for flood forecasting, but with some components scheduled for improvement when additional information becomes available. If the decision is taken to install new monitoring equipment, this introduces a delay into the model development programme whilst site permissions are obtained, site works are completed, and a period of reliable data is collected to use in model calibration. Some options for accelerating this process include adding a telemetry link to existing instruments which are known to perform reliably, or upgrading existing equipment to resolve known problems, such as with the high flow end of stage discharge relationships (see Chapter 2). The approach used will depend on the time and budget available, and Chapter 11 describes a range of approaches to prioritisation of investment. For the telemetry and modelling solution aspects, some additional guidance can also be found in Sene et al. (2006) and Tilford et al. (2007).

6.2 6.2.1

Rainfall Runoff Models Introduction

Rainfall runoff models aim to estimate flows in a river channel from observations or forecasts of rainfall, and are sometimes called hydrological or hydrologic models. If observed values of rainfall are used, the maximum lead time provided is typically similar to the average response time of the catchment to rainfall, although may be influenced by factors such as initial catchment conditions, snowmelt, storm speed and direction, and reservoir storage. In many cases, the lead time provided by using observed rainfall from raingauges or weather radar can be sufficient, but can be extended by using rainfall forecasts, although usually with a trade off between increasing lead time and decreasing accuracy and a coarser spatial resolution. Some general categories of rainfall runoff model include: ●

Process based models – which typically attempt to model catchment processes in some detail, using partial or ordinary differential equations and possibly simpler, more conceptual relationships, to represent overland flows, soil infiltration and percolation, groundwater flow, evapotranspiration, and other factors. Models are often grid based, using regular or irregular grids, ideally with the grid scale small relative to the scale of the catchment. The gridded approach is well suited

6.2 Rainfall Runoff Models

●

●

133

to use with weather radar or satellite data, and nowcasting and Numerical Weather Prediction model forecasts, although rainfall fields derived from raingauge observations can also be used. Models of this type can, in principle, represent the spatial variations in runoff which occur as a storm passes over a catchment, including modelling the influence of storm path and direction on flood response. Some alternative names which are used include distributed models, and physically based models (although the degree to which processes are modeled can vary between each approach). Conceptual models – sometimes called lumped conceptual models, also use a physically based approach in the sense that they attempt to capture the main features of catchment response through empirical and simple physically-based equations. However, models are typically formulated at the scale of the whole catchment, and use rainfall inputs averaged (or lumped) at that scale, hence losing some of the detail of the flood response provided by a fully distributed approach. Data based models – such as transfer functions and artificial neural networks typically view the rainfall-flow or rainfall-level forecasting problem from a systems analysis perspective, with the aim often being to optimise forecasts at one or more lead times (1 hour ahead, 2 hours ahead, for example), making best use of all real time data available on current and past rainfall and river conditions. Rainfall inputs are typically used as received, and are not pre-processed to catchment or grid scale. Some alternative names for this type of model include Black Box or Metric models. One of the earliest types of model to be used was the unit hydrograph approach (Sherman 1932), although this has various drawbacks for real time application, and is little used nowadays in flood forecasting applications.

In rainfall runoff forecasting applications, one distinguishing feature of the data based approach is that models are event-based, in that they are usually only operated when required during a flood event. By contrast, process based and conceptual models usually need to save some measure of catchment conditions at the end of each run, or be able to reconstruct that information by retrieving historical data if the model has not been in continuous operation, or by receiving an external feed of observations or other estimates of catchment state. However, the distinction between these types of model is not always clear cut; for example, process based models can include conceptual components for some processes, whilst conceptual models can be applied to individual subcatchments, or to hydrological response units based on catchment characteristics such as vegetation, soil, geology and topography (e.g. Beven 2001), thereby providing a form of semi distributed model. Various hybrid forms of model are also available; for example, conceptual models incorporating transfer function components, and data based models representing slow and fast response timescales for catchment response, which can be interpreted in terms of groundwater and surface flow pathways, and which are sometimes called hybrid-metric-conceptual or grey box models. More generally, a widely used option for conceptual and data based models is to interconnect a number of models for individual tributaries via flow routing models to represent, to some extent, the

134

6 Rivers

variations in runoff as a storm passes across a catchment (see Section 6.1 and Chapter 8 for examples of this approach) There have been several major intercomparison studies of the performance of different types of rainfall runoff model, including the following two studies: ●

●

European Flood Forecasting Study (EFFS) – this major collaborative research study into improved flood forecasting techniques for large catchments included a comparison of the performance of eight process based and conceptual rainfall runoff models on 14 catchments in Europe (European Flood Forecasting System 2003). World Meteorological Organisation 1992 study – this was the third such intercomparison exercise organised by WMO, and involved 14 models from 11 countries (previous studies were in 1974 and 1983). Models were compared using data for catchments with areas of 104, 1,100 and 2,344 km2, using real time updating where available, and a range of performance statistics (World Meteorological Organisation 1992).

However, due to the number of factors to consider, conclusions from this type of study can sometimes be mixed, and Reed (1984), for example, suggests that a thorough assessment of rainfall runoff methods for flood forecasting might need to consider: ● ● ● ● ●

●

At least four distinct approaches Perhaps four different model structures in each approach Several methods of real time correction Perhaps several alternative types of rainfall forecast Various objective functions both for model calibration and performance assessment Application to a range of flood forecasting problems

However, despite these difficulties, intercomparisons can provide useful insights into model performance, and some general conclusions on factors which can influence the performance of rainfall runoff models in forecasting applications include: ●

●

●

Calibration – the approach used to calibration, the calibration criteria used, and more qualitative factors, such as the skill of the modeller. Also, for real time applications, it is often appropriate to use additional performance measures, based on threshold crossing and the model performance at different lead times, in addition to the classical measures used for simulation modelling (see Chapter 5). Initial conditions – except in some locations (e.g. some desert mountains), the runoff generated by a catchment often depends strongly on catchment initial conditions (soil moisture, reservoir and lake levels, snowcover etc.). The degree to which a model is able to capture this effect is an important consideration for rainfall runoff models. Real time updating – given the many uncertainties in measuring and forecasting rainfall, and estimating the resulting flows, updating (or data assimilation) can be particularly effective for rainfall runoff models, provided that the data quality

6.2 Rainfall Runoff Models

135

is sufficient. It is important to examine the performance of the model both with and without updating, and perhaps to evaluate performance with a range of approaches to updating (see Chapter 5). The following sections discuss these points further, whilst additional background on rainfall runoff modelling techniques for a range of applications can be found in Singh (1988, 1989), Beven (2001) and Moore et al. (2006), for example, and in papers and conference proceedings from the MOPEX and IAHS-PUB research studies for ungauged catchments which are described later.

6.2.2

Process-Based Models

Process based, distributed or physically based hydrological models have been used in research and simulation studies for many years (e.g. Vieux 2004) and are gradually being adopted for real time forecasting applications. Historically, one of the obstacles to real time use has been that models of this type are data hungry, in the sense that they require detailed spatial information on rainfall, temperature, evaporation, land use, catchment state and other factors, and model run times may be too long. Increasingly, however, these problems are being overcome through improvements in computing power, and the resolution of weather radar data and weather forecasting models, and in remote sensing and spatial analysis techniques for assessing factors such as topography, vegetation, land use, river networks etc. In particular, in recent years, for short forecast lead times, the typical horizontal resolution of the nowcasting and Numerical Weather Prediction models used for weather forecasting (see Chapter 2) has started to reach grid scales which are of interest for hydrological modelling, which are often a few kilometres or less, depending on catchment size. One distinguishing feature of the process based approach is that often the model parameters are defined to be within specified ranges depending on soil type, slope, land use, river channel hydraulic characteristics, and other factors. Values are typically derived from laboratory or field experiments, or datasets from previous studies on other catchments, and the initial model calibration consists of choosing the most appropriate parameter set given the characteristics of the catchment. Parameter values are then fine tuned to achieve a good match between observed and forecast flows, either by trial and error, or by using optimisation algorithms for those parameters for which values are uncertain. Bayesian and other approaches may also be used to provide an assessment of model and parameter uncertainty. This approach contrasts with the conceptual and data based approaches described later in which model parameters are typically estimated primarily by minimising one or more measures of model fit. Many types of process-based model have been proposed, and some typical features can include one or more of the following components:

136 ● ●

● ●

● ●

●

●

6 Rivers

Surface and sub-surface flow paths Kinematic Wave and similar approaches to routing water between cells, possibly also considering channel characteristics such as roughness coefficient and channel geometry (see Section 6.3) Flow paths inferred from Digital Elevation Model datasets Multi-layer soil moisture models operating at grid scale and, sometimes, at subgrid scale Evapotranspiration linked to vegetation cover, Leaf Area Index etc. Runoff processes linked to soil and land cover properties, typically obtained from remote sensing The option to include additional processes such as snowmelt, lake/reservoir influences etc. A mixture of storage functions and partial differential equations used to represent individual processes

Alternative grid configurations can also be used, in which flows are either translated between grid cells, or routed directly to the catchment outlet (e.g. Moore et al. 2006). For flood forecasting applications, the choice of model will depend on many of the factors described already, including the forecasting requirement, catchment response, forecasting system, real time data availability etc., and the information available for model set up and calibration. Some real time flood forecasting applications of process-based models include the following examples: ● ● ● ● ● ● ●

G2G (Bell et al. 2007) HL-RMS (Koren et al. 2004) LISFLOOD (De Roo et al. 2000; Van Der Knijff et al. 2004) MGB-IPH (Collischonn et al. 2007) MIKE-SHE (Butts et al. 2005) REW (Reggiani and Schellekens 2003) TOPKAPI (Liu et al. 2005)

An important consideration in assessing the performance of process based models is the extent to which parameter values can be specified in advance based on land surface characteristics (soil, vegetation, topography etc.), and used on ungauged catchments. Two major international studies on parameter estimation for ungauged catchments are: ●

●

IAHS PUB – the “Predictions in Ungauged Basins” study – is an International Association of Hydrological Sciences (IAHS) initiative, from 2003 to 2012, aimed at uncertainty reduction in hydrological practice (Sivapalan et al. 2006). MOPEX – The “Model Parameter Estimation Experiment” – which is an open collaborative study funded by the NOAA Office of Global Programs (USA), whose aim is to develop techniques for the a priori estimation of the parameters used in land surface parameterisation schemes of atmospheric and hydrological models (Andrssian et al. 2006).

6.2 Rainfall Runoff Models

137

As studies like these progress, the options for forecasting on ungauged catchments should improve, and this topic is discussed further in Chapters 3 and 8.

6.2.3

Conceptual Models

Conceptual rainfall runoff models have many similarities to process based models, but operate at a catchment scale, and place less reliance on a physically based description of response. Typically, models represent the flow of water using a series of conceptual stores, which fill and empty depending on the rainfall inputs, and lose water from the system due to evaporation (from open water), evapotranspiration (from vegetation), and to river flows. Many types of conceptual model have been proposed, and some typical components can include one or more stores of the following types: ●

●

● ●

Interception store – in which rainfall falling on a catchment initially enters a store representing water held by vegetation (forests, grass etc.), with some representation of evaporation and evapotranspiration processes Surface store – which represents flow through the river network and overland flows Soil store – which represents storage of soil moisture in the catchment Subsurface store – which represents recharge to groundwater, and subsequent outflows to the river network

Individual stores may also be represented in a variety of ways, with options including modelling of outflows as a function either of volume stored (e.g. linear or power law functions), or using stores with fixed volumes which pass water downstream once they fill and overflow. For flood forecasting applications, conceptual models are typically calibrated by trying to achieve a good representation of observed flows across several flood events, and possibly also for the intervening flow periods, to achieve a long term water balance. The criteria for optimisation can include factors such as the shape of the hydrograph, the peak levels reached, mean square error, bias, and various real time related statistics such as those described in Chapter 5. Some types of models may have 15–20 or more parameters to consider, and a typical approach is to optimise a few parameters at a time (e.g. for the baseflow component), holding all others constant, and perhaps placing bounds on the permitted values. The optimisation scheme may also explicitly, or implicitly, include the parameters for the following two additional modelling components: ●

Catchment averaging component – one consequence of using the conceptual approach with raingauge data is the need to derive catchment average values for rainfall, and the parameters of the averaging process can also be viewed as part of the optimisation process. Chapter 2 discusses a range of approaches to catchment averaging, including Thiessen Polygon and geostatistical methods.

138 ●

6 Rivers

Real time updating – can be particularly important for conceptual rainfall runoff models to compensate for uncertainties in rainfall values, initial conditions and other factors. As noted in Chapter 5, error prediction, state updating and parameter updating techniques can all be used although, given the process based flavour of the approach, parameter updating is less widely used.

Many different types of conceptual model have been developed for flood forecasting and other applications (e.g. World Meteorological Organisation 1994), and the intercomparison experiments described earlier present several examples. Some approaches which have been used in flood forecasting applications include: ● ● ● ● ● ● ●

HBV model (Bergstrom 1995) MIKE11/NAM model (Madsen 2000) PDM model (Moore 1985, 2007) Sacramento catchment model (Burnash 1995) TANK model (Sugarawa 1995) URBS model (Malone 1999) Xin’anjiang model (Zhao 1992)

Some particular model configuration options worth noting include: ●

●

●

●

●

No subsurface store – for models used primarily for flood forecasting, one option is to use an event based approach and to omit the subsurface component entirely, on the basis that runoff during a flood event occurs primarily by surface flow. Instead, a fixed or variable runoff coefficient is introduced to represent the proportion of rainfall which goes to surface runoff, such as in the Isolated Event Model, for example. Enhanced subsurface stores – additional submodels can be included, or updating can be based on real time monitoring of groundwater levels, to represent the influence of groundwater levels and storage on river runoff where this is a significant factor. Soil store – use of a probability distribution for soil moisture storage to take account of the variations in soil storage which occur in a catchment (e.g. Moore 1985). Time delays – although the store outflow parameterisations typically provide some time delay between inflows and outflows, it can be helpful to include additional lag parameters to assist with fitting the overall model to represent the time delays in surface and subsurface flow pathways. Complicating features – some models include the facility to model reservoirs (including control rules), irrigation schemes, abstractions and discharges related to water supply, and other factors which influence river flows.

Given the range of modelling possibilities, an increasingly common approach in developing conceptual models is to provide a range of options for the types of stores which are included, and for how they are interconnected and parameterised. This has led in recent years to the development of modelling toolkits, which allow users to select from a range of sub-models to represent antecedent conditions, subsurface flows, runoff response etc., whilst providing an overall framework for model calibration and optimisation. The model can then be configured in the most appropriate way for each problem.

6.2 Rainfall Runoff Models

139

As with process based models, attempts have been made to relate the parameters of conceptual models to catchment characteristics, although results are often dependent on the type of model. However, this approach can help to reduce the work required in model calibration, and also opens the way to forecasting flows on ungauged catchments (i.e. catchments with no river gauging telemetry).

6.2.4

Data-Based Methods

Data-Based methods are used in many technical fields for the real time forecasting and control of complex systems. Examples can be found in the transportation, industrial process, aviation and financial forecasting sectors, and include transfer functions, artificial neural networks, and fuzzy logic techniques. In flood forecasting applications, the main use is usually for rainfall runoff modelling, although data-based methods have also been used for river flow routing, estuary forecasting, and coastal surge forecasting as described in later sections and chapters. Data-based rainfall runoff models can also be used to forecast river levels directly (i.e. rainfall-level models), although a cautionary note is that this may only work well if there is a unique relationship between levels and flows at the selected Forecasting Point (for example, if there are no significant backwater influences). In the early years, one of the drivers for development was the limited computing power available at that time, and data-based models generally run more quickly than a more physically based approach. With current processor speeds this is nowadays much less of a consideration, except in the case of ensemble forecasting, where data based techniques have potential as emulators for models with longer run times (e.g. process based models). However, the data based approach also views flood forecasting from more of a systems perspective, in which the main aim is to develop a model to infer future conditions at one or more locations (Forecasting Points), making best use of all of the real time and historical information available at the time of the forecast, and often providing a measure of uncertainty in the resulting outputs. In particular, models are often optimised to provide forecasts of the particular parameter of interest for flood warning, at the lead times which are most useful operationally. This contrasts with the more physically based approaches, in which the focus for calibration is often reproducing the shape and timing of the whole hydrograph across a number of events in simulation (off-line) mode, although both threshold and lead time based criteria are increasingly being used in model development (see Chapter 5). Perhaps the most widely used data-based techniques are the transfer function and artificial neural network approaches. Transfer functions are one of a wide range of time series analyses techniques used in a range of industries (Box and Jenkins 1970) and can be combined into networks to form integrated catchment models (e.g. Beven et al. 2005). Various related autoregressive techniques have also been considered for forecasting applications, although have been less widely used

140

6 Rivers

(although are used extensively in some of the real time updating approaches described in Chapter 5). For flood forecasting applications, a simple transfer function formulation might relate river flows at time t (Qt) to flows and/or rainfall (P) at previous time steps as follows: Qt = a1Qt-1 + a2Qt-2…amQt-m + b0Pt-T + b1Pt-1-T + b2Pt-2-T …… + bnPt-n-T

(6.1)

where ai and bi are the parameters of the model, and T is a fixed time delay. A model with m flow parameters and n rainfall parameters is said to be of structure or order (m, n, T). A random noise component may also be included, although is not shown here. An important objective in transfer function modelling is often to derive a model which uses the minimum number of parameters possible (i.e. is parsimonious). The use of a time delay, whilst not essential, can help in reducing the overall number of parameters required. Some potential issues with the basic linear approach represented by Equation (6.1) are that flows are unconstrained and may be oscillatory or negative, and that the use of observed (total) rainfall can fail to capture the influence of initial catchment state on runoff. One way to reduce the risk of oscillatory response is to minimise the number of parameters used. Alternatively, the mathematical formulation can be redefined to provide a constrained and stable output; for example, in the Physically Realisable Transfer Function (PRTF) approach of Han (1991). Various techniques have also been developed to account for non-linear influences on river flows such as initial catchment conditions, and these include: ●

●

●

Effective rainfall inputs – using only a proportion of the total rainfall to drive the model, with rainfall separation techniques including use of a variable, thresholdbased or fixed runoff coefficient, perhaps related to initial catchment state, or functions based on real time observations of parameters which may influence runoff (e.g. air temperature), or continuous soil moisture accounting (e.g. Moore 1982; Lees 2000) Flow based initialisation – using the current observed flow as a surrogate for catchment state (for example, a dry catchment will usually have lower flows than a wet catchment) (e.g. Young and Tomlin 2000) Parallel pathway representation – interpretation of the model in terms of one or more linear pathways in parallel which represent the different time responses of key catchment processes, such as surface and groundwater flows (e.g. Beven 2001; Young 2001)

Although some of these techniques may seem to introduce conceptual modelling ideas, the data-based approach is generally retained, with the model structure simply being extended to capture different timescales and rainfall distribution effects which are known to exist in the forecasting problem. The extent to which these enhancements are included will depend on the nature of the catchment for which forecasts are required. For example, a single pathway

6.3 River Channel Models

141

model may be sufficient for a small, fast response catchment, but a multiple pathway model might be more appropriate to represent runoff in a large complex catchment with significant groundwater influences. Real time updating techniques may also be used to help to account for the differences between observed and forecast flows and can include flow substitution, adaptive gain, transfer function noise, genetic algorithm, and Kalman filter methods (including extended and ensemble versions), and are intrinsic to some approaches. Artificial neural networks also use a parallel pathway approach, although with many more layers or pathways in the formulation. The building blocks of a network typically include components such as neurons, summing junctions, and activation functions. For flood forecasting applications, one aspect of the model development (often called ‘training’) is to achieve a compromise between including too few neurons, thereby failing to capture the full range of river flow response, and using too many neurons, focussing on the noise rather than the underlying river flow signal (see Section 7.3 for more information). The potential for rainfall runoff modelling applications has been the subject of several review studies (e.g. Dawson and Wilby 1999; ASCE 2000), and the technique has been trialled in various flood forecasting applications (e.g. Solomatine and Price 2004; Abrahart et al. 2004). One active area of research is into how models can be formulated to provide reliable predictions of extreme events, beyond those in the calibration or training dataset.

6.3 6.3.1

River Channel Models Introduction

Whilst rainfall runoff models represent the translation of rainfall into inflows to a river network, river channel models represent the flow of water within that network, and include the following approaches: ●

●

●

Process based – hydrodynamic models which use one-dimensional (1D), 2D or 3D approximations to the mass and momentum equations for both flows and river levels (sometimes called physically based models) Conceptual models – various hydrological flow routing techniques which can range from empirical techniques to approximations to the full mass and momentum equations for river flow Data-based approaches – which use transfer function, artificial neural network, and other approaches to represent river flows or levels (sometimes called black box methods)

All of these approaches have been used successfully in river flow forecasting applications. Note that level to level, and flow to flow, correlations may also be regarded as a type of river forecasting model, but for convenience are discussed in Chapter 3, together with a range of other simpler empirical flood forecasting techniques.

142

6 Rivers

In considering the most appropriate approach to use, it is useful to consider the forecasting requirement (see Table 6.1), and the nature of river flows at or near the Forecasting Points of interest and in the river network upstream. If the main modelling requirement is for the likely timing or magnitude of a peak, or exceedance of a flow threshold, then a wide range of approaches may give satisfactory results. Here the focus is primarily on river flows, although levels may be estimated if a suitable stage-discharge relationship is available. However, some complicating factors which can influence the choice of modelling approach, and may require use of a more complex type of model (e.g. a hydrodynamic model), include: ●

●

●

●

●

●

Backwater influences – river levels at a Forecasting Point are influenced by river levels downstream of that point; for example, from tidal influences, inflows from a tributary, or operations at a river control structure. Often this leads to a non-unique relationship between river levels and flows, which simpler models may not be able to capture. Spillage – river levels may exceed the height of natural river banks, or of flood defences. Flows may be permanently lost to the river system, or may re-enter further downstream, or as river levels drop. Natural floodplains – flows onto floodplains will tend to reduce (attenuate) the magnitude of flow peaks further downstream, and delay the arrival of the peak. Lakes, wetlands and marshes have a similar influence. Embanked river channels – flows may spill and be lost permanently from the river network unless there is a return route via pumping or gravity drainage. However, return flows may not occur until river levels have dropped considerably. Artificial influences – influences from dams, river control structures, off-line storage (polders, washlands), pumped or gravity fed flows and other factors may affect the timing, magnitude and duration of flood peaks. Tributary inflows – inflows may contribute to peak flows in the main river channel, and may have different response characteristics to main channel flows (faster rate of rise etc.).

The following sections describe the extent to which these various factors can be represented by the three main types of modelling approach.

6.3.2

Process Based Models

The velocity and depth of flows in a river reach depends upon the inflow of water to the reach, friction losses, and changes in river slope, width and shape along the channel (e.g. Chow 1959; Chanson 2004). The unsteady water flow in a natural river is governed by the principles of conservation of mass and momentum. Except in certain simplified situations, the resulting equations cannot be solved analytically, and must normally be solved numerically. Solutions can be obtained using one-dimensional (1D), two-dimensional (2D) or three-dimensional (3D) approaches, although 1D and, to a lesser extent, 2D approaches are most widely used in river flood forecasting applications. Typically, solutions are obtained on a regular or irregular grid using finite difference approaches, although finite element

6.3 River Channel Models

143

and finite volume approaches can be used. One widely used approximation for one dimensional flows is the assumption of a gradual variation of flow and a hydrostatic pressure distribution, which results in the well-known Saint-Venant equations. The friction terms are typically parameterised using an empirically based roughness coefficient (e.g. Mannings n) which may be fixed or vary along the river reach, and may be estimated from look up tables or a database of values for channels with similar characteristics (bed type, vegetation etc.). Hydraulic controls on river levels such as bridges, weirs and river control structures may be represented by sub-models, typically parameterised using a loss coefficient. Logical rules may be included for barrier, gate and other operations. The information required to develop a model typically includes: ●

●

● ●

●

Survey data for the channel dimensions at a sufficient number of locations to capture the hydraulic characteristics of the river channel, any structures to be included in the model, and possibly of flood defences (dikes/levees) and the floodplain Estimates or measurements for inflows from the upstream end of the reach, and any tributary inflows Estimates for the friction loss terms A well-defined downstream boundary condition; for example, from observed or forecast levels, or a normal depth or observed stage-discharge relationship Information on structure control rules (if applicable)

Hydrodynamic models are particularly well-suited to applications where precise estimates of river levels are needed, such as forecasting the overtopping of flood defences, or real time control of structures such as pumps and gates, and in situations with tidal, confluence, multiple channel and other backwater influences, and for estimating depths and (possibly) velocities on the floodplain. ‘What if’ scenarios, such as culvert blockages, and defence breaches, can also be considered. Changes in river bed profiles might also be included by incorporating more up to date survey data. The physical basis also suggests that model outputs can be extrapolated to flows and levels outside the range of calibration, provided that there are no significant changes in river characteristics at these higher levels, and that stage-discharge relationships remain valid. For flood forecasting applications, even for a well designed and calibrated model, some potential issues for real time applications include (e.g. Chen et al. 2005): ●

●

●

●

Model run times – models initially developed for off-line flow simulation can run too slowly to be useful for real time operation Stability – numerical problems can arise from the discretisation in time and space causing models to exhibit unstable or oscillatory behaviour, or even to stop operating during a model run Convergence – failure of the model to achieve the required accuracy within a specified number of iterations of the solver scheme Initial conditions – may need to be saved between model runs, and carefully specified to avoid stability problems

144 ●

●

6 Rivers

Survey requirements – the required survey data may not be available, or too expensive to capture given available budgets, or of insufficient resolution Ungauged inflows – if there are significant ungauged inflows, then flows may need to be estimated using techniques for ungauged catchments (see Chapter 8)

In addition, the model may not have initially been developed for estimating high flows, and have performance, stability, and other problems (e.g. with the high flow end of stage discharge relationships) under these conditions. These problems are all potentially solved, although can require considerable expertise to achieve a fast stable model. The following options are possibilities (Huband and Sene 2005; Chen et al. 2005): ●

●

●

●

Removal of model complexity (nodes) in locations where detail is not required; for example, away from Forecasting Points, and areas of significant hydraulic control. This might include replacing hydrodynamic modelling reaches with simpler flow routing reaches, in which flow is conserved but river levels are not required, or using so-called sparse hydrodynamic modelling techniques, in which only key cross section locations are retained, and the details of river response are not considered. Removal or aggregation of hydraulic structures which do not exert a significant influence on river levels in the areas of interest, and using simpler or alternate representations for any control structures which cause stability or convergence problems. Checking for problems which can cause poor stability, including poor initial or boundary conditions, river channels running dry, inappropriate spacing of river cross sections, surcharging at bridges and other structures, operations at river control structures, transitions from subcritical to supercritical flow, and other factors. More detailed studies of the accuracy of the high flow end of stage discharge relationships, possibly using 2D or 3D modeling, commissioning additional survey data near gauging stations, and additional measurement campaigns (if practicable).

Figure 6.3 shows an example of how both the spacing and lateral extent of river cross sections might be tailored to modelling river levels or flows for a Flood Warning Area, with sparse hydrodynamic modelling elsewhere except in the vicinity of a telemetry site (where an estimate for the high flow rating is required). Improvements to model stability and convergence, and reductions in complexity, will generally also improve model run times, both through reductions in the number of calculations required per iteration, and the number of iterations required. Another consideration with hydrodynamic models is whether to use real time updating. The simplest option is to split the model at key river telemetry sites, and to update the model outputs using an independent error prediction algorithm. However, this can be undesirable since it removes the ability for forecasts of downstream influences to feed back to locations upstream of the telemetry site,

6.3 River Channel Models

145 INFLOW Gauge

Flood Warning Area Gauge

Fig. 6.3 Illustration of sparse hydrodynamic modelling techniques (not to scale)

and mass and momentum conservation will not necessarily be maintained throughout the river network. The main alternative is to attempt to update the model at internal model nodes, and methods which have been developed include: ●

●

State updating – adjustments to main channel or tributary inflows to distribute errors in flow volume noted at telemetry sites Parameter updating – real time adjustments to model parameters such as the roughness coefficient to ‘fine tune’ the model performance

This topic is an active area of research since both state and parameter updating can cause unwanted transient behaviour to propagate within the model domain.

6.3.3

Conceptual Models

Hydrological flow routing models provide a simplified representation of river flows compared to a hydrodynamic modelling approach, yet perform well in some flood forecasting applications, and can offer the following advantages: ● ● ●

Models run quickly, with stability and convergence problems less likely River cross section information is not necessarily required The techniques can work well on steep river sections, where a hydrodynamic model might fail

The main restriction is that models work in terms of flows, and do not compute levels, other than through optional application of a stage-discharge relationship. There is also typically no representation of backwater influences, and the approach is less suited to shallow sloping rivers. Some of the more complex river flow phenomena described in the previous section also cannot easily be represented.

146

6 Rivers

One of the first methods to be developed was the Muskingum method. The approach uses the concept of a triangular ‘wedge’ and rectangular ‘prism’ to represent storage in the river channel and flood wave, and aims to maintain a water balance between upstream and downstream reaches. More recent developments have usually adopted a more physically based approach, and include the following techniques: ● ●

●

Muskingum-Cunge method (Cunge 1969) Variable Parameter Muskingum-Cunge method (VPMC) (Price 1977; Tang et al. 1999) Kinematic Wave approaches (e.g. Lighthill and Whitham 1955; Jones and Moore 1980)

Cunge (1969) showed that, with an appropriate choice of length and time steps, the Muskingum Cunge method provides a good approximation to the convectivediffusion equation, which in turn is a simplification of the full St Venant equations. Some studies have shown that, for a simple river channel with a floodplain but no artificial influences, there is sometimes little to choose between fixed and variable parameter routing models in terms of predicting peak flows, but that variable parameter models can perform better on the rising limb of the hydrograph. This can be important in flood warning applications, where the interest may be in success at crossing a flood warning threshold. Although stability problems are unlikely, most types of model still require a numerical solution scheme, in which the river reach is divided into discrete sections. For example, Tang et al. (1999) showed that there can be some numerical issues with more complex types of flow routing models (oscillations, lack of volume conservation etc.) when used in compound channels, which can be greatly reduced through choice of an appropriate computational scheme. Some types of routing model also require wavespeed and attenuation parameters to be estimated. Typically, wavespeeds increase with increasing discharge until the river level reaches a value close to bank full, then decrease as water spills onto the floodplain, but increase again once the floodplain flow exceeds a significant depth. Both fixed wavespeed, and discharge dependent (variable parameter) values, may be assumed. Values must either be supplied by the modeller or, in some cases, can be estimated by the calibration software from typical river cross sections. As with hydrodynamic models, it may be necessary to represent the inflows from tributaries in a river reach, and flows into or out of the reach due to abstractions, floodplain flows, and other factors. Lateral flows of this type are easily included, and can either be modelled directly, or represented as a proportion of flows in the main river, weighted by catchment area, mean annual rainfall, or some other indicator of runoff (see Chapter 8).

6.3.4

Data Based Methods

Data Based methods can also be used for river flow modelling, with the main difference compared to rainfall runoff models (see Section 6.2.4) being that flows are esti-

6.3 River Channel Models

147

mated from telemetered values for a location further upstream, rather than from rainfall values (e.g. World Meteorological Organisation 1994). For example, a simple linear transfer function formulation for flow routing might be of the following form: Qt = a1Qt-1 + a2Qt-2…amQt-m + b1qt-1-T + b2qt-2-T …… + bnqt-n-T

(6.2)

where Q is the flow at the Forecasting Point, q is the inflow to the reach, ai and bi are the parameters of the model, and T is an optional fixed time delay. Some formulations also include a random noise component. Artificial Neural Network techniques may also be used (see Section 7.3 for more information). Table 6.4 provides some examples of transfer function and artificial neural network flow routing approaches which have been used or trialled in real time flood forecasting applications. Table 6.4 Examples of real time flood forecasting applications of data based flow routing models Model

Full name

Example applications

References

Transfer function General

Data based mechanistic Artificial neural networks Artificial neural networks

River Nith, Scotland River Necker, Germany

Lees et al. (1994) Shrestha et al. (2005)

Rivers in NE England

Kneale et al. (2000)

Neural network

One interesting feature of the data based approach is that models can also be formulated in terms of river levels, rather than flows. This avoids the need to have well defined, accurate stage-discharge relationships for the inflow and outflow locations, although with the possible difficulty that backwater and other influences may cause errors if they lead to a non-unique relationship between levels and flows at either location.

Chapter 7

Coasts

Coastal flood forecasting models are used to estimate conditions at or near locations which may be at risk from flooding, such as towns, ports and harbours, and coastal roads and railways. Forecasts may also be required at structures, such as tidal barriers, to assist with operations to reduce the risk of flooding. Models can range from simple empirical relationships to complex process-based models combining offshore, nearshore, wave overtopping and flood inundation components. Forecasts may be based primarily on coastal observations, or also make use of the surge, wind and wave forecasting outputs provided by national meteorological services and coastal observatories. This chapter describes some of the issues in selecting an appropriate modelling approach and then discusses a range of process based and data based techniques for coastal flood forecasting.

7.1

Model Design Issues

Coastal flooding can arise from several factors, including high tides, and the impacts of storms on conditions at the sea surface, generating surge and wave action. Heat exchange between the ocean and atmosphere can also cause storms to intensify, as with hurricanes, typhoons and tropical cyclones, for example. Tidal effects are generated by the gravitational attraction of the sun and the moon, usually leading to a twice-daily maximum in tidal levels along the coastline, with the peak values varying depending on the relative alignment of the sun and the moon. Maximum and minimum tidal ranges occur twice each month when the earth, sun and moon are aligned so that their gravitational pull combines (high, or spring, tides) or are approximately at right angles (low, or neap, tides). The influence of the moon on tides is roughly twice that of the sun and perturbations in the orbits of the sun, the moon and the earth lead to additional impacts at a range of timescales which are significant up to a period of 18.61 years. Surge is generated by a combination of low atmospheric pressure and wind friction at the ocean surface; for example from tropical cyclones, hurricanes and typhoons, or mid-latitude storms. Surges may develop locally, or can propagate from distant areas, and can lead to an increase or decrease in sea levels, although it K. Sene, Flood Warning, Forecasting and Emergency Response, © Springer Science + Business Media B.V. 2008

149

150

7 Coasts

is increased levels which are of concern for coastal flood forecasting. In the deep ocean, a 100 N m−2 (1 mbar) drop in pressure causes roughly a 0.01 m rise in still water levels (the so-called inverted barometer effect), although wind related effects are often considerably more important. Onshore winds can cause water to accumulate at the shoreline and this effect can be amplified by reflections in estuaries (deltas) and local channelling effects. Surges can develop rapidly over timescales of a few hours or less, although more typically last between a few hours and 2–3 days, depending on the scale and duration of the storm. At the shoreline, surge effects are higher for a shallow sloping continental shelf or sea bed. Some notable locations for storm surge impacts from tropical cyclones include the Gulf of Mexico, the Caribbean and the Bay of Bengal, and the eastern coastlines of Japan and China. For example, surge heights generated in Hurricane Katrina in August 2005 were reported to have reached 7–9 m along the Mississippi coastline (Knabb et al. 2006). Note that storms of this type are called hurricanes in the Atlantic and Eastern Pacific Oceans, tropical cyclones in the Indian Ocean and typhoons in the Western Pacific. Typical storm sizes are in the range 100–1,000 km, with maximum recorded wind speeds of the order of 300 km hour−1. Waves usually develop as a result of wind action, and can be generated both locally by wind effects (wind waves), or may propagate from distant storms (swell). The height and frequency of waves generated by a storm can depend on wind speeds and on the size and duration of the storm. The eventual wave heights at the shoreline depend on the distance over which waves develop (often called the fetch), the slope of the sea-bed, channelling by local topography (bathymetry), and other local factors, such as reflections at cliffs or reefs. At high latitudes, wave development may also be affected by sea ice. Deep sea (or swell) waves are generally not as high as locally generated waves, but have longer periods, and can potentially cause more damage and overtopping when they encounter sea defences and other structures. Tsunami waves may also cause flooding, although are generated by subsea landslides, volcanic activity and earthquakes (see Chapter 8). Whilst these effects generally arise from distinct processes, the resulting impacts on sea water levels are not necessarily independent. For example, tide and surge effects may interact, particularly in shallow water, affecting both the timing and magnitude of peak levels at the shoreline. Similarly, heavy rainfall may occur inland during depressions and tropical cyclones, leading to high river flows as well as surge, particularly on small coastal catchments. The likelihood of high river flows and tidal levels coinciding can depend on many factors, including bathymetry, the storm intensity, track and duration, and catchment response times (e.g. Environment Agency 2005) Local factors, such as headlands, breakwaters and estuaries (or deltas) can also have a strong influence on tides, surge and waves. For example, resonance effects may amplify the tidal range in estuaries, or cause a second high tide peak on each cycle. Also, operational problems, such as tidal gates failing to close, or breaches occurring at sea defences, may lead to flooding whilst gate operations (e.g. at tidal barriers) can also influence tidal levels. One example of a location with a wide range of factors is the southwest coastline of England, which is a peninsula approximately 250 km long, and 100 km across at

7.1 Model Design Issues

151

its widest point. The northeastern coastline lies within the estuary of the River Severn, which has one of highest tidal ranges in the world, exceeding 10–15 m in spring tides. Wave and surge influences are relatively small along this coastal reach, although can be important if they coincide with high tides. By contrast, on the opposite side of the peninsula, the tidal range is considerably lower, whilst surge remains of a similar magnitude, and maximum wave heights are now higher, particularly when they originate from the open waters of the Atlantic Ocean, and can reach heights of 5 m or more due to local shoaling and wind driven effects. Other examples include the wide variations in coastal inundation even over short distances during the December 2004 Tsunami, depending on coastal topography and nearshore bathymetry relative to the direction of travel of the wave, and the differences in the impacts of hurricanes and tropical cyclones in locations like the Gulf of Mexico, where surge effects dominate, and Hawaii, where wave effects dominate, and can inundate land to heights of 9 m (e.g. Cheung et al. 2003). For coastal forecasting applications, the modelling approach which is used needs to be tailored to the types of flooding mechanisms which are most important for the locations at which forecasts are required. Some examples of possible locations for Forecasting Points (sometimes called Coastal Cells or Units) include: ● ● ● ●

Flood Warning Areas (to assist with issuing warnings) High Risk Locations (ports, harbours, refineries, coastal transport routes etc.) Tidal Barriers (to assist with structure operations; see Chapter 8) Tide Gauges (for real time evaluation and updating of forecasts)

In addition to the choice of modelling approach, some other factors to consider for a real time application (see Chapter 5) include: ● ● ● ●

Forecasting Requirements – the intended use of the forecasts Data Availability – the availability of data and other information in real time Forecasting System – the system(s) on which models will be operated Performance Targets – for model run time, accuracy, and other considerations

Forecasting system and performance considerations are discussed in Chapter 5, whilst the main sources of real time coastal data are described in Chapter 2. A typical coastal forecasting requirement might be to estimate tidal (still water) levels plus the additional impacts of surge and wave action. Wave overtopping estimates may also be required. Some potential requirements (e.g. Environment Agency 2004) are to inform decisions related to the: ● ● ● ● ● ● ●

●

Areas most likely to flood Defence lengths most likely to breach Number of people in danger Vulnerability of the endangered people Extent of likely damage to property Extent of impact of individual failed defences on the number of people in danger Extent of impact of individual failed defences on the likely extent of damage to property The most beneficial areas to target emergency resources

152

7 Coasts

The complexity of the modelling approach used will depend on the precise requirements in each application. As with many other types of modelling, the design of a coastal flood forecasting model is often a compromise, and the overall approach may also be constrained by non-technical factors such as costs and system limitations. For example, the level of risk (e.g. the number of properties) can be a major factor in deciding on the complexity of approach. These points are discussed further in Chapter 11 when considering the economic benefits of flood warning schemes. Coastal forecasting models can be considered to fall into two main categories: ●

●

Process Based models – which represent to varying degrees the main physical processes which determine tidal levels, surge, and wave action (sometimes called physically based models) Data Based models – which attempt to capture the main features of the response, but do not represent physical processes directly (e.g. artificial neural networks)

Chapter 6 includes some additional discussion of the relative merits of these approaches in the context of rainfall runoff modeling. Threshold based methods are also widely used in coastal flood warning applications and are described in Chapter 3. In selecting appropriate coastal flood forecasting techniques, it is also helpful to introduce the following idealisation of the coastal flooding process (Environment Agency 2004): Sources ● Offshore Zone – tides, surges, wave generation and the interaction of waves with each other ● Nearshore Zone – water levels and shallow water effects such as shoaling, depth refraction, interaction with currents and depth induced wave breaking Pathways ● Shoreline Response Zone – surf zone/beach response, wave structure interaction, overtopping, overflowing and breaching ● Flood Inundation Zone – flow of flood water over the flood plain area The boundaries between these zones are indicative, and will vary between different forecasting situations. In many cases the starting point for model development will be an offshore forecasting model for the open ocean of the type operated by many national meteorological services or coastal observatories. Models of this type usually provide forecasts on a gridded basis for which, depending on model resolution, the closest nodes may be some distance from the Forecasting Point(s) of interest. Additional models may then be required to represent the details of nearshore and shoreline coastal processes, and these sub-models will often be site specific, requiring a separate set of calibration factors, and possibly alternate types of model, for each location. The overall set of models form an integrated coastal flood forecasting system, in which each model output acts as the input to the next model (or models) in the chain. With modern computing power, it is feasible to operate systems of this type in real time for large numbers of coastal Forecasting Points, although inevitably

7.1 Model Design Issues

153

some approximations may be required if model run times are too slow to be of operational use for flood warning applications. Empirical and data-based techniques offer one approach to improving model run times. The development of integrated coastal flood forecasting models, combining offshore, nearshore and shoreline components, is an active area of research with both deterministic (e.g. Cheung et al. 2003) and probabilistic (e.g. Tozer et al. 2007) methods under development. There are also many international initiatives in the area of ocean observation, modelling and operational response, and Table 7.1 gives several examples, of which the longest established is the World Meteorological Organisation Tropical Cyclone Programme (see Box 7.1). Some active areas for research and development include: ●

●

●

Ensemble/probabilistic forecasting – assessing the uncertainty in surge and wave components, including transformations from offshore to nearshore and wave overtopping, and using this information for improved operational decision making Data assimilation – making use of real time observations from the ocean (buoys, gauges, ferries, offshore platforms etc.), the atmosphere, and satellites to improve model initial and boundary conditions Performance monitoring – developing improved ways to assess and compare the outputs from different models

Chapter 5 discusses the general principles of these techniques in more detail, whilst the following sections include several examples of practical applications in coastal flood forecasting.

Table 7.1 Some international programmes in coastal modelling and forecasting Programme TCP – WMO Tropical Cyclone Programme

Scope or objective

To assist member states in tropical cyclone observation, forecasting and response JCOMM – IOC/WMO Joint Promoting appropriate technical Technical Commission for standards and procedures for a Oceanography and Marine fully integrated marine observMeteorology ing, data management and services system (open ocean) GODAE – Global Ocean Data Marine observation, data assimilaAssimilation Experiment tion, forecasting, performance monitoring GOOS – IOC, UNEP, WMO, Observations, modelling and analyICSU Global Ocean sis of marine and ocean variables Observing System to support operational ocean services worldwide (coastal regions) e.g. EUROGOOS and NOOS in Europe

Example references Holland (2007)

JCOMM (2007) WMO (1998)

http://www.godae.org/

http://www.ioc-goos.org/

154

7 Coasts

Box 7.1 The WMO Tropical Cyclone Programme The WMO Tropical Cyclone Programme (Fig. 7.1) is part of the World Weather Watch Applications Department, and aims to encourage and assist the members of WMO to: ●

● ●

●

●

Provide reliable forecasts of tropical cyclone tracks and intensity, and related forecasts of strong winds, quantitative forecasts or timely assessments of heavy rainfall, quantitative forecasts and simulation of storm surges, along with timely warnings covering all tropical cyclone-prone areas Provide forecasts of floods associated with tropical cyclones Promote awareness to warnings and carry out activities at the interface between the warning systems and the users of warnings, including public information, education and awareness Provide the required basic meteorological and hydrological data and advice to support hazard assessment and risk evaluation of tropical cyclone disasters Establish national disaster preparedness and prevention measures

Fig. 7.1 Structure of the Tropical Cyclone Programme (Reproduced from Twenty Years of Progress and Achievement of the WMO Tropical Cyclone Programme (1980–1999), courtesy of WMO)

(continued)

7.1 Model Design Issues

155

Box 7.1 (continued) The Programme was established in 1980, and works closely with regional and international disaster relief organisations including the United Nations Economic and Social Commission for Asia and the Pacific (ESCAP), UN ISDR (United Nations International Strategy on Disaster Reduction), UNEP (United Nations Environment Programme), UNDP (United Nations Development Programme) and the International Federation of Red Cross and Red Crescent Societies. Activities of the Programme are implemented through coordination with six Regional Specialised Meteorological Centres (Fig. 7.2) and five Regional Tropical Cyclone Warning Centres which cover the following ocean basins: ● ● ● ● ● ●

Southwest Indian Ocean North Atlantic Ocean, Caribbean Sea and Gulf of Mexico Eastern North Pacific South Pacific and southeast Indian Ocean Bay of Bengal and Arabian Sea Western North Pacific and the South China Sea

The main activities within the Programme include sharing of best practice and training in operational meteorology and hydrological techniques, and initiatives to encourage member organisations to assess risks from tropical cyclones, and to establish structural and non-structural measures to reduce property damage and loss of life to a minimum. The Programme also facilitates the transfer of technology between member states, including satellite reception, weather radar, telecommunication, data processing and monitoring equipment, and atmospheric, surge and hydrological models.

Fig. 7.2 Regional Specialised Meteorological Centres within the WMO Tropical Cyclone Programme (Reproduced courtesy of WMO)

(continued)

156

7 Coasts

Box 7.1 (continued) Operational Tropical Cyclone Plans/Manuals have also been prepared by each region covering topics which include station duties, addresses, telephone and other communication numbers, communication procedures, terminology, definitions, procedures, tropical cyclone naming conventions, unit conversions, coordination, analysis requirements, radar and satellite observations and dissemination, aircraft reconnaissance (where applicable), and wording of warnings (Holland 2007). Note: Any text/material regarding TCP/WMO does not imply the expression/endorsement of any opinion whatsoever on the WMO Secretariat concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries.

7.2 7.2.1

Process-Based Models Astronomical Tide Prediction

Of the various factors which influence coastal flooding, astronomical tide levels are in principle the easiest to estimate, although in practice estimates are subject to many uncertainties. Gravitational forces from the sun and moon produce a twicedaily tidal cycle, with the highest and lowest values (spring and neap tides) occurring twice each (synodic) month linked to the orbit of the moon around the earth, with annual extremes linked to the orbit of the earth around the sun. Superimposed on this pattern are other effects arising from variations in the relative orbits of the sun, the moon and the earth. The longest period of motion which is usually considered is an 18.61 year cycle arising from variations in the moon’s orbit. Tide prediction methods use wave theory in which the twice-daily cycle is combined with other harmonics arising from the perturbations in the orbits of the sun, moon and earth. More than 30–50 harmonics can be used, although a point is reached at which the incremental improvements in accuracy are negated by other influences. For example, the tidal response can be modified in estuaries, where the reflected component and depth effects may affect the timing and amplitude of the response considerably, depending on the shape, size and bed profile of the estuary, in some cases leading to a tidal bore. More generally, factors which can affect the response (Hicks 2006) include: ●

● ● ● ● ●

The restrictive depths of the oceans not allowing the generated tidal wave to be in equilibrium with the rotation of the earth Irregular ocean depths over which the waves must travel Reflections and interactions of the waves from irregularly shaped continents Bottom friction Turbulence Viscosity of the water

7.2 Process-Based Models

157

Given these uncertainties, a semi-empirical technique called harmonic analysis is often used, and is the process of identifying site-specific coefficients (amplitude, phase etc.) from observed tidal records. For maximum accuracy, a record of at least the length of the longest period (18.61 years) is required. Future values can then be estimated by recombining the harmonics. Hence, although process-based, in practice the methods used are strongly data-based in implementation, and purely data-based techniques, such as artificial neural networks, are also sometimes used (see later). The accurate determination of gauge datums (or benchmarks) is also an important component in this type of analysis, with long term monitoring also required to account for changes in land levels over time.

7.2.2

Surge Forecasting

Surge forecasting, by contrast, is often performed using hydrodynamic models for the response of the oceans to atmospheric influences. Astronomical tide and wave estimation components are also often included in the analysis. Ocean models may also serve a range of purposes other than coastal flood forecasting, including marine forecasting (e.g. for shipping and oil platform operators), and as part of the ocean-atmosphere component within Numerical Weather Prediction models. Models of this type can be global in extent (albeit sometimes at a coarse resolution), or cover smaller areas in a greater level of detail (e.g. the continental shelf, or specific coastlines). Global scale models are typically operated by national meteorological services, whilst limited extent versions are often developed for local forecasting applications by coastal observatories and other organisations. A widely used approach in coastal flood forecasting is to take a real time feed of offshore forecasts from a global or continental scale model and then to develop additional models or empirical relationships to translate these results to the points of interest, possibly including more detail on bathymetry and other local factors to improve the model accuracy. However, in some situations, the offshore forecasts alone may be sufficient for the application, with no need for additional model development. Table 7.2 provides some examples of regional or local hydrodynamic models, and Box 7.2 describes the approach used in the United Kingdom’s Surge Tide Forecasting Service: Hydrodynamic models typically use a grid-based approach, in which the grid extent or domain extends along the entire coastal region or reach under consideration and into the open ocean (e.g. to the edge of the continental shelf). Typical horizontal grid lengths might be of the order 10–100 km in the open ocean, and possibly of higher resolution (e.g. 0.1–10 km) for specific coastal reaches or features. In three dimensional (3D) models, the vertical division of layers may be fixed, or may be varied according to the dominant processes under consideration; for example, Chassignet et al. (2006) describe an ocean model which uses density tracking layers in the deep ocean, fixed depth (or pressure) coordinates near the surface within the mixed layer, and terrain following coordinates in shallow coastal regions.

158

7 Coasts

Table 7.2 Examples of operational surge, tide and wave forecasting systems Country

Operator

Reference

Australia

Bureau of Meteorology, CSIRO

Brassington et al. (2005)

Hong Kong The Netherlands

Hong Kong Observatory Royal Netherlands Meteorological Institute (KNMI) and National Institute for Coastal and Marine Management (RIKZ) Met Office, Proudman Oceanographic Laboratory National Weather Service (Hydrometeorological Prediction Centre, National Hurricane Centre, Ocean Prediction Centre, Honolulu Weather Forecast Office); NOAA Centre for Operational Oceanographic Products and Services

Lam et al. (2005) Verlaan et al. (2005)

United Kingdom USA

Flather (2000) Chassignet et al. (2006) Jelesnianski et al. (1992), Berg et al. (2007)

Box 7.2 Storm Tide Forecasting Service, United Kingdom The Storm Tide Forecasting Service (STFS) is operated jointly by the UK Met Office and the Proudman Oceanographic Laboratory. The service provides forecasts of surge magnitudes around the coastline of the UK (e.g. Fig. 7.3) and has been in operation since 1978 (Flather 2000). Forecasts are provided to a lead time of 36 hours ahead, with a 6 hour hindcast period to initialise the model run. The current model (CS3X) was developed by the Proudman Oceanographic Laboratory and covers the entire continental shelf from the west of Ireland and into the North Sea and the Bay of Biscay and English Channel. The model extent was increased in 2006 from an earlier version (CS3) to better account for surge generating conditions over the westernmost parts of the continental shelf (i.e. the Bay of Biscay and the Rockall shelf). The model uses a two-dimensional finite difference scheme with a 12 km horizontal resolution. The model is forced with 26 tidal constituents, and wind and atmospheric pressure fields from the North Atlantic Extended (NAE) component of the Met Office’s Unified Numerical Weather Prediction model. The lower boundary is provided by detailed bathymetry for the sea bed and the coastal margins. Wave forecasts are obtained from a 2D spectral model, again running on a 12 km grid, and the output from the surge model is used as an input to this model to allow for wave-current interactions. The model also allows for drying and inundation of inter-tidal areas. The model runs four times per day and, for each forecast, two model runs are used; a tide-only version using standard atmospheric pressure and no wind at (continued)

7.2 Process-Based Models

159

Box 7.2 (continued)

Fig. 7.3 Example of an STFS simulation of progression of a storm surge around the coast of the UK (National Tidal and Sea Level Facility (NTSLF), Proudman Oceanographic Laboratory, http://www.pol.ac.uk/ntslf)

the ocean boundary, and a version with the atmospheric forcing included. The surge is then estimated from the difference between the outputs from these two model runs. Total water levels at specific monitoring sites (e.g. tide gauges) are then estimated by adding the surge to the site-specific estimate of tidal levels based on harmonic analysis (e.g. Fig. 7.4). Finer mesh models have also been developed for some areas; for example, to provide better resolution of surge-tide interactions in the strongly tidally influenced Bristol Channel and Severn Estuary in the southwest of England. Data assimilation techniques are also being evaluated, including 3D-Var and Optimal Interpolation methods. A prototype 24 member ensemble surge product is also being developed, based on multiple CS3X model runs using ensemble estimates for atmospheric pressure and wind speed and direction from the Met Office NAE model. In collaboration with the Environment Agency, which is responsible for issuing coastal flood warnings in England and Wales, a demonstration project is providing probabilistic coastal forecasts to a coastal

(continued)

160

7 Coasts

Box 7.2 (continued)

Fig. 7.4 Surge model residuals and astronomical tidal elevation predictions for Sheerness for a 2.5–3 m North Sea surge event on 9 November 2007 (National Tidal and Sea Level Facility (NTSLF), Proudman Oceanographic Laboratory, http://www.pol.ac.uk/ntslf)

location, combining ensemble and Monte Carlo techniques to account for uncertainties in surge, wave transformations, beach profiles, sea defence condition, wave overtopping, bathymetry, and other factors.

7.2 Process-Based Models

161

Various solution schemes can be used, including finite difference, finite element, and finite volume methods. Some approaches support use of irregular grids or curvilinear grids, in addition to regular grids. Irregular grids typically require more complex solution schemes, but have the advantage that the resolution and shape of the grid can be tailored to situations where more detail is required (e.g. in bays and estuaries and around islands and headlands). Models of this type aim to solve approximations to the three dimensional mass, momentum and energy conservation equations for fluid motion. Forecasting may be in terms of total water levels (combing surge and tidal influences), or the surge residual may be calculated separately and then added onto the results from harmonic analysis at tide gauges (e.g. Horsburgh and Wilson 2007). Some processes which can be represented include: ● ● ● ● ● ● ●

Turbulence, friction, temperature and salinity effects Surge generation (from atmospheric pressure and wind shear) Energy transfer (from the ocean to the atmosphere, and vice versa) Freshwater inflows and river levels at fluvial/tidal boundaries Rotational effects (e.g. storm scale effects) Wave generation and transformation Ice formation, movement and dissipation

In practice, some simplifications must usually be made, since the details of processes may not be fully understood, and model run times would not be practicable for a real time forecasting application. Some general types of approximation can include: ● ● ● ●

Distance (depth or width) averaging (e.g. shallow water equations) Time averaging (e.g. wave spectral analysis, turbulence effects) Exclusion of secondary effects (depending on the application) Introduction of statistical or empirical sub-grid models

Temperature and density effects may be approximated also. Another key approximation is often to reduce the number of dimensions of the model, and Table 7.3 illustrates some possible options if using Cartesian coordinates. Table 7.3 Examples of applications of simplified hydrodynamic models Form of model

x

y

z

Potential coastal forecasting applications

One-dimensional (1D)

Yes

No

No

Two-dimensional Yes – vertical (2DV) Two-dimensional Yes – horizontal (2-DH) Three-dimensional Yes

No

Yes

Yes

No

Yes

Yes

Shallow narrow estuaries, offshore-onshore processes where depth effects and alongshore processes are insignificant. Deep narrow estuaries where influences from the width of the estuary are insignificant. Wide shallow estuaries, bays etc. where the vertical transport is insignificant. Also, models for the open ocean and some coastal reaches. Detailed three dimensional modelling. Options include barotropic and baroclinic models.

162

7 Coasts

In the table, the x-dimension is from offshore to onshore, the y-dimension is alongshore, and the z-dimension is for depth (although different conventions may be used). The issue of model run times is of course a factor for real time applications, and run times often increase with increases in the number of processes represented and the dimensions of the model. Apart from selecting a simpler model, with a coarser resolution, another solution is to run models off-line for a large number of scenarios, then to derive simpler regressions, transfer functions or transformation matrices based on these results for real time implementation. Some examples of this approach are described in Section 7.3. The wind and pressure conditions at the open water boundary are often taken from Numerical Weather Prediction models of the atmosphere (see Chapter 2). Various approaches can be used to represent the wind stress at the ocean surface, including simple quadratic functions using a drag coefficient, and methods allowing for the relative velocity of the airflow compared to tidal currents. The pressure and wind fields derived from these models are usually available on a regular grid, typically with horizontal dimensions of the order 10–100 km for regional or global scale models, although with values of 1–10 km becoming more common for national or local (mesoscale) models. The atmospheric and ocean models may also be coupled, to allow for interactions between these two components. Also, given that large extents of ocean may need to be modelled, it is likely that only a few grid points in the modelling domain will have real time available for data assimilation into model outputs data (e.g. tide gauges, wave buoys, offshore platforms, boat observations, weather stations), so further approximations need to be made when using real time data. Chapter 5 discusses data assimilation techniques further. Due to the complexities of modelling the atmospheric component, parametric inputs are also often used for hurricanes, tropical cyclones and typhoons. Techniques which have been developed include trajectory tracking, statistical, and expert system approaches (e.g. Holland 2007), and the resulting wind and pressure fields can then be used as inputs to surge and wave forecasting models. However, as the accuracy and resolution of models improves, the direct outputs from Numerical Weather Prediction models are increasingly being used in the modelling of tropical cyclones (e.g. Davidson et al. 2005; Sheng et al. 2005). Also, since surge estimates, in particular, are sensitive to the predicted track and locations of landfall, ensemble and probabilistic techniques are also used in hurricane forecasting; for example, using historical error characteristics in speed, radius, intensity etc., or ensemble forecasts from fully dynamic atmospheric forcing models (Box 7.3).

7.2 Process-Based Models

163

Box 7.3 SLOSH model, National Hurricane Centre, USA The Sea, Lake and Overland Surges from Hurricanes (SLOSH) model (Jelesnianski et al. 1992) is used by the National Hurricane Centre in Miami to calculate surge magnitudes from hurricane events. The model is used in planning mode to calculate maximum surge envelopes for input to the hazard assessment component of Hurricane Evacuation Studies, and operationally in the 24-hour period leading up to the estimated time of hurricane landfall. The model solves the depth averaged, shallow water equations of motion, typically on a curvilinear, polar grid (Fig. 7.5). The smallest grid length is typically about 250 m, although can be less, and recent developments have introduced elliptic and hyperbolic grids, allowing a finer resolution near to the shoreline (Massey et al. 2007). Approximately 40 individual models have been developed for locations and embayments along the Gulf of Mexico, the Florida coastline and the eastern seaboard of the USA. A number of SLOSH basins have been developed for countries such as India, South Korea, and China. The model resolves the effects of estuaries, bays and structures on surge propagation,

Fig. 7.5 Example of SLOSH surge height output (National Oceanic and Atmospheric Administration/National Weather Service, http://www.nhc.noaa.gov/)

(continued)

164

7 Coasts

Box 7.3 (continued) and also calculates overland flows, including wetting and drying of cells, and the influence of barriers such as levees and sub-grid resolution channels. To operate the model, the main inputs required are values for initial water levels and for the forecast track of the eye of the hurricane, sea level barometric pressure in the eye, and the radius of the maximum winds (which is typically of the order 10–100 km). Input values are required at 6 hourly intervals for a model run period of 72 hours, whilst the main model outputs are surge heights. In the planning mode, maximum values for each grid cell are estimated by performing model runs assuming a range of typical hurricane categories, speeds, and paths, based on historical information. The resulting Maximum Envelopes of Water (MEOW) and Maximum of MEOW (MOM) maps are a valuable planning tool for emergency managers. Up to 15–20,000 model runs can be required to generate each MEOW map, and the results can be categorised by Saffir-Simpson scale, forward speed, direction etc. as required. In operational use, estimates for hurricane speed, track and size are obtained from the National Hurricane Centre official advisory forecasts. Typically one

Fig. 7.6 Illustration of probabilistic hurricane storm surge product (experimental version) (National Oceanic and Atmospheric Administration/National Weather Service, http:// www.weather.gov/mdl/psurge/)

(continued)

7.2 Process-Based Models

165

Box 7.3 (continued) to two basin models are operated based on the estimated location of landfall, although sometimes up to five may be needed for hurricanes tracking along the coast. Analysis of historical events suggests that the accuracy in predicted surge heights is about 20%, based on 30 years of records, with uncertainties in the track of the hurricane being one of the main potential sources of error, the other being post-storm observations. To provide an indication of this uncertainty, cumulative probabilities and exceedance heights for surge are also published on the NOAA/NWS Meteorological Development Laboratory website (Fig. 7.6). These estimates are derived by performing multiple model runs (more than 1,000/hour) on a supercomputer, in which the storm track, speed, radius and intensity are varied over plausible ranges based on historical errors. Future gains in forecast accuracy are likely to arise from improved atmospheric modelling of hurricane development, improved bathymetry and topography, and the use of coupled offshore-onshore wave transformation models. To assist with real time interpretation of outputs, an ongoing programme of gauge improvements aims to increase the network density of tide gauges in some locations, to harden gauges to better withstand extreme surge and winds, and to monitor and refine datum values for gauges Source: (National Hurricane Centre website http://www.nhc.noaa.gov/ and Dr. Stephen Baig, personal communication)

7.2.3

Wave Forecasting

Surge forecasting models often include a wave modelling component, and the surge and wave models may be coupled to allow the interactions between tidal currents and waves to be represented. Waves can occur at many different scales and frequencies, and can be affected by changes in water depth, wind speed and direction, interactions with other waves, and other factors, such as coastal features and structures. The main processes which might be considered are wave generation and wave transformation and Table 7.4 briefly describes these factors (e.g. World Meteorological Organisation 1998; Environment Agency 2004; Holthuijsen 2007). The two main approaches to wave modelling are: ●

Phase averaging – in which the sea state at any location is considered as a statistical process resulting from the sum of many individual waves, whose amplitudes, directions, phase, wavelengths and frequencies are represented by a wave energy spectrum. Typically the energy balance is expressed in terms of the energy inputs (from the wind field), transfer (e.g. wave-wave

166

7 Coasts

Table 7.4 Summary of some key wave generation and transformation processes Category

Sub-process

Description

Wave generation

Wind shear, pressure

Generation and growth of waves linked to wind friction and pressure effects at the sea surface Wave generation due to sub-sea factors such as earthquakes and land slides (e.g. Tsunami) Whitecapping (overturning) of waves due to wave growth The overturning of waves in shallow water due to depth related differences in wave propagation speed The transfer of wave energy along a wave crest due to interactions with headlands, islands and coastal structures (breakwaters, harbour walls etc.) The change in direction of wave propagation and subsequent wave interactions due to changes and variations in water depth (e.g. as waves approach the shoreline) or interactions with tidal currents The reflection and interaction of waves as they meet headlands, islands, coastal structures etc. The onshore transfer of momentum by waves leading to increased water levels in the surf zone The increase in wave height, and decrease in wave length, as waves propagate into shallow water (or encounter currents travelling in the opposite direction to wave propagation)

Seismic (geotechnical) Wave transformation Breaking (offshore) Breaking (onshore)

Diffraction

Refraction

Reflection Set-up

Shoaling

●

interactions) and dissipation (e.g. from wave breaking), including representation of the non-linear transfer of energy between frequencies (e.g. from wavewave interactions) and from nearshore effects (e.g. currents, water depth, bottom friction). Key characteristics such as significant wave height, mean wave period and mean wave direction can be derived by integration. Both two-dimensional (excluding direction) and three-dimensional representations can be used, and grid based (e.g. finite difference) or ray tracing solution schemes. Phase Resolving techniques – in which deterministic hydrodynamic models are used to model the motion of individual waves, or waves within a given spectral band (direction and frequency), as they propagate. A spectrum of waves can be considered by running multiple realisations of inputs through the model. Any or all of the processes listed in Table 7.4 may be represented, although in some cases may need to be approximated by empirical relationships.

7.2 Process-Based Models

167

Phase averaging techniques are widely used in operational wave and surge forecasting models. A series of nested models may be used, at global, regional and local scales, with each model providing the boundary conditions to the next; for example, swell waves generated in the global model might be propagated into the regional model. Due to the smaller spatial extent, local models typically have a higher grid resolution, and it may also be possible to increase the run frequency and the number of processes represented whilst retaining a satisfactory run time for flood warning applications. Some well known examples of the phase averaging approach which are used by national meteorological and coastal services include the WAM family of models (Wave Model; Komen et al. 1994) and the WAVEWATCH III (Tolman 1999) and SWAN models (Simulating Waves Nearshore; Holthuijsen 2007). Many other types of model are also available through research projects or commercially. Phase resolving techniques provide more detailed information on changes in wave height and direction, particularly in areas of complex bathymetry or interactions with coastal structures. However, they are more computationally intensive and are generally not suitable for modelling large regions, or for direct use in real time modelling, unless wave directions and frequencies fall within a narrow band (as on some parts of the western coast of the USA, for example). Kirby et al. (2005) and Shi et al. (2001) describe examples of this approach.

7.2.4

Shoreline Processes

Shoreline models consider the processes of wave run up on a beach and overtopping of defences and natural features (e.g. dunes). Additional factors, such as breaches, may also occur, although these are discussed in Chapter 8. Overtopping (or splash-over) can be expressed in terms of the mean overtopping discharge or peak volumes. Peak volumes apply to the maximum likely value in a single wave, whilst discharge values may be calculated over a given period (e.g. hourly) or number of waves (e.g. 1,000 waves). Volumes are important when considering the risk from a single wave (e.g. the loading on a building), whilst discharge values can be used to estimate the likely depth and extent of inundation. Models are usually calibrated using estimates of overtopping rates from video images, photographs and site surveys. The rate or volume of overtopping depends on the wave propagation process at the shoreline and sea defences, and is highly sensitive to both of these factors. For off-line design applications, in addition to empirical approaches and laboratory testing, a phase resolving approach is sometimes used (see previous section), in which the motion of waves is modelled using a 1D, 2D or 3D hydrodynamic model describing the conservation of mass and momentum. Atmospheric and free surface influences may also be included and non-linear shallow water (depth averaging) approximations may also be made.

168

7 Coasts

Models of this type operate by propagating waves of a given height, period and shape (or significant wave height, frequency and spectral shape) to generate a time series of wave elevations at the seaward boundary of the defence, from which overtopping discharges and volumes can be estimated (Environment Agency 2004). The input data requirements can include information on beach profiles, the geometry and condition of sea defences (such as those illustrated in Fig. 7.7), and of structures in the shoreline region (e.g. breakwaters, groynes). Some examples of models of this type are described by Hubbard and Dodd (2002) and Causon et al. (2000). There are of course many potential uncertainties in this type of approach, particularly for the interactions between waves and coastal structures, and physical model tests may be required to estimate the required calibration factors. Depth averaged models tend to run faster, so perhaps have the greatest potential for real time use, but have limitations when vertical fluid motions are important (e.g. for steep or vertical defences). Ensemble and probabilistic approaches provide one possible route to assessing the uncertainty in model outputs, whilst data based techniques of the type described in the next section provide an alternative approach. More generally, in wave overtopping research, the focus is to extend the range of methods to more complex situations, such as steeper sea walls, different types of armouring on structures, and a wider choice of structure geometries, as well as developing artificial neural network and process-based approaches to estimation of overtopping, and the risk of defence breaches.

Fig. 7.7 Examples of sea defences (Kevin Sene, Springer)

7.3 Data-Based Methods

169

In addition to the uncertainties in the approach, model run times are another factor which to date has prevented real time applications for operational coastal forecasting. However, models of this type have been used off-line to calibrate simpler data-based methods for operational use, as described in the next section.

7.3

Data-Based Methods

Data-Based methods aim to capture the main features of the coastal response, but do not attempt to model physical processes directly. Methods include transfer functions, artificial neural networks, and other artificial intelligence techniques, and are widely used in other fields, including river flow forecasting (see Chapter 6), industry, finance and transport. Due to their fast run times, and ability to represent non-linear processes, models of this type are also candidates for use as emulators for more complex models which are too time consuming to run in real time, particularly for ensemble forecasting, as described in Chapter 5. In coastal forecasting applications, artificial neural networks are perhaps the most widely used approach, although some other techniques which have been applied include: ●

●

●

Bayesian techniques – application of fuzzy Bayesian modelling techniques to estimating the propagation of surge along the East Coast of the United Kingdom (Randon et al. 2007) Chaos theory – the application of linear and non-linear time series analysis techniques to current and surge forecasting, with applications in the Netherlands, for example (Solomatine et al. 2001) Transfer functions – application of the methods described in Chapter 6 to water level modelling in tidal zones; for example, for a tidal (estuarine) river reach in Scotland (Lees et al. 1994)

Although the term data-based is usually used to describe artificial intelligence and related techniques, it is also convenient to discuss several simpler empirical or analytical techniques in this section, including transformation matrices and wave overtopping formulae.

7.3.1

Artificial Neural Networks

Artificial Neural Networks were originally devised as a tool for helping to understand the human brain, but have since developed into a powerful technique for solving complex non-linear multivariate problems. A network is usually constructed from individual neurons, whose inputs are adjusted by weighting factors, and are transformed by a function (a so-called activation or transfer function) into the output. Networks are often constructed in the

170

7 Coasts Input 1

We

igh t1 Input 2 Weig ht 2 ht 3 Input 3 Weig t 4 igh We Input 4

Input Layer

Hidden Layer(s)

Function Output

Output Layer

Fig. 7.8 Example of a three layer artificial neural network

form of an input layer, an output layer, and one or more hidden layers which join the outputs to the inputs, as shown by the example in Fig. 7.8. Networks can adapt, or be ‘trained’, by adding or removing neurons, and changing the strength of the interactions between neurons (in this case, via the weighting factors). Training is performed with reference to an assessment (or optimisation) function, often chosen to be the mean square error between input and output values. Apart from the choice of assessment function, some distinguishing features between different networks can include the number of neurons and layers, and the choice of activation functions and training techniques. Given the potentially large number of configuration options, number of neurons, weighting factors, activation functions and other variables, much research has been performed on training techniques, including development of Bayesian methods, genetic algorithms, and stochastic approaches, such as simulated annealing. A compromise is also needed between having too complex a network (operating slowly, and over-parameterised), and over simplifying the network (potentially losing useful information). Artificial neural networks have been developed and tested for a wide range of coastal forecasting and related applications and Table 7.5 shows a few examples. In coastal forecasting applications, artificial neural networks can be trained using a range of inputs (depending on the application), including information on sea defence geometry and condition, historic databases of wave overtopping rates, real time measurements of water levels, wind speeds, wind directions, river levels,

7.3 Data-Based Methods

171

Table 7.5 Examples of research and other applications of artificial neural networks to coastal flood forecasting and related problems Type

Location

Reference

Tide level predictions Surge forecasting

Western Australia Gulf of Mexico, USA

Wave forecasting Estuary forecasting

Portugal Texas, USA Long Island, NY, USA United Kingdom

Makarynskyy et al. (2004) Patrick et al. (2002), Prouty et al. (2005) Makarynskyy et al. (2005) Tissot et al. (2003), Huang and Murray (2003) Wedge et al. (2005)

Sea wall overtopping

significant wave heights and periods, and barometric pressure, and forecasts of wind speeds, wind directions and atmospheric pressure from Numerical Weather Prediction models. Models are typically optimised for one or more forecast lead times, and evaluated against a range of performance statistics, including statistical and categorical measures, as described in Chapter 5.

7.3.2

Other Techniques

Despite advances in computing power, it can be impractical to operate some types of process-based coastal forecasting models in real time. This has led to the development of a range of techniques which use the outputs from off-line simulations to guide the development of simpler forecasting tools for real time use. Also, for some situations (e.g. sea defence breaches), it may be desirable to run multiple scenarios off-line for later use during a flood event as required. For example, coastal forecasting models are often run at time intervals linked to the daily tidal cycle (e.g. every 6 or 12 hours) which, after allowing for data gathering, preprocessing, post-processing and decision times (see Chapter 5), may give a much smaller time window for the computational aspect of the model run, including any ensemble analyses (if used). Also, some types of coastal flood event (e.g. surge) can develop in shorter timescales, or may have little tidal influence, in which case the potential lead time (and time available for model runs) may only be 1–2 hours or less. Various approaches can be used for capturing the essential features of processbased model runs and/or historical observations, including nomograms, look-up tables, carpet plots, multiple regressions, and decision support tools (e.g. World Meteorological Organisation 1998). One example is the operational coastal flood forecasting system used in parts of the United Kingdom (e.g. Environment Agency 2004; Hu and Wotherspoon 2007). Conditions at the shoreline are estimated using transformation matrices derived from off-line modelling using an offshore-nearshore wave transformation model and a shoreline wave-overtopping model. For each coastal Forecasting Point (or

172

7 Coasts

cell), the model is run using a range of scenarios for the offshore values of the directional wave spectra (mean wave direction, wave height, and wave period), and wind field (speed and direction). Site specific transformation coefficients are derived for each location. In real time operation, the resulting matrices are used to transform the offshore wave and wind forecasts from the UK’s Storm Tide Forecasting Service (see Box 7.2) to the shoreline. As with all modelling systems, the forecast performance needs to be monitored regularly, and after each major event, and the coefficients updated if there are reductions in performance, or changes to the physical characteristics at each Forecasting Point (e.g. erosion following a coastal flood event). This type of approach has also been used for forecasting the coastal impacts of tropical cyclones and hurricanes. For example, Holland (2007) and others describe the use of nomograms for estimating the likely surge impacts of tropical cyclones based on hydrodynamic model simulations for a range of idealised coastal basins and hypothetical cyclone conditions. The calculations exclude small islands and assume a regular unbroken mildly curved coastline with no inland transfer of water (e.g. steeply rising terrain). Tropical cyclones are assumed to travel in straight lines at a constant angle of attack to the coast and with constant parameters (e.g. speed). Surge estimates are derived for a range of possible tracks and parameters, and basin slopes and coastal depths. Nomograms based on these simulations provide estimates of surge for a range of possible input parameters, such as radius of maximum winds and ambient pressure drop. Correction factors may be included for factors such as the angle of attack to the coast, and shoaling effects. Another technique which is described is to prepare an atlas of pre-computed surges based on the historical characteristics of tropical cyclones which have affected a coastal basin. Cyclones are categorised by preferred track directions, intensities, and sizes, perhaps assuming that the speed, pressure and size remain constant. Values are computed for a range of likely tracks, pressures, sizes and speeds. This approach can be taken a step further by calculating the Maximum Envelopes of Water (MEOW) to give an indication of the likely worst-case surge for a given set of cyclone profiles and conditions. Computer based versions can provide an interactive tool which, near the time of landfall, can be used to select the most likely range of scenarios, with the MEOW distribution constructed from a stored set of values computed earlier. This method is used in the USA for example, as a complement to other more sophisticated approaches (see Box 7.3). Restrictions on model run times, and uncertainties in model formulation, have also led to simpler regression and other approaches being adopted for estimating wave overtopping. For example, some methods provide an estimate of likely maximum wave heights at the toe of a structure based on functions incorporating key descriptors for the characteristics of the structure, and offshore parameters such as mean significant wave height and period, together with information on the sea bed slope and the water depth. The information required on structures can include crest height, geometry for the seaward side, surface roughness etc., and model parameters may be derived from physical model tests or hydrodynamic modelling.

7.3 Data-Based Methods

173

Although these methods were usually developed for design applications (e.g. to estimate optimum values of sea defence crest levels, slope angle, required roughness), some of these approaches can also be used operationally to estimate wave overtopping volumes and discharges at a sea defence, based on offshore or nearshore forecasts of key wave parameters. For example, Tozer et al. (2007) describe an application for a rail operator in which empirical overtopping formulae are combined with a wave transformation model and offshore forecasts to provide forecasts to coastal rail lines with up to a 36 hour lead time, together with a prototype flood warning service for a coastal development. In the latter case, four hazard levels were proposed as follows: ● ● ●

● ●

Hazard level 0 – Safe use for all areas behind seawall. Hazard level 1 – Promenade immediately behind seawall shut to public. Hazard level 2 – Promenade behind seawall is unsafe for staff. Storm gates in secondary defences should be shut. Hazard level 3 – Public to be excluded from areas behind secondary defences. Hazard level 4 – Areas behind secondary may be unsafe. All protection devices should be secured.

The categories correspond to different wave overtopping rates in the range 0.03 l per second per m of sea wall, to more than 1.0 l per second per m.

Chapter 8

Selected Applications

Previous chapters have described the main techniques which are used in river and coastal flood forecasting, whilst this chapter presents a selection of forecasting applications. These include integrated catchment forecasting models, and forecasting techniques for flash floods, snowmelt, ice jams, dams and reservoirs, control structures, urban drainage flooding, and geotechnical risks, such as Tsunami, debris flows, and dam break. The chapter also includes several examples in fields which are closely related to flood forecasting, such as the real time control and optimisation of reservoir and urban drainage systems. Some themes which run throughout the chapter are the use of ensemble and probabilistic techniques to provide information on risk and uncertainty, and the use of process-based, conceptual and data-based modelling approaches.

8.1 8.1.1

Integrated Catchment Models Introduction

Real time integrated catchment models seek to model whole catchments using a range of rainfall runoff and river flow routing components. Sub-models for additional features may also be included as required, such as dams, control structures, and flood defence systems. Models may be used for a range of applications, including: ●

●

●

●

Flood forecasting – forecasting of flood flows at various Forecasting Points in the catchment Water resources – real time modelling of river and reservoir conditions, and of the impacts of abstractions and discharges for water supply, agriculture and industry Navigation – real time forecasting of water levels and velocities to assist shipping authorities and recreational boat users Pollution control – real time forecasting of the passage of pollutants through a catchment following a spillage, or following a major rainfall event; for example, linked to bathing water quality at beaches

K. Sene, Flood Warning, Forecasting and Emergency Response, © Springer Science + Business Media B.V. 2008

175

176

8 Selected Applications

The focus of this discussion is on flood forecasting, although the techniques described are also relevant to other applications. In addition to the issues of real time data availability, run time, cost, systems, and model performance which have been discussed in previous chapters, some additional factors which need to be considered in designing a real time integrated catchment model include: ● ●

●

●

The key processes which influence river flows and levels in the river network The forecasting requirements for floods, water resources, and other applications Whether the model should be optimised for the full flow range, or only parts of the flow regime (e.g. flood flows) Opportunities for simplifying the model to improve run times and stability

The modelling strategy also needs to consider whether a process-based, conceptual or data-based approach is to be used (or some combination of these approaches), whether the model will run all year round (in both low and high flow periods), or just as required (e.g. when there is a flood event), opportunities for real time updating of outputs, and the calibration criteria to be used (and whether a multi-criteria approach needs to be considered). Integrated catchment models could potentially make use of process-based, conceptual and data-based rainfall runoff and flow routing models in various combinations, perhaps also combined with a coastal forecasting model for the lower tidal boundary condition. Some types of process-based rainfall runoff models may also include flow routing components, and can be viewed as a type of integrated catchment model. However, although modern forecasting systems allow a range of model types to be operated together, the usual approach is to combine models of similar type, resolution and complexity. For example, there may be little advantage in developing a sophisticated hydrodynamic model if inflows are derived from a correlation of doubtful accuracy. Also, simpler models can have a role, both as a back up to more complex approaches, and to extend the functionality of the underlying models; for example, by relating levels at Forecasting Points to levels at key locations nearby which are not covered by the main model. As an example of an integrated catchment approach, Huband and Sene (2005) describe a real time model developed for the Environment Agency for a catchment in Eastern England with many complicating influences on flows, particularly in the lower reaches, including pumped and gravity fed discharges and abstractions, offline storage reservoirs (e.g. ‘washlands’), tidal influences, and manual and automatic flow control structures. In addition to providing flood forecasts, the model was developed for a range of other real time applications including: ● ● ● ● ●

Management of raw water transfers Management of river support Management of drought and license control Management of pollution incidents Management of navigation and of strong stream warnings

8.1 Integrated Catchment Models

177

The original version of the model included 55 lumped conceptual rainfall runoff models, operating from raingauge and weather radar data, and rainfall forecasts, feeding over 400 km of hydrodynamic network, with almost 500 structures including flood storage reservoirs, siphons, pumps and complex gates, bridges, weirs and culverts, with operations at automatic structures represented by logical rules. River levels and flows were simulated in real time at 52 Forecasting Points in the catchment.

8.1.2

Modelling Approach

When developing an integrated catchment model, some options for model development include: ●

●

Develop an integrated model specifically for the application of interest, with the minimum complexity needed to meet the requirement Develop a ‘best possible’ model for the catchment, from which other more specialised models can be derived as required

The latter approach is adopted by the Environment Agency for some catchments in the UK, for example (Huband and Sene 2005), with the best possible model called a ‘Parent Model’. In this approach, an overall model is constructed which can either be used unchanged for a range of applications, or from which simplified/optimised models can be developed for other applications (such as flood forecasting), or for studies of specific parts of the catchment. The advantage of the ‘Parent Model’ approach is that, at any one time, there is only one ‘best attempt’ model for the catchment, which can form the basis for all ‘Child Models’ used for specific applications. This simplifies the process of maintaining and improving the model, and makes it easier to document and audit the history of model development. However, the initial development time and costs can be higher than in the more classical approach of developing a range of models for different applications, although overall costs may be lower in the long term. The usual issues of data availability and type of forecasting system described in Chapter 5 also need to be considered. Some possibilities for model configuration include the following options: ● ●

●

Single model – one model for the whole catchment Multiple models – two or more models running in sequence or in parallel for all or part of the catchment Nested models – single or multiple models with more complex models nested within them for specific forecasting issues

Table 8.1 presents some of the strengths and limitations of these approaches. The choice of approach will depend on the particular modelling and forecasting issues for each catchment.

178

8 Selected Applications

Table 8.1 Some strengths and limitations of various model configuration options for hydrodynamic and flow routing models (Huband and Sene 2005) Method

Strengths

Limitations

Single model

Simple to manage future development of the model No complex decisions to make about which forecast to use No discrepancies between forecasts for the same location No boundaries between models to define Models can be run independently (not requiring the full model run) Parallel processing of model runs is possible giving faster overall run times A staged delivery of models is possible The splits between models allow independent updating algorithms to be used

Possible node limitations for hydraulic models Run times can be long

Multiple models

Nested models

8.1.3

Model must be built and tested in one operation Updating must be internal to the model Boundary conditions must be defined explicitly at model joins Downstream information may not be correctly transferred to models upstream e.g. backwater influences, tidal effects An overall catchment water balance may not be preserved Greater complexity for Duty Officers during an event (if the model run sequence is not automated) More detailed model outputs can be More difficult to maintain and update provided in areas of interest models Nested components can be used Greater complexity for Duty Officers ‘on demand’ so do not normally during an event impact upon overall model run Conflicting forecasts possible at the times same location An alternative simpler model will be The physical representation of the catchment may no longer apply available as a fall back in case of (e.g. model reaches may need to model failure overlap) Parallel processing is possible giving faster run times

Ungauged Inflows

Another issue which may require particular consideration is the representation of ungauged inflows to river reaches. These are flows for sub catchments, or areas adjacent to river reaches, which are not monitored by real time telemetry, or only partially monitored (e.g. if a gauge is some distance upstream from the confluence). Some sources of ungauged inflows can include: ● ● ● ● ● ●

Tributary inflows Surface runoff along a river reach Gravity fed abstractions or discharges Pumped inflows or outflows Spillage or seepage at river banks or flood defences Groundwater inflows or recharge

8.1 Integrated Catchment Models

179

A

B

C

D

E

F

Fig. 8.1 Illustration of ungauged and lateral flow modelling issues (circles indicate telemetered river gauges) based on a catchment in NE England (not to scale)

The need to represent these contributions will depend on the likely magnitude and timing of inflows or losses compared to flows in the main river channel. Also, although, in general, it is desirable to develop a model which performs well throughout the flow range, in practice the emphasis of development may be on the areas of particular interest (e.g. high flows, for flood forecasting, or low flows, for drought forecasting), and the relative importance of the inflow components will vary between applications. Figure 8.1 shows an example of the upper reaches of a catchment in which there are five sub-catchments with telemetry sites (shown shaded), one ungauged subcatchment (unit A), and five lateral inflow areas (units B–F). Ideally, the contributions from these areas require modelling individually, although one simplifying option might be to model various combinations of units B–F, whilst being careful not to introduce unwanted transient and other effects from introducing unrealistically high flows at some locations along the river network. Note that, even for the gauged catchments, there may be areas downstream of instruments which also need to be modelled. Some reasons for not installing river gauges at confluences can include the requirement to position the gauge in a location upstream of any backwater influences from the main river, and a range of other factors, such as lack of a suitable site or telemetry connections. If backwater effects are a factor then, when a hydrodynamic approach is being used, one option is to extend the hydrodynamic component of the model into the tributaries affected. Table 8.2 shows some typical approaches to modelling ungauged and lateral inflow components. For tributary inflows, if scaling or correlation techniques are used, the main criteria which determine flood flows are often the catchment area and some measure of rainfall (for example, mean annual rainfall), although many other factors might

180

8 Selected Applications

Table 8.2 Some possible approaches to modelling ungauged and lateral inflow and outflow components Model component

Some typical approaches

Tributary inflows (see Chapter 6 also)

Scaling from a nearby telemetered gauge based on catchment area, and possibly other parameters (e.g. rainfall, effective rainfall), possibly including a timing difference Conceptual rainfall runoff modeling using parameters calibrated from flow data from another similar catchment or historical (non-telemetered) values for the same catchment (if available) Distributed or process based hydrological modeling using parameters linked to catchment characteristics Trial and error as part of the overall calibration, possibly based on scaling inflows or outflows to the reach Assume a constant inflow per unit river length Typical daily, monthly or seasonal profiles

Gravity fed abstractions or discharges

Lumping values for many locations Pumped inflows or outflows Typical daily, monthly or seasonal profiles Lumping values for many locations Spillage or seepage at river Stage discharge relationship derived from theory or off-line banks or flood defences modelling Side weir representation Main channel inflow-outflow relationship Real time hydrodynamic model Correlations related to river flow, river depth, rainfall and/or boreGroundwater inflows or recharge hole levels Conceptual or process based rainfall runoff models with a subsurface component Data-based techniques (artificial neural networks etc.) Typical daily, monthly or seasonal profiles 3D numerical groundwater models

also need to be considered, including soil type, mean slope, topography etc. However, when used operationally, this approach does not allow the influence of local variations in catchment rainfall to be considered and, if the model is to include this effect, then normally a rainfall runoff model will be required. If there are no historical records for the sub-catchment (e.g. from an existing river gauge without telemetry), then some options for developing a rainfall runoff model include transferring model parameters from a similar nearby catchment, or attempting to derive a pseudo inflow record by taking the difference between measurements upstream and downstream of the location(s) at which flows enter the main stream, allowing for lag times, attenuation, rating curve errors etc. Both approaches have their disadvantages and, in practice an iterative, trial and error is often used to obtain the best calibration of rainfall runoff and flow routing models in combination. An integrated catchment model can also include a number of sub-models for other features which can influence river flows, including dams, reservoirs and control structures, and some of these components are described in later sections.

8.2 Flash Flood Forecasting

8.2

181

Flash Flood Forecasting

Flash floods can be damaging due to the short time in which they develop, and the high depths and velocities which may be reached. Flows may also contain significant quantities of debris and sediment, potentially causing blockages and further raising river levels at structures such as bridges, weirs and river control structures. Chapter 10 discusses some of the issues that can arise in responding to this type of event, and provides an example for a flash flood which occurred in the UK in 2004, whilst Chapter 3 discusses a range of rainfall threshold based approaches for early warning of events such as flash floods. The definition of a flash flood varies between countries, with a common theme being that flooding of properties and infrastructure may occur in locations where there is no recent flooding history, and develop sufficiently rapidly that normal flood warning dissemination and emergency response procedures do not have time to operate effectively. Flash flooding is therefore often defined in relation to the locations at risk and local response procedures, as in the following example (ACTIF 2004): “A flash flood can be defined as a flood that threatens damage at a critical location in the catchment, where the time for the development of the flood from the upstream catchment is less than the time needed to activate warning, flood defence or mitigation measures downstream of the critical location. Thus with current technology even when the event is forecast, the achievable lead-time is not sufficient to implement preventative measures (e.g. evacuation, erecting of flood barriers).” Some common features which appear in many definitions of flash floods include the following items (e.g. World Meteorological Organisation 1982; Meon 2006) although, in many cases, only some of these factors may apply: ● ● ● ● ● ●

Tend to be caused by short duration, localised, intense storms (e.g. thunderstorms) Develop rapidly in response to rainfall May have significant mud/debris content Tend to occur on small, steep and/or urban catchments May be strongly influenced by catchment antecedent conditions Tend to occur in locations with no recent experience of such events

If a particular catchment response time is specified, then the values for events classified as flash floods vary widely between countries and can range from a few minutes to a few hours (typically 6 hours e.g. World Meteorological Organisation 1982), with catchment areas typically of at most a few hundred square kilometres, although sometimes much less than this. Various indicators such as catchment area, topography, mean slope, soil type, response times and flood risk may also be used to identify flash flood prone catchments. Fast response events from other sources, such as ice jams (Section 8.3), urban drainage systems (Section 8.5) and dam breaks, landslides, and Glacial Lake Outburst Floods (Section 8.6), are sometimes also classified as types of flash flood. The main difficulty with forecasting for flash floods is the speed at which they develop, compared to other types of flood event, and the uncertainty about which

182

8 Selected Applications

particular locations will be affected in a region. The rarity of events may also mean that there has been no need in the past for river level monitoring in the catchment to confirm that an event is developing (i.e. the catchment is ungauged). These factors also have implications for issuing flood warnings, and for providing an effective response, and this topic is discussed in Chapter 10. Some techniques used for flash flood forecasting include: ●

●

●

●

Rainfall Thresholds using information on observed or forecast rainfall, and possibly current catchment conditions (see Chapter 3) Meteorological Indicators of flash flood generation potential from observations, historical databases, or Numerical Weather Prediction models (see Chapter 3) River Level Thresholds in which warnings are issued using decision criteria based on increasing river levels or flows (see Chapter 3) Rainfall Runoff Models using observed rainfall and, possibly, forecast rainfall as inputs

As indicated, Chapter 3 describes a range of approaches which can potentially be used for flash flood forecasting and warning, including rainfall depth-duration, Flash Flood Guidance and probabilistic rainfall threshold techniques, geopotential, vorticity, precipitable water and lightning meteorological indicators, and river level threshold crossing, rate of rise and correlation methods (Box 8.1). Rainfall Threshold and Meteorological Indicator approaches have the advantages of providing additional lead time, and that they can be used even if there is no river telemetry available. However, sometimes there can be considerable uncertainties in the precise location, timing and magnitude of flooding, and this point needs to be emphasised when issuing advisories, pre-warnings and other alerts based on these techniques. Flash flood hazard maps, based on indicators of flash flood risk, such as steepness, vulnerable locations, public awareness, and soil type, can also provide guidance on catchments at particular risk, together with flash flood risk catalogues at a national scale. By contrast, River Level Thresholds have the advantage of being based on observations of current river conditions, although provide less lead time than the other two approaches. If a Rainfall Runoff Modelling approach is used, the techniques used for flash flood forecasting are similar to those for other types of flooding, and include process-based, conceptual and data-based models (see Chapter 6). However, one potential difficulty may sometimes be a lack of historical and real time data for the calibration and operation of models. Rainfall inputs can be obtained from raingauges, weather radar, satellite, and rainfall forecasts, with a typical problem for raingauges being that there may be no gauge within or close to the catchment and, for the other methods, that the resolution may sometimes be too coarse compared to the scale of the catchment to provide estimates which are representative. For catchments where soil moisture or snow cover is a factor in flash flood generation, there will often also be uncertainties about the initial catchment conditions. If the catchment is gauged, then the model can be calibrated to historical river flow records, if these are available. Given the rapid rate of rise of many flash floods, the focus of model calibration and development may be more on success at predict-

8.2 Flash Flood Forecasting

183

ing the crossing of thresholds on the rising limb of the hydrograph, rather than on accurate prediction throughout the event. Rainfall forecasts may be also used to extend lead times, although the accuracy of forecasts decreases with increasing lead time. Real time updating (data assimilation) is also usually desirable to help to account for differences between observed and forecast flows, provided that the data quality is sufficient. Modelling (and monitoring) may also be complicated by the risks of debris and sediment, and in some cases that the river may be dry for part of the year (e.g. on wadis). In urban areas, a range of extra factors may also need to be considered, such as the capacity of the drainage system, blockage risks, and the performance of flood storage and detention areas (see Section 8.5). For ungauged catchments, there is the additional uncertainty arising from having no river flow data for calibration, and views differ on whether warnings should be issued on the basis of model forecast outputs alone in flash flood situations. Much will depend on confidence in the model itself, and whether verification studies show that the model outputs are a reliable predictor of flooding. Also, considerable advances are being made in developing techniques for rainfall runoff modelling in ungauged catchments, as described in Chapter 6. In addition to these various techniques, the warning time available can sometimes be increased by reducing the various time delays in the overall flood warning process (see Chapter 5). Some possibilities include making decision making procedures more efficient, adopting faster approaches to warning dissemination, and improving the speed of operation of telemetry and forecasting model systems. Automated linkages between telemetry observations or flood forecasting model outputs might also be implemented to reduce the time required to issue warnings, such as automated signs or barriers on roads (although with the potential disadvantages discussed in Chapter 4). Probabilistic and ensemble techniques (see Chapters 1, 5 and 10) also have the potential to improve operational decision making during the development of flash flood events, and to assist with developing a more risk-based approach to issuing flash flood warnings. Given the destructive nature of some flash floods, and recent developments in modelling and monitoring techniques, the issue of flash flooding is an active research area, and Table 8.3 summarises some major research programmes on this topic. Table 8.3 Some international research and collaboration programmes in flash flood forecasting Project WMO flash flood guidance system Central America Flash Flood Guidance (CAFFG) system

Location

International covering all WMO regions Belize, Costa Rica, El Salvador, Guatemala, Honduras, Nicaragua, and Panama ICIMOD flash flood projects Afghanistan, Bangladesh, Bhutan, China, India, (capacity building, early warning, satellite detection etc.) Myanmar, Nepal, Pakistan Project URBAS (prediction and Germany management of urban flash floods)

Reference World Meteorological Organisation (2007) Georgakakos (2005) Sperfslage et al. (2005) Erikkson (2006)

Castro et al. (2006); (see Section 8.5)

184

8 Selected Applications

Box 8.1 WMO Flash Flood Guidance System The WMO Flash Flood Guidance System (FFGS) aims to assist National Meteorological and Hydrological Services to improve their capability in providing warnings for flash flooding. The project was launched as part of the WMO Flood Forecasting Initiative following the first International Workshop on Flash Flood Forecasting held in Costa Rica in March 2006. The project would be implemented by the Hydrologic Research Centre (HRC) in collaboration with NWS and funded by the U.S. Agency for International Development/Office of Foreign Disaster Assistance (USAID/OFDA). The aim of the Initiative is to enable access to satellite information which provides warnings for catchments with areas in the range 100–300 km2 for lead times of 1–6 hours. The FFGS follows a similar approach to that adopted for the Central America Flash Flood Guidance system (CAFFG), implemented by the Hydrologic Research Centre (HRC). The CAFFG system has been operational in seven countries in Central America since 2004 (Belize, Costa Rica, El Salvador, Guatemala, Honduras, Nicaragua, and Panama), and builds upon the Flash Flood Guidance approach developed and used by the National Weather Service in the USA since the 1970s (Fig. 8.2).

Fig. 8.2 Sample graphical system output for Nicaragua from the CAFFG System showing Flash Flood Guidance (FFG) and Flash Flood Threat (FFT) (reproduced from the WMO Prospectus for Implementation of a Flash Flood Guidance System, courtesy of WMO)

Flash Flood Guidance is defined as the amount of rainfall of a given duration over a small basin needed to create minor flooding (bankful) conditions at the outlet of the basin. Threshold rainfall values can be estimated by inputting historical events or typical storm profiles to a catchment rainfall runoff model for a range of durations and soil moisture conditions. In real time operation, rainfall observations or forecasts are combined with the outputs from physically-based soil moisture accounting models to estimate the Flash Flood (continued)

8.2 Flash Flood Forecasting

185

Box 8.1 (continued) Threat, which is the amount of rainfall of a given duration in excess of the corresponding Flash Flood Guidance value. A phased approach to implementation is envisaged in which regional centres are established at selected National Meteorological and Hydrological Services, equipped with the necessary computer, database, hydrological modelling, display and product generation facilities. Considerable flexibility will be provided in choice of hydrological models and rainfall observations and forecast products, including the option to use satellite estimates of precipitation, automatically corrected for bias based on telemetered raingauge data (where available). The facility will be included for forecasters to make adjustments to values received from regional centres based on experience since this has been shown to improve success at forecasting flash floods, and to reduce false alarm rates. Data and products will be transferred using existing WMO telecommunications systems and protocols, with appropriate training, verification, capacity building and documentation all forming important components of the overall project.

Box 8.1 provides more information on the first two of these initiatives. Flash flooding is also a major driver for research into detection techniques, including the following methods which are discussed in Chapter 2: ● Satellite based techniques – analysis of cloud type, extent and rainfall generating potential, and of catchment conditions (snow cover, soil moisture etc.) ● Weather radar – improved signal processing algorithms and hardware, particularly for automated detection of heavy rainfall in mountainous areas ● Microwave techniques – methods for rainfall detection along microwave paths between locations several kilometers (or more) apart ● Raingauges – low cost devices allowing greater network densities, and a greater range of options for use in mountainous areas ● River monitoring – small, low cost devices allowing larger numbers to be installed within a given budget, automated analysis of CCTV images, radar and ultrasonic methods for remote detection of water levels

8.3 8.3.1

Snow and Ice Snowmelt Forecasting

In catchments where snowfall accumulates, snowmelt has the capacity to cause significant flooding, and events can be notable both for their magnitude and long duration.

186

8 Selected Applications

Flooding can arise directly from the melting of fallen snow from rainfall or rising temperatures, or from radiation effects, or enhanced runoff from frozen snow, or some combination of these mechanisms. Snowmelt events may also raise river levels so that the capacity to absorb even moderate rainfall events is reduced. Radiation melt often dominates in high latitude and high mountain areas, whereas warm, moist turbulent airflows are an important factor in mid latitude, lower lying regions. Snowmelt forecasting models can range from simple empirical approaches relating snowmelt to temperature, through to models which attempt to represent both the mass and energy balances of the snowpack and melted snow. Real time data inputs (see Chapter 2) can include measurements of meteorological and hydrological conditions, satellite observations of snow cover, ground based measurements (e.g. ablation stakes, snow pillows), and forecasts for air temperature, humidity, wind speed, cloud cover, and rainfall. Some simple empirical approaches include linear or non-linear regressions between flow volumes and various indicators for factors which can influence the rate of snowmelt, such as air temperature and rainfall, the water equivalent depth of accumulated snow, soil moisture, and the depth of frozen soil (e.g. World Meteorological Organisation 1994). Another commonly used approach is the temperature index or degree-day method in which the rate of snowmelt is represented as a function of the difference between mean daily air temperature and a threshold value above which snowmelt is considered to occur (often zero centigrade). The function is usually a simple constant factor and is site specific, and different values may be required for open and forest areas. The term degree-day refers to an integrated value over a day, although other periods, such as hourly or monthly values, can also be used. Usually the only real time input required is for air temperature, although more advanced forms may also include wind speed and radiation inputs, and may allow for elevation influences, and cloudy and sunny conditions. However, known shortcomings are that the accuracy decreases as the time interval chosen is reduced, and that it is difficult to account for spatial variability in snowmelt due, for example, to topographic influences (Hock 2003). Conceptual models are also used in which the snow store is separated into snow and melt components, and a simple water balance is used to keep account of each component, with empirical or simple physically based equations to relate snowmelt rates to air temperature. Typically, these models use air temperature relative to a threshold (e.g. zero centigrade) to help to decide whether to partition precipitation into rainfall or snow, and when to trigger the snowmelt component of the overall model. Effects such as the proportion of snow cover in the catchment (areal depletion curves), windspeed, and the influences from altitude and aspect (the direction in which major hill slopes face) may also be included. One or more stores may be used to represent the current status of melted snow (e.g. stored within the snowpack, lost as runoff), and the catchment divided into elevation zones to better account for differences in temperature with elevation. Other advances include the use of data assimilation through state updating based on point measurements by snow pillow or snow cores (e.g. Bell et al. 2000).

8.3 Snow and Ice

187

Forecasting models have also been developed using process-based rainfall runoff models of the type described in Chapter 6 (e.g. Koren et al. 1999; Dunn and Colohan 1999). Factors such as elevation and aspect can be included directly, and the models can also account for other variations across a catchment, such as in wind speed, the accumulation of snow due to wind effects (e.g. snow drifts) and variations in snow density. The most detailed models use a full set of mass and energy balance equations to describe snowmelt, including real time observations of net radiation, air temperature, wind speed and humidity (e.g. World Meteorological Organisation 1994). Terms which are included in the energy balance can include energy storage in the snow layer, net radiation, latent heat fluxes, the heat flux to the ground or soil, and advection losses to wind flow over the snowpack, whilst the water balance may allow for two or more layers to represent snow in various stages of melting or freezing. Figure 8.3 illustrates some terms in the overall energy and mass balances for a melting snow layer; note that the directions shown for energy and water fluxes are indicative, and in some cases can be in the opposite direction. The net radiation depends on several factors including time of day, season, forest cover, and snow surface conditions (which are typically parameterised via albedo and emissivity). Other less significant factors may also contribute to the energy balance, such as heat input from rainfall.

ENERGY BALANCE

WATER BALANCE Precipitation

Net Radiation

Sensible Heat

Latent Heat

Stored Energy

Evaporation Wind

Snow Layer

Recharge

Ground Heat Flux Runoff

Melting Surface Layer

Fig. 8.3 Some key terms in the energy and water balance for a melting snow layer

188

8 Selected Applications

Historically, one barrier to application of this type of model has been the availability of real time data on both meteorological conditions and snow depth and cover. However, improvements in the accuracy and resolution of Numerical Weather Prediction models, and satellite based observation techniques, have helped to make the application of this type of model more practicable. Ensemble techniques can also be used; for example, the Extended Streamflow Prediction System (ESP) used by the National Weather Service River Forecast Centres in the United States since the 1970s (see Chapter 5). The method uses an ensemble technique to create probabilistic river stage forecasts for periods of up to a few months ahead. The main use is to provide forecasts for the Spring snowmelt, by using the state variables of models at the time of forecast and up to 40 years of historical time series data for model inputs (precipitation, temperature, potential evaporation). The outputs from the model include probabilistic forecasts for peak flows and volumes at multiple forecast points. Kuchment and Gelfan (2005) describe a similar technique which has been developed for the Sosna River Basin in Russia, and incorporates a process based rainfall runoff model which represents snow accumulation and snowmelt, soil freezing, soil moisture, and runoff generation. Long term deterministic and stochastic (Monte Carlo) estimates for daily air temperature and precipitation are used as input to the model. Shorter range probabilistic flood forecasting techniques are also under development and are described in Chapter 5. Various international comparisons have also been performed of snowmelt models; for example, in the first phase of the Snow Model Intercomparison Experiment Project (SNOWMIP), more than 20 models from ten countries were compared, with case study sites in the USA, Canada, France and Switzerland, and the project has continued into a second phase (SNOWMIP2; Rutter and Essery 2006).

8.3.2

River Ice Forecasting

River ice can lead to an increased risk of flooding through mechanisms which include: ●

●

●

Ice formation – impeding flows in river channels and at bridges and other structures, reducing conveyance capacity, possibly for periods of weeks or months Ice jam – site specific blockages following the break up of ice cover further upstream, particularly at bridges and flow control structures, causing a local shorter term risk of flooding, and with the potential for significant structural damage Ice break up – flood waves generated by the sudden release of water following melting or release of an ice jam, layer or dam

Ice formation tends to occur over periods of hours or days whilst ice jams can develop rapidly, causing backwater influences and significant flooding in periods of as little as an hour or less (and, for that reason, are sometimes called flash floods). Ice jams may also act as an additional location for ice formation as floating ice accumulates and ice forms at the point of constriction.

8.3 Snow and Ice

189

The presence of ice may also cause uncertainties in the inputs to (and hence outputs from) flood forecasting models, by affecting the operation of river instrumentation and control structures, and causing a change in the stage-discharge relationship (or rating curves) used to convert levels to flows at gauging stations. River levels may also rise and fall as ice forms and breaks up, sometimes rapidly in the case of ice jam movement. Some techniques for monitoring the formation and break up of ice are described briefly in Chapter 2. Two main categories of ice can be identified; thermally grown (or sheet) ice and frazil ice. Thermally grown ice tends to occur in slow moving or still water from temperature influences whilst frazil ice forms in moving water, and is a common type of ice formation in rivers. Some particular locations where frazil ice occurs include turbulent regions such as at river confluences, sudden reductions in river slope, and downstream of rapids or in turbulent flow related to structures. Techniques for estimating ice formation can range from simple correlation approaches to process-based techniques of the type first developed for ocean ice formation (e.g. World Meteorological Organisation 1994; Snorrason et al. 2000; Kubat et al. 2005). Ice melt and break up can occur from thermal effects such as increases in water or air temperature, and direct radiation from sunlight, or from mechanical influences from the hydraulic loading of river flows, or impacts from floating ice, particularly during flood flows. For thermal melt from water temperatures, river ice tends to melt at the upstream edge of the ice cover, and the pattern of break up along a river reach depends on the mean river slope, and changes in slope along the reach. Procedures for forecasting ice formation and break up are used in several countries where ice-related flooding is a problem, although are subject to many uncertainties. For estimating the rate of formation of ice, one simple approach is to use a temperature index approach similar to that described earlier for snowmelt modelling (e.g. World Meteorological Organisation 1994). The rate of ice formation is assumed to be a function of a degree-day total (cumulative temperature over threshold measure) calculated since the start of freezing. Similar techniques can also be used for estimating ice melt, in which melt is assumed to be proportional to a function of the river discharge and the air temperature above a threshold value (typically zero degrees). The time required to melt a given reach of ice can then be estimated from the melt rate and the volume of ice estimated from the average ice thickness and width. Hydraulic models of the types described in Chapter 6 also provide a way to model potential locations for ice formation, and the impact of ice jams, and both steady state and unsteady models might potentially be used (e.g. Blackburn and Hicks 2003). Some formulations also include dam break type models, and sub-models for ice formation and ice melt. Operationally, the challenge in using these models is to obtain or estimate up to date values for river ice formation, transport and break up for input to the models; also, the nature of ice jams and break up can be site specific, making it difficult to develop general approaches (e.g. White 2003; Hom et al. 2004). However, the timescale required for observations will depend on the nature of the risk so that occasional (e.g. daily) observations may be sufficient in some cases, for example using visual, CCTV, webcam or video-camera based observations. For example, in Alaska, an extensive programme of aerial observations, called

190

8 Selected Applications

River Watch, operates when the flood risk from ice is high, making use of voluntary contributions by air taxi operators, private pilots, and others. Various statistical and regression techniques have also been developed, sometimes linking into extensive databases of historical records (e.g. Mahabir et al. 2006). Some other current areas of research into data-based methods include multivariate statistical methods, artificial intelligence (e.g. neural networks), and Kalman filtering, in some cases combined with a hydraulic modelling approach (e.g. Daly 2003; Morse and Hicks 2005).

8.4

Control Structures

Control structures can be used in rivers and along coastlines for a variety of purposes including: ● ● ● ● ● ●

Regulation of river flows Controlling water levels for navigation, irrigation etc. Protecting against high river or tidal levels Hydropower generation Diverting flows for flood mitigation Storing water for water supply, irrigation etc.

Types of structure can include dams, gates, sluices, weirs, barrages, locks and siphons. Structures may be uncontrolled, in the sense that they always operate when certain criteria are met (e.g. levels exceeding a certain threshold), or controlled either manually or automatically. For flood forecasting applications, a given structure may influence flows or levels sufficiently that if possible it needs to be included in a forecasting model, and this section describes some typical types of model for the following applications: ● ● ●

Dams and reservoirs River control structures Tidal barriers

The resulting forecasts may also be used as inputs to real time control and decision support systems and several examples are described in the following sections.

8.4.1

Dams and Reservoirs

Dams and reservoirs can be constructed for a range of applications, including hydropower generation, water supply, navigation, and flood control. The impounding structure may be in the main river channel (in-line, or on-stream), or adjacent to the channel (off-line, or off-stream). In-line structures typically regulate or control flows using gates, valves or sluices at low to medium flows, but allow flows to pass freely over spillways during flood events.

8.4 Control Sructures

191

Some schemes, such as hydropower plants, may also have emergency drawdown arrangements. Smaller scale reservoirs built specifically for flood mitigation, such as flood detention ponds, may have no control at low flows (for example, using a free flowing tunnel as an exit), but start to hold back water as levels rise towards the peak. Flood flows may be discharged over a spillway built into the front of the dam wall, or using side channels, tunnels or bellmouth (shaft) spillways (Fig. 8.4). Self-priming siphons may also included to rapidly draw down water levels if they approach the dam crest. Off-line reservoirs, sometimes known as washlands, typically have an earth, rockfilled or concrete wall around the perimeter, possibly with one or more internal barriers with gates or sluices to control which areas of land are flooded. Flows in the main river channel are diverted at weirs and sluices as required to reduce flows further downstream, or for irrigation of the enclosed areas. Since floods may only occur

Fig. 8.4 Examples of overflow and bellmouth reservoir spillways

192

8 Selected Applications

occasionally, the land may be used for farming or recreation at other times. Reservoirs of this type are sometimes called polders although, for a true polder, the main purpose is to protect reclaimed land within the boundaries of the polder, rather than to act as a flood storage area, and sophisticated drainage and pumping systems may be used to help to avoid flooding from rainfall, seepage, groundwater, rivers and the sea. Figure 8.5 shows some examples of the potential impacts on flood flows downstream of fully regulated and unregulated in-line reservoirs, and an unregulated off-line reservoir (World Meteorological Organisation 1994). Reservoirs are usually operated using control rules (steering rules) linked to reservoir levels, and possibly river levels, river flows, and other parameters (e.g. rainfall). Rules may be static (e.g. seasonally dependent), or updated dynamically based on observations and the outputs from forecasting models. Real time forecasting systems can be used to assist with decision making and optimising reservoir operation throughout the flow range (droughts, floods etc.), particularly when there are several interconnected reservoirs to consider (multi-reservoir systems). Some key decisions to take during a flood event include: ●

●

●

●

●

●

Whether the predefined flood buffer (if any) will be sufficient for the anticipated event and, if not, how much to draw down levels to protect the dam and locations further downstream, particularly if multiple flood peaks are anticipated (e.g. during a succession of storm events) Whether levels or flows are likely to reach values which might damage or overtop the dam wall or operating equipment Whether normal operations should be suspended and for how long (e.g. for water supply) Whether to warn or evacuate people downstream of the reservoir if it is likely to spill or breach When, and to what extent, to divert flows into an off-line storage area, particularly if people need to be warned or evacuated, or further flood peaks are forecast The optimum fill and drawdown sequence for a chain of reservoirs in a multireservoir system

Fig. 8.5 Effects of reservoirs on floods (a) regulated storage (b) unregulated on-stream storage (c) unregulated off-stream storage (Reproduced from the WMO Guide to Hydrological Practices – Data Acquisition and Processing, Analysis, Forecasting and Other Applications, courtesy of WMO)

8.4 Control Sructures

193

Also, there will often be an economic dimension to consider as part of the decision making process; for example, if water is spilled from a hydropower or water supply reservoir in advance of a flood event, this may incur an opportunity loss for electricity generation or future water supply. Similarly, for off-line reservoirs, there may be penalty payments to land users when flows are diverted onto farmland. Various local arrangements may also be in place; for example, a flood warning authority may rent the upper part of reservoir storage (a flood buffer) at a fixed annual rate, but incur penalty payments if additional drawdown is required for flood protection. Dynamic programming, stochastic simulations, artificial neural networks and other techniques can be used to assist in the optimisation process (e.g. Bhattacharya et al. 2003; Lobbrecht et al. 2005; Nandalal and Bogardi 2007). Forecasting models for reservoirs can range from simple correlation and water balance models through to complex integrated catchment models incorporating rainfall runoff, snowmelt, flow routing, water transfer, hydrodynamic, and evaporation models, including representation of the reservoir control rules. Some key factors to consider in selecting an appropriate model for the reservoir component include: ●

●

●

●

The availability of real time information on reservoir levels for model initialisation since, if the initial storage is not known, the forecast outflows can be considerably in error. The availability of historical and real time information on reservoir control rules and gate settings (if relevant), and the extent to which these can be encapsulated into a model (e.g. logical rules). Also, the influence of pumping, water transfer and other operations. The extent to which the reservoir influences flows downstream during flood events. For example, if the impact is simply to attenuate and delay flows a simple flow routing model may be sufficient but, if outflows are controlled, or sensitive to forecast levels (e.g. siphon flows), then a full hydrodynamic model may be required. Whether components in the water balance such as open water evaporation, and controlled releases for water supply, environmental or ecological purposes, are of sufficient magnitude to require inclusion in the model.

The availability of real time data can be a particular issue for reservoir forecasting, particularly for the reservoir levels required for model initialisation. Ideally, in addition to reservoir levels, all key inflows, gate settings and outflows would be monitored in real time, but this is often not the case unless the monitoring network has been designed specifically to support a real time control and forecasting application. Also, for a reservoir with a large surface area, gauging of all major inflows may be impracticable. However, approximate modelling solutions can often be developed, for example, parameterising control rules, using ungauged catchment forecasting techniques, and other methods. The control rules used in practice may also differ from the published control rules, which can require considerable investigation of historical records and discussions with operators to determine the actual rules which are used. Probabilistic techniques (see Chapters 5 and 10) are increasingly being developed to assist in decision making during reservoir operations. This can bring several benefits, including increased transparency in decision making, awareness of the likely

194

8 Selected Applications

worst case scenarios given model uncertainty, and the ability to estimate probabilities of occurrence for input to optimisation routines using cost-loss or utility or penalty functions (see Chapter 10 for a discussion of these terms). Some examples of optimisation problems might include the trade-offs between opportunity losses due to draw down compared to the potential flood damage to locations downstream, the likely cost of repair work at the dam if damage occurs, impacts at the reservoir (recreational, ecological etc.), or the costs of any penalty payments which are incurred. Stakeholders might also choose to receive warnings at different risk thresholds, where risk can be defined as the combination of probability and consequence. For example, a dam operator might be interested in receiving a flood warning at a much lower probability or risk level (with a higher number of false alarms) than, say, a community downstream of the reservoir. Table 8.4 gives some examples of research and operational studies into the use of forecasting models and decision support systems for real time reservoir operation and flood control. Table 8.4 Examples of real time decision support systems for reservoir control Model inputs and structure

Location

Reservoir uses

Lake Como, Italy

Irrigation, hydropower

Stochastic

Lenne River, Germany

Multiple reservoirs for hydropower, water supply Multiple polder systems

Deterministic

The Netherlands

Hydropower Powell and Lois rivers, Canada Folsom project, Flood protection California, to Sacramento, USA hydropower, water supply, recreation China Reservoir flood forecasting and control system Ebro river Multi reservoir basin, Spain systems (41 reservoirs) Hydropower, flood Paranaiba control river basin, Brazil Feitsui Hydropower, water supply Reservoir, Taiwan

Optimisation problem

Reference

Opportunity losses from drawdown versus flood damage downstream Optimisation of a multiple reservoir system for flood control Optimisation of water level management

Todini and Codeluppi (1998)

Deterministic, Ensemble

Flood control, Emergency Response

Bowles et al. (2004)

Deterministic

Flood control

Guo et al. (2004)

Deterministic

Flood control, water management

Garcia et al. (2005)

Deterministic

Hydropower operation and flood control

Deterministic

Flood control in typhoons

Collischonn et al. (2007a,b) Nandalal and Bogardi (2007)

Artificial neural network Ensemble, statistical

Göppert et al. (1998)

Bhattacharya et al. (2003) Hydropower generation Howard (2004, 2007)

8.4 Control Sructures

8.4.2

195

River Control Structures

River control structures can include gates, sluices, weirs, barrages, locks, and siphons. It is also convenient to discuss pumps in this section. Some typical applications include: ● ● ● ● ● ●

Diverting flows to off-line storage, water transfer schemes, farms etc. Maintaining levels upstream for irrigation or water supply Reducing river flows to mitigate flooding downstream Micro-hydropower schemes River flow gauging Control of river levels and flow velocities for navigation

Another application is to protect against high tides although tidal barriers are discussed in the following section. For flood forecasting applications, in addition to the usual considerations of cost, data availability and the functionality of the forecasting system etc., the decision on which, if any, structures to include in a model will depend on the proposed application of the model, the locations of Forecasting Points, real time data availability, information on control rules, and other factors. In some cases, major simplifications may be possible. For example, for pumps or gravity fed offtakes, large numbers may be grouped together with a combined operating rule, perhaps linked to seasonal control rules rather than relying on real time data. A group of pumps might be considered to divert a fixed flow above a certain threshold, and zero flow at all other times, or to pump or discharge flows at a rate which depends on river levels. Also, some structures may be distant from the required Forecasting Points and have minimal influence on the timing and magnitude of levels and flows at those locations, and hence can be omitted entirely. The decision on which structures to include will vary from case to case. The most appropriate type of model will depend on the mode of operation of each structure. In some cases, a simple flow routing or water balance approach may be sufficient to represent the effect of the structure on downstream flows. However, a real time hydrodynamic model may be required if the structure has complex control rules, can impound significant volumes of water, or if its operation depends strongly on forecasts of levels upstream and/or downstream of the structure (e.g. backwater influences). If the structure is within or downstream of a Flood Warning Area, then a detailed model may be required for its influence on both upstream and downstream river levels. The modelling of structures can become complex when there are several structures interacting and controlling levels or flows further downstream. In this case, feedback effects can develop, and multiple scenarios or a probabilistic approach may need to considered to achieve an optimum response as described in the previous section. Figure 8.6 gives an example of the type of situation which can arise. In this example, river levels in a flood defence system protecting a town are controlled by diverting flows to off-line storage reservoirs. The flows are diverted at weirs with diversion channels, and the decision making process includes estimating how much flow to divert, and at what times in the flood event. For example, one consideration is that, if water is diverted too early, the reservoirs may not have the capacity required to

196

8 Selected Applications

Flood Defences

Off-line storage 1

Weir 1

Weir 2

Off-line storage 2

Fig. 8.6 Illustration of a simple real time control problem

protect against the peak of the event. If multiple peaks are forecast (e.g. due to a succession of heavy rainfall events), then decisions need to be made on whether to delay diverting flows to preserve capacity for later events (perhaps incurring some flooding downstream), and on the extent to which reservoirs can be drained down between peaks. Of course, in a worst case scenario, the forecasting model needs to alert operators that the flood defences are likely to be overtopped, and a flood warning issued. The choice of modelling approach will depend on the availability of real time data, the control rules at the weirs, and the availability of survey data (if a hydrodynamic model is required), as well as the usual factors of cost, system environment etc. Also, the required accuracy at the flood defences may be a major consideration; for example, if levels are be controlled during a flood event to within a typical freeboard allowance of 0.1–0.5 m, then a hydraulic model would probably be required. A possible forecasting approach could include rainfall runoff models for the tributary and lateral inflows, a hydrodynamic model extending from upstream of the weirs to the flood defence system, and water balance or hydrodynamic models for the offline storage. Some other factors to consider include possible numerical transient effects which may appear in model outputs if a hydrodynamic model is used (e.g. when gates operate), whether there are any backwater influences from downstream of the flood defence system (e.g. tidal influences), and whether to use a probabilistic or deterministic approach. Also if real time updating (data assimilation) is used at river gauging stations, this can further complicate development of the real time control algorithms. Hence, although at first sight this is a simple configuration, this is a potentially complex optimisation problem, which might require some exploratory investigations to develop a solution.

8.4 Control Sructures

197

Box 8.2 Thames Barrier (Environment Agency) The Thames Barrier (Fig. 8.7) is situated in East London and protects London from coastal flooding. The barrier consists of six rising sector gates which, when lowered, lie in sills flat against the river bed allowing river users (ships, boats etc.) to pass but, when raised, prevent levels rising on the upstream side of the barrier due to tidal influences further downstream in the Thames Estuary. Four simple radial gates are also included in the structure. When necessary, the gates are typically closed four to six hours before high tide, requiring the use of coastal forecasts to provide sufficient early warning to allow the barrier to be closed in time to avoid flooding. The control rules are based upon observed river levels upstream of London, and forecasts for tide levels and surge in the tidal reaches of the Thames Estuary. These rules are based on detailed hydraulic modelling performed at the time that the barrier was being designed, combined with experience gained since completion of the barrier.

Fig. 8.7 Thames Barrier (© Environment Agency copyright and/or database right 2007. All rights reserved)

Coastal forecasts are obtained from the UK’s Storm Tide Forecasting Service (STFS), which operates models on a 12 km grid for the entire coastline of the UK and the North Atlantic continental shelf, including the Bay of Biscay. The model provides hourly surge forecasts up to 36 hours ahead at 6hourly intervals. A reduced version of the model, for is also operated locally (continued)

198

8 Selected Applications

Box 8.2 (continued) at the Thames Barrier which allows data to be assimilated at the model boundaries to improve model accuracy. Astronomical tide predictions are obtained from the Proudman Oceanographic Laboratory (POL). In the tidally influenced river reaches, a real time hydrodynamic model is used for forecasting tidal levels, supplemented by rainfall runoff and onedimensional hydrodynamic models for forecasts of river flow in reaches further upstream. These models include representation of major tributary inflows and gates and barriers along the Thames estuary shoreline, and discharges from key waste water treatment works. An empirical look-up table approach is also used as a backup, and provides forecast levels at 14 locations in the estuary. The performance of forecasts for the barrier is regularly reviewed, together with horizon scanning of research developments which may assist in the future; for example, increased use of data assimilation, ensemble surge forecasting, and use of artificial intelligence techniques (such as artificial neural networks).

8.4.3

Tidal Barriers

Tidal barriers and barrages have many similarities to river control structures, with the added complication of a tidal influence further downstream. In flooding applications, they are usually used to prevent river or estuary levels rising above flooding threshold levels due to high tidal levels, including situations when river flows are also high. Other applications include hydropower generation, and amenity use (e.g. marinas, harbours etc.). The operating rules for a tidal barrier typically guide the opening and closing of the barrier gates according to a range of combinations of upstream river levels (or flows), and tidal levels downstream. Control rules may be presented in the form of charts, look up tables, or encapsulated within a computer-based decision support system, possibly combined with real time forecasting models. Rules are often developed using a combination of experience and detailed hydrodynamic and other modelling for a large number of scenarios. The design of a tidal barrier needs to allow for the possibility of river levels upstream of the barrier rising to flooding levels during a river flood event whilst the barrier is closed to provide protection from coastal flooding. The rules may allow for this accumulated river flow to be released during low tide periods, even when a major coastal event is in progress. The need for river flow releases will depend on the magnitude of the storage upstream of the barrier in river channels and estuaries, compared to the flow volumes likely to accumulate during a tidal cycle.

8.5 Urban Drainage

199

Although barriers can often be operated on the basis of observed levels alone, the use of river flow and coastal forecasts can extend the lead time available to operators, allowing more time for mobilisation of staff, and closure of the barrier (if the time taken is a constraint). Shipping operators can also be warned of possible barrier closures (if relevant) and local authorities of any potential for flooding, if this cannot be entirely mitigated by the barrier. However, it is often desirable to minimise the number of closures per year, leading to an interesting optimisation problem; for example, balancing the need to provide flood protection against the cost of operating the barrier (staffing, power etc.), the impact on maintenance and whole-life costs, interruptions to shipping and other river users, and other factors. An ensemble or probabilistic approach may help in optimising the decision and weighing up the costs and risks involved. Some examples of tidal barriers include the Maeslant barrier near Rotterdam in the Netherlands, which provides protection against surge events in the North Sea, the Thames Barrier in London (Box 8.2) and the St Petersburg Flood Protection Barrier, which is situated in a low lying area where the Neva River meets the Gulf of Finland. The St Petersburg barrier is due for completion in around 2009–2010 and a decision support system using observed data and meteorological and hydrodynamic forecasts is under development to assist with barrier operations (Villars et al. 2007).

8.5

Urban Drainage

Urban drainage systems typically consist of a network of pipes, open channels and culverts draining into sewers which carry flood flows to wastewater treatment works or river or sea outfalls. Combined systems may be used for foul water (sewerage) and surface runoff, or the flows may be handled by separate networks. Systems may also include flood detention ponds, storm storage tanks and other types of storage. Urban flood events tend to be characterised by a rapid response, with complicating factors from debris and blockages, and restrictions on flood flows at major obstacles on the floodplain (road or rail embankments etc.). The percentage of rainfall which appears as runoff may be high due to the impermeable nature of some surfaces, such as roads, car parks and pedestrian areas. Urban flooding mechanisms may include surface runoff before water enters the drainage system (pluvial flooding), outflows from combined sewer systems and from the drainage network where flows exceed capacity (at manholes, for example), flows developing along paths of least resistance (e.g. roads), and ponding where normal drainage paths are blocked. River flooding can also occur in urban areas, although the response may be complicated by factors such as culverts and bridges, and may lead to additional drainage related flooding (for example, where high river levels impede drainage into the river network). River flows typically respond on a longer timescale than urban runoff, so may occur some time after the local urban response, although this will depend on the timing and distribution of rainfall in the catchment both within and upstream of the urban area.

200

8 Selected Applications

For planning and design, hydraulic models are widely used and can model drainage networks in great detail, including pipes, pumps, valves, flood detention areas, and other features. In principle, models of this type could also be used in real time to forecast surface flooding such as surcharging, outflows at outfalls, and other factors, although this is rarely done at present. Modelling can be at district, street or property level, and may include models for surface runoff into the drainage network, although identification of drainage routes and contributing areas (drainage catchments) can be difficult in urban areas, and sensitive to small changes in elevation (e.g. at roadside kerbs). The hydrological inputs to such models usually assume idealised design storms whereas, for real time use, details of storm speed, intensity and distribution are required, and would ideally be available at a high resolution, comparable to typical drainage catchment areas. Storm direction is of particular interest, since the load on the drainage network can be higher if rainfall moves down the network, as runoff in the later stages of the event adds to water already in the system from higher elevation areas. Options for monitoring rainfall in urban areas include raingauges and weather radar. For raingauges, the network density required is usually higher than for river catchment monitoring, and may not be feasible on cost grounds, although dense networks are operated by some authorities; for example, Harris County, Texas operates approximately 100 real time raingauges within the boundaries of Houston. Weather radar can potentially provide more information on the spatial distribution of rainfall although, as noted in Chapter 2, performance will depend on distance from the nearest radar, and the accuracy in urban areas may be affected by tall buildings, masts and other factors. Cheaper, low cost radar, and microwave techniques, are also other possibilities for monitoring rainfall in urban areas. Information on antecedent conditions may also be required, with runoff characteristics defined individually, or in aggregate, for a wide range of features, including roofs, roads, gardens, fields, and features in sustainable drainage systems (SUDS), together with gullies, culverts and other local drainage routes. Geographical Information Systems combined with Digital Terrain Models offer the possibility of modelling at this level of detail, although obtaining calibration data and parameters for all combinations of conditions can be difficult. The requirements for urban flood forecasting models have many similarities to river forecasting techniques, but in addition to rainfall runoff and flow routing components, there may be a need for models for the drainage network and other local effects, such as temporary flow paths along roads and other open areas. River and urban models can also be coupled dynamically, allowing the interactions between systems to be represented (e.g. of river levels on drainage, or river spill into a drainage system). However, although modern computing systems have the capability to run very detailed models, even in real time, in practice there is usually a need to simplify and conceptualise models to focus on locations and factors which have the most influence on flooding, perhaps linked to surface water flooding maps. In addition to providing information to guide warnings of potential flooding, models can also be used for adaptive or predictive real time control of urban

8.5 Urban Drainage

201

drainage networks (e.g. Cluckie et al. 1998; Vitasovic 2006). For flood events, the objective is typically to reduce flooding and pollution incidents by making use of spare capacity in the drainage network by operating pumps, gates, control valves, weirs etc., or temporarily diverting water to storage areas. Potential flooding may therefore be reduced or avoided in flood prone locations, whilst some wider objectives in normal operation can include reducing pumping costs, optimising the performance of water treatment plant through providing forecasts of future water and pollution loads, and decision support if problems arise with parts of the network. Schilling (1989) and Schütze et al. (2004) present reviews of approaches to the real time control for urban drainage systems, whilst Table 8.5 presents some examples of systems with a flood control component which have been used operationally or in research studies. The key components in this type of system include monitoring equipment (and related telemetry), automated or remotely controlled actuators (on valves, gates etc.), and a defined real time control strategy or set of rules. Some options for developing control rules include: ● ● ●

Heuristic techniques – based on experience and trial and error Off-line simulation – to develop a pre-defined set of rules Real time predictive control – optimising performance in real time

A range of optimisation criteria might be considered, or multi-criteria approaches used. Current research themes in real time control for urban drainage systems have many similarities to the more general themes in river and coastal flood forecasting which have been described in earlier chapters, and include: ●

Objectives – better definition of control objectives with respect to national and international legislation, and the roles and responsibilities of organisations with an interest in, or responsibility for, urban flooding

Table 8.5 Examples of studies into real time control systems incorporating flood control aspects Location

Key features

Reference

Haute-Sûre Reservoir, Luxembourg

Raingauge and radar inputs to sewer model

Henry et al. (2005)

Quebec urban community Multi-objective optimisation to minimise RTC, Canada overflows, maximise treatment plant use, minimise accumulated volumes etc.

Schutze et al. (2004)

Brays Bayou, Houston, USA

Weather radar inputs to a fine mesh processbased rainfall runoff model

Vieux et al. (2005)

URBAS project, Germany

Improved radar and 2D modelling techniques for urban flood forecasting with case studies for 15 German municipalities

Castro et al. (2006)

202 ●

●

●

●

8 Selected Applications

Monitoring – improved techniques, particularly for weather radar in urban areas, automated fault detection and diagnosis (including blockages), and devices for monitoring surface runoff (more reliable, lower cost, lower visual impact etc.) Modelling – improved ways of modelling surface runoff in urban areas, including higher resolution models, automated generation of flow paths from digital terrain and urban land use data, and alternative approaches such as artificial neural networks Probabilistic methods – using ensemble rainfall forecasts, stochastically generated scenarios etc., and risk based approaches to decision making (combining probability and consequence) Control – improved optimisation techniques, allowance for human factors, improved information gathering and display systems

8.6

Geotechnical Risks

Geotechnics is a general expression for a range of specialist areas including the study of natural hazards such as earthquakes, landslides, mudflows and avalanches, and of the performance of structures where they interact with the earth surface. Table 8.6 illustrates some potential causes of flooding to people and property from geotechnical risks. In flood forecasting applications, the extent to which these various types of event can be modelled will depend on how well the underlying cause can be monitored and predicted, including its influence on river, reservoir or coastal levels. Interactions between flooding mechanisms may also need to be considered. Within an overall flood forecasting model, additional sub-models can be developed for these components, or purpose made models developed specifically for these types of risk. A common theme in this type of event is that estimation of the forcing component requires an understanding of geotechnical processes, and that

Table 8.6 Some potential causes of flooding from geotechnical risks General category

Type

Examples

Structural risks

Dam break Failure of river or coastal defences Glacial lake outburst floods/ jökulhlaups, landslide dam outburst floods

Can occur to any dam or defence if overtopped, poorly maintained or designed, damaged by earthquake or landslide etc.

Tsunami

December 2004 Tsunami

Landslides or debris flows into rivers or reservoirs

Western USA (particularly after wildfires), Central Asia and the Caucasus

Earth movements

A risk in some mountain areas e.g. the Himalaya, Canada, Iceland

8.6 Geotechnical Risks

203

the flooding impact can potentially be extensive and extreme, and may require hydrodynamic modelling techniques to predict. Some other types of flood event which might be included in this category (although this is debatable) include: ●

●

Groundwater flooding – due to raised water levels in aquifers, causing increased flows at springs and in rivers, and sometimes additional flow routes developing that are not usually observed (e.g. springs appearing within properties) Blockage risks – flooding caused by the accumulation of debris (trees, vegetation etc.) at structures such as bridges, culverts and trash screens

For groundwater flooding, there are well developed hydrodynamic modelling techniques for simulation of aquifers, and these can in principle be applied in near real time if sufficient real time input data are available on borehole levels, river flows, rainfall and snow cover, pumping operations etc. Given the often slow nature of groundwater flooding, a model run frequency of once per day or less may be sufficient; the problem being to obtain enough up to date data to initialise the model. Other simpler correlation, data based and conceptual or process-based rainfall runoff modelling techniques may also be used (see Chapter 6). For blockage risks, although these can pose a considerable hazard, they are intrinsically difficult to forecast, unless there are known problem locations at which debris usually tends to accumulate, or up to date information can be obtained on blockage locations (e.g. from at site river level instrumentation, or observations by CCTV or webcam). This information can then be input into a flow routing or hydrodynamic model, possibly trying various scenarios for the extent of blockage, and for the discharge coefficient at the structure. Simpler threshold based techniques may also be used (see Chapter 3).

8.6.1

Structural Risks

Structural risks include dam break, defence (levee) breaches or failure, and failure of naturally occurring moraine in mountain regions. This type of event can cause extensive and rapid flooding, possibly with depths, velocities and extents beyond any previous experience. Some examples have included the failures of two major dams and dozens of smaller dams in Henan Province in China during a typhoon event in August 1975, in which about 85,000 people died, and millions were displaced, and the failure of the Vaiont Dam in Northern Italy in October 1963, due to the wave caused by a landslide into the lake, in which about 2,000 people died (Graham 2000). The levee failures during and following Hurricane Katrina in August 2005 also inundated large parts of New Orleans. The mechanisms for dam and defence failures can include: ●

Piping/seepage – flow routes developing through or under the structure, causing progressive erosion and eventual failure

204 ●

● ●

8 Selected Applications

Overtopping or wave action – eroding the crest of the structure in places, again with progressive increases in erosion and flows Liquefaction – for earth structures, for example Earth movements – earthquakes and landslides causing direct damage to structures

For flood defences (or levees or dikes), failure modes can depend on the materials used for construction (earth, rockfill, concrete etc.), the hydraulic loading, and the geometry of the structure. The hydraulic loading can arise from water level differences across a structure, wave action/overtopping, and flows along the length of the structure. Many sub categories and secondary influences can also be identified. As part of contingency arrangements, particularly for dam failure, some organisations maintain maps of likely flood extents, and emergency plans detailing the actions which should be taken if a failure occurs. During a flood event, rapid flood mapping exercises might also be commissioned if there are concerns about the integrity of a structure. Detailed monitoring can also be performed of the structure using in-situ (e.g. accelerometers, piezometers) and remote techniques (e.g. webcams, laser scanning). Many studies have been performed into simulation (off-line) modelling of structural failures. Models for the breach component can range from simple weir equations assuming a fixed breach size, through to fully dynamic process-based simulations in which the hydraulic and soil erosion processes are coupled, and modelled in two or three dimensions. These types of technique could potentially be applied in real time for scenario modelling if the location of the breach (or likely location) is known; however, forecasting likely locations remains a research area at present. For example, the probabilistic risk assessment techniques developed for flood risk modelling (see Chapter 1), which combine information on probability of levels (loading) and defence fragility curves, might be used to provide guidance on potential breach locations. Multiple breach scenarios are also calculated off-line to provide one of the sources of information available to emergency managers in the Netherlands, for example (see Chapter 10). Model calibration is complicated by the lack of reliable calibration data, due to the rarity and unpredictability of occurrence of this type of event, although a number of large-scale experiments have been performed to gain an insight into typical parameter values (e.g. IMPACT 2005). For debris blockages in rivers from landslides, added complications include the irregular, and probably unknown, nature of the materials making up the barrier. For Glacial Lake Outburst Floods and Landslide Dam Outburst Floods, the conditions leading up to failure of the debris/moraine/ice dam can often be recognised, although the timing is difficult to predict, and may be influenced by volcanic and earthquake activity (e.g. Snorrason et al. 2000). For dam breaks, in addition to breach models, flood propagation models are usually also required for flows downstream, and are usually of the types described in Chapter 6 for river flows. However, some added complications may include high sediment loads, the need to model structures and obstacles which are normally above extreme river flood levels, and uncertainty about the appropriate values to use for some model parameters (e.g. roughness coefficients) and for likely flow paths.

8.6 Geotechnical Risks

8.6.2

205

Earth Movements

Earthquakes, landslides, volcanic and other events can affect river and coastal conditions either indirectly (by damage to structures, for example; see the previous section), or directly, by generation of flood waves, such as Tsunami in coastal waters, and by causing surge waves, flow blockages, and an increase in debris and sediment content in river waters. For rivers, debris or mud flows often result from heavy rainfall, and are in principle predictable using similar techniques to those used for flash flooding. The ‘flows’ consist mainly of water, soil, rocks, slurry and other debris (e.g. Mirtskhoulava 2000). Operational forecasting systems typically use rainfall depth duration thresholds based on observed rainfall, and sometimes forecast rainfall, and these methods have been reported for Puerto Rico, Hong Kong, Taiwan, and locations in Italy and the USA, for example (e.g. NOAA-USGS 2005). A prototype Debris Flow Warning System for the western USA is also described using a rainfall threshold approach combined with predictive debris-flow-volume models, but with a long term development path towards distributed near real time process based models combining digital terrain models, rainfall distributions, and models for rainfall infiltration, slope stability, and channel bed erosion and deposition (e.g. Rickenmann and Chen 2003). By contrast, in the oceans, perhaps the main hazard is from the waves generated by subsea earthquakes, volcanic activity or landslides, which are usually known as Tsunami. In addition to the catastrophic December 2004 event in the Indian Ocean, other major Tsunami in recent decades have included the 1993 Hokkaido Tsunami in Japan, which reached heights of 5–10 m, and the 1992 Flores Tsunami in Indonesia, with heights of about 4–7 m (Satake 2000). In the open ocean, Tsunami waves typically travel at speeds of about 500– 1,000 km per hour, depending on ocean depth, with a wave length of about 5–10 km or more (and sometimes hundreds of kilometres). Here, they pose little threat since wave magnitudes are small, and typically only of the order of 1 m. The destructive effect of a Tsunami arises in shallow coastal waters, when raised water levels, combined with the large volumes of water involved, can cause extensive flooding inland. The modelling techniques for Tsunami are essentially those already described in Chapter 7 for surge propagation. For open ocean propagation, the non-linear shallow water equations provide a reasonable approximation to wave motion due to the long wavelength, which is comparable to typical ocean depths. However, techniques are less well developed for modelling the initial wave formation, and the run up process at shorelines, and high grid resolutions, and three dimensional modeling techniques, are possibly required to obtain accurate estimates (and remain an active area of research). Tsunami monitoring and forecasting is performed both nationally and internationally; for example, in Japan, and at the Pacific Tsunami Warning Centre in Hawaii. The December 2004 Tsunami has also led to major efforts to install and upgrade warning systems. Forecasts rely both on ocean modelling, and detection of

206

8 Selected Applications

seismic activity. Since the speed of seismic waves is about an order of magnitude more than for Tsunami waves, seismic detectors can also be used to trigger an increased state of monitoring and mobilisation, if the conditions appear likely to have caused an event. For example, in some locations, such as the Pacific Ocean, the time for a Tsunami to cross the ocean is several hours, so forecasts can be based on sea level observations as well (e.g. Satake 2000). However, locally generated events can occur only a few minutes after the initial seismic event so that is the main indicator of likely flooding.

Chapter 9

Preparedness

As with other types of natural hazard, the effectiveness of response to a flood event can be improved if an emergency plan has already been prepared, so that all participants understand their roles and responsibilities, including the overall chain of command. The potential disruption from flooding also needs to be considered, including the possibility of communication, instrumentation, computer and other systems failing, and access and evacuation routes being cut by flood water. Risk assessment techniques are also increasingly used to assess the resilience of response procedures, together with developments in information technology for the spatial analysis and visualisation of flood extent relative to properties, infrastructure and transport routes. This chapter provides an introduction to these issues, and discusses the general trend towards multi-hazard systems, which share systems and resources across many types of threat.

9.1 9.1.1

Flood Emergency Planning General Principles

Flood Emergency Plans describe the actions to take between, during and following flood events, and typically cover operational procedures, emergency response assets, contact details for key staff, health and safety issues, procedures for liaison with the media and the public, and information on safe access and evacuation routes and shelters. Some guidelines on developing flood emergency plans include US Army Corps of Engineers (1996), NOAA/NWS (1997) and Emergency Management Australia (1999) for river flooding, and Holland (2007) for tropical cyclone forecasting. Depending on the type of flooding, lead time available, and population affected, examples of actions which may need to be taken in the run up to and during a flood event include (USACE 1996): ● ● ●

Providing search, rescue, and evacuation services Scheduling closure of schools and transportation of students Curtailing electric and gas service to prevent fire and explosions

K. Sene, Flood Warning, Forecasting and Emergency Response, © Springer Science + Business Media B.V. 2008

209

210 ●

● ● ● ●

● ● ●

9 Preparedness

Establishing traffic controls to facilitate evacuation and prevent inadvertent travel into hazardous areas Dispersing fire and rescue services for continued protection Establishing emergency medical services and shelters Closing levee openings Moving public and private vehicles and equipment from areas subject to flooding Relocating or stacking contents of private structures Initiating flood-fighting efforts (e.g. sandbagging etc.) Establishing security to prevent looting

For tropical cyclones (e.g. World Meteorological Organisation 2006a; Holland 2007), at about 36–48 hours from landfall, activities can include fishing boats returning to home ports, setting up emergency operation centres, preliminary precautions being taken by residents, and starting to provide 6 hourly updates. Then, within about 24–36 hours from landfall, evacuation of exposed properties begins, businesses and industry commence shutdown procedures, and updates move to 3 hourly intervals. Most preparations should be complete within about 6–8 hours of landfall, leaving the emergency services to check for any remaining vulnerable people and secure community lifelines. Warnings continue for about 12 hours after landfall, including for river flooding from heavy rainfall, and any changes in cyclone strength or track. Information on roles and responsibilities, and inter-agency coordination, is often an important consideration in developing flood emergency plans, since a river or coastal flooding event can cut across administrative and political boundaries, and may affect more than one country, so many organisations may be involved in the response. The organisational structure and terminology used varies widely between countries, but Table 9.1 illustrates in general terms some of the organisations which may be involved in responding to a flood event, and some typical roles and responsibilities. In emergency planning guides, some key points which are often emphasised include: ●

●

●

●

●

Command – a clear chain of command is needed to avoid duplication of effort and missing vital actions, and for communicating information to the public and the media. Key Contacts – regular communication between agencies through exercises and training, and personal contacts, pays dividends under the pressure of a major event. Resilience – plans should be sufficiently adaptable to cope with extreme events beyond those in recent memory, with contingency plans in case of failure of any component. Vulnerable Groups – special arrangements are often needed for people with physical disabilities or a medical condition, the elderly (in some cases), and others who are dependent, such as children. Transient Populations – plans need to allow for temporary residents of areas at risk, including tourists, business travellers, road users, student populations, and others.

Table 9.1 Organisations which may be involved in responding to a flood event Type

Group/department etc.

Environmental, Scientific

Meteorological Service Flood Warning Service (if separate)

Emergency Services/Civil Protection

National, State, Local Authorities

Communities

Utilities

Transport Operators

Private Sector Voluntary Sector

Media

Typical roles and responsibilities

Weather forecasting and monitoring; river and coastal monitoring and forecasting; flood warning dissemination; and possibly operating flow control structures, flood defence repairs and reinforcement etc. Police Road closures; evacuation of properties; dissemination of flood warnings (in some countries); law enforcement Social and Health Treatment and evacuation of injured people; services precautionary evacuation of locations such as nursing homes and hospitals; providing food, water, clothing, counselling etc. Fire & Rescue Rescues from flooded properties and elsewhere; pumping of flooded water; fire fighting Emergency Planners, Overall coordination of activities; establishing Disaster Managers, evacuation centers; arranging distribution of others sandbags; temporary defences; levee repairs and reinforcement; financial assistance etc. Highways, Contractors Road closures; temporary works; debris removal Army, Navy, Air Force, Rescue operations; provision of specialist Coastguard equipment and additional staff resources, boats, hovercraft, helicopters, trucks, amphibious vehicles etc. Local representatives Coordinating community response for floodfighting and evacuation; obtaining relief funding and additional resources (people, equipment etc.) Community members Assisting with evacuation; issuing warnings to neighbours/friends etc.; temporary measures to reduce flooding and protect property etc. Water Installation of temporary defences at treatment works; switching supplies between works (if possible); provision of temporary water supplies Electricity and Gas Installation of temporary defences at substations, gas works and power stations; precautionary close down and rerouting of power supplies; safety inspections and advice; drawdown of hydropower reservoirs Telecommunications Installation of temporary defences at communication hubs; rerouting of networks if needed Canal/Port operators, Operating canal and other gates; reinforcing Navigation Authorities defences; assisting ship and boat owners; closing tidal barriers Rail, Road Inspection or closure of flooded rail lines; provision of buses, trucks etc. to help in evacuation and recovery Airports Closure of flooded runways, diversion of flights Contractors, vehicle Provision of transport, emergency repair equipment, owners etc. additional staff, fuel, food etc. Charities, Relief Assisting in own specialist areas (the aged, Organisations, others the young, animals, food and drink, clothing, counselling etc.) Television, radio, other Reporting, relaying information and warnings

212 ●

●

●

9 Preparedness

Community Engagement – it is important to involve community members (or their representatives) in developing emergency plans, and ensuring that these are tailored to their own needs and resources. Health and Safety – apart from the risk of drowning, flood waters present a range of other risks to emergency workers, including waterborne diseases, hazardous materials (fuel, chemicals, sewage etc.), and electrocution from exposed cables, so strict procedures, safety equipment and decontamination facilities are ideally required. Continuous Improvement – an ongoing process of validation, testing, review and updates is important, particularly to account for changes in staff, organisational structures, suppliers, contractors etc.

Social, political, and cultural differences between countries can lead to a wide range of approaches to command and control structures, but often national disaster managers, local authorities, the police or the military will assume the role of overall coordinators, depending on the scale of the event. In a major event, regional and national command centres may also be established to ensure that resources, funds and equipment are made available to flooded areas. The resources available for flood response can include extra staff and private sector contractors (e.g. for emergency repairs and debris removal), whilst equipment can include vehicles, medical supplies, earthmovers, pumps, generators, boats, sandbags, and mobile communications. Procedures also need to be established for accepting assistance from third parties. For example, in rapidly flowing water, swift water rescue skills are needed, and offers of assistance may need to be refused. The establishment of a national system of response and rescue competencies (sometimes called ‘team typing’), provides one way of ensuring that volunteers and others have the necessary training, equipment and procedures to avoid becoming victims themselves Community engagement can cover both education on actions to take when receiving warnings, and direct inputs to emergency plans from community members or their representatives. Much of the literature on early warning systems (e.g. Handmer 2002; World Meteorological Organisation 2006b; ISDR 2006; United Nations 2006) emphasises a community based or people-centred approach, in which those at risk are engaged in the planning process, helping to decide on the most effective forms of response, and the best way to present and receive warning messages. Four key elements to consider (ISDR 2006) include: ●

●

●

Risk Knowledge – Are the hazards and the vulnerabilities well known? What are the patterns and trends in these factors? Are risk maps and data widely available? Monitoring & Warning Service – Are the right parameters being monitored? Is there a sound scientific basis for making forecasts? Can accurate and timely warnings be generated? Dissemination and Communication – Do warnings reach all of those at risk? Are the risks and warnings understood? Is the warning information clear and useable?

9.1 Flood Emergency Planning ●

213

Response Capability – Are response plans up to date and tested? Are local capacities and knowledge made use of? Are people prepared and ready to react to warnings?

Cultural issues relating to gender, income, language and other factors also need to be considered, such as informal developments on floodplains, with plans tailored to meet the needs of specific groups (e.g. do people have livestock or pets which they may be reluctant to leave behind). In particular, this may increase the success rate of evacuating people during a flood, with failure to leave property leading to the emergency services needing to engage in time consuming and potentially high risk rescues at later stages in the event. Chapter 4 discusses some of the techniques which can be used for raising public awareness of flooding, and of the actions to take on receiving flood warnings, including leaflet campaigns, seminars, meetings, radio and television appearances, open days, flood fairs, school visits, and educational films. A distinction can also be made between plans at individual (or family) level, community level, and at site specific, organisational, local, regional or national level. For example, individuals can develop (perhaps with assistance) personal action plans describing actions to take in the event of flooding, including consideration of locations at risk from flooding, how to protect property, checklists of what to take if evacuated (food and water, medical supplies, phones, radios, food, flashlights etc.), key contacts, information on where to go in an emergency, and the safest escape routes (e.g. FEMA 1997). Other items might include batteries, blankets, protective gloves, disinfectant, a first aid kit, personal documents, insurance policies, and switching off gas, water and electricity. In the Netherlands, for example, residents can view neighbourhood flood emergency response plans and maps on a website by entering their property location. At community/village level (World Meteorological Organisation 2006a), plans might include: ● ● ●

●

●

● ● ●

Identifying and maintaining of safe havens, safe areas and temporary shelters Putting up signs on routes or alternate routes leading to safe shelters Informing the public of the location of safe areas and the shortest routes leading to them Having all important contacts ready: district or provincial and national emergency lines; and having a focal point in the village Making arrangements for the damage and needs assessment team and health team Setting up community volunteer teams for a 24-hour flood watch Improving or keeping communication channels open to disseminate warnings Distributing the information throughout the community

In rural areas, cattle, poultry and other livestock might also be moved to safety and supplies of firewood, drinking water and animal folder secured. The Stormwatch programme in the USA is another example at community level, and a community can be designated as StormReady (FEMA 2005) for weather hazards by completing the following steps:

214 ● ●

● ● ●

9 Preparedness

Establish a 24-hour warning point and emergency operations center Have more than one way to receive severe weather warnings and forecasts and to alert the public Create a system that monitors weather conditions locally Promote the importance of public readiness through community seminars Develop a formal hazardous weather plan, which includes training severe weather spotters and holding emergency exercises

Site specific plans may discuss particular hazards and needs at individual locations (e.g. water treatment works, power stations) whilst, at local authority, regional and national level, flood emergency plans usually also cover the actions needed in the aftermath of an event (the recovery phase), including which organisation(s) will assume responsibility for repairs, debris removal, reuniting families, emergency funding arrangements, and providing shelter, food, water, medical care, counselling, support to businesses, and restoration of services if interrupted (power, water, communications etc.). However, these topics fall outside the scope of this book and are not considered further, except for the topic of post event analysis of flood warning performance, which is discussed in Chapter 11.

9.1.2

Risk Assessments

One key stage in developing a flood emergency plan is to assess the flood risk, and to tailor the response to the level of risk. Flood risk can be expressed in terms of the probability of flooding and the likely impacts (e.g. the number of people or properties at risk, or economic value, or the combination of exposure and vulnerability). Here vulnerability concerns not just the threat of flooding, but the ability of people to cope with a flood event, including mobility, age, type of residence, awareness of flooding, and access to transport and flood resilience measures. Chapter 1 describes techniques for estimating the likely extent of inundation, which can include compiling historical observations, using local knowledge, and performing hydrodynamic modelling studies of various levels of complexity. If available, the resulting inundation maps can be overlain on street maps and satellite or aerial photographs, perhaps combined with terrain models based on digital elevation datasets. Additional potential sources of flooding may also need to be considered, including flood defences which are known to be in poor condition, locations where debris or ice jams may cause raised water levels, and dam breaches. In addition to identifying properties at risk, high-risk locations such as hospitals, schools and nursing homes may also need to be considered, and specific procedures developed for these locations within the overall plan. For a hospital, for example, this might include defining the criteria for evacuation, and the additional resources which health workers would require in moving patients to a place of safety (vehicles, people, equipment etc.). For some situations, there can be a difficult balance of risks to consider, each requiring a separate risk assessment; for example, for

9.1 Flood Emergency Planning

215

Fig. 9.1 Illustrative example of an evacuation sequence

nursing homes and hospitals, evacuating patients may cause injury or distress, and this needs to be balanced against the risks from flooding. Specific assessments may also be needed for other high-risk locations such as power stations, water treatment works, prisons, high-rise buildings, underground car parks, major roads, airports and rail stations, shopping centers and subway systems (e.g. Ishigaki et al. 2006), industrial plant, and sites with hazardous materials. Evacuation sequences can also be prepared as shown in the simple example in Fig. 9.1. The figure shows a hypothetical example of a flood warning area with five zones or districts defined. An indication is provided of the numbers of properties which need to be evacuated as river levels rise, and the time available before the onset of flooding (which is shown by grey shading). Links are provided to a map at each river or tidal level showing the estimated extent of flooding at that level, the properties affected, and safe evacuation routes. During a flood event, plans would of course need to be adapted as the situation unfolds, and real time decision support systems are increasingly being considered for use in updating emergency plans in real time as described in Chapter 10. As noted in Chapters 3 and 5, several factors need to be considered when estimating the actual time available for emergency actions, including the time delay between observations being made and being ready to use (polling time), the various time delays in performing forecasting model runs, the decision times needed by operational staff, and the time required to issue warnings to all recipients. For example, for tropical cyclone forecasting, the time delays up to the time of issuing a warning (i.e. excluding the dissemination time) include the time taken for observational data to arrive at the Tropical Cyclone Warning Centre, the time taken for data to be processed and then presented to the forecaster, forecaster analysis, assessment and prediction time, and the time needed for message preparation (Holland 2007). Estimates such as these can guide response strategies, and help to gain a better idea of the trade off between delaying a decision (hoping for better information), and reducing the time available for emergency actions to be taken, such as evacuating properties. This point is discussed further in Section 10.3. Perhaps the most highly developed approach to evacuation planning is that used in the USA for hurricane evacuations (see Box 9.1). For example, during Hurricane Katrina in 2005, it is estimated that some one million people left their properties. Evacuation plans are typically developed as part of Hurricane Evacuation Studies

216

9 Preparedness

Box 9.1 Hurricane evacuation procedures Hurricanes, typhoons and tropical cyclones can cause widespread damage, and evacuation is one of the main options for reducing risk to coastal populations. Flood risks arise from storm surge and heavy rainfall, with the added complication of high winds and the possibility of tornadoes. However, the decision on whether to issue an evacuation order is a difficult balance between the need to protect people, and to avoid the economic and other impacts of false alarms. Most states in the USA follow similar procedures as a hurricane approaches land, although the details and terminology used can differ (e.g. Smith and Ward 1998; Wolshon et al. 2001, 2005a). A typical five stage alert might initiate the following actions: ● ●

●

●

●

Level V – represents normal, routine operation. Level IV – is a preliminary state of alert which is activated if a tropical storm is developing. A team is formed to monitor developments and report to key government officials, emergency responders, and the Federal Emergency Management Agency (FEMA). Level III – is activated if a hurricane strike appears likely in the state, with actions to clear evacuation routes of obstructions, monitor traffic volumes, and to liaise with the National Guard and officials in neighbouring states. Level II – is the stage at which information on evacuation and shelters is issued to the public, and a Declaration of Emergency is requested. Level I – is the highest state of alert, at which recommendations to issue the order or recommendation to evacuate are made to local authorities, and arrangements are made (if required) to evacuate vulnerable people; for example in nursing homes, hospitals, and prisons.

The timing of events varies but typically the move to Level IV might be up to 1 week ahead of the estimated time of landfall, Level III 3 or more days ahead, Level II 2–3 days ahead, and Level I within 1 day of landfall. Minimum lead times required for evacuation can be in the range 9–72 hours ahead of landfall depending on the category of hurricane and road network and population densities, but are typically in the range 12–24 hours. Evacuation routes are typically closed shortly before the hurricane makes landfall so that any residents, media representatives etc. remaining can be directed to local shelters or refuges of last resort. Evacuation orders can be voluntary, recommended or mandatory. Voluntary orders are used for offshore workers and others who require long lead times to move, whilst recommended orders are issued if there is a high probability of a hurricane causing a threat to people in at risk areas. Mandatory orders are only used in some states. Public awareness campaigns between events play an important role in improving the effectiveness of evacuation orders when a hurricane occurs, with the aim to maximise the number of people at risk moving to shelters, and minimise the number of so-called shadow evacuations, where people move although they may not be directly at risk.

9.1 Flood Emergency Planning

217

(USACE 1995), initiated since the 1980s by the Federal Emergency Management Agency (FEMA), and available to download from the internet. These studies typically incorporate a hazard analysis, a vulnerability analysis, an evacuee behavioural analysis, a sheltering analysis, and a transportation analysis, guided by estimates for maximum inland water levels from the National Weather Service (see Chapter 7), and including plans to evacuate vulnerable groups.

9.1.3

All-Hazard Approaches

Flood emergency plans are often developed as part of an all-hazard (or multi-hazard) approach to emergency planning, and the methods used may be formalised in national legislation. A multi-hazard approach brings economies of scale, sustainability and efficiency, and opportunities for more frequent use and testing than for single-hazard systems (e.g. ISDR 2006). For example, in the United States, the Federal Emergency Management Agency (FEMA http://www.fema.gov/) uses a standard approach called the National Incident Management System for incident management for all hazards and across all levels of government. The six main components are: ●

●

●

●

●

●

Command and Management – covering Incident Command Systems, MultiAgency Coordination Systems, and Public Information Systems Preparedness – covering planning, training, exercises, personal qualification and certification, equipment acquisition and certification, mutual aid and publications management Resource Management – covering standardized mechanisms and establishing requirements for processes to describe, inventory, mobilise, dispatch, track and recover resources over the life cycle of an incident Communications and Information Management – covering Incident Management Communications and Information Management Supporting Technologies – including voice and data communications systems, information management systems, and data display systems Ongoing Management and Maintenance – covering routine review and continuous refinement over the long term

Emphasis is placed on a flexible, adaptable approach, which can be easily adopted and understood by a wide range of types and scale of organisation, including governmental, private sector and voluntary sector organisations. FEMA also produces the national flood maps used in flood risk assessments and emergency planning, typically using hydraulic modelling approaches to assess inundation extent. Some other examples of generic approaches to incident management include those developed by the French Organisation de la Réponse de Securité Civile (plan ORSEC: Direction de la Défense et de la Sécurité Civiles 2006), Emergency Management Australia (1999), the United Nations/International Strategy for Disaster Reduction (ISDR 2006) and the Civil Contingencies Secretariat in the UK (see Box 9.2).

218

9 Preparedness

Box 9.2 United Kingdom – Civil Contingencies Act (2004) The Civil Contingencies Act defines two classes of responder to emergency situations which, depending on the situation, can include: ●

●

Category 1 responders – emergency services, local authorities, Health Bodies, Environment Agency/Scottish Environment Protection Agency (SEPA) Category 2 responders – Health and Safety executive, transport operators (rail, road, airport, harbour, underground), utilities (electricity, gas, water, telecommunications), Strategic Health Authorities

Category 1 responders are typically first on the scene of an event, and take responsibility for managing the crisis, and include the two main organisations responsible for issuing flood warnings (the Environment Agency and SEPA). The main civil protection duties for Category 1 responders are as follows (HM Government 2005): ● ● ● ●

● ●

Risk assessment Business continuity management (BCM) Emergency planning Maintaining public awareness and arrangements to warn, inform and advise the public Cooperation and information sharing Provision of advice and assistance to the commercial sector and voluntary organisations (Local Authorities only)

Collaboration is also facilitated by a system of local and regional Resilience Forums, which meet regularly to review emergency plans for a range of hazards, including flooding, where appropriate. At a national level, control is coordinated by the Civil Contingencies Committee which convenes as required to deal with national and regional scale emergencies, including river and coastal flooding. Other mechanisms for collaboration include visits, seminars, joint projects and exercises, via bilateral and multilateral groups and forums. The Act distinguishes between generic plans, covering a range of hazards, and specific plans, to deal with particular kinds of emergency. Plans can be prepared for single organisations or across groups of organisations (MultiAgency plans). The minimum level of information to be contained in a specific plan is (HM Government 2005): ● ●

●

●

Aim of the plan, including links with the plans of other responders Information about the specific hazard or contingency or site for which the plan has been prepared Trigger for activation of the plan, including alert and standby procedures Activation procedures (continued)

9.1 Flood Emergency Planning

219

Box 9.2 (continued) ● Identification and roles of multi-agency strategic (gold) and tactical (silver) teams ● Identification of lead responsibilities of different responder organisations at different stages of the response ● Identification of roles of each responder organisation ● Location of joint operations centre from which emergency will be managed ● Stand-down procedures ● Annex: contact details of key personnel and partner agencies ● Plan validation (exercises) schedule ● Training schedule In addition to the needs of victims and the safety of emergency workers, particular emphasis is given to the needs of vulnerable groups, who may require special assistance during an incident. Crown copyright material is reproduced with the permission of the Controller of HMSO and the Queen’s Printer for Scotland

9.1.4

Validation and Testing of Plans

Plans usually also consider arrangements for validation, testing and regular reviews. Emergency response exercises are often office-based (e.g. table top exercises), and may occur in ‘accelerated’ time. Typically, the exercise coordinator will introduce complications as the event unfolds, including escalating the situation, and introducing issues needing immediate response. Public relations skills may also be tested, including simulated questions from the public, media, and politicians. Larger scale exercises may also include simulated television and radio news broadcasts, and computer-based simulations of the types of output which would be available in a real event, such as forecasting model outputs, and weather radar displays. As an example of the complexity of these exercises, Exercise TRITON held in the UK in 2004 simulated an extreme flood event in England and Wales, and involved some 1,000 representatives from more than 60 organisations and agencies, based at 35 locations, over a period of 3 days. The exercise explored the effectiveness of interagency coordination, command structures, response assets etc. during both the response and recovery phases of the event. In the USA, the Hurricane Pam exercise in 2004 was based around a fictional Category 3–4 hurricane, with sustained winds of 120 mph, a major storm surge, and up to 20 in. of rainfall in places (US House of Representatives 2006). The aim was to help officials develop joint response plans for a catastrophic hurricane in Louisiana. The exercise was performed over four workshops, and the first workshop, held in July 2004, involved some 300 emergency officials from 50 parish, state,

220

9 Preparedness

federal, and volunteer organizations over a period of 8 days. The topics considered included search and rescue operations, temporary shelters, medical care, debris management, and other factors, including dealing with hazardous materials, power, water and ice distribution, and drainage of flood waters. The exercise was performed as part of a wider initiative by the Federal Emergency Management Agency (FEMA) in conducting catastrophic disaster planning, and was the first in a series of 25 disaster scenarios to be considered (prioritised on the basis of risk). The lessons learned from the exercise helped with the response to Hurricane Katrina during 2005 although, as with any exercise, some additional factors occurred during that event which were not anticipated, and work was still in progress in implementing some of the lessons learned from the exercise at the time of the event.

9.2 9.2.1

Resilience Introduction

There is an increasing trend towards viewing flood emergency management as an exercise in risk management and reduction. Except in certain cases, such as closing a flood gate, or raising a temporary barrier, flood warnings can only reduce the adverse impacts of flooding on people and property so, during a flood event, the task is therefore to minimise the risks as effectively and safely as possible. One key aspect of risk management is to assess the likelihood of failure, and some common sources of problems in flood incident response include: ● ● ● ● ● ● ● ●

Equipment – a shortage or breakdown of vehicles, high volume pumps, boats etc. People – key staff out of contact, insufficient training, insufficient people Power – interruptions to electricity or gas supplies and telecommunications Fixed Communications – damage to land lines and telecommunications hubs Mobile Communications – interference, damage to towers, poor signal strength Drinking Water – flooding of treatment works, contamination by flood waters Medical Care – shortages of medical staff and equipment if many casualties Fuel – shortages due to garages being flooded or inaccessible, oil supplies being disrupted

Many of these factors are interdependent, with electricity failures in particular potentially affecting both fixed and mobile communications, operations at water treatment works, and hospitals and other key infrastructure. Flooding can also be exacerbated by problems at river control structures, tidal gates, temporary barriers, and other control structures and contingency plans need to be in place for emergency repairs or manual operation of structures. Communications failures can also seriously hamper rescue operations. For example, during Hurricane Katrina, in August 2005, damage to radio and cell phone communication towers, telephone switching hubs, power supplies, and other

9.2 Resilience

221

problems, resulted in the loss of three million telephone lines in the states of Louisiana, Mississippi and Alabama, 38 emergency phone call centres, some police and fire department communication networks, up to 2,000 cell phone sites, and 37 out of 41 radio stations in New Orleans and surrounding areas. Backup generators for radio networks also failed in places due to flooding, or the difficulty in refuelling them. Satellite phones, either hand-held or in Mobile Emergency Response Support vehicles, remained the only reliable form of communication in some locations. Critical infrastructure affected included the New Orleans Police Department headquarters, and six of the eight police districts’ buildings (US House of Representatives 2006). To give an indication of the range of technical and communication problems which can occur, Table 9.2 illustrates some issues for a number of flood events in

Table 9.2 Illustration of some of the technical and communications difficulties which can occur during flood events Topic

Point of failure

Control centres/ emergency operations centres Rescue centres, shelters

Communications, power or access, or the site itself, may be affected by flood waters or traffic hold ups Sites may be at risk from flooding, or inaccessible to evacuees and staff and to vehicles bringing food, water and clothing Access to locations expected to flood may be affected by flood waters or the weight of traffic Roads may be blocked by floodwater or the weight of traffic; fuel may be inaccessible or in short supply

Access routes

Evacuation routes

Utilities

Flooding may interrupt power, water, and gas supplies, including locations distant from the areas affected by flooding

Warning dissemination systems

Failure of telephones, cell phones, sirens, blocked access for loud hailer routes, overloading of websites and phone based warning services Systems may not be compatible

Communications between organisations

Some examples of possible contingency arrangements One or more backup locations in areas away from flooding; permanently relocate if the flood risk is significant Choose locations for rescue centres away from flood waters and with good access even in flooding conditions Pre-position key staff, vehicles, and equipment, install temporary barriers (if used) before routes are cut off Plan a phased evacuation, possibly assisted by computer modelling of scenarios. Make arrangements for access to boats, helicopters, hovercraft, amphibious vehicles etc. Have backup supply arrangements planned as far as possible, and alternatives such as bottled water, clothing etc. (or evacuate areas at risk). Consider flood proofing or relocation Develop alternative warning dissemination arrangements which do not rely on mains power or access to specific locations; perform load testing for extreme call volumes etc. Investigate the frequencies and systems used by other organisations, particularly for radio communications, and agree on interagency communication arrangements. (continued)

222

9 Preparedness

Table 9.2 (continued) Topic Communication failures

Point of failure

Some examples of possible contingency arrangements

Flooding of key telecommunications hubs or of the power supplies to cell phone and radio networks Poor signal strength on site for cell phones and two way radio

Arrange back up communication methods, and floodproofing or relocation of key communication infrastructure Mobile Investigate the coverage in flood communication prone areas, arrange alternatives and problems backups if necessary Media Television and radio broadcasts Have additional warning communications may fail due to power problems dissemination arrangements in place Widespread Regional flooding may also affect Develop links with organisations located further afield (bilaterally, flooding organisations normally relied upon to assist at a local level or through national coordination), with the possibility of international assistance Telemetry Key instruments or telemetry links Decide which alternative instruments may fail so that information on would be used in case of failures, meteorological, river or coastal install additional instruments and conditions is telemetry links as a backup, and raise not available electrical equipment above likely peak water levels Hazardous material Flood waters may affect sewage Procure protective equipment for staff treatment works, and contain (dry suits etc.), and develop standard fuel, chemicals, animal carcasses, decontamination procedures, develop human waste and other toxins policies for advising the public, and media on risks Equipment available Limited availability of high Explore the availability within other volume pumps, communication organisations and nationally, with equipment, sandbags, temporary arrangements in place for requesting barriers, boats, vehicles etc. assistance if needed; also, stockpiling of equipment at strategic locations Staff welfare Health and Safety issues from Review training, equipment, duty floods, waterborne diseases and rotas, with backup arrangements for hazardous materials, wind, heavy staff etc. rain; potential issues with hours on duty, friends and family affected by flooding, access to home, site or office in a flood event

Europe and the USA in recent years, based on various sources referred to in this and other chapters, together with some simple examples of possible contingency arrangements. In addition to evacuation of properties, and mobilising staff, some preparatory actions which can be taken by the emergency services and others if warnings are received in time include:

9.2 Resilience

223

Dam

Railway Major road

Police patrol

Village

Town

Pre-position 2 crews

Town

Town Power Station

Farms Caravan Park

Precautionary evacuation

Discuss contingency plans with duty manager

Chemical Factory Town

Flood risk map outline

Fig. 9.2 Simple illustration of operational response early in a developing river flood event

●

● ●

●

Installation of temporary or demountable barriers around key infrastructure, or rerouting of power, water, communications links etc. Controlled shutdown of industrial processes Pre-positioning of emergency response assets in locations likely to be cut off by flood water Stockpiling of food, water, fuel, sandbags and other supplies at strategic locations

For example, for the hypothetical flood warning scenarios presented in Fig. 1.5, a flood risk mapping exercise might show that, in a heavy rainfall event, the main areas at risk include the main town, power station, motorway, caravan park, and isolated farms. An initial assessment (Fig. 9.2) might suggest a range of options in the early stages of the event before flooding occurs, including placing residents of the caravan park on standby to evacuate the site, assessing the readiness of the power station operator to install temporary defences, or switch to alternate power supplies, placing a police patrol at the main road bridge ready to close the road if flooding looks likely, and positioning a number of search and rescue crews, with mobile communications, food and water supplies, in the town in case access routes are cut. The timing of actions would depend on the probability of flooding, and levels of risk (which, for simplicity, are not shown on the figure). Increasingly, spatial analysis tools are being used for this type of assessment, and are described in Section 9.3.

224

9.2.2

9 Preparedness

Analysis Techniques

Many other problems can occur in the chain of detection, forecasting, warning, dissemination and response, and the analysis of possible causes of failure, and devising alternative plans, is often called contingency planning or business continuity management. The initial assessment of risks is usually performed based on experience and lessons learned from previous flood events, and many analyses may proceed no further than this stage. However, more formal techniques are widely used in other sectors, such as the chemical, aviation and oil and gas industries, and the aim is usually to consider the following questions (e.g. Federal Aviation Authority 2000): ● ● ● ● ●

What can fail? How it can fail? How frequently will it fail? What are the effects of the failure? How important, from a safety viewpoint, are the effects of the failure?

Some widely used methods include: ●

●

●

●

Failure Mode, Effects and Criticality (FMECA) Analysis – which systematically considers the potential modes of failure of each component or system, the likely impacts, the chances of detecting each type of failure, and the overall risks presented by each scenario Probabilistic Risk Assessments – which combine the probabilities and consequences of various scenarios to estimate the risk from each combination, and are often performed using a Fault or Event Tree analysis, or Monte Carlo simulations Fault or Event Tree Analysis – in which a graphical representation is produced of the chain of faults or events which can lead to a given outcome, typically linked by logical AND or OR gates. The probabilities of occurrence for each individual component in the chain can be combined to estimate the overall probability for each combination of circumstances Scenario Modeling – using computer simulation and numerical modeling techniques to investigate alternative scenarios (which in a flooding context might include factors such as flood defence breaches, pump failures, failures at river control structures etc.)

Analyses may start from a single event, and explore all outcomes, or consider a single outcome, and explore what faults or events could realistically have led to that problem. These are known as top down or bottom up approaches. Systems may be probabilistically safe, inherently safe, failsafe, or fault tolerant. For example, a probabilistically safe system has enough redundancy (computers, instruments etc.) that a failure is unlikely to cause problems, whilst a fault tolerant system can continue to operate if problems occur, although not as effectively. Thus an analysis of flood emergency response would need to consider all likely causes of failure (‘weak links’) and critical paths, including problems due to equipment, communications, and human factors, either for individual subsystems, or for the entire

9.2 Resilience

225

Fig. 9.3 Representation of the flood incident management process for a site at Pill near Bristol (Environment Agency 2006, © Environment Agency copyright and/or database right 2008. All rights reserved)

system. The analyses could be used as a simple screening tool, to help to explore potential sources of failure, or taken further to derive quantitative estimates of risk. These types of analysis are little used at present in flood incident management, but are being developed by some authorities. To take a simple example, in probabilistic risk assessment the risk presented by a pump failure could be estimated as the probability of failure per event, multiplied by the number of events in the period being considered (e.g. a year), and the probability of this leading to property flooding, to give an overall risk score. Figure 9.3 shows a slightly more complex example for a tidal sluice gate. The gate needs to be closed during exceptional tides to protect the village of Pill in southwest England from flooding, and this example was developed to illustrate the potential use of Bayesian networks in risk assessments for flood incident management. For individual components in the chain of people, equipment and systems the situation being considered may be reasonably self contained and amenable to analysis (and these techniques may already be used by suppliers of some forms of emergency response equipment, for example). However, when considering wider issues, there are numerous interacting factors, both technical and non-technical, which can affect

226

9 Preparedness

how the response to a flood event unfolds. For example, the response and expectations of individuals may need to be considered; including the extent to which people will help neighbours and the emergency services, or may take risks trying to rescue animals or valuable possessions. Also, whether people will agree to leave their homes when issued with a warning. For analysis of systems with high levels of uncertainty, and complex interactions, other more qualitative methods are available in which user views on probability (e.g. ‘quite high’) and consequence (e.g. ‘serious’) are translated into a form which can be analysed. Techniques include fuzzy set logic and linguistic reasoning in which scoring and ranking systems are used to quantify risks, based on interviews and expert opinion (e.g. Environment Agency 2007). Other approaches, such as Bayesian Networks, Artificial Neural Networks, and Agent Based Modelling, might also be considered. The focus of the analysis could be at a range of levels, including at strategic, operational or tactical level, covering multiple organisations and stakeholders, or covering just individual organisations, or single locations.

9.3 9.3.1

Role of Information Technology Introduction

Information technology plays an increasingly important role in the management of flood emergencies, and several examples of the components within a typical integrated flood warning and response system have already been discussed, including: ●

●

●

●

Flood Risk Assessments – hydrodynamic modelling to derive quantitative estimates of flood risk, combining probability and consequence (Chapter 1) Telemetry Systems – to manage the collection of real time data for river levels, rainfall, tide levels, and other parameters (Chapter 2) Dissemination Systems – automated systems for sending flood warnings to the public and first responders, by telephone, email and other methods (Chapter 4) Flood Forecasting Systems – to operate real time rainfall runoff, flow routing, hydrodynamic, surge, wave and other forecasting models (Chapter 5)

In the area of emergency response, some additional types of system which are increasingly being used include: ●

●

●

Geographical Information Systems – which can display and analyse information on flood risk, emergency assets etc. Decision Support Systems – which can provide guidance on optimum response strategies when developing emergency response plans, and in real time (see Chapter 10) Simulators – which can be used in training exercises and to help to develop emergency response plans

9.3 Role of Information Technology

227

For all applications, if systems are integrated into operational procedures, then some important factors to consider include resilience, staff training, data security, data quality, metadata, and access to confidential information. Also, for interagency systems, there are many issues to consider concerning funding, terminology, standards, and the interoperability of systems. Adoption of formal risk assessments, and software design and acceptance testing procedures, is essential.

9.3.2

Geographical Information Systems

Geographical Information Systems (GIS) are increasingly used to assist both with planning for emergencies, and in the recovery phase (e.g. MacFarlane 2005; van Oosterom et al. 2005). Information can be combined from a wide range of sources and presented in a consistent and intuitive way. In flooding applications, some information of interest can include rivers, coastlines, flood hazard maps, census data, administrative boundaries, roads, streets, topography, critical installations, industrial hazards, commercial properties, utility data, and other features, as well as information specific to the emergency response (e.g. control rooms, evacuation shelters, medical facilities). Searches can also be performed for specific sectors of the community who may require specific assistance during an event (e.g. people with medical conditions, or who speak a foreign language). Sources of information can include local authorities, the emergency services, central government, and the private sector. To combine datasets from many different sources, the location and extent of each feature is needed. For records which are not already geographically referenced, information on street addresses, post codes, or road names or may be sufficient to generate the required datasets. The functionality available in a modern system typically includes: ●

●

●

●

●

●

●

Pan, zoom and overlay of map layers and images (e.g. aerial or satellite photographs) Other presentation options (e.g. transparent layers, three dimensional views, animations, graphs, reports) Overlay, Neighbourhood and Buffering analyses (e.g. all properties which lie within a flood risk area, or within a given distance of a river) Boolean analysis allowing complex searches to be performed (e.g. the operating bases for all helicopters with the required load capacity within 20 minutes flying time of a site) Network analyses (e.g. for travel times of emergency response vehicles from one location to another) Terrain analysis (e.g. for assessing the line of sight of radio and cell phone communications to and from a flood prone area) Data modelling using the outputs from models to generate new datasets (e.g. flood hazard maps)

228

9 Preparedness

Systems can be operated on a desktop computer, or made available over a corporate intranet, and increasingly selected results are published on the internet (e.g. the flood hazard maps in the UK and USA). Web-based GIS can also be used to facilitate the exchange of information between different organisations involved in emergency response, although may lack the full functionality of a desktop system. Most systems can also link to an external database, containing additional information on specific features, such as key data, images or video clips. For example, the US FEMA National Incident Management System IRIS database, launched in 2007, provides the facility to store detailed information on approximately 120 types of emergency response asset, including equipment, communications, contracts, facilities, responders, services, supplies and teams. The system allows emergency responders to inventory, categorize type, locate, request, and track all internal and external resources from a single integrated system to help facilitate the response of requested resources during an incident (FEMA 2007). Geographical Information Systems and Decision Support Systems can also assist with the response during a flood event and this topic is discussed in Chapter 10.

9.3.3

Visualisation and Simulation

Visualisation and simulation tools are increasingly used to present complex information in an intuitive way to non-specialists, and for training exercises and assisting in developing emergency response plans. In flood emergency response applications, two examples of interest are: ●

●

Flood Maps – three dimensional views and animations of flood extent against a backdrop of topography, buildings, mountains etc. Simulators – virtual reality effects incorporated into systems used for emergency response exercises, and in decision support systems

This is an active area of research (e.g. Pajorova et al. 2007), and practical applications are already in use in other emergency response applications; for example, in training search and rescue teams to deal with major fires and nuclear incidents. Figure 9.4 shows an example of a virtual reality representation of flooding in a residential area produced by the Virtual Environmental Planning Project (http:// www.veps3d.org). River levels are shown in bank and for a major flood event. The flood scenario is hypothetical, and extreme, but illustrates how flood extents can be viewed in context. The software also allows users to animate the flood sequence, and to view the scene from different directions, and at different magnifications. More generally, search and rescue services are increasingly using simulators for training exercises or to test emergency response plans, often building on techniques developed for computer games, and these techniques also have potential

9.3 Role of Information Technology

229

Fig. 9.4 Virtual reality representation of flooding (Copyright 2007 VEPS)

for flood response applications. Some typical functionality (e.g. Gyorfi et al. 2006) includes: ●

●

●

●

Virtual Reality – the trainer and students can move around and interact with a virtual representation of the control room or incident, including buildings, rescue equipment, vehicles, personnel etc. Simulations can be replayed as part of the training exercise. Multimedia – video links are provided to actual or simulated personnel for interviews, consultation etc., simulated television news, and synthetic or actual radio and cell phone discussions. Networking – the facility is available for multiple participants to join the simulation over the internet, and to influence the course of events. Artificial Intelligence – is used to encapsulate and represent the behaviour of other participants at the scene of the event (the public, casualties, emergency services etc.).

The nature of the simulations can be tailored to national incident protocols and linked into national standards. For example, the Incident Commander training simulator software developed for the National Institute of Justice, which is the research arm of the U.S. Department of Justice, can be used by emergency managers in testing response, and training, to hurricane, flood and other small to medium scale incidents (up to 500,000 residents). The severe storm recovery component provides the option to deal with many types of problem, included obstructed roads, casualties, gas leaks, and downed trees blocking roads.

Chapter 10

Response

Flood warnings provide local authorities, the emergency services, the public and others with time to take actions to reduce the risk from flooding, and information on the likely extent and locations of flooding. Actions which can be taken before a flood starts include installation of temporary defences, operation of flow control structures, protection of personal property, evacuation of people from areas at risk, and positioning of emergency vehicles and other assets in locations which may become inaccessible due to flood waters. Increasingly, decision support systems are also being developed to assist in responding to flood events, and can provide advice on strategies for evacuating property, casualty management, and emergency repairs to flood defences. This chapter considers these issues, together with the more general topic of dealing with uncertainty in decision making during flood events.

10.1 10.1.1

Flood Event Management Preparatory Actions

One of the key benefits of flood warnings is that, if time permits, a number of actions can be taken in advance of flooding to reduce the extent of damage to property and the risk to life. The key stages in issuing a flood warning have been described in earlier chapters and include: ● ● ● ●

Detection – of meteorological, river and coastal conditions (Chapter 2) Thresholds – identification of conditions likely to cause flooding (Chapter 3) Dissemination – issuing of flood warnings (Chapter 4) Forecasting – prediction of future river and coastal conditions (Chapters 5–8)

If sufficient time is available, warnings are normally escalated from an initial advisory that flooding is possible, through to a full flood warning. The receipt of an advisory or warning is usually the trigger to activate a flood emergency plan (if one exists), and to commence preparatory actions. As described

K. Sene, Flood Warning, Forecasting and Emergency Response, © Springer Science + Business Media B.V. 2008

231

232

10 Response

Fig. 10.1 Federal Emergency Management Agency (FEMA) Incident Command Centre at a flood event; Kingfisher, Oklohoma, August 20, 2007 (FEMA/Marvin Nauman)

in Chapter 9, these can include mobilisation of staff and equipment, establishing a command centre, and issuing of initial warnings to the media and public. Pre-positioning of people and equipment can also be started, particularly if it is thought likely that access routes may be cut off by floodwater. For example, emergency services increasingly use temporary or mobile command centers (e.g. Fig. 10.1) and these need to be placed in locations suitable for local communications, and clear of any likely flooding or access problems. Various actions can also be taken to reduce or even prevent flooding, including: ●

●

●

Emergency works – reinforcement of weak spots in flood defences, and at locations where existing river or coastal works are underway (and patrols to inspect defences and other structures), clearance of drains and blocked watercourses Temporary defences – raising temporary or demountable barriers, placing sandbags along flood defences and river banks, and at individual properties Flow control operations – diversion of river flows, closing (or opening) gates, emergency draw-down of reservoirs etc.

The time available to prepare depends on the type of flood event. For the case of flash floods from rainfall, ice jams and landslides, only a few hours at best may be available. However, for events such as floods in the lower reaches of large rivers, or storm surge, a day or more of warning may be possible. For tropical cyclones, typhoons and hurricanes, sometimes up to a week of advance warning can be provided, although with considerable uncertainty about the location, timing and severity of the event at that lead time.

10.1 Flood Event Management

233

As noted in Chapters 5 and 9, the actual time available is reduced by various time delays in the detection, forecasting and warning process, and these delays should be considered in planning actions as the event develops, and in particular any ‘points of no return’ beyond which a course of action (e.g. evacuation) becomes unfeasible. In some types of event, the numbers of people to consider can be large; for example, during Hurricane Katrina in the USA in 2005, more than one million people are estimated to have left their properties in advance of landfall and, during 1995 in the Netherlands, a precautionary evacuation of more than 200,000 people took place in advance of a forecast flood event (although the subsequent flooding was not as serious as expected). The issue of false alarms also needs to be considered, which can lead to unnecessary costs (e.g. closing down businesses or industrial plant), and can also incur a risk (e.g. if hospital patients are moved). These considerations of optimum decision making, and flood warning performance, are discussed further in Section 10.2 and Chapter 11. Another factor to consider is the cooperation likely to be received from people in flood affected areas; for example, will people leave their properties if asked to, what assistance can people provide in helping neighbours and the emergency services, and will people interpret warnings correctly? These questions often cannot be resolved during the pressure of a flood event, highlighting the important role that community engagement and awareness has to play in maximising the effectiveness of emergency response (see Chapter 9). In particular, studies have shown that the format and wording of messages is crucial and this topic is discussed further in Chapters 4 and 11. During the onset of a flood event, several actions can also usefully be taken (if resources are available) to help with the subsequent post event analysis including: ●

●

●

Photography – taking photographs and videos of the flood extent from the ground, and from helicopters and aircraft (if available) River gauging – making spot measurements of high flows to help in calibrating river monitoring equipment (see Chapter 2) Flood inundation sensors – reading staff gauges, noting maximum levels, and sometimes installing equipment (e.g. maximum level recorders) to record the flood depths reached (see Chapter 2)

Many hydrological services routinely perform tasks of this type during flood events, with the staff requirements and procedures written into warning procedures to ensure that they are not overlooked. Similarly, decisions to pre-position people and equipment in locations likely to be cut off by flooding are much easier to take if the requirements have been identified and agreed in advance. For example, there may be a need to preposition a high volume pump or fire truck at a location expected to flood severely, even though minor flooding is already occurring at another location where the equipment is also needed. If a risk-based assessment has already been performed at the planning stage, then this helps to avoid local conflict when the decision needs to be taken.

234

10 Response

Much of the focus in the run up to the onset of flooding is also on avoiding the need to rescue people, pets or livestock. Compared to other alternatives, such as evacuation, or installing temporary barriers, rescues can be risky and time consuming, and limit the resources available to respond to other problems as they arise. For example, in the UK, procedures can require a team of five to six people to rescue one person (or more) trapped in fast flowing river water, including one person upstream to spot debris flowing towards the rescue site, two people downstream to spot the casualty if they are swept away by water, and two to three people on site for the rescue itself (e.g. by boat or rope). Similarly, rescues from vehicles can be hazardous, and survival rates are much lower than for people trapped in property, so closing roads early to avoid the situation arising is the preferred option. Once flooding has started, and flood warnings have been issued, the role of flood warning and forecasting starts to assume less importance, although forecasting models (see Chapters 5–8) can continue to assist throughout the event in answering questions on the likely magnitude and timing of the peak, flooding extent, and when waters are likely to drop, and roads can be reopened and people allowed to return to their properties.

10.1.2

Timelines

One widely used concept in emergency response is that of the event timeline. This describes the sequence of incidents, emergency calls, responses, actions etc. which occur during an event. Timelines are used in post event assessments of response, and may also be available in real time to assist other responders in understanding the situation. Modern incident management and decision support systems can automatically log occurrences from a wide range of sources and organisations, and make this information available to all responders over secure websites (see Section 10.2). As an illustration of a timeline, Fig. 10.2 provides a hypothetical example, from the perspective of a flood warning service, for a short lived flash flooding event affecting a small town called Newtown during the early hours of the morning. The roles, staff job titles, actions etc. will vary widely between organisations and specific flooding events so this example is just for illustration. The sequence would also continue into post event actions, reporting, assisting residents to return to properties and assess and repair damage, reopening roads, debriefs etc. Also, this event proceeds predictably with no problems arising such as flow control structures not operating as expected, or problems being encountered with contacting key people, or access to equipment etc. As another example, Box 10.1 shows the actual timeline for a flash flooding event, with a focus on the emergency response (adapted from North Cornwall District Council 2004). The event occurred in the village of Boscastle in South West England in the summer of 2004.

02:00

02:45

02:55

03:00

03:15 03:20 03:25 03:30 04:10

04:15 04:15 04:30 04:35

04:50

05:10 05:40

06:10 06:40

07:30

The 2 am routine discussion with the duty weather forecaster suggests a general risk of heavy rainfall in the region, although it is difficult to be specific on locations at this stage Flash warning received of a 70% probability of heavy rainfall over the Newtown catchment in the next 1–2 hours. Discussions with the duty weather forecaster suggest that a major rainfall event is possible in the catchment Duty officer completes reviews of weather forecast, weather radar, raingauge, river level and flood forecasting model outputs (including two model runs for different rainfall scenarios) Duty officer issues a Flood Watch advisory message, incident room opened, duty and operations managers informed by telephone at home First heavy rainfall recorded by a raingauge in the catchment Duty manager arrives at the incident room Briefing of duty manager completed; discussions with the town emergency planning officer. Operations manager arrives. Emergency workforce instructed to deploy to Newtown Flood warning threshold levels reached; flood warning formally issued to the properties at risk, the local authority and police; loud hailer patrol started Emergency workforce completes closing of the two flood gates in Newtown Hydrometry team deployed to Newtown gauge for high flow spot gaugings On-site briefing between local authority and police representatives, and the emergency workforce Flood incident declared; police control centre established, designated evacuation centre opened, social service and voluntary staff called onto site Duty weather forecaster informs flood warning duty officer that rainfall should stop in the catchment within the next 20–30 minutes Completion of evacuation of the 125 properties likely to be affected; nursing home evacuated Area likely to be affected cleared of all people and vehicles; main access roads closed to the public. Sandbagging of properties completed where feasible Flood waters start to overtop river banks Flood levels reach a maximum depth of 1 m; 25 properties affected; hydrometry team completes highest flow gauging achieved to date at Newtown gauge Flood levels falling rapidly at Newtown gauge; local authority advised that the flood threat is receeding

Fig. 10.2 Illustration of a flood event timeline up to the time that flood levels start to drop from peak values

236

10 Response

Box 10.1 Boscastle flood event 2004 The village of Boscastle is on the north coast of Cornwall at the bottom of the 23 km2 catchment area of the rivers Valency and Jordan. On 16 August 2004, in mid summer, the village experienced its worst flooding on record. The peak hourly rainfall in the area exceeded 80 mm at one raingauge, with a 24 hour total of about 200 mm. Swift action by local people and the emergency services meant that no lives were lost and there was only one minor injury. Approximately 1,000 residents and visitors were affected by the event, with about 100 people rescued by helicopter from rooftops, cars and trees, 58 properties flooded, and 116 cars swept out to sea. Roads, sewers, bridges and other infrastructure were badly damaged (Fig. 10.3).

Fig. 10.3 Helicopter rescue in main street of Boscastle (Pam Durrant; text based on a description by Heulyn Lewis, North Cornwall District Council)

Some key actions in the timeline for the event were: 1215

Reports of heavy rainfall in the upper catchment but none in the middle or lower reaches (continued)

10.2 Decision Support Systems

237

Box 10.1 (continued) 1230 Heavy rainfall starts in the middle reaches of catchment 1415 Rain still persists but seems to be easing 1500 Heavy persistent rain starts again 1515 River Valency approaching bankfull flow 1530 River starts to spill over bank 1535 First call received by fire and rescue service 1545 Visitor car park in village starts to flood 1546 Call to coastguard from local representative saying that the river has risen about 2 m in the past hour 1600 1–3 m wall of water sweeps across visitor car park trapping people in the visitor centre 1603 Helicopter rescue coordination centre put on ‘standby’ 1617 Inshore lifeboat launched 1630 All access roads closed by police 1636 Air ambulance put on standby 1700 Flood approaching peak levels 1712 Major incident declared by emergency services 1723 First helicopter winch rescue completed 1755 Two regional hospitals put on standby 2000 Water levels back within river banks 2100 Helicopters start returning to base

10.2

Decision Support Systems

Information is a key requirement during flood events, particularly widespread events, and computer systems are increasingly being used to assist in guiding the response. Some particular characteristics of decision making in emergencies (MacFarlane 2005) include uncertainty, complexity, time pressure, a dynamic event that is innately unpredictable, information and communication problems (overload, paucity or ambiguity), and the heightened levels of stress for participants, coupled with potential personal danger, whilst some general advantages of using Geographical Information Systems (GIS) during emergency incidents include: ● ●

●

● ● ●

Support for tasking and briefing Producing hard copy maps which remain a key information product for responders and planners Integrating data from multiple sources that may flow in during the course of an emergency Developing a Common Operational Picture for multi-agency staff Supporting two way flows of information through mobile GIS Assessing likely consequences and supporting forward planning

238 ● ●

● ●

10 Response

Managing assets and resources for current and projected future demands Keeping the public and other affected parties informed through internet or intranet mapping systems Establishing one element of an audit trail Supporting the transition to recovery with a baseline database that also integrates a full picture of the emergency itself

Here, mobile-GIS is a term used to describe hand-held Personal Digital Assistant (PDA) or laptop devices able to display maps, possibly linked to real time data feeds showing the present locations of key assets. Vehicle and personnel tracking systems using Global Positioning System (GPS) devices can also be used to relay information on, for example, the positions of rescue helicopters, and of fire and police vehicles and other assets. This approach is used by some fire service mobile command centres, for example. Systems may be designed specifically for flooding, or form part of a wider allhazards approach. This includes the concept of Virtual Emergency Operations Centres for major events, making extensive use of interactive GIS displays for incident scenes, video links, emergency response assets (pumps, generators etc.), satellite and other imagery, with automated sharing of information between vehicles, personnel and command centres (e.g. Prendergast 2007) Compared to the flood risk mapping applications described in Chapter 2, and the planning role described in Chapter 9, there are of course a number of issues to consider if relying on computerised systems for information and situational awareness during an event, including the availability of trained operators, resilience to communications and power failures, and the likely availability of sufficient information in real time. It may also be useful to develop interfaces to other systems, such as telemetry systems and flood forecasting models, to provide updates on current and future estimates for flood extent. Some examples of real time use of GIS in flood related applications include: ●

●

●

The Shelter Navigation System – a map based system which allows authorised staff to monitor the status of designated shelters during a hurricane, including keeping track of evacuees, and which allows the public to look up the shelters closest to their home (South Carolina Department of Health and Environmental Control). Surrey Alert – a map based system for sharing information during emergencies of all types, including flooding, in the county of Surrey in the UK. The system consists of a secure extranet, accessible only to local authorities, the emergency services and other responders, and a publicly available website, which includes news, media and travel updates, and the option to show flood warnings in force on a map. The extranet component includes an emergency contacts database, information on key facilities (medical, control centers/emergency operations centres, rest centres etc.), and an incident log, giving times and descriptions for actions and decisions taken during the event (http://www.surreyalert.info). Flood Forecasting Systems – modern forecasting systems (see Chapter 5) also increasingly include the facility to map flood inundation in real time, including

10.2 Decision Support Systems

239

running ‘what if’ scenarios (e.g. for defence breaches, and structure operations) and sometimes the option to generate lists of properties at risk for output to paper based or automated warning dissemination systems, thereby having many of the characteristics of a GIS system. Whilst Geographical Information Systems might be viewed as a simple type of decision support system, more sophisticated systems are available which provide guidance on actions to take during an emergency, and these can be used both at the planning stage and during an event. The level of guidance provided can range from presentation of information in a range of formats (overlays, visualisation etc.) through to recommendations on optimum response strategies. These types of system are used in the oil and gas, chemical and nuclear industries, for example, and often combine the outputs from sophisticated computer models (e.g. of gas dispersion) with optimisation algorithms or logical rules (for example, IF the spill is ACID and the ACID is in the GASEOUS state THEN…..), and links to external databases on equipment characteristics and histories, personnel records, emergency checklists and other information. Recent developments in this area include the use of probabilistic techniques, and artificial intelligence methods, such as artificial neural networks, fuzzy logic and genetic algorithms. Internet based applications are also being investigated, allowing decision support/expert systems to ‘learn’ and update rules from the vast amount of information available on the world wide web. Several systems of this type are also under development for use in flood emergency applications, although their use is not yet widespread. The base data (static data) can include information on: ● ● ● ● ● ●

● ● ● ● ● ● ●

Flood defence locations and geometry River control structures Topography Property and census information Locations of vulnerable people who may require assistance Critical assets such as power stations, water treatment works, telecommunication hubs, and associated supply/communication networks Transient populations such as at campsites, caravan parks, hotels Emergency equipment depots (sandbags, earthmovers etc.) Fixed emergency response assets (fire and police stations, command centres etc.) Medical facilities (doctors, pharmacies, hospitals etc.) Emergency shelters and safe areas for escape Road network characteristics (evacuation and access routes, capacities etc.) Background mapping (rivers, coastlines etc.)

whilst dynamic data, updated during a flood event, can include: ● ● ● ●

Water depths, velocities and flows Flood defence condition including the locations and size of breaches River control structure settings Temporary defences (barriers, sandbags etc.)

240 ● ●

10 Response

Locations of vehicles, personnel and other resources (e.g. helicopters, boats) Traffic density and flow data

Systems may include data gathering and modelling components, or import this information from external sources, such as flood forecasting models, telemetry systems, and dynamic traffic flow models. Dynamic scenarios have the advantage of making use of the latest information, with the potential disadvantages of lack of real time information due to communications failures, model uncertainties, and the risk of models failing to operate correctly, or outputs being misinterpreted by non-experts. Alternatively, stored scenarios may be used, and have the advantage that many more options can be considered than would be feasible during a flood event (e.g. for defence/levee/dike breaches, time of day or day of week, or traffic routing), possibly also using more sophisticated modelling approaches (e.g. two-dimensional rather than one-dimensional flood models). Scenarios can also be reviewed and audited before use in a real event although, as in many aspects of flood modelling, with the difficulty of a lack of calibration and validation data for the more extreme events (e.g. mass evacuations, widespread flooding). Perhaps the most widespread application which has been considered to date has been for evacuation planning and management. The aim of evacuation modules is to provide guidance on how to optimise the number of people who leave an area in a given time, taking account of the likely delays due to weight of traffic, and routes and safe havens becoming inaccessible due to flood water, together with the access requirements of the emergency services. For example, in the USA, systems which are available (e.g. Fu 2004; Wolshon et al. 2005) include: ●

●

Hurrevac – a GIS based software package developed on behalf of FEMA by the US Army Corps of Engineers that has been used since 1988 by government emergency managers to help with managing major evacuations during hurricanes, and includes import of surge scenarios from the National Weather Service, a shelter module, a traffic and evacuation tool, the facility for what-if scenarios for hurricane track, winds etc., and the option to display data from river and tide gauges. The system also allows hypothetical storms to be input for use in training exercises (http://www.hurrevac.com). Evacuation Traffic Information System (ETIS) – a web-based hurricane evacuation travel demand forecast model for which inputs include the category of hurricane, tourist occupancies, and anticipated percentages of people leaving property and arriving in given locations. Outputs include estimates for traffic volumes, cross-state flows of traffic, locations for congestion, and the numbers of vehicles arriving at specific locations (http://www.fhwaetis.com).

More generally, dynamic approaches to traffic management are increasingly being explored, including use of contraflow systems and intelligent transport systems (e.g. Wolshon et al. 2005). In other applications, Rodrigues et al. (2006) describe a prototype internet based decision support tool called DamAid to assist emergency managers with evacuation of properties and emergency response in case of dam failure, whilst Simonovic and Ahmad (2005) describe a prototype computer simulation module for flood evacuation planning for areas at risk from flooding in the Red River basin in Canada, combining social, psychological, policy and other factors.

10.2 Decision Support Systems

241

To allow for applications other than evacuation, increasingly a modular approach is being adopted to development of decision support systems, including modules for advising on evacuation routes, responses to defence breaches, likely casualties, search and rescue strategies, management of the recovery phase (e.g. return to properties), and emergency response (e.g. road closures, placing sandbags, reinforcing flood defences). For example, Van der Vaat et al. (2007) propose a methodological framework for flood event management decision support systems consisting of hazard, exposure, vulnerability, consequence and risk modules, with inputs from management response, external driver and tools modules. Systems may also use web-based displays of data, with dynamic generation and updating of emergency response plans, and the facility to view plans at a range of scales (national, local, site specific e.g. power stations). The relative importance attached to the evacuation component depends to some extent on the scale and rate of development of a flood event; for example, in a widespread slowly developing coastal event, the focus may be on evacuation whilst, in a flash flood, the focus may be on optimum access routes for the emergency services, search and rescue operations, and management of the recovery phase. Table 10.1 summarises several examples of research and operational projects which have developed (or are developing) decision support tools for flood event management.

Table 10.1 Examples of research and operational decision support tools for flood event management Features or Project Country applications Reference ANFAS

DAMAID FLIWAS

GDH NISFCDR OSIRIS

China, France, Slovakia, Greece, UK Portugal The Netherlands, Germany, Ireland The Netherlands China

PACTES

France, Poland, Germany France

PREVIEW

Europe-wide

RAMFLOOD

Spain, Germany, Greece and three others China

Web GIS

Flood forecasting component

Prastacos et al. (2004)

Dam break forecasts See Box 10.2

Rodrigues et al. (2006) See Box 10.2

Stored scenarios Flood forecasting component Scenarios

Flikweert et al. (2007) Huaimin (2005)

Flood forecasting component All hazards system Artificial neural network

Costes et al. (2002)

Emergency levee repairs

Erlich (2007)

PREVIEW (2007) http://www. cimne.upc. es/ramflood/ Zhou et al. (2004)

242

10 Response

Box 10.2 provides a more detailed description for the FLIWAS system, developed as part of a major collaborative study between several organisations in the Netherlands, Germany and Ireland.

Box 10.2 Flood Information and Warning System (FLIWAS) The Flood Information and Warning System (FLIWAS) is an information and communication system to assist operational personnel, coordinators, information services and decision makers in flood event management (e.g., Langkamp et al. 2005). The system is one of the outputs from the Interreg IIIB-Project NOAH and is coordinated by the Foundation for Applied Water Research (STOWA) in the Netherlands in collaboration with several Dutch, German and Irish organisations, and project observers from France, Poland, Scotland and Germany.

Fig. 10.4 Flood defence overtopping (FLIWAS News, August 2007, http://www. noah-interreg.net/)

FLIWAS is internet-based and is designed to collect and process information to assist in managing actions before, during and after a flood event. Typical actions might include the activation of special dike watches, reinforcing of flood defences, communication with other organisations and the public, and preparatory measures for evacuation. Target users include water managers, local, regional and national authorities, civil protection units, and the general public and media. The system has been developed in close consultation with potential users and builds upon several existing systems in Germany and the Netherlands, including the High Water Information System (HIS) developed by the Ministry of Transport, Public Works and Water Management in the Netherlands (Ritzen 2005). Real time information can be imported from a wide range of sources, including observations of water levels and flows, weather radar, satellite and helicopter data, rainfall forecasts, and the outputs from flood forecasting models. Emergency (continued)

10.2 Decision Support Systems

243

Box 10.2 (continued) response plans can be defined and viewed for individual locations and organisations, and updated dynamically as an event occurs, with all actions taken logged automatically, and locally applicable information available in the field on palmtop computers and Personal Digital Assistants (PDAs). Standard situation reports can also be predefined, and updated during the event, and can be sent automatically to selected persons and organisations. Selected information can also be made available for viewing on the public access website. The system can also replay historical events for use in training and flood response exercises.

Fig. 10.5 Illustration of information flow within FLIWAS (http://www.noahinterreg.net/)

The baseline information required includes data on the geometry and condition of flood defences, action plans, key contacts within organisations, flood hazard maps, and information on properties at risk, critical locations such as hospitals and nursing homes, livestock, transport and diversion routes (e.g. railways), emergency response assets (e.g. sand depots, machinery, fire trucks), and sites with dangerous substances. The system is generic with the first modules to be developed covering generation and monitoring of calamity plans, resource management (people, equipment etc.), damage and casualty assessment, and evacuation planning. (continued)

244

10 Response

Box 10.2 (continued) FLIWAS will be completed with an evacuation module to advise on optimum evacuation routes and likely travel times, and to help with difficult choices such as whether there is enough time for people to leave an area, or they should be advised to move to shelters of last refuge (e.g. high rise buildings). Other modules will be included as required for individual applications, such as for assessing the impact of defence breach scenarios.

The issue of how to present and handle uncertainty is also increasingly being considered, with one related example being the applications for optimisation of hydropower operations discussed in Chapter 8. More generally, flood related decision support systems might be linked with information systems managed by other organisations, such as traffic management, communications, and power supply systems, and guidance provided on likely consequential effects from failure of any component (e.g. traffic hold-ups, cell phone failures, power cuts).

10.3

Dealing with Uncertainty

One of the challenges in responding to a flood event is the uncertainty both in current conditions, and what will happen next. The need to appraise recipients of the uncertainty in warnings is also widely emphasised (e.g. World Meteorological Organisation 1994; Emergency Management Australia 1999; Martini and De Roo 2007). Some typical sources of uncertainty which have been discussed in other chapters include: ● ●

●

Flood risk – uncertainty in the locations at risk from flooding (see Chapter 1) Detection – uncertainty in observations and forecasts of rainfall and other meteorological conditions, river levels, tidal levels, river flows, reservoir levels etc. (see Chapter 2) Flood forecasts – uncertainty in estimates for the likely magnitude, timing and extent of flooding, particularly for extreme events outside recent experience (see Chapters 5–8)

For the flood response component, there can be additional uncertainties in the specific risks to people and property during an event, in how people will respond to warnings, and in which secondary risks, such as power failures or communication breakdowns, might occur. Time of day or year is also a factor, with people better able to cope with flooding on a warm, summer day than in winter, and to respond to warnings during daytime than in the middle of the night. Factors such as traffic flows, numbers of properties occupied, and the ability to contact people, will also vary within and outside normal business hours.

10.3 Dealing with Uncertainty

245

Decision support systems can assist by helping to improve situational awareness during an event, and providing guidance on optimum decisions, whilst flood forecasting models, possibly also including real time inundation mapping, help with understanding likely future conditions. However, information on uncertainty can also be useful when deciding how to respond to a flood event and some potential users of this information include: ●

●

●

●

Flood forecasters – in deciding on the accuracy of the forecast, and what message to convey to people who will use the forecast Expert users – who can use probabilistic information directly for input to their own decision support systems Emergency response organisations – if they have the appropriate skills and training to interpret information on uncertainty The public – who may find some types of probabilistic information useful

For the first user group, flood forecasters, scenario and ‘what if’ modelling techniques have been widely used for many years, whilst probabilistic and ensemble forecasting techniques for both river and coastal flood forecasting are actively under development in several organisations, as discussed in Chapter 5. For expert users, earlier chapters provide several examples of the use of probabilistic information to assist in flood response, including for hydropower applications (see Chapter 8), and emergency management for polder areas in the Netherlands (see Chapter 3). The basis of the techniques used is often to compare the cost of taking action with the expected losses if no action is taken. The range of outcomes can be summarised in a contingency table, as illustrated by the simple example shown in Table 10.2. If the probability of flooding is p then, over the long term, taking mitigating action is cost effective if the cost of taking action C is less than pL, or p > C/L. The so-called cost/loss ratio then provides an indication of the forecast probability above which action should be taken, with the expectation that following this strategy over a number of events will result in total costs being less than total losses. More complex formulations can also be devised, for example taking account of partial mitigation of losses, and the costs which would have been incurred anyway in the absence of the event (e.g. Pierce et al. 2005). For flood warning applications, costs and losses are also dependent on lead time; for example, apart from the risk to life, it is usually more expensive to rescue people from properties than to evacuate them, and property damage is also usually higher when people have less time to prepare. Probability thresholds might therefore be defined for different stages in the run up to an event, perhaps linked to different flood warning stages (flood watch, flood warning etc.), with cost-loss relationships changing during the event (e.g. Roulin 2007). Table 10.2 Decision table for cost loss approach example Flooding

No flooding

Mitigating action No mitigating action

Cost (C) No cost

Cost (C) Loss (L)

246

10 Response

More generally, a utility or penalty function approach may be taken for situations where costs or losses cannot easily be defined in monetary terms (e.g. intangible losses) or the relationship between costs and losses is more complex. Utility (sometimes called Value) is a concept widely used by economists, in which typically a numerical score or ranking is given to the required outcomes, based on expert judgement, and possibly weighting of different outcomes (with weights elicited from multi-criteria analysis). Functions may be defined in such a way that severe outcomes, such as a dam overtopping, have exceptional scores, so that the optimum solution is steered away from or prevented from reaching that situation. The issue of how to present uncertainty information to emergency responders, and the public, is a new and developing area and one that has received much attention by meteorological, hydrological and social and behavioural researchers (e.g. National Research Council 2006; Demeritt et al. 2007; Demuth et al. 2007). Some conclusions from these types of study can include (e.g. Environment Agency 2007): ●

●

●

●

The best way to present information will vary between users, depending on their interests, technical expertise and roles. Given the wealth of information available, several alternative types of presentation may be useful, focussing on spatial, site specific and temporal information. Approaches should be simple and intuitive (at least in the initial stages), although a demand often arises for more sophisticated approaches as skills and experience develop. Forecast products are best developed as a joint exercise between forecasters and end-users.

In particular, the importance of consultations with end users is often emphasised (e.g. using focus groups, pilot studies), as is the value of working with probabilistic forecasts operationally to gain an intuitive feel for how to interpret them. In addition, Collier et al. (2005) note that there is a need to keep a clear distinction between the needs of hydro-meteorological services and flood emergency operations. In the former case the interest is in getting the best possible forecast, whereas in the latter case the interest is in making the best possible decision. Requirements can range from the wish for a simple yes/no answer on whether to perform an action, through to the need for a detailed understanding of the risks involved in taking a decision, and a wish to see all of the information that the forecaster has available, including that on uncertainty in forecasts; for example, in the following situations: ●

●

Mass evacuations – an emergency manager needs to balance the risks of an unnecessary evacuation against the risks of failing to evacuate properties if flooding occurs. There is also the trade off between waiting to make a decision, by which time the forecast will hopefully be more certain, and the time lost for starting the evacuation. For hurricane evacuations, the use of cost-loss approaches has been proposed to help in optimising these time based decisions (e.g. Regnier and Harr 2006). Flood defence operations – in the run up to a flood event, emergency managers may make decisions on the height to which defences need to be raised (e.g. using sandbags) to protect property or, in extreme situations, may decide to deliberately

10.3 Dealing with Uncertainty

247

breach parts of the defence network to flood, for example, agricultural land to protect towns and cities further downstream. The example of the Red River of the North Flood in 1997 is often quoted as an example of a situation where information on uncertainty could possibly have changed the course of the event (see Box 10.3)

Box 10.3 Red River of the North Flood, Grand Forks, April 1997 The Red River of the North catchment is located in North Dakota and Minnesota in the United States and southern Manitoba in Canada. Following record snowfall in the winter months, the most severe floods since 1826 occurred in April and May 1997 during the spring snowmelt (Glassheim 1997; Krzysztofowicz 2001; National Research Council 2003). The flood affected the cities of Fargo and Winnipeg, and in particular the two towns of Grand Forks, North Dakota, and East Grand Forks, Minnesota, where levees were overtopped and floodwaters reached over 3 miles (5 km) inland, inundating virtually everything in the twin communities and causing almost US$4 billion in damages and affecting about 5,000 properties, although fortunately with no lives lost. Nearly 90% of the area was flooded and three neighbourhoods were completely destroyed. Flood predictions were estimated using the seasonal forecasting technique described in Chapter 8. The first forecast of a major event was issued nearly 2 months before, on February 28, with a peak forecast of just under 49 ft under average precipitation conditions. This value was subsequently revised to 50 ft on April 14, then 52 and 54 ft on April 17 and 18. In Grand Forks and East Grand Forks, a previous major event had reached a peak of just under 49 ft and, based on the available information, city officials decided to prepare the city for a 52-ft river crest, whilst the actual peak reached exceeded 54 ft (National Research Council 2003). One comment following the event was that “If someone had told us that these estimates were not an exact science, or that other countries predict potential river crest heights in probabilities for various levels, we may have been better prepared.” (Glassheim 1997; Krzysztofowicz 2001). As with many flood events, post event studies showed a range of technical, communication and organisational issues which were quickly addressed, and of course the science of flood forecasting has improved significantly since the time of that event, including major improvements in flood forecasting techniques. However, the event was influential in changing views on including information on uncertainty in forecasts, and on the interactions between forecasters, the public and decision makers, with one conclusion (National Research Council 2006) being that unclear communication of uncertain forecast information can hinder decision-making and have significant negative consequences.

248

10 Response

Public awareness of techniques is also likely to increase as products become more widely available. For example, some meteorological services make information on uncertainty freely available on the internet and other media (e.g. television), with examples including: ●

●

●

Hurricane strike probability and cone of uncertainty forecasts – used by the National Weather Service in the USA to show the estimated spread of estimates in forecasts for hurricane tracks Rainfall forecast plumes – the approach used in the Netherlands by some television broadcasters of showing the range of estimates derived from Numerical Weather Prediction models Rainfall probabilities – maps for the percentage probability of heavy rainfall presented in weather forecasts on some meteorological service websites

Qualitative guidance, using terms such as ‘may’, ‘probably’ and ‘likely’, is also suggested in some applications, provided that messages also include advice on the actions to take (e.g. Emergency Management Australia 1999), whilst ISDR (2006), in the context of all-hazard warning systems, give the examples of using phrases such as “if present conditions continue…” or that “there is an 80% chance that…”. In Chapter 5, it was also noted how some recipients of warnings might, given the option, choose to receive warnings at lower levels of risk, defined as the combination of probability and consequence, and modern automated flood warning dissemination systems have the capability to target warnings to individual property owners, if required. Some examples of situations where targeted warnings might be useful include (e.g. Environment Agency 2007): ● ● ● ● ●

●

● ●

●

Local authorities who can close riverside and coastal footpaths to walkers Large businesses who can prevent customers parking in areas at risk Outdoor event organisers who can reschedule or relocate an event Farmers who can move livestock between fields Residents in frequent flooding locations who can install flood boards or sandbags Emergency managers who can plan staff rotas and check that equipment is ready Operators of temporary defences who can mobilise staff and equipment Hospitals who can reschedule operations and alert staff to the possibility of flooding Utility operators who can invoke contingency plans for flood events

Often, the interest is in receiving a personalised warning in advance of the official warning or whilst uncertainty still remains high. Techniques from the field of risk communication and perception can also assist in defining requirements. Some operational considerations with this approach can also include deciding on who is best placed to define risk thresholds, and to take risk based decisions, and how this affects the relative roles and responsibilities between the forecasting and warning authority, and the recipients of warnings.

Chapter 11

Review

Reviews of flood warning systems are often required following major flood events, and may form part of a programme of continuous improvement, sometimes linked to performance targets for different aspects of the system. Performance monitoring should ideally cover all aspects of the system, including detection, forecasting, dissemination, and response to warnings, together with feedback from users on satisfaction with the system. The lessons learned from post event assessments, and recommendations from regular reviews, can then guide future investments, and provide baseline information for use in economic assessment and prioritisation exercises. However, improvements need not necessarily require significant investment, and much can be gained from improving operating procedures, and closer collaboration between the various participants in the flood warning process, including communities and their representatives. This chapter discusses these various issues, and highlights some common themes from earlier chapters on ways of improving flood warning, forecasting and emergency response systems.

11.1

Performance Monitoring

Performance monitoring usually consists of a process of reviews, recommendations, implementation of findings, and continuous assessment to check that recommendations are being acted upon, and improvements are being made. Also, flood warning services increasingly need to demonstrate the benefits that they bring, and that improvements are being achieved over time. Routine reviews may be performed against benchmarks or targets, and of areas which may have changed since the time of the last review (key staff, equipment, procedures etc.). As noted in Chapter 9, regular tests and exercises can also help to identify problems, and keep staff trained in use of systems. Reviews can be performed for individual flood warning schemes, or on a regional, national, organisational, or multi-agency basis. Many organisations also routinely perform formal reviews of performance after major flood events, and sometimes near misses, both to answer immediate questions from the public, media, and politicians, and to identify improvements for the future. K. Sene, Flood Warning, Forecasting and Emergency Response, © Springer Science + Business Media B.V. 2008

249

250

11 Review

Two high profile examples were the various reports following Hurricane Katrina in August 2005 (e.g. US House of Representatives 2006) and the December 2004 Tsunami event (e.g. ISDR 2006). Ideally, each study should refer back to previous studies, and a record should be maintained of findings over the years, which can be referred to periodically to check that issues are not being overlooked, including changes which may influence flood response and flood warnings (e.g. to flood defences, instrumentation, control structures etc.). Some topics which are often covered in post event or lessons learned reports include: ●

●

●

●

●

●

●

●

●

Flooding causes – analysis of the meteorological and hydrological or coastal conditions leading to the event, and an assessment of the severity of the event compared to previous events Flooding impacts – assessments of the number of people injured, evacuated or rescued, the number of properties flooded, and of infrastructure and businesses affected, including a timeline for the event Flood warnings – summaries of the warnings issued, the lead time provided, and the number of people who received and acted upon those warnings Flood forecasting – performance of any flood forecasting models during the event (and related weather forecasts) Flood defence – performance of any flood defence and flow control structures, a summary of emergency works undertaken, and remaining problem areas Coordination – assessments of the coordination between the various agencies involved in the event, policy implications, and liaison with the media and the public Recovery – actions taken to make property safe, restore utilities (if applicable), remove debris, decontaminate sites, return people to their properties etc. Systems – a summary of how well detection, communication, telemetry, forecasting and other systems performed during the event Lessons learned – a summary of key findings and an action plan

The precise topics covered will depend on the scope of the review and organisational responsibilities. The review may also include interviews with residents, the emergency services and community representatives, and the findings from site visits to survey flood boundaries, and to inspect structures and flood defences. Routine assessments may involve workshops and research studies to share best practice, annual reporting against targets, and independent reviews of performance both within the organisation, and externally in collaboration with other emergency response organisations. Some questions on the content and delivery of messages might include (Emergency Management Australia 1999): ● ● ● ●

Did the target audience receive the warnings in time? Did they understand the warning message? Were their responses appropriate? If not, why not? What evidence is there for the answers to these questions?

11.1 Performance Monitoring

251

For example, in England and Wales, surveys of flood warning recipients are commissioned both annually, and after major events, to determine how many people received flood warnings, and what their opinion was of the information provided. Targets, performance measures, indicators and/or benchmarks are another way of driving improvements to performance, and are identified as an important component in flood warning systems, and disaster management systems generally (e.g. Handmer et al. 2001; Elliott et al. 2003; Andryszewski et al. 2005; Basher 2006). Some high level targets might be that there should be little or no loss of life from flood events, and that damages and disruption from flooding should be minimised. More specific performance measures may relate to the individual components in the flood warning process, such as the Accuracy, Reliability and Timeliness of flood warnings, the number of properties receiving flood warnings, flood forecasting accuracy, the frequency of emergency response exercises, the performance of flood emergency plans, and the number of health and safety incidents. Chapter 3 presents some examples of statistical and other approaches to assessing the performance of flood warnings, including contingency measures such as the Probability of Detection, and False Alarm Ratios, whilst Chapter 5 discusses a range of flood forecasting model performance measures. Various social response factors may also be considered, related to the ability of people to respond to flood warnings, their satisfaction with the warning service provided, and awareness of actions to take. When considering lead time targets, the values which are selected typically depend on the type of flooding anticipated, and the detection, forecasting and response systems which are available. For example, for tropical cyclones, typhoons and hurricanes, events may become apparent several days in advance, and warnings may be issued with 24 hours or more of notice. For coastal surge events from widespread, less intense storms, a few hours or more may be possible whilst, for flash floods, sometimes only a few minutes might be available. However, for the flash flood example, small numbers of people in a mountain village might only need a short time to move to the safety of higher ground whilst, for the tropical cyclone, typhoon and hurricane example, a major evacuation of residents might take a day or more. The speed and effectiveness of the response will also depend on the efficiency of dissemination systems, the local capabilities of emergency services, and on public awareness about how to respond to warnings. In estimating timeliness (i.e. the time between issue and receipt of a warning), the various time delays in the system also need to be accounted for as described in Chapters 5 and 9. False alarm rates are another measure where views on acceptable values differ widely (e.g. Barnes et al. 2007). If events only happen infrequently, then the occasional false alarm may be viewed as beneficial as a way of maintaining public awareness, and rehearsing and testing emergency response systems (e.g. Emergency Management Australia 1999). Similarly, if a property floods frequently, the owner may have well rehearsed procedures to protect the building (e.g. installing flood boards, moving vehicles), and view the occasional false alarm as a small price to pay for being able to continue living at that location. The analogy is sometimes

252

11 Review

drawn with other types of warnings, such as fire alarms and bomb alerts, where there can be a high public tolerance to occasional testing of systems and other false alarms (provided that the reasons are explained). However, if the action on receiving a warning is a widespread evacuation, or closing down of critical or expensive installations (e.g. oil refineries), then a low false alarm rate is often desirable. False alarms are caused in part by the various uncertainties in the detection, forecasting and warning process, and this topic is discussed in more detail in Chapter 10. When defining targets or performance indicators, the usual approach is to consider each stage in the flood warning, forecasting and emergency response process, and to devise suitable indicators of performance. Overall performance might then be assessed using summary tables or graphs, or by combining values using weighting or multi-criteria approaches. The information obtained can also help in understanding the areas in which future investment will be most beneficial (see Section 11.2). However, the choices made should consider the feasibility of collecting the supporting information required, the likely effort and costs, and how realistic it will be maintain the performance monitoring system over a period of years. Also, whether the best approach is to collect high quality information for a small number of indicators, or more comprehensive but less complete information across a larger number. Many different indicators could potentially be envisaged; for example, some possible descriptors considered in one research study included preparedness, forecasting, warning and promoting response, other communication, coordination, media management, equipment provision, environmental damage, economic damage, injuries, loss of life, victim trauma and reputation (Environment Agency 2007). The concept of levels of service might also be introduced as a way of monitoring performance and assisting with the design of flood warning systems. For example, for England and Wales, the Environment Agency (Andryszewski et al. 2005) uses this approach to ensure consistency in the implementation of flood warning schemes, and that schemes are prioritised according to risk, calculated from the probability of flooding and the number of properties at risk. The approach defines maximum, intermediate and minimum levels of service provision in the following areas according to the level of risk: ●

●

●

Detection and forecasting – requirements for raingauge and weather radar coverage, river gauge network density, and for false alarm rates and probability of detection Dissemination – recommended methods for providing indirect and direct warnings, including community level methods (e.g. sirens, loudhailers), and warnings via the media Public communications – recommended actions according to the level of risk, including direct mailing (at least annually, or up to every 3 years), public notices on noticeboards and in local media, and other activities, such as flood fairs, leaflets, newsletters, displays in libraries and newspaper articles

Some studies have also used reliability analysis techniques of the type described briefly in Chapter 9 to examine the various trade offs between indicators such as flood warning lead times, the success rate of warnings, and false alarm rates.

11.2 Performance Improvements

253

For example, Krzysztofowicz et al. (1994) describe a Bayesian approach applied to two case studies in Pensylvania, and Norouzi et al. (2007) describe application of a similar approach to assessing the reliability of a flood warning system in Iran.

11.2

Performance Improvements

Post event reviews, and regular performance monitoring, can lead to a range of recommendations for improvement, which might include: ●

●

●

●

●

●

Detection – improvements to meteorological, river and coastal observations (e.g. site locations, types of instrument, network density, resilience, accuracy), and meteorological forecasts Thresholds – revision of threshold levels to reduce false alarm rates, improve success rates, provide more lead time, provide backup in case of failure etc. Dissemination – improvements to systems and procedures to increase the proportion of people at risk receiving warnings, change the wording of messages, update flood risk assessments, raise public awareness etc. Forecasting – improvements to models and systems, such as recalibration of models, use of more sophisticated models, use of data assimilation, use of ensemble techniques Preparedness – revisions to flood emergency plans, more frequent and detailed emergency response exercises etc. Response – improvements to inter-agency coordination, information and communication systems, liaison with the media etc.

In the integrated or total flood warning system approach, all components need to be improved if the ultimate aim of minimising risks to people and property is to be achieved (e.g. Emergency Management Australia 1999; Andryszewski et al. 2005). More general requirements may also be identified at an organisational or national level; for example, the need to extend the flood warning service to new locations, to introduce greater consistency in procedures, and to provide warnings for additional types of flooding, such as urban flooding, or for fast response catchments. The decision may also be made to introduce or improve flood warning targets and improved performance monitoring systems to help to drive future improvements. International reviews and comparisons may also highlight potential changes; for example, Parker et al. (1994) used the following 14 criteria to compare flood forecasting, warning and response systems (FFWRS) between several European countries: flood warning philosophy, dominance of forecasting vs warning, application of technology to FFWRS, geographical coverage, laws relating to FFWRS, content of warning messages to public, methods of disseminating flood warning, attitudes to freedom of risk/hazard information, public education about warnings, knowledge of FFWRS effectiveness, dissemination of lessons learned, performance targets and monitoring, national standards, and organisational culture.

254

11 Review

Opportunities may also exist to widen the scope of the flood warning service into other types of forecasting and warning, and to share costs with other organisations or departments with common interests. For example, the requirement for real time monitoring of rivers and coastal waters is often shared by other groups with an interest in, or responsibility for, managing water resources, pollution incidents, navigation etc. Similarly, integrated catchment forecasting models can be used for forecasting across the full range of flows, including drought forecasting (see Chapter 8, for example), whilst the requirements for decision support systems, GIS systems, dissemination systems, and other emergency response equipment are common to many types of natural and technological hazard. The range of possible areas for improvement is huge and it is only possible to discuss a few common themes here. The various guidelines and reviews cited in previous sections and chapters provide more information on potential ways of improving aspects of the flood warning, forecasting and emergency response process.

11.2.1

Detection

Improvements in detection can include monitoring at more locations, using new techniques and technology, and improving existing instruments and telemetry systems. For flood warning applications, optimum network densities depend on the type of flood response, the level of risk, and other factors, such as the need to provide information for flood forecasting models, and backup instruments in case of failures. Chapters 2 and 5–8 discuss some of these issues further. There can also sometimes be advantages in sharing instrumentation across more than one flood warning scheme to reduce costs, for example by installing raingauges near to catchment boundaries if there are flood risk areas in two adjacent catchments. Network densities might also be linked to required levels of service and flood risk (e.g. Andryszewski et al. 2005; Sene et al. 2006). Technological developments (see Chapters 2 and 5) can also offer opportunities for improvement, and some notable developments in recent years include: ●

●

●

●

Acoustic Doppler Current Profilers (ADCP) – making high flow gaugings more feasible during flood events to assist with the subsequent development of stage discharge relationships Nowcasting – improved techniques for high resolution short term rainfall forecasts Ensemble forecasting – probabilistic estimates of rainfall, river flows, surge and other variables Remote sensing – improvements in accuracy and resolution, and the range of parameters which can be monitored by satellite, and increased availability of products

11.2 Performance Improvements ●

255

Forecasting systems – improvements in the functionality and usability of systems for running forecasting models, and for interfacing to other systems

Some emerging technologies, such as low cost sensor networks, disdrometers for rainfall measurement, and remote techniques for measuring river velocity (and hence flow), also show potential (see Chapter 2). Resilience is also an important factor in flood warning applications, and Chapter 2 highlights the need to ensure that instruments are sited in locations where they will not be damaged by flood water or debris, and with electronic equipment above likely flood levels. Backup instruments and telemetry can also be provided at the same site (e.g. for raingauges and coastal instrumentation) or further upstream (for river level or flow gauges). Issues of site access during flood events also need to be considered, and the health and safety of staff.

11.2.2

Thresholds

Flood warning thresholds define the conditions (or criteria) under which flood warnings are issued (or considered for issue) and are described in Chapter 3. Although values may be defined based on past experience, and sometimes by sophisticated computer modelling, post event reviews and performance monitoring may show the need for improvement. Some indicators of the possible need to revise thresholds include: ●

●

●

Missed warnings – problems with not issuing warnings when flooding occurs, or after the start of flooding, may require a detailed investigation into the likely causes and areas for improvement (e.g. in instrumentation, threshold levels, forecasting models) False alarms – an unacceptable rate of false alarms may indicate that thresholds are too conservative or, possibly, that an acceptable success rate is only possible at the expense of a high false alarm rate Insufficient lead time – problems with warnings being issued later than would ideally be required for an effective emergency response

Some approaches to increasing lead times include the use of new or improved flood forecasting models, additional thresholds on gauges upstream or distant from the flood risk area (preferably whilst continuing to maintain the existing thresholds) and, possibly, adjusting existing thresholds so that they achieve more lead time, whilst still maintaining an acceptable false alarm rate. All adjustments to thresholds need to be made with care, and fully tested and documented before implementation, preferably also consulting with those affected to confirm that the approach is acceptable. More sophisticated approaches might also be considered, including use of computer modelling to support the development and testing of thresholds, probabilistic techniques, and other methods (see Chapter 3). Improvements can also aim to reduce the various time delays in the detection, dissemination and response process (see Chapter 5) through improved systems, procedures, and training.

256

11.2.3

11 Review

Dissemination

The principles for issuing flood warnings are similar to those in many other areas of risk communication and Chapters 4 and 10 discuss this topic. Some key points to consider (e.g. Emergency Management Australia 1999; Drabek 2000; Handmer et al. 2001; Martini and de Roo 2007), beyond the need for warnings to be accurate, reliable and timely, and reach the intended recipients, include: ●

●

●

●

●

●

To give people time to prepare and plan for flooding, warnings should be staged (if there is time), starting from advisory/watch/warning alerts (or similar), before escalating to a full warning Messages should be consistent, clear and concise, and tailored to the audience, ideally in their own language Messages should be specific about the threat; for example in terms of flood timing, depth, duration etc. Messages should also include advice on actions to take to protect people and property, and distinguish between forecasts and warnings Messages should be received from a single, authoritative (and trusted) source, whilst acknowledging that in practice people may seek information from multiple sources, both formal and informal, before deciding to act Multiple means of dissemination should be used in case of failure in any one route, and to improve the effectiveness of response

Potential recipients (communities, emergency services) should also be actively involved in the choice of appropriate ways of disseminating information and the wording of messages. A particular challenge can be deciding on methods for issuing warnings to vulnerable groups, and to transient or mobile populations (e.g. tourists, vehicles, business travellers). Some other factors which may need to be considered in message design (Elliot and Stewart 2000) include: ● ● ●

● ● ●

Degree of flood exposure, severity of impact Degree of flood experience Financial or emotional ‘stake’ in the flood-prone area (e.g. residents as opposed to tourists) Household structure (e.g. age, health status) Language and Employment status (e.g. likelihood of being at home during the day)

New technologies are also increasing the range of methods which can be used, allowing better targeting of warnings, and reducing the time taken to issue warnings. Some recent developments include: ● ●

● ●

Computerised multimedia systems for issuing warnings Internet based systems combining real time data, flood extent and advice on actions to take Cell phone broadcasting techniques to all subscribers within range of a mast Digital radio warnings to drivers of vehicles

11.2 Performance Improvements ● ●

257

Remotely activated barriers and electronic road signs for roads and footpaths Low cost battery free radios and satellite transmission of warnings

As with all components in a flood warning system, dissemination techniques should be able to continue working under flooding conditions and associated problems, such as high winds and heavy rainfall, with backup methods in case of failure. For techniques which require access to potential flooding areas (e.g. loud hailer, door knocking) access may also be interrupted by flood waters, or not be possible due to health and safety considerations.

11.2.4

Forecasting

Improvements to flood forecasting can focus on improving existing models, developing new models, and improving the robustness and speed of operation of models. Performance monitoring techniques can also be used to guide future improvements (see Chapter 5). Changes to existing models can include recalibrating models to account for recent flood events, or changes to instrumentation, flood defences, river channels, and coastal conditions, and adding in new functionality. Where new models are to be developed, the opportunity may also be taken to try out new modelling techniques; for example, using different types of model, or introducing data assimilation and probabilistic techniques. Such changes may also be linked to improvements to forecasting systems. Opportunities may also arise to combine models across a number of flood warning schemes; for example, using an integrated catchment approach. A spin-off benefit from this approach is often that the need to consider the catchment as whole may suggest improvements to instrumentation that would not otherwise be obvious from examination of records for single gauges. The choice of models should be appropriate to the level of risk and a range of other factors, and various guidelines are available to help in deciding on an appropriate approach (e.g. World Meteorological Organisation 1994; USACE 1996; Environment Agency 2002, 2004). For example, Tilford et al. (2007) describe a structured approach to the selection of river forecasting models which considers: ● ● ● ● ●

The physical characteristics of the catchment and river(s) The varying levels of data availability and quality The cost and time of development The technical and economic risks associated with the investment Performance targets

The method uses a combination of flowcharts, risk assessment matrices and cost benefit analyses to decide on an appropriate choice of model. The modelling options provided include empirical, data-based, conceptual and process-based models for a range of river modelling problems (e.g. floodplains, reservoirs, structures, snowmelt, tidal influences). The method also includes consideration of a range of practical

258

11 Review

issues, such as how long it might take to install instrumentation, the level of modelling expertise needed, and the need for additional exploratory investigations. Throughout, room is left for expert judgement and the method avoids being prescriptive in the choice of modelling solution.

11.2.5

Preparedness

Techniques for developing flood emergency plans are well established (see Chapter 9), with issues such as inter-agency collaboration, a clear chain of command, interoperability of equipment, team typing and other factors increasingly emphasised, together with the involvement of communities and their representatives in formulating plans. The needs of vulnerable groups in particular need to be considered. Increasingly, an all-hazards approach is being adopted by many organisations, providing benefits of scale and allowing plans to be rehearsed more frequently. For plans to remain effective, they need to be regularly rehearsed and reviewed and updated. The issue of resilience is also important for all aspects of the flood warning process, and probabilistic and risk based approaches from other sectors may become more widely used in future in assessing potential points of failure. Information technology also provides the opportunity for more realistic training exercises, combining multimedia simulations, and animation of flood extents in computer models of towns and cities. Geographical Information Systems can also help during the planning phase in examining how access routes, infrastructure, key facilities and other factors will be influenced by flood water, and in refining risk assessments.

11.2.6

Response

Previous chapters have discussed the important of clear presentation of flood warnings (Chapter 4), active engagement by communities (Chapter 9), and the use of a range of approaches for providing information to people (e.g. ISDR 2006; United Nations 2006a). A clear statement can also help with understanding the objectives of a flood warning and forecasting system; for example (Defra 2004) that: “Flood warning is the provision of advance warning of conditions that are likely to cause flooding to property and a potential risk to life. The main purpose of flood warning is to save life by allowing people, support and emergency services time to prepare for flooding. The secondary purpose is to reduce the effects and damage of flooding. This might include moving property to a safer location such as upstairs or putting in place temporary measures to prevent floodwater entering properties such as flood boards or sandbags. In addition flood warning informs operating authorities who need to take action such as closing floodgates or other control structures in advance of flooding conditions.” Although local authorities and emergency services can take many actions to reduce or mitigate flooding, ultimately, if flooding does occur, then the success of

11.2 Performance Improvements

259

the warnings issued will also depend on residents understanding the risks from flooding and taking appropriate actions in time to protect people and property. Some principles which can improve the success of flood warning systems (Handmer 2001; Betts 2003) include: ● ●

● ●

The public’s access to both formal and informal sources of warning information The value of ‘shared understanding’ between the public and emergency managers about the warning message and process Inter-organisation cooperation The recognition of local needs

An ongoing programme of education, training and capacity building helps to keep the awareness of flood risk high, and improves the likelihood that people will respond appropriately in the next flood event. Informal warning systems also have a valuable role to play during flood events (e.g. Parker 2003) and reinforce and add credibility to formal warnings. Techniques and expertise from the social sciences, market research, education, and health care promotion, can all be used to help improve the effectiveness of campaigns. Avoidance of risks to people is a primary objective for many flood warning schemes, and some factors which can lead to an increased risk to loss of life include (Environment Agency 2003): ● ●

● ●

●

● ●

●

● ●

●

Where flow velocities are high Where flood onset is sudden as in flash floods, for example the Linton/Lynmouth floods in 1952, Big Thompson flood, USA, in 1976 and flash floods in Southeast China in 1996 Where flood waters are deep Where extensive low lying densely populated areas are affected, as in Bangladesh, so that escape to high ground is not possible Where there is no warning (i.e. where there is less than, say, 60 minutes of warning) Where flood victims have pre-existing health/mobility problems Where natural or artificial protective structures fail by overtopping or collapse. Flood alleviation and other artificial structures themselves involve a risk to life because of the possibility of failure, for example dam or dike failure Where poor flood defence assets lead to breaches or flood wall failure, leading to high velocities and flood water loadings on people in the way Where there is debris in the floodwater that can cause death or injury Where the flood duration is long and/or climatic conditions are severe, leading to death from exposure Where there is dam failure

Other causes include building collapse and related circumstances (e.g. mobile homes, campsites), being swept away, falling down manholes or similar, and being trapped in buildings or vehicles. For example, Jonkman and Kelman (2005) suggest that drowning in vehicles seems to be a worse problem in the US than in Europe, and that significant numbers of flood deaths are attributable to unnecessary risky

260

11 Review

behaviour, whilst in Australia almost 10% of flood related deaths result from people trying to retrieve property or animals. Some estimates (e.g. Henson 2001) suggest that about half of all flash flood related deaths in the USA occur to people in vehicles. Risks can be reduced by raising public awareness on the dangers of driving through flood water, precautionary closing of roads in advance of flooding, and dissemination techniques aimed specifically at drivers, such as digital radio alerts, and remotely or locally activated barriers or electronic signs (see Chapter 4). Flood risk assessments may also need to account for flow velocities since this can be a significant factor in whether vehicles are swept away by floodwater. For the particular example of non-residential properties (e.g. shops, businesses, factories), some criteria for the effectiveness of flood warnings which have been proposed (Defra 2005) include: ● ● ● ● ●

● ●

●

●

They have a long lead time (preferably at least 8 hours) Management have confidence in the warning and the issuing authority The warnings give specific information on the timing and likely level of flooding Staff are aware of and trained in the actions to take There are enough able bodied staff or contractors available to move equipment and goods and take mitigating actions Equipment and goods are able to be moved (e.g. not too large or heavy) There is enough space on upper floors or storage areas, in an alternative location, or on higher ground, to move equipment and goods to Appropriate refrigeration is available for storing perishable foodstuffs, drinks, pharmaceuticals etc. elsewhere Surrounding areas and roads are not flooded to facilitate evacuation and movement of equipment

To help deal with the complexity of a flood event, Decision Support Systems and Geographical Information Systems can also assist emergency managers in assessing risks, improving situational awareness, sharing information between people and organisations, automated logging of actions, and (in some cases) providing guidance on optimum decisions, such as the requirements for evacuating properties. Hand held units may also be used by staff on site to view locally relevant information. Probabilistic and cost loss approaches may also provide one way of optimising decisions to take account of uncertainties in factors such as flooding extent, flooding impact, and the response of individuals to flooding. Again, as with all other improvements, resilience needs to be considered from flooding and associated rainfall and high winds etc., together with staff training, interoperability of systems, and a range of other factors (see Chapter 9).

11.3

Prioritising Investment

The outcome from regular or post event reviews is often a series of recommendations, some of which may require investment in new equipment, procedures and other resources. The requirements to justify expenditure vary widely between

11.3 Prioritising Investment

261

organisations, but there will often be a need to consider where priorities lie, and the relative costs of different options. Some techniques which can be used for prioritising investment include: ●

●

●

Cost benefit analysis – in which costs are compared to the estimated benefits from improving flood warnings Multi-criteria analysis – which uses a range of weighting techniques across a number of factors to rank the relative merits of proposals Risk based approaches – in which investment is targeted to the areas of greatest risk, taking account of factors such as the frequency of flooding, the number of people affected, risk to life, risk to critical infrastructure, or the risk of flow control structures not being operated effectively

Each method has its advantages and limitations. For example, cost benefit analyses focus mainly on the economic aspects of investment, and it can be difficult to bring in other more intangible quantities, such as loss of life and the long-term health costs from people affected by flooding who, with sufficient warning, might have moved to safety. People may also place more value on saving personal items (documents, memorablia etc.) and domestic animals and pets than on high cost items. By contrast, Multi-Criteria Analysis (MCA) methods do not consider the economic case in detail, and are more subjective in the way that decisions are reached, although can easily be combined with economic analyses. Other priorities may also influence the overall decision, such as pressure from local residents and politicians to improve flood warning schemes, and reputational issues, related to not having issued a warning before an event, or high false alarm rates. Risk based approaches are also often incorporated into the other two techniques, as described later. The scope of the economic analysis also needs to be defined; for example, flood warning is increasingly seen as a key aspect of overall flood risk management, or one component in a multi-hazard approach. Economic analyses may therefore need to be tied into a wider assessment covering flood defences, development on floodplains, catchment management and risks from sources other than flooding. Other complicating factors can include situations where the costs and/or benefits are accrued by different organisations, some of which may be outside the flood warning process, and in trans-national river basins, where several countries may participate in the flood warning scheme. Sensitivity studies, or probabilistic techniques, may also be used to help to account for uncertainty in inundation extents, depths, velocities and impacts.

11.3.1

Cost Benefit Analysis

Cost benefit analysis is widely used in a number of fields, including flood risk management (e.g. ISDR 2006; World Meteorological Organisation 2007), and has also been applied to flood warning systems. The cost element is built up from systematic analysis of the individual (unit) costs of the items which make up the

262

11 Review

system, whilst benefits are typically estimated in terms of the damage avoided due to issuing flood warnings. Some approaches also attempt to assign values to lives saved, for example in terms of lifetime earnings, or compensation payments, although this is a problematic area which is often excluded from the analysis. The scope of the analysis will depend on the level at which financial decisions are made, and could be just for an individual department or organisation, or across a number of organisations. Analyses can also be at the scale of individual flood warning schemes, regional systems, or at national level. Care is needed to avoid double counting of benefits when considering multiple schemes and organisations. Depending on the objectives of the analysis, the cost element of the analysis can consist of a wide range of items, including: ● ●

●

●

Start up costs – feasibility studies, design studies etc. Capital costs – instrumentation, computers, models, dissemination equipment, communication equipment etc. Operating and maintenance costs – for office accommodation, public awareness campaigns, emergency response, post event studies etc. Review costs – for periodic reviews and reporting on performance

Depending on the scope of the analysis, staff costs may just relate to the flood warning and forecasting service, or extend more widely into the emergency services, and local authorities, and businesses. Future investments also need to be converted to a common basis; for example using net present value techniques. Additional costs, such as those incurred in taking mitigating actions (e.g. temporary closing of businesses), may also need to be considered. Estimates for the benefits from flood warnings usually focus on the damage avoided to property by moving items to safety, and perhaps through preventing flooding by installing temporary measures in time, such as flood boards, or sandbags. Some techniques for estimating potential damages include (e.g. World Meteorological Organisation 2007): ●

●

●

●

Historical damages – information on actual damages from previous flood events (e.g. insurance claims, surveys) Unit area method – in which losses are estimated based on the floor area of properties Percentage of property values – in which losses are estimated from property values (ideally for the building alone and excluding land values) Weighted average annual damages – based on the frequency and severity of flooding

These methods all have various advantages and limitations; for example, information on previous floods may be incomplete, or biased by the approach used to collect data, whilst unit area approaches may be more suitable for commercial properties than residential properties. The percentage of property value approach uses data which is widely available, but it may be difficult to separate out land values from building values. Also, building contents can vary widely, particularly

11.3 Prioritising Investment

263

for commercial properties, whilst the average annual damage method requires long term reliable information on flooding histories and damages. Various combined and synthetic approaches are also used. The benefit from flood warnings are often separated into direct and indirect (or intangible) losses in a number of areas; for example (USACE 1996): ● ●

●

● ●

●

Reduced threat to life – barricades, evacuations, rescues, public awareness Reduced property loss – removal or elevation of residential and commercial structure contents and vehicles Reduced social disruption – traffic management, emergency services, public awareness Reduced health hazards – evacuations, public information, emergency services Reduced disruption of public services – utility shutoffs, emergency services, supplies, inspection supplies, inspection, public information Reduction in inundation – flood fighting, temporary flood damage reduction measures, technical assistance

Here, flood fighting means taking actions to reduce flooding, such as repairing breaches, clearing channels of debris, sandbagging etc. USACE (1994) additionally notes the benefits arising from a number of other factors, including temporary flood proofing of properties, reductions to recovery costs, and temporary suspension of industrial and other processes. Many of these items can in principle be estimated, although some with difficulty due to lack of data and other problems, such as the loss of life issue referred to earlier. Values may also depend on factors such as weather conditions (temperature, wind chill etc.), the time of day that the warning is received, and the time elapsed since the last flood (e.g. World Meteorological Organisation 1973). More detailed discussions of methods for estimating flood warning benefits can be found in World Meteorological Organisation (1973), USACE (1994), Carsell et al. (2004), and Parker et al. (2005). In flood warning applications, the annual average damage approach is perhaps the most widely used method. For general classes of property (both residential and commercial), annual average damage curves can be estimated for a range of flood depths and, possibly, velocities. These values can be estimated from post event data across a number of flood events and types/ages of property, and from demographic characteristics, although there can be considerable uncertainty in the estimates derived (e.g. Merz et al. 2004). Values can also be probability weighted by integrating damage-depth and depth-frequency curves. For example, a common approach is to assess typical depth (stage) damage relationships for various classes of property, and then to assess the reductions in damage for different flood warning lead times, as illustrated in Fig. 11.1 for a specific community and lead time. Studies of this type have shown that, as might be expected, the damage avoided increases with increasing lead time up to a point of diminishing returns, beyond which any additional lead time becomes of little benefit in reducing damages. Other benefits might also be included, such as the damage avoided by operating flow

264

11 Review

Fig. 11.1 Example of a community stage-damage relationship (World Meteorological Organisation 1973, 1994) (Reproduced from the WMC Guide to Hydrological Practices - Data Acquisition and processing Analysis, Forecasting and other Applications, courtesy of WMO)

control structures, or installing temporary defences (barriers, flood boards, sandbags etc.) to protect communities or individual properties. For example, in England and Wales, the following equation (CNS Scientific and Engineering Services 1991; Parker et al. 2005; Tilford et al. 2007) forms the basis of the method used to estimate flood warning benefits from reductions in damage to residential property and road vehicles: FDA = R x Pi x Pa x Pc x PFDA where: FDA = Flood Damages Avoided PFDA = Potential Flood Damages Avoided R = Service Effectiveness Pi = Probability that the individual will be available to be warned Pa = Probability that the individual is physically able to respond Pc = Probability that the individual knows how to respond effectively The Service Effectiveness is the proportion of properties which were sent a flood warning whilst the Potential Flood Damages Avoided is calculated from the average annual damages (AAD) as follows: PFDA = DR x C x AAD Here, DR (the Damage Reduction factor) is the proportion of damages which can realistically be avoided by flood warning (since some damage is unavoidable), and depends on warning lead time, and C is the Coverage of the flood warning

11.3 Prioritising Investment

265

service (i.e. the proportion of properties which receive a flood warning service). An alternative version (e.g. Environment Agency 2002; Tilford et al. 2007) includes an allowance for the costs of mitigating actions by property owners; for example, when taking time off work to protect properties. Similar techniques can in principle also be used for commercial properties, although some sites may need to be considered on a case by case basis (e.g. major chemical works, oil refineries, distribution warehouses etc.). In addition to providing a basis for estimating flood warning benefits, the various factors also provide a focus for improvements to individual components in the flood warning service. For example, in England and Wales, the 2012/13 target values are in the range 80–90% for most parameters except for the Damage Reduction factor (e.g. Parker et al. 2005). Additional factors are used to monitor progress in the percentage of people making preparations in advance of flooding (e.g. individual flood plans), and the proportion of people at risk signing up for a direct flood warning service. Progress is assessed through post event reviews, independent market surveys of flood warning recipients, and other approaches. However, the extent to which improvements are possible depends in part on the nature of the flood risk in individual locations, the scope to take actions to prevent flooding, how frequently flooding occurs, and socioeconomic and other factors. Computer simulation tools of the type described in Chapter 10 might also be used to explore the effectiveness of improvements to individual components, including social, vulnerability, psychological and policy aspects (e.g. Simonovic and Ahmad 2005). Many social and behavioural studies have also been performed into losses from factors other than damage to property, and on the general effectiveness of flood warnings, including studies on loss of life (Jonkman and Kelman 2005), public response to flood warnings (Drabek 2000; Pfister 2002), health impacts (Parker et al. 2005), and risks to people in vehicles (e.g. Henson 2001). Other examples include studies on the real time assessment of hurricane losses (Dixon et al. 2006), and the benefits from flood forecasting for reservoir flood control and short and long term flood forecasting (National Hydrologic Warning Council 2002). Various estimates have also been derived for the damage reduction component of flood warnings, with values typically in the range from a few percent to 30–40% or more (e.g. ISDR 2006; World Meteorological Organisation 1989; Parker et al. 2005). However, care is needed in interpreting estimates to see which types of losses have been included in the analysis, and the various other assumptions made. For example, some studies have focussed mainly on the monetary benefits to residential property owners, and much less is known about intangible benefits, and the varying reasons why some people do not take effective action even after receiving a warning.

11.3.2

Multi Criteria and Risk Based Analysis

Rather than working in monetary terms, multi criteria analyses aim to evaluate a range of options against criteria or objectives agreed with key stakeholders, and can

266

11 Review

be used as an initial screening tool, before proceeding to a more detailed analysis, or as perhaps the only way forwards when there are many conflicting objectives. The technique builds on the experience and knowledge of stakeholders, and can help to make the decision process more transparent, although inevitably includes an element of subjectivity. There are many forms of multi criteria analysis technique, typically involving assigning weights and scores to a range of options and issues, and then combining the results to reach an overall conclusion (e.g. World Meteorological Organisation 2007). Multi criteria techniques may also be combined with cost benefit techniques; for example, a methodology proposed for evaluating river and coastal flood defence schemes (Ash et al. 2005; Defra 2005) involves the following steps: ● ● ● ● ● ● ● ● ● ● ●

Definition of problem, the objectives and identification of all options Elimination of unreasonable options Structuring the problem (high level screening) Qualitative assessment of impacts Quantitative assessment of impacts Determine the tangible benefits and costs of options (economic analysis) Scoring of options Weight elicitation, as appropriate Comparison of options using expanded decision rules Testing the robustness of the choice Selecting the preferred option

Impact assessments are performed using a structured approach covering a range of economic, environmental and social issues which need to be considered. Numerical or descriptive scores can be used, with scoring by key experts or a committee, whilst weights reflect the preferences of stakeholders. The comparison of options stage combines the outputs from the multi-criteria and cost benefit analyses. In another example, Sene et al. (2006) describe a simple scoring approach to supplement cost benefit analyses for the design for telemetry networks for flood warning applications, in which the scoring criteria were: ● ● ● ● ●

Risk category for the flood warning area (high, medium, low etc.) Cost of implementation Cost per property of implementation Benefit-cost ratio at a flood warning area level Views from consultations (e.g. flooding ‘hotspots’)

More generally, risk categories can be used as a guide to the choice of appropriate instrumentation, performance criteria, and flood warning dissemination methods in a new or upgraded flood warning scheme (e.g. Andryszewski et al. 2005).

Glossary

A Action Table – a table of actions to take as meteorological, river and/or coastal conditions exceed predefined threshold values Antecedent Conditions – the state of wetness of a catchment prior to an event or period of simulation (Beven 2001) Antecdent Precipitation Index – the weighted summation of past daily precipitation amounts, used as an index of soil moisture. The weight given each day’s precipitation is usually assumed to be an exponential or reciprocal function of time, with the most recent precipitation receiving the greatest weight (UNESCO/WMO 2007) Automated Voice Messaging (AVM) – automated telephone system for issuing flood warnings Automatic Weather Station (AWS) – an instrument for automatically measuring climate data in real time including (typically) wind speed and direction, solar radiation, air temperature, humidity, and rainfall, and possibly other parameters, such as soil temperature

B Baseflow – part of the discharge which enters a stream channel mainly from groundwater, but also from lakes and glaciers during long periods when no precipitation or snowmelt occurs (UNESCO/WMO 2007) Basin – see Catchment Black Box Model – a model that relates only an input to a predicted output by a mathematical function or functions without any attempt to describe the processes controlling the response of the system (Beven 2001) Boundary Conditions – constraints and values of variables required to run a model for a particular flow domain and time period (Beven 2001) Business Continuity Management – a management process to identify and manage the hazards or threats which can disrupt the smooth running of an organization or delivery of a service

267

268

Glossary

C Calibration – adjustment of the parameters of a model, either on the basis of physical considerations or by mathematical optimization, so that the agreement between the observed data and estimated output of the model is as good as possible (see Model Calibration, UNESCO/WMO 2007) Cascade Warning – an approach to dissemination of warnings in which warnings are passed from one person or organization to others, who in turn pass on the warning following predefined procedures. Includes Call Trees and Telephone Trees Catchment – drainage area of a stream, river or lake, or area having a common outlet for its surface runoff (see Basin or Catchment, UNESCO/WMO 2007) Conceptual Hydrological Model – simplified mathematical representation of some or all of the processes in the hydrological cycle by a set of hydrological concepts expressed in mathematical notations and linked together in a time and space sequence corresponding to that occurring in nature (UNESCO/WMO 2007) Contingency Table – a table usually summarizing the relationship between the frequencies of occurrence of two or more variables, at the simplest level consisting of a 2 × 2 matrix Cost Benefit Analysis – a decision making technique which compares the likely costs of an action or investment with the expected benefits Cost Loss Analysis – an analysis technique which compares the cost of taking an action with the likely losses if that action is not taken, which can include dependence on lead time, the influence of only partial protection against losses, and other factors

D Damage Avoidance – the potential financial benefit from providing a flood warning taking into account the maximum damage which could be avoided and possibly the costs of property owners acting upon the warning Data Assimilation – the use of current and recent real time observations of meteorological, river and/or coastal conditions to improve a forecast (e.g. a flood forecast) Data Collection Platform – automatic measuring device with a radio transmitter to provide contact via a satellite with a reception station (UNESCO/WMO 2007) Debris Flow/Mud Flow – flow of water so heavily charged with earth and debris that the flowing mass is thick or viscous (UNESCO/WMO 2007). A high-density mud flow with abundant coarse-grained materials such as rocks, tree trunks, etc. (IDNDR 1992) Decision Support System – in emergency management, usually a computerized system for collating and displaying real time information of many types (spatial, time series, descriptive etc.), and sometimes for advising on optimum decisions

Glossary

269

Degree-Day – algebraic difference, expressed in degrees C, between the mean temperature of a given day and a reference temperature (usually 0°C). For a given period (months, years) algebraic sum of the degree-days of the different days of the period (UNESCO/WMO 2007) Deltas – see Estuaries Deterministic Model – a model that with a set of initial and boundary conditions has only one possible outcome or prediction (Beven 2001) Dike – see Flood Defence Dissemination – in flood warning applications, the issuing of warnings by a range of direct, community based and indirect methods Distributed Model – a model that predicts values of state variables varying in space (and normally time) (Beven 2001)

E Effective Rainfall – that part of rainfall which contributes to runoff. In some procedures the prompt subsurface runoff is entirely excluded from direct runoff and then effective rainfall is equal to rainfall excess (UNESCO/WMO 2007) Ensemble Forecast – a number of alternative realisations of future meteorological, river or coastal conditions based on alternative values for initial conditions, model parameter values etc., which reflect the inherent uncertainties in observations and forecasting models Estuary – the tidal reaches of a river as it outfalls to the sea, where fresh and sea water mix. Sometimes called a Delta or River Delta (although this term describes the sediment deposited by some rivers within the tidal zone) Evapotranspiration – quantity of water transferred from the soil to the atmosphere by evaporation and plant transpiration (UNESCO/WMO 2007)

F False Alarm – in flood warning applications, a warning which is issued but for which no subsequent flooding occurs. Can also include ‘near misses’ Fetch – area in which ocean, lake and reservoir waves are generated by the wind. The length of the fetch area is measured in the direction of the wind (UNESCO/ WMO 2007) Finite Difference – the approximate representation of a time or space differential in terms of variables separated by discrete increments in time or space (Beven 2001) Finite Element – the approximate representation of time or space differentials in terms of integrals of simple interpolation functions involving variables defined at nodes of an irregular discretization of the flow domain into elements (Beven 2001)

270

Glossary

Flash Flood – flood of short duration with a relatively high peak discharge (UNESCO/WMO 2007). Alternatively, a flash flood can be defined as a flood that threatens damage at a critical location in the catchment, where the time for the development of the flood from the upstream catchment is less than the time needed to activate warning, flood defence or mitigation measures downstream of the critical location. Thus with current technology even when the event is forecast, the achievable lead-time is not sufficient to implement preventative measures (e.g. evacuation, erecting of flood barriers) (ACTIF 2004) Flood Defence, Dike or Levee – water-retaining earthwork used to confine streamflow within a specified area along the stream or to prevent flooding due to waves or tides (UNESCO/WMO 2007). Can be constructed from a range of materials, including concrete, steel and rockfill Flood Fighting – emergency response operations to reduce or prevent flooding, including reinforcing flood defences, sandbagging, installation of temporary defences, and other measures Flood Forecasting System – a computer system for managing the operation of one or more flood forecasting models, include automated collection and validation of real time data, post processing of model outputs, and possibly automated alerting facilities if thresholds are exceeded Flood Risk Area – an area at risk from flooding which may or may not have an existing warning service and whose extent is typically estimated from historical information, modeling, or other methods Flood Risk Assessment – an assessment of the likely extent and probability of flooding at one or more locations Flood Warning Area – an area defined for use in flood warning procedures, within which people receive flood warnings Flow Routing (or Flood Routing) – a technique used to compute the movement and change of shape of a flood wave moving through a river reach or a reservoir (UNESCO/WMO 2007) Forecasting Point – a location at which it is useful to have a forecast of future river or coastal conditions (e.g. a Flood Warning Area, a river or coastal monitoring site, a control structure) Freeboard – vertical distance between the normal maximum level of the surface of a liquid in a conduit, reservoir, tank, canal, etc., and the top of the sides of the retaining structure (UNESCO/WMO 2007)

G Geographical Information System – computer software for the graphical presentation and analysis of spatial datasets and the associated hardware, procedures, equipment etc. Glacial Lake Outburst Flood – a flood caused by the sudden release of water from a lake formed by moraine, ice or similar

Glossary

271

H Hurricane – see Tropical Cyclone Hydrodynamic Model – a solution to the equations expressing mass, momentum and energy conservation of water, sediment, heat and other parameters in a river, estuary or coastal reach Hydrograph – graph showing the variation in time of some hydrological data such as stage, discharge, velocity, sediment load, etc. (UNESCO/WMO 2007)

I Ice Jam – accumulation of ice at a given location which, in a river, restricts the flow of water (UNESCO/WMO 2007) Initial Conditions – values of storage or pressure variables required to initialize a model at the start of a simulation period (Beven 2001) Intangible Losses – losses which cannot easily be expressed in economic terms, including impacts on health, business disruption, stress, impacts on tourism etc. Isochrone Map – map or chart of a drainage basin in which a series of lines (isochrones) gives the times of travel of water originating on each isochrone to reach the outlet of the basin (UNESCO/WMO 2007)

K Kalman Filter – a time series analysis technique which seeks to provide an improved forecast of future conditions accounting for differences between previous observations and forecasts. Also extended Kalman Filter and ensemble Kalman Filter variants

L Lead Time – warning lead time is the time between receipt of a flood warning and the time of the onset of flooding; forecast lead time is the maximum lead time at which forecasts can be provided to an acceptable accuracy Levee – see Flood Defence

M Monte Carlo Simulation – simulation involving multiple runs of a model using different randomly chosen sets of parameter values or boundary conditions (Beven 2001) Multi-Criteria Analysis (MCA) – a structured decision making technique widely used for evaluating alternative options where multiple criteria and priorities are involved, perhaps including social, environmental, financial, political and other factors

272

Glossary

N Nowcast – a meteorological modelling technique which combines the outputs from weather radar observations and possibly Numerical Weather Prediction model outputs to produce short term (typically 0–6 hour ahead) forecasts of rainfall and other parameters Numerical Weather Prediction (NWP) – computer modelling technique in which the atmosphere, oceans and land surface are modelled on a three dimensional grid to produce forecasts of future conditions based on data assimilated from a wide range of sources (ground based observations, satellite, ships, aircraft etc.)

O Objective Function – a measure of how well a simulation fits the available observations (Beven 2001) Orographic Precipitation – precipitation caused by the ascent of moist air over orographic barriers (UNESCO/WMO 2007)

P Parameter – a constant that must be defined before running a simulation (Beven 2001) Polder – a mostly low-lying area artificially protected from surrounding water and within which the water table can be controlled (UNESCO/WMO 2007) Process Based Model – models which to varying degrees solve the partial differential equations representing catchment and coastal processes, typically on a gridded basis, perhaps including empirical or conceptual representations for some components of the model Public Switched Telephone Network (PSTN) – the telecommunications equipment and infrastructure which connects land line telephones

Q Quantitative Precipitation Forecasts – precipitation (rainfall, snow, hail etc.) forecasts typically based on nowcasting or Numerical Weather Prediction techniques

R Rainfall Runoff Model – a model which converts observed or forecast rainfall into estimated river flows Rating Curve –see Stage Discharge relationship Real Time Updating – see Data Assimilation River Gauging Station – a measuring location where observations of water level (river, reservoir) and discharge are made

Glossary

273

Resilience – the capacity of a system, community or society potentially exposed to hazards to adapt, by resisting or changing in order to reach and maintain an acceptable level of functioning and structure. This is determined by the degree to which the social system is capable of organizing itself to increase its capacity for learning from past disasters for better future protection and to improve risk reduction measures (UN/ISDR 2004)

S Saffir/Simpson – five categories indicating the damage potential of tropical cyclones (Holland et al. 2007) Set-Up – water forced inshore by breaking waves (Holland et al. 2007) Situation Report – a brief report that is published and updated periodically during a relief effort and which outlines the details of the emergency, the needs generated and the responses undertaken by all donors as they become known (IDNDR 1992) Snow Pillow – device filled with antifreeze solution and fitted with a pressure sensor which indicates the water equivalent of the snow cover (UNESCO/WMO 2007) Soil Moisture Deficit (SMD) – a state variable used in many hydrological models as an expression of water storage. SMD is zero when the soil is at field capacity and gets larger as the soil dries out. It is usually expressed in units of depth of water (Beven 2001) Stage Discharge Relationship – or Stage Discharge Relation – relation between stage and discharge at a river cross section and which may be expressed as a curve, table or equation(s) (UNESCO/WMO 2007) Stochastic – a model is stochastic if, for a given set of initial and boundary conditions, it may have a range of possible outcomes, often with each outcome associated with an estimated probability (Beven 2001) Surge – or Storm Surge –a sudden rise of sea as a result of high winds and low atmospheric pressure; sometimes called a storm tide, storm wave, or tidal wave. Generally affects only coastal areas but may intrude some distance inland (IDNDR 1992) Swell – smooth, regularly spaced waves that have propagated long distances from their initial generation region (Holland et al. 2007)

T Threshold – the meteorological, river or coastal conditions or forecasts which initiate (or escalate) the flood warning dissemination process. Sometimes called triggers, criteria, warning levels, alert levels or alarms Trigger – see Threshold

274

Glossary

Tropical Cyclone – a synoptic-scale to mesoscale low pressure system which derives its energy primarily from evaporation from the sea in the presence of high winds and low surface pressure and condensation in convective clouds concentrated near its center (Holland et al. 2007), usually for maximum sustained surface winds of 64 knots (33 m s−1) or more (although sometimes defined for winds of 34 knots or more). The term tropical cyclone is used in the Indian Ocean, hurricane in the Atlantic and Eastern Pacific Oceans, and typhoon in the Western Pacific Tsunami – a series of large waves generated by sudden displacement of seawater (caused by earthquake, volcanic eruption or submarine landslide); capable of propagation over large distances and causing a destructive surge on reaching land (IDNDR 1992) Typhoon – see Tropical Cyclone

U Ungauged Catchment – a catchment or subcatchment in which flows are not recorded to the extent required for the application (e.g. in real time for flood forecasting applications)

V Vulnerability – the conditions determined by physical, social, economic, political and environmental factors or processes, which increase the susceptibility of a community to the impact of hazards

W Wadi – or Ouedd – channel which is dry except in the rainy season (UNESCO/ WMO 2007) Watershed – see Catchment Wave – disturbance in a body of water propagated at a constant or varying speed (celerity), often of an oscillatory nature, accompanied by the alternate rise and fall of surface fluid particles (UNESCO/WMO 2007) Weather Radar – an instrument for detecting cloud and precipitation using microwaves typically with wavelengths in the range 3–10 cm Wind Waves – choppy and chaotic waves generated locally by the wind (Holland et al. 2007)

References

Chapter 1

Introduction

Alexander D (2002) Principles of Emergency Planning and Management. Oxford University Press, Oxford Andryszewski A, Evans K, Haggett C, Mitchell B, Whitfield D, Harrison T (2005) Levels of Service Approach to Flood Forecasting and Warning. ACTIF international conference on innovation advances and implementation of flood forecasting technology, 17–19 October 2005, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Basher R (2006) Global early warning systems for natural hazards: systematic and people-centred. Phil. Trans. R. Soc. A 364: 2167–2182. doi:10.1098/rsta.2006.1819 Beven K J (2008) Environmental Modeling: An Uncertain Future. Routledge, London Drabek T E (2000) The social factors that constrain human responses to flood warnings. In Parker D J (Ed.), Floods. Routledge, London Emergency Management Australia (1999) Guide 5 – Flood Warning. Australian Emergency Manuals Series, Part III Emergency Management Practice, Volume 3 (2nd edition). http:// www.ema.gov.au/. Accessed 15 January 2008 FEMA (2005) Reducing Damage from Localized Flooding: A Guide for Communities. Chapter 6: Warning and Emergency Services, Report 511-07. http://www.fema.gov/hazard/flood/pubs/ flood-damage.shtm. Accessed 15 January 2008 Fortune D (2006) Floods and Hydroinformatics: New Challenges. 7th International Conference on Hydroinformatics, HIC 2006, Nice, France, Volume 1, pp. 122–127 Guha-Sapir D, Hargitt D, Hoyois P (2004) Thirty Years of Natural Disasters 1974–2003: The Numbers. Centre for Research on the Epidemiology of Disasters (CRED) Report. UCL Presse Universitaires de Louvain. http://www.cred.be/cred1/publicat/online.htm. Accessed 15 January 2008 Handmer J (2002) Flood warning reviews in North America and Europe: statements and silence. Australian Journal of Emergency Management, 17(3): 17–24. http://www.ema.gov.au/agd/ EMA/emaInternet.nsf/Page/Publications. Accessed 15 January 2008 Henson R (2001) U.S. flash flood warning dissemination via radio and television. In Gruntfest E, Handmer J (Eds.), Coping with Flash Floods. Kluwer, The Netherlands, pp. 243–252 Holland G (Ed.) (2007) Global Guide to Tropical Cyclone Forecasting. Bureau of Meteorology Research Centre (Australia) WMO/TC-No. 560, Report No. TCP-31, World Meteorological Organization; Geneva, Switzerland. http://www.bom.gov.au/bmrc/pubs/tcguide/globa_guide_ intro.htm. Accessed 15 January 2008 Howard C D D (2004) Ensemble Optimization for Hydroelectric Operations. Workshop on the Hydrological Ensemble Prediction Experiment (HEPEX), ECMWF, 8–10 March. http:// ecmwf.int/newsevents/meetings/workshops/2004/HEPEX/presentations.html. Accessed 15 January 2008

275

276

References

Huaimin G (2005) Flood Forecasting and Management in China. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 ISDR (2006) Developing Early Warning Systems: A Checklist. EWC III Third International Conference on Early Warning from Concept to Action, 27–29 March, Bonn, Germany. http:// www.unisdr.org/ppew/. Accessed 15 January 2008 Krzysztofowicz R (2001) The case for probabilistic forecasting in hydrology. J. Hydrol. 249: 2–9 Kundzewicz Z W, Jun X (2004) Editorial – towards an improved flood preparedness system in China. Hydrolog. Sci. J. 49(6): 941–944 Martini F, De Roo A (Eds.) (2007) EXCIFF Guide: Good Practice for Delivering Flood Related Information to the General Public. European Commission/Joint Research Centre Report EUR22760EN. http://exciff.jrc.it/downloads/exciff-related-documents/EXCIFF_guide.pdf. Accessed 15 January 2008 National Research Council (2006) Completing the Forecast: Characterizing and Communicating Uncertainty for Better Decisions Using Weather and Climate Forecasts. National Academies Press, Washington, DC. http://www.nap.edu/catalog.php?record_id=11699. Accessed 15 January 2008 Pappenberger F, Beven K J (2006) Ignorance is bliss: or seven reasons not to use uncertainty analysis. Water Resour. Res. 42(5) doi:10.1029/2005WR004820 Pappenberger F, Beven K, Frodsham K, Romanowicz R, Matgen P (2007) Grasping the unavoidable subjectivity in calibration of flood inundation models: a vulnerability weighted approach. J. Hydrol. 333(2–4): 275–287 Parker D J (Ed.) (2000) Floods. Volumes 1 and 2. Routledge, London/New York Parker D J (2003) Designing Flood Forecasting, Warning and Response Systems from a Societal Perspective. Paper Presented at the International Conference on Alpine Meteorology and Meso-Alpine Programme, 21 May 2003, Brig, Switzerland. http://www.map.meteoswiss.ch/ map-doc/icam2003/Presentation/10.1/Brig_document.pdf. Accessed 15 January 2008 Sayers P B, Hall J W, Meadowcroft I C (2002) Towards Risk-Based Flood Hazard Management in the UK. Proc. ICE 150: 36–42 Smith K (2004) Environmental Hazards. Assessing Risk and Reducing Disaster (4th edition). Routledge, London/New York Smith K, Ward R (1998) Floods: Physical Processes and Human Impacts. Wiley, Chichester, UK Thieken A H, Kreibich H, Müller M, Merz B (2007) Coping with floods: preparedness, response and recovery of flood-affected residents in Germany in 2002. Hydrolog. Sci. J. 52(5): 1016–1037 Todini E, Alberoni P, Butts M, Collier C, Khatibi R, Samuels P, Weerts A (2005) ACTIF Best Practice Paper – Understanding and Reducing Uncertainty in Flood Forecasting. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October, Tromsø, Norway. http://www.actif-ec.net/conference2005/ proceedings/index.html. Accessed 15 January 2008 UNESCO (2006) Tsunami teacher: an information and resource toolkit, building global capacity to respond to and mitigate tsunamis. IOC Manuals and Guides, 47. http://unesdoc.unesco.org/ images/0014/001496/149617e.pdf. Accessed 15 January 2008 United Nations (2006a) Global survey of early warning systems. An Assessment of Capacities, Gaps and Opportunities Towards Building a Global Early Warning System for All Natural Hazards. United Nations, Geneva, Switzerland. http://www.unisdr.org/ppew/. Accessed 15 January 2008 United Nations (2006b) Guidelines for reducing flood losses. International Strategy for Disaster Reduction. http://www.unisdr.org/eng/library/isdr-publication/flood-guidelines/isdr-publicationfloods.htm. Accessed 15 January 2008 U.S. Army Corps of Engineers (USACE) (1996) Hydrologic Aspects of Flood WarningPreparedness Programs. Report ETL 1110-2-540, Office of Chief of Engineers, Washington,

References

277

DC. http://www.tpub.com/content/USACETechnicalletters/ETL-1110-2-540/index.htm. Accessed 15 January 2008 Wade S D, Ramsbottom D, Floyd P, Penning Rowsell E, Surendran S (2005) Risks to People: Developing New Approaches for Flood Hazard and Vulnerability Mapping. 40th Defra Flood and Coastal Management Conference, York, England World Meteorological Organisation (1994) Guide to Hydrological Practices (5th edition). WMO No. 168, Geneva, Switzerland World Meteorological Organisation (2006a) Strategy and Action Plan for the Enhancement of Cooperation Between National Meteorological and Hydrological Services for Improved Flood Forecasting. WMO, Geneva, Switzerland. http://www.wmo.ch/pages/prog/hwrp/documents/ FFInitiativePlan.pdf. Accessed 15 January 2008 World Meteorological Organisation (2006b) Preventing and Mitigating Natural Disasters. WMO No. 993, Geneva World Meteorological Organisation (2006c) Social Aspects and Stakeholder Involvement in Integrated Flood Management. WMO No. 1008, Geneva, Switzerland

Chapter 2

Detection

Beneviste J, Berry P (2004) Monitoring river and lake levels from space. ESA Bulletin 117, July 2004. http://www.esa.int/esapub/bulletin/bullet117/chapter5_bul117.pdf. Accessed 15 January 2008 Bowler N E, Pierce C E, Seed A W (2006) STEPS: a probabilistic precipitation forecasting scheme which merges an extrapolation nowcast with downscaled NWP. Quart. J. R. Meteorol. Soc.132: 2127–2155 Cluckie I D, Rico-Ramirez M A (2004) Weather radar technology and future developments. GIS and Remote Sensing in Hydrology, Water Resources and Environment (Proceedings of ICGRHWE held at the Three Gorges Dam, China, September 2003). IAHS Publications, 289 Collier C G (1996) Applications of Weather Radar Systems: A Guide to Uses of Radar Data in Meteorology and Hydrology. 2nd edition. Wiley, Chichester Creutin J D, Obled C (1982) Objective analyses and mapping techniques for rainfall fields: an objective comparison. Water Resour. Res. 18(2): 413–431 Creutin J D, Muste M, Bradley A A, Kim S C, Kruger A (2003) River gauging using PIV techniques: a proof of concept experiment on the Iowa river. J. Hydrol. 277(3–4): 182–194 Crow W T, Bindlish R, Jackson T J (2004) Improving Flood Forecasting Via the Integration of Remotely Sensed Soil Moisture Observations Derived from Passive Spaceborne Observations. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October 2005, Tromsø, Norway. http://www.actif-ec.net/ conference2005/proceedings/index.html. Accessed 15 January 2008 Environment Agency (2002) Fluvial Flood Forecasting for Flood Warning - Real Time Modelling. Defra/Environment Agency Flood and Coastal Defence R&D Programme. R&D Technical Report W5C-013/5/TR Environment Agency (2004) Best Practice in Coastal Flood Forecasting. Defra/Environment Agency Flood and Coastal Defence R&D Programme, R&D Technical Report FD2206/TR1 Environment Agency (2007) Probabilistic Flood Forecasting Scoping Study. Joint Defra/ Environment Agency Flood and Coastal Erosion Risk Management R&D Programme, R&D Technical Report FD2901/TR Franklin J L, McAdie C J, Lawrence M B (2003) Trends in track forecasting for tropical cyclones threatening the United States 1970–2001. Bull. Am. Meteorol. Soc. 1197–1203. http://www. nhc.noaa.gov/pdf/03franklin.pdf. Accessed 15 January 2008 Garcia Moya J A, Callado A, Santos C, Santos Muñoz D, Simarro J (2006) Multi-model Ensemble for Short Range Predictability. Proceedings of the 1st MAP D-Phase Scientific Meeting, 6–8

278

References

November 2006, Vienna. http://www.univie.ac.at/IMG-Wien/meetings/map_d-phase/proceedings. pdf. Accessed 15 January 2008 Golding B W (2000) Quantitative precipitation forecasting in the UK. J. Hydrol. 239: 286–305 Goovaerts P (2000) Geostatistical approaches for incorporating elevation into the spatial interpolation of rainfall. J. Hydrol. 228: 113–129 Grimes D I F, Coppola E, Verdecchia M, Visconti G (2003) A neural network approach to real time rainfall estimation for Africa using satellite data. J. Hydrometeorol. 4: 1119–1133 Grimes D I F, Diop M (2003) Satellite based rainfall estimation for river flow forecasting in Africa. I: rainfall estimates and hydrological forecasts. Hydolog. Sci. J. 48(4): 567–584 Harrison D, Scovell R, Lewis H, Matthews S (2006) The Development of the EUMETNET OPERA Radar Data Hub. Proceedings of ERAD 2006. http://www.grahi.upc.es/ERAD2006/ index.php. Accessed 15 January 2008 Hershey R W (1999) Hydrometry: Principles and Practices. Wiley, New York Holt A R, Collier C G, Upton G J G, Rahimi A R, Cummings R J, Robbins G L (2005) The Use of Dual-Frequency Microwave Links for Measuring Path-Averaged Rain. Paper for the InterAgency Committee on the Hydrological Applications of Weather Radar Hughes D, Greenwood P, Blair G, Coulson G, Pappenberger F, Smith P, Beven K (2006) An intelligent and adaptable grid-based flood monitoring and warning system. UK eScience All Hands Meeting Intergovernmental Oceanographic Commission (1994) Manual on sea level measurement and interpretation. Volume II - Emerging Technologies, UNESCO International Standards Organisation (1996, 1998) Measurement of Liquid Flow in Open Channels. Part 1: Establishment and operation of a gauging station and Part 2: Determination of the stage-discharge relation. ISO 1100-1, 1100-2, Geneva Leijnse H R, Uijlenhoet R, Holleman I (2008) Sources of error in microwave link rainfall estimation. International Symposium on Weather Radar and Hydrology, Grenoble, France Love T B (2006) An Overview of Satellite-Based Precipitation Estimation for Flash Flood Modeling and Forecasting. International Workshop on Flash Flood Forecasting, 13–17 March 2006, Costa Rica, CA. http://www.nws.noaa.gov/iao/iao_FFW.php. Accessed 15 January 2008 Massel S R (1996) Ocean Surface Waves; Their Physics and Prediction. World Scientific Publisher, Singapore/NJ/London/Hong Kong Meischner P (2005) Weather Radar: Principles and Advanced Applications. Series: Physics of Earth and Space Environments. Springer, New York. Moore R J, Jones A E, Jones D A, Black K B, Bell V A (2004) Weather Radar for Flood Forecasting: Some UK Experiences. Sixth International Symposium on Hydrological Applications of Weather Radar, Melbourne, Australia, 2–4 February 2004 Nemeth K (2006) Enhanced Precipitation Identifier and New Generation of Present Weather Sensor. International Workshop on Flash Floods Forecasting, 13–17 March 2006, Costa Rica, CA. http://www.nws.noaa.gov/iao/iao_FFW.php. Accessed 15 January 2008 NOAA/NWS (1997) Automated Local Flood Warning Systems Handbook, Weather Service Hydrology Handbook No. 2. National Oceanic and Atmospheric Administration, National Weather Service, Office of Hydrology, Washington DC. http://www.weather.gov/oh/docs/ alfws-handbook/. Accessed 15 January 2008 Oi H, Opadevi J (2006) Development of Equipment for Community Flash Flood Warning in the Caribbean and the Central America. International Workshop on Flash Flood Forecasting, 13–17 March 2006, Costa Rica, CA. http://www.nws.noaa.gov/iao/iao_FFW.php. Accessed 15 January 2008 Park J, Oh B, Jeon S, Kim J (2006) The Automatic Discharge Acquisition & Management System (ADAMS) in Korea. 7th International Conference on Hydroinformatics, HIC 2006, Nice, France Pedersen L, Jensen N E, Madsen H (2007) Network Architecture for Small X-Band Weather Radars - Test Bed for Automatic Intercalibration and Nowcasting. Paper 12B.2, 33rd Conference on Radar Meteorology, Cairns, Australia. http://ams.confex.com/ams/33Radar/ techprogram/programexpanded_423.htm. Accessed 15 January 2008

References

279

Price C, Yair Y, Mugnai A, Lagouvardos K, Llasat M S, Michaelides S (2007) FLASH: a new EU project related to Mediterranean flash floods. Geophys. Res. Abstracts 9, 02638. http://www. copernicus.org/EGU/gra/volume_9.html. Accessed 15 January 2008 Rahimi A R, Holt AR, Upton G J G, Cummings R J (2003) Use of dual-frequency microwave links for measuring path-averaged rainfall. J. Geophys. Res. 108(D15): 4467 Rebora N, Ferraris L, von Hardenberg J, Provenzale A (2006) Rainfall downscaling and flood forecasting: a case study in the Mediterranean area. Nat. Hazards Earth Syst. Sci. 6: 611–619 Rohr P C, Husebye S (2005) Flood Forecasting Technology in Norway, Innovation and Advances. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, Tromsø, Norway, 17–19 October 2005. http://www.actif-ec.net/ conference2005/proceedings/index.html. Accessed 15 January 2008 Rotach M W, Arpagaus M, Dorninger M, Hegg C, Montani A, Ranzi R (2007) The D-PHASE Operations Period (DOP). 3rd HEPEX Workshop, 27–29 June 2007, Stresa, Italy Schaake J, Restrepo P, Seo D J, Demargne J, Limin W, Perica S (2005) Hydrological Ensemble Prediction: Challenges and Opportunities. ACTIF International conference on innovation advances and implementation of flood forecasting technology, Tromsø, Norway, 17–19 October 2005. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Schaefer G L, Paetzold R F (2000) SNOTEL (SNOwpack TELemetry) and SCAN (Soil Climate Analysis Network). Paper Presented at “Automated Weather Stations for Applications in Agriculture and Water Resources Management: Current Use and Future Perspectives”, March 2000, Lincoln, NB Sene K J., Gash J H C, McNeil D (1991) Evaporation from a tropical lake; comparison of theory with direct measurements. J. Hydrol. 127: 193–217 Sene K J, Jones D A, Meigh J R, Farquharson F A K (1998) Rainfall and flow variations in the Lesotho Highlands. Int. J. Climatol. 18: 329–345 Sene K J, Ainsworth E J, Graham G (2006) Risk Based Design of Telemetry Networks for Flood Warning Applications. 41st Defra/Environment Agency Flood and Coastal Management Conference, July 2005, York, England Stow D, Bradley S G, Farrington K E, Dirks K N, Gray W R (1998) A Rain Gauge for the Measurement of Finescale Temporal Variations. J. Atmos. Ocean. Tech. 15(1): 127–135 Strangeways I (2007) Precipitation: Theory, Measurement and Distribution. Cambridge University Press, Cambridge Todini E (2001) A Bayesian technique for conditioning radar precipitation estimates to rain-gauge measurements. Hydrol. Earth Syst. Sci. 5(2): 187–199. http://www.hydrol-earth-syst-sci. net/5/187/2001/hess-5-187-2001.html. Accessed 15 January 2008 U.S. Army Corps of Engineers USACE (1996) Hydrologic Aspects of Flood WarningPreparedness Programs. Report ETL 1110-2-540, Office of Chief of Engineers, Washington, DC. http://www.tpub.com/content/USACETechnicalletters/ETL-1110-2-540/index.htm. Accessed 15 January 2008 Wilson J W (2004) Precipitation Nowcasting: Past, Present and Future. Sixth International Symposium on Hydrological Applications of Weather Radar, Melbourne, Australia. http:// www.bom.gov.au/bmrc/basic/old_events/hawr6/qpf/WILSON_KEYNOTE.pdf. Accessed 15 January 2008 World Meteorological Organization (1980) Manual on Stream Gauging. Volumes I and II, Operational Hydrology Report No. 13. WMO No. 519, Geneva World Meteorological Organisation (1994a) An Overview of Selected Techniques for Analysing Surface Water Data Networks. WMO No. 806, Geneva World Meteorological Organisation (1994b) Guide to Hydrological Practices. 5th edition. WMO No. 168, Geneva World Meteorological Organisation (1998) Guide to Wave Analysis and Forecasting. 2nd edition. WMO No. 702, Geneva World Meteorological Organisation (2000) Precipitation Estimation and Forecasting. Operational Hydrology Report No. 46, Geneva

280

References

World Meteorological Organisation (2003) The Role of Satellites in WMO programmes in the 2010s. WMO Technical Document No. 1177, Geneva World Meteorological Organisation (2005) WHYCOS Guidelines. WMO/TD No. 1282, Geneva. http://www.whycos.org/IMG/pdf/TD.pdf. Accessed 15 January 2008 Xu K, Zhang J, Watanabe M, Sun C (2004) Estimating river discharge from very high-resolution satellite data: a case study in the Yangtze River, China. Hydrol. Process. 18(10): 1927–1939 Zakharova E A, Kouraev A V, Cazenave A, Seyler F (2005) Amazon discharge estimated from TOPEX/Poseidon altimetry. C.R. Geosci. 338: 188–196. doi:10.1016/j.crte.2005.10.003

Chapter 3

Thresholds

Bernard P (2004) Hydrometeorological Forecasts for Hydropower Generation. Workshop on the Hydrological Ensemble Prediction Experiment (HEPEX), Reading, March 8–10. http://ecmwf.int/ newsevents/meetings/workshops/2004/HEPEX/presentations.html. Accessed 15 January 2008 Collier C G, Rihan F, Davies F, Robbins G L (2005) On the Challenges Arising from the Use of High Resolution NWP Models to Forecast Heavy Rain. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Collier C G, (2007) Flash flood forecasting: what are the limits of predictability? Q.J.R. Meteorol. Soc. 133: 3–23 Cox P M, Betts R A, Bunton C B, Essery R L H, Rowntree P R, Smith J (1999) The impact of new land surface physics on the GCM simulation of climate and climate sensitivity. Clim. Dynam. 15: 183–203 Environment Agency (2002) Fluvial Flood Forecasting for Flood Warning - Real Time Modelling. Defra/Environment Agency Flood and Coastal Defence R&D Programme. R&D Technical Report W5C-013/5/TR Environment Agency (2004a) Best Practice in Coastal Flood Forecasting. Defra/Environment Agency Flood and Coastal Defence R&D Programme, R&D Technical Report FD2206/TR1 Environment Agency (2004b) Performance Measures for Flood Forecasting: Review and Recommendations. Defra/Environment Agency Flood and Coastal Defence R&D Programme, R&D Technical Report W5C-021/2b/TR Environment Agency (2007) Extreme Event Recognition Phase 2. Joint Defra/Environment Agency Flood and Coastal Erosion Risk Management R&D Programme, R&D Technical Report FD2208/TR Fouchier C, Arnaud P, Lavabre J, Mizzi J P (2007) AIGA: an operational tool for flood warning in southern France. Principle and performances on Mediterranean flash floods. Geophys. Res. Abstracts 9, 02843. http://www.copernicus.org/EGU/gra/volume_9.html. Accessed 15 January 2008 Georgakakos K P (2005) Modern Operational Flash Flood Warning Systems Based on Flash Flood Guidance Theory: Performance Evaluation. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, Tromsø, Norway, 17–19 October. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Georgakakos K P (2006) Analytical results for operational flash flood guidance. J. Hydrol. 317: 81–103 Graham G, Johnson T (2007) Simple Triggers for Flood Warning. 42nd Defra Flood and Coastal Management Conference, July 2007, York, England Jolliffe I T, Stephenson D B (2003) Forecast Verification: A Practitioner’s Guide in Atmospheric Science. Wiley, Chichester, UK, pp. 240 Kok C J, Vogelezang D H P, Wichers Schreur B, Holleman I (2008) Description and use of the automated warning system for the Dutch water boards. Royal Netherlands Meteorological Institute (KNMI), Netherlands

References

281

Martina M L V, Todini E, Libralon A (2006) A Bayesian decision approach to rainfall thresholds based flood warning. Hydrol. Earth Syst. Sci. 10: 413–426. http:// www.hydrol-earth-syst-sci. net/10/413/2006/. Accessed 15 January 2008 National Weather Service (2003) Flash Flood Guidance Improvement Team. Final Report: 6 February, Washington, DC. http://www.nws.noaa.gov/oh/rfcdev/docs/ffgitreport.pdf. Accessed 15 January 2008 NOAA-USGS (2005) NOAA-USGS Debris-Flow Warning System. Final Report: U.S. Geological Survey Circular 1283. http://pubs.usgs.gov/circ/2005/1283/. Accessed 15 January 2008 Obled C, Datin R (1997) Rainfall Information Requirements for Mediterranean Flood Operational Forecasts. In Proc. Workshop on River Basin Modelling, Management and Flood Mitigation, Padua, Italy, 25–26 September Price C, Yair Y, Mugnai A, Lagouvardos K, Llasat M C, Michaelides S (2007) FLASH: A New EU Project Using Lightning Data to Study Mediterranean Flash Floods. 13th International Conference on Atmospheric Electricity, August 13–18, Beijing Reed S, Schaake J, Zhangl Z (2006) A Statistical-Distributed Hydrologic Model for Flash Flood Forecasting. International Workshop on Flash Flood Forecasting. 13–17 March, San Jose, Costa Rica. http://www.nws.noaa.gov/iao/iao_FFW.php. Accessed 15 January 2008 Stanski H R, Wilson L J, Burrows W R (1989) Survey of Common Verification Methods in Meteorology. World Weather Watch Technical Report No.8, WMO/TD No.358, World Meteorological Organisation, Geneva, Switzerland Sweeney T L (1992) Modernized Areal Flash Flood Guidance. NOAA Technical Report NWS HYDRO 44, Hydrology Laboratory, National Weather Service, NOAA, Silver Spring, MD Tissot P E, Duff S, Michaud P, Rizzo J, Jeffress G (2005) DNR-TCOON: An Integrated Observation and Operational System for the Gulf of Mexico. American Meteorological Society’s (AMS) 85th Annual Meeting, 9–13 January, San Diego, CA. http://ams.confex.com/ ams/Annual2005/techprogram/paper_83842.htm. Accessed 15 January 2008 Torfs P J J F (2004) Black Box Forecasting and Uncertainty. ACTIF Workshop on Identification, Quantification, Propagation and Communication of Uncertainties in Flood Forecasting. 22–23 November, Delft, The Netherlands. http://www.actif-ec.net/news.htm. Accessed 15 January 2008 U.S. Army Corps of Engineers USACE (1996) Hydrologic Aspects of Flood WarningPreparedness Programs. Report ETL 1110–2–540, Office of Chief of Engineers, Washington, DC. http://www.tpub.com/content/USACETechnicalletters/ETL-1110–2–540/index.htm. Accessed 15 January 2008 World Meteorological Organisation (1994) Guide to Hydrological Practices (5th edition). WMO No. 168, Geneva, Switzerland

Chapter 4

Dissemination

Andryszewski A, Evans K, Haggett C, Mitchell B, Whitfield D, Harrison T (2005a) Levels of Service Approach to Flood Forecasting and Warning. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology. 17–19 October 2005, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Andryszewski A, Evans K, Haggett C, Mitchell B, Whitfield D (2005b) Flood Warning – The Next Step Change. 40th Defra Flood and Coastal Management Conference, York, England Emergency Management Australia (1999) Guide 5 – Flood Warning. Australian Emergency Manuals Series, Part III Emergency Management Practice. Volume 3, 2nd edition. http://www. ema.gov.au/. Accessed 15 January 2008 Environment Agency (2002) Fluvial Flood Forecasting for Flood Warning - Real Time Modelling. Defra/Environment Agency Flood and Coastal Defence R&D Programme. R&D Technical Report W5C-013/5/TR

282

References

Environment Agency (2004) Best Practice in Coastal Flood Forecasting. Defra/Environment Agency Flood and Coastal Defence R&D Programme, R&D Technical Report FD2206/TR1 Holland G (Ed.) (2007) Global Guide to Tropical Cyclone Forecasting. Bureau of Meteorology Research Centre (Australia) WMO/TC-No. 560, Report No. TCP-31, World Meteorological Organization, Geneva, Switzerland. http://www.bom.gov.au/bmrc/pubs/tcguide/globa_guide_ intro.htm. Accessed 15 January 2008 ISDR (2006) Developing Early Warning Systems: A Checklist. EWC III Third International Conference on Early Warning from Concept to Action, 27–29 March 2006, Bonn, Germany. http://www.unisdr.org/ppew/. Accessed 15 January 2008 Martini F, De Roo A (Eds.) (2007) EXCIFF Guide: Good Practice for Delivering Flood Related Information to the General Public. European Commission/Joint Research Centre Report EUR22760EN. http://exciff.jrc.it/downloads/exciff-related-documents/EXCIFF_guide.pdf. Accessed 15 January 2008 Morel G, Despres S (2006) ISLAND: Improving the Communication of Information on Flooding and Other Risks to Local Communities in South East Asia. http://www.isted.com/programmes/ island/anglais/documentation.htm. Accessed 15 January 2008 NOAA/NWS 1997 Automated Local Flood Warning Systems Handbook. Weather Service Hydrology Handbook No. 2. National Oceanic and Atmospheric Administration, National Weather Service, Office of Hydrology, Washington DC. http://www.weather.gov/oh/docs/ alfws-handbook/. Accessed 15 January 2008 Oi H, Opavedi J (2006) Development of Equipment for Community Flash Flood Warning in the Caribbean and the Central America. WMO International Workshop on Flash Flood Forecasting, 13–17 March 2006, Costa Rica, CA. http://www.nws.noaa.gov/iao/iao_FFW.php. Accessed 15 January 2008 Parker D J (2003) Designing Flood Forecasting, Warning and Response Systems from a Societal Perspective. Paper presented at the International Conference on Alpine Meteorology and Meso-Alpine Programme, 21 May 2003, Brig, Switzerland. http://www.map.meteoswiss.ch/ map-doc/icam2003/Presentation/10.1/Brig_document.pdf. Accessed 15 January 2008 Sponberg K (2006) RANET Dissemination and Communication of Environmental Information for Rural and Remote Community Development. WMO International Workshop on Flash Flood Forecasting, 13–17 March 2006, Costa Rica, CA. http://www.nws.noaa.gov/iao/iao_FFW.php. Accessed 15 January 2008 Tilford K A, Sene K J, Khatibi R (2007) Flood forecasting model selection: a structured approach. In Begum S, Stive M J F, Hall J W (Eds.), Flood Risk Management in Europe. Springer, Dordrecht, The Netherlands UNESCO (2006) Tsunami Teacher: An Information and Resource Toolkit, Building Global Capacity to Respond to and Mitigate Tsunamis. IOC Manuals and Guides 47. http://unesdoc. unesco.org/images/0014/001496/149617e.pdf. Accessed 15 January 2008 U.S. Army Corps of Engineers (USACE) (1996) Hydrologic Aspects of Flood Warning-Preparedness Programs. Report ETL 1110–2–540, Office of Chief of Engineers, Washington, DC. http://www.tpub. com/content/USACETechnicalletters/ETL-1110–2–540/index.htm. Accessed 15 January 2008 World Meteorological Organisation (1994) Guide to Hydrological Practices. 5th edition. WMO No. 168, Geneva World Meteorological Organisation (2006) Preventing and Mitigating Natural Disasters. WMO No. 993, Geneva

Chapter 5

General Principles

Anderson M G, Bates P D (Eds.) (2001) Model Validation: Perspectives in Hydrological Science. Wiley, Chichester, UK Beven K J (2001) Rainfall Runoff Modelling – The Primer. Wiley, Chichester, UK

References

283

Butts M B, Falk A K, Hartnack J, Madsen H, Klinting A, van Kalken T, Cadman D, Price D (2005) Ensemble Based Methods for Data Assimilation and Uncertainty Estimation in the FLOODRELIEF Project. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October, Tromsø, Norway. http:// www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Carsell K M, Pingel N D, Ford D T (2004) Quantifying the Benefit of a Flood Warning System. Natural Hazards Review, ASCE, August 2004: 131–140 Chen Y, Sene K J, Hearn K (2005) Converting Section 105 or SFRM Hydrodynamic River Models for Real Time Forecasting Applications. 40th Defra Flood and Coastal Defence Conference, York, England Cranston M, Werner M, Janssen A, Hollebrandse F, Lardet P, Oxbrow J, Piedra M (2007) Flood Early Warning System (FEWS) Scotland: An Example of Real Time System and Forecasting Model Development and Delivery Best Practice. 42nd Defra Flood and Coastal Defence Conference, York, England El Serafy G Y, Mynett A E (2004) Comparison of EKF and ENKF in Sobek River: Case Study Maxau ljssel. HYDROINFORMATICS Vol. 1, Proceedings of the 6th International Conference, Singapore, 21–24 June 2004 Elliott J, Catchlove R, Sooriyakumaran S, Thompson R (2005) Recent Advances in the Development of Flood Forecasting and Warning Services in Australia. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, Tromso, Norway, 17–19 October2005. http://www.actif-ec.net/conference2005/proceedings/ index.html. Accessed 15 January 2008 Environment Agency (2002) Fluvial Flood Forecasting for Flood Warning - Real Time Modelling. Defra/Environment Agency Flood and Coastal Defence R&D Programme. R&D Technical Report W5C-013/5/TR Environment Agency (2004) Performance Measures for Flood Forecasting: Review and Recommendations. Defra/Environment Agency Flood and Coastal Defence R&D Programme, R&D Technical Report W5C-021/2b/TR Environment Agency (2007) Probabilistic Flood Forecasting Scoping Study. Joint Defra/EA Flood and Coastal Erosion Risk Management R&D Programme, R&D Technical Report FD2901/TR Fortune D (2006) Floods and Hydroinformatics: New Challenges. 7th International Conference on Hydroinformatics, HIC 2006, Nice, France, Volume 1, pp. 122–127 Georgakakos K P, Seo D-J, Gupta H, Schaake J, Butts M B (2004) Towards the characterization of streamflow simulation uncertainty through multimodel ensembles. J. Hydrol. 298: 222–241 Goniadzki D (2006) Institutional Aspects of the Hydrological Warning System in the Del Plata Basin. WMO International Workshop on Flash Flood Forecasting, 13–17 March, San José Costa Rica. http://www.nws.noaa.gov/iao/iao_FFW.php. Accessed 15 January 2008 Goswami M, O’Connor K M, Bhattarai K P, Shamseldin A Y (2005) Assessing the performance of eight real time updating models and procedures for the Brosna River. Hydrol. Earth Syst. Sci. 9(4): 394–411. http://www.hydrol-earth-syst-sci.net/9/394/2005/. Accessed 15 January 2008 Huaimin G (2005) Flood Forecasting and Management in China. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October 2005, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Jolliffe I T, Stephenson D B (2003) Forecast Verification. A Practitioner’s Guide in Atmospheric Science. Wiley, Chichester, UK Krzysztofowicz R (2001) The case for probabilistic forecasting in hydrology. J. Hydrology 249: 2–9 Laio F, Tamea S (2007) Verification tools for probabilistic forecasts of continuous hydrological variables. Hydrol. Earth Syst. Sci. 11:1267–1277. http://www.hydrol-earth-syst-sci. net/11/1267/2007/. Accessed 15 January 2008 Moore R J (1999) Real-time flood forecasting systems: perspectives and prospects. In: R. Casale and C.Margottini (Eds.), Floods and Landslides: Integrated Risk Assessment, Chapter 11. Springer-Verlag, Berlin and Heidelberg pp. 147–189

284

References

Pappenberger F, Matgen P, Beven K J, Henry J-B, Pfister L, de Fraipont P (2004) The influence of rating curve uncertainty on flood inundation predictions. Proc. IMA International Conference on Flood Risk Assessment, Bath, 7–8 September, England Pappenberger F, Beven K, Frodsham K, Romanowicz R, Matgen P (2007) Grasping the unavoidable subjectivity in calibration of flood inundation models: a vulnerability weighted approach. J. Hydrol. 333(2–4): 275– 287 Paudyal G N (2002) Forecasting and warning of water-related disasters in a complex hydraulic setting—the case of Bangladesh. Hydrolog. Sci. J. Special Issue: Towards Integrated Water Resources Management for Sustainable Development 47(S), August: S5–S18 Rebora N, Ferraris L, von Hardenberg J, Provenzale A (2006) Rainfall downscaling and flood forecasting: a case study in the Mediterranean area. Nat. Hazards Earth Syst. Sci. 6: 611–619. Reed D W (1984) A Review of British Flood Forecasting Practice. Institute of Hydrology, Report No. 90 Richardson (2003) Economic Value and Skill. Chapter 8 in Jolliffe I T, Stephenson DB (Eds.) (2003) Forecast Verification. A Practitioner’s Guide in Atmospheric Science. Wiley and Sons Ltd Røhr P C, Husebye S (2005) Flood Forecasting Technology in Norway, Innovation and Advances. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October, Tromsø, Norway. http://www.actif-ec.net/ conference2005/proceedings/index.html. Accessed 15 January 2008 Romanowicz R J,Young P C, Beven K J (2006) Data assimilation and adaptive forecasting of water levels in the river Severn catchment, United Kingdom. Water Resour. Res. 42(6) doi:10.1029/2005WR004373 Rotach M W, Arpagaus M, Dorninger M, Hegg C, Montani A, Ranzi R (2007) The D-PHASE Operations Period (DOP). 3rd HEPEX Workshop, 27–29 June, Stresa, Italy Roulin E (2007) Skill and relative economic value of medium-range hydrological ensemble predictions. Hydrol. Earth Syst. Sci., 11, 725–737. http://www.hydrol-earth-syst-sci. net/11/725/2007/hess-11-725-2007.html. Accessed 15 January 2008 Rungo M, Refsgaard J K, Havno K (1989) The updating procedure in the MIKE 11 modelling system for real time forecasting. In Cluckie I D, Collier C G (Eds.), Hydrological Applications of Weather Radar, Ellis Horwood, Chichester, West Sussex, England. Schaake J, Restrepo P, Seo D J, Demargne J, Wu L, Perica S (2005) Hydrological Ensemble Prediction: Challenges and Opportunities. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October, Tromso, Norway. http://www. actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Serban P, Askew A J (1991) Hydrological forecasting and updating procedures. IAHS Publ. No. 201: 357–369 Sperfslage J A, Georgakakos K P, Carpenter T M, Shamir E, Graham N E (2005) Design and Implementation of a Real-Time Regional Flash Flood Guidance System for Central America Driven by Adjusted GOES Satellite Rainfall. 21st International Conference on Interactive Information Processing Systems (IIPS) for Meteorology, Oceanography, and Hydrology. http:// ams.confex.com/ams/Annual2005/techprogram/session_17767.htm. Accessed 15 January 2008 Thielen J, De Roo A, Niemeyer S, Gouweleeuw B, Van der Knijff J, Franchello G (2004) Development and testing of a Pan-European Flood Alert System (EFAS). Abstracts of 2nd European Flood Alert System (EFAS) Workshop, Ispra, 10–12 November 2004. http://efas.jrc. it/workshop2004/. Accessed 15 January 2008 Tozer N, Saulter A, Hawkes P (2007) Coastal Flood Forecasting. 42nd Defra Flood and Coastal Management Conference, York, England U.S. Army Corps of Engineers (USACE) (1994) Framework for Estimating National Economic Development Benefits and Other Beneficial Effects of Flood Warning and Preparedness Systems. Institute for Water Resources Report IWR-94–3, Alexandria, VA. Vehviläinen B, Huttunen M, Huttunen I (2005) Hydrological Forecasting and Real time Monitoring in Finland: The Watershed Simulation and Forecasting System (WSFS). ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting

References

285

Technology, 17–19 October, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Verlaan M, Zijderveld A, de Vries H, Kroos J (2005) Operational storm surge forecasting in the Netherlands: developments in the last decade. Phil. Trans. Royal Society 363(1831): 1441–1453 Weerts A H, El Sarafy G (2005) Particle Filtering and Ensemble Kalman Filtering for Input Correction in Rainfall Runoff Modeling. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Werner M, Self K (2005) Measuring Performance in Flood Forecasting: A Question of Model Reliability or of Forecast Reliability. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Werner M, van Dijk M (2005) Developing Flood Forecasting Systems – Examples from the UK, Europe and Pakistan. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October, Tromsø, Norway. http:// www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Wöhling Th, Lennartz F, Zappa M (2006) Technical Note: Updating procedure for flood forecasting with conceptual HBV-type models. Hydrol. Earth Syst. Sci., 10:783–788. http://www. hydrol-earth-syst-sci.net/10/783/2006/hess-10-783-2006.html. Accessed 15 January 2008 World Meteorological Organisation (1994) Guide to Hydrological Practices. Fifth Edition. WMO No. 168, Geneva Young P C, Tomlin C M (2000) Data-based Mechanistic Modelling and Adaptive Flow Forecasting. In Lees M, Walsh P (Eds.), Flood Forecasting: What does Current Research Offer the Practitioner, BHS Occasional Paper 12. British Hydrological Society, London

Chapter 6

Rivers

Abrahart R J, Kneale P E, See L M (Eds.) (2004) Neural Networks for Hydrological Modelling. Taylor & Francis, London Andrssian V, Hall A, Chahinian N, Schaake J (Eds.) (2006) Large Sample Basin Experiments for Hydrological Model Parameterization: Results of the Model Parameter Experiment-MOPEX, IAHS Publication, 307 Arduino G, Reggiani P, Todini E (2005) Recent advances in flood forecasting and flood risk assessment. Hydrol. Earth Syst. Sci. 9(4): 280–284. http://www.hydrol-earth-syst-sci. net/9/280/2005/hess-9-280-2005.html. Accessed 15 January 2008 ASCE (2000) Artificial neural networks in hydrology II: Hydrologic applications. J. Hydrol. Eng. 5(2): 124–137 Bell V A, Kay A L, Jones R G, Moore R J (2007) Development of a high resolution grid-based river flow model for use with regional climate model output. Hydrol. Earth Syst. Sci. 11(1): 532–549. http://www.hydrol-earth-syst-sci.net/11/532/2007. Accessed 15 January 2008 Bergstrom S (1995) The HBV model. In: Singh V P (Ed.), Computer Models of Watershed Hydrology. Water Resources Publications, Littleton, CO Beven K J (2001) Rainfall Runoff Modelling – The Primer. Wiley, Chichester Beven K, Romanowicz R, Pappenberger F, Young P, Werner M (2005) The Uncertainty Cascade in Flood Forecasting. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, Tromsø, Norway, 17–19 October 2005. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Box G E P, Jenkins G M (1970) Time Series Analysis Forecasting and Control. Holden Day, San Francisco, CA

286

References

Burnash R J C (1995) The NWS river forecast system – catchment modeling. In Singh V P (Ed.), Computer Models of Watershed Hydrology, Water Resources Publications, Highlands Ranch, CO Butts M B, Overgaard J, Viaene P, Dubicki A, Stroñska K, Szalinska W, Lewandowski A, Olszewski T, Kolerski (2005) Flexible Process-based Hydrological Modelling Framework for Flood forecasting – MIKE SHE. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, Tromsø, Norway, 17–19 October 2005. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Chanson H (2004) Hydraulics of Open Channel Flow: An Introduction. 2nd edition. ButterworthHeinemann, Oxford Chen Y, Sene K J, Hearn K (2005) Converting Section 105 or SFRM Hydrodynamic River Models for Real Time Forecasting Applications. 40th Defra Flood and Coastal Defence Conference, York, England Chow V T (1959) Open Channel Hydraulics. McGraw-Hill, New York Collischonn W, Allasia D G, Silva B C, Tucci C E M (2007) The MGB-IPH model for large scale rainfall runoff modeling. Hydrol. Sci. J. 52(5): 878–895 Cunge J A (1969) On the subject of a flood propagation method. J. Hydraul. Res. IAHR 7: 205–230 Dawson C W, Wilby R L (1999) A comparison of artificial neural networks used for flow forecasting. Hydrol. Earth Syst. Sci. 3: 529–540 De Roo A D P, Wesseling C G, Van Deursen W P A (2000) Physically based river basin modelling within a GIS: the LISFLOOD model. Hydrol. Process. 14(11–12): 1981–1992 Environment Agency (2002) Fluvial Flood Forecasting for Flood Warning - Real Time Modelling. Defra/Environment Agency Flood and Coastal Defence R&D Programme. R&D Technical Report W5C-013/5/TR European Flood Forecasting System EFFS (2003) Final Report WP8. Deliverable 8.3, WL/Delft Hydraulics. http://effs.wldelft.nl/. Accessed 15 January 2008 Han D (1991) Weather Radar Information Processing and Real Time Flood Forecasting. Ph.D. thesis, University of Salford Huband M, Sene K J (2005) Integrated catchment modelling issues for flow forecasting applications. Scottish Hydraulics Study Group, Catchment Modelling for Flood Risk Management, 18 March 2005 Jones D A, Moore R J (1980) A simple channel flow routing model for real time use. Proc. Oxford Symposium Hydrological Forecasting, IAHS Publication, 129 Kneale P E, See L, Kerr P (2000) Developing a Prototype Neural Net Flood Forecasting Model for Use by the Environment Agency. Proceedings 35th MAFF conference of River and Coastal Engineers, Keele University, England Koren V, Reed S, Smith M, Zhang Z, Seo D J (2004) Hydrology Laboratory Research Modeling System (HL-RMS) of the National Weather Service. J. Hydrology. 291(3/4): 297–318 Lees M, Young P, Ferguson S, Beven K J, Burns J (1994) An adaptive flood warning system for the River Nith at Dumfries. In White W R, Watts J (Eds.), 2nd International Conference on River Flood Hydraulics. Wiley, Chichester Lees M J (2000) Advances in transfer function based flood forecasting. In Lees M, Walsh P (Eds.), Flood Forecasting: What does Current Research Offer the Practitioner. BHS Occasional Paper 12, British Hydrological Society, London Lettenmaier D P, Wood E F (1993) Hydrologic forecasting. In Maidment D R (Ed.), Handbook of Hydrology. Chapter 26, McGraw-Hill, New York Lighthill M J, Whitham G B (1955) On Kinematic Waves. I – Flood Movement in Long Rivers. Proc. Roy. Soc. London Series A 229: 281–316 Liu Z, Martina M L V, Todini E (2005) Flood forecasting using a fully distributed model: application of the TOPKAPI model to the Upper Xixian Catchment. Hydrol. Earth Syst. Sci. 9(4): 347–364. http://www.hydrol-earth-syst-sci.net/9/issue4.html. Accessed 15 January 2008

References

287

Madsen H (2000) Automatic calibration of a conceptual rainfall-runoff model using multiple objectives. J. Hydrol. 235(3–4): 276–288 Malone T (1999) Using URBS for Real Time Flood Modelling. Water 99: Joint Congress; 25th Hydrology & Water Resources Symposium, 2nd International Conference on Water Resources & Environment Research Moore R J (1982) Transfer functions, noise predictors and the forecasting of flood events in real time. In. Singh, V.P. (Ed.), Statistical Analysis of Rainfall and Runoff. Water Resources Publications, Littleton, Colorado pp. 229–250. Moore R J (1985) The probability distributed principle and runoff production at point and basin scales. Hydrol. Sci. J. 30(2): 273–297 Moore R J (2007) The PDM rainfall runoff model. Hydrol. Earth Syst. Sci. 11(1): 483–499 Moore R J, Cole S J, Bell V A, Jones D A (2006) Issues in flood forecasting: ungauged basins, extreme floods and uncertainty. In: Tchiguirinskaia I, Thein K N N, Hubert P (Eds.), Frontiers in Flood Research, 8th Kovacs Colloquium, UNESCO, Paris Price R K (1977) FLOUT – A River Catchment Flood Model, Report No. IT 168. Hydraulics Research Station, Wallingford Reed D W (1984) A Review of British Flood Forecasting Practice. Institute of Hydrology, Report No. 90 Reggiani P, Schellekens J (2003) Modelling of hydrological responses: the representative elementary watershed approach as an alternative blueprint for watershed modelling. Hydrol. Process. 17: 3785–3789 Rotach M W, Arpagaus M, Dorninger M, Hegg C, Montani A, Ranzi R (2007) The D-PHASE Operations Period (DOP). 3rd HEPEX Workshop, 27–29 June 2007, Stresa, Italy Sene K J, Ainsworth E J, Graham G (2006) Risk Based Design of Telemetry Networks for Flood Warning Applications. 41st Defra/Environment Agency Flood and Coastal Management Conference, York, England Sherman L K (1932) Streamflow from rainfall by the unit graph method. Eng. News Record 108: 501–505 Shrestha R R, Theobald S, Nestmann F (2005) Simulation of flood flow in a river system using artificial neural networks. Hydrol. Earth Syst. Sci. 9(4): 313–321. http://www.hydrol-earthsyst-sci.net/9/313/2005. Accessed 15 January 2008 Singh V P (1988) Hydrologic Systems Volume I: Rainfall Runoff Modelling. Prentice-Hall, Englewood Cliffs, NJ Singh V P (1989) Hydrologic Systems Volume II: Watershed Modelling, Prentice-Hall, Englewood Cliffs, NJ Sivapalan M, Wagener T, Uhlenbrook S, Liang X, Lakshmi V, Kumar P, Zehe E, Tachikawa Y (eds.) (2006) Predictions in Ungauged Basins: Promises and Progress. IAHS Publication, 303, May 2006 Solomatine D P, Price R K (2004) Innovative Approaches to Flood Forecasting Using Data Driven and Hybrid Modelling. 6th International Conference on Hydroinformatics, 21–24 June 2004, World Scientific Publishing Company, Singapore Sugarawa, M (1995) Tank model. In: Singh, V P (Ed.), Computer Models of Watershed Hydrology. Water Resources Publications, Littleton, CO, pp. 165–214 Tang X, Knight D W, Samuels P G (1999) Volume conservation in variable parameter Muskingum Cunge method. J. Hydraulic. Engineering, 610–620 Tilford K A, Sene K J, Khatibi R (2007) Flood Forecasting Model Selection: A Structured Approach. In Begum S, Stive M J F, Hall J W (Eds.), Flood Risk Management in Europe, Springer, Netherlands Van Der Knijff J, De Roo A, Franchello G (2004) The LISFLOOD Model. Abstracts of 2nd European Flood Alert System (EFAS) Workshop, Ispra, 10–12 November 2004. http://efas. jrc.it/workshop2004/. Accessed 15 January 2008 Vieux B (2004) Distributed hydrologic modeling for flood forecasting. GIS for Remote Sensing in Hydrology. Water Resources and Environment. Proceedings IGRHWE, Three Gorges Dam, China, September 2003. IAHS Publication 289

288

References

World Meteorological Organisation (1992) Simulated real time intercomparison of hydrological models. Operational Hydrology Report No. 38, Geneva World Meteorological Organisation (1994) Guide to Hydrological Practices. 5th edition. WMO No. 168, Geneva Young P C, Tomlin C M (2000) Data-based Mechanistic Modelling and Adaptive Flow forecasting. In Lees M, Walsh P (Eds.), Flood Forecasting: What does Current Research Offer the Practitioner, BHS Occasional Paper 12, British Hydrological Society, London Young P (2001) Data Based Mechanistic modeling and validation of rainfall flow processes. In Anderson M G, Bates P D (Eds.), Model Validation: Perspectives in Hydrological Science. John Wiley & Sons Ltd Zhao R J (1992) The Xin’anjiang model applied in China. J. Hydrology. 135: 371–381

Chapter 7

Coasts

Berg R, Clark J, Roth D, Birchard T (2007) The National Weather Service Unified Surface Analysis. Paper 6A-4, 22nd Conference on Weather Analysis and Forecasting/18th Conference on Numerical Weather Prediction, 25–29 June, UT. http://ams.confex.com/ams/22WAF18NWP/ techprogram/programexpanded_414.htm. Accessed 15 January 2008 Brassington G B, Warren G, Smith N, Schiller A, Oke P R (2005) BLUElink Progress on operational ocean prediction for Australia. Bull. Aust. Meteorol. Oceanogr. Soc. 18: 104–109. http://www.cmar.csiro.au/bluelink/scientific/index.htm. Accessed 15 January 2008 Causon D, Ingram D, Mingham C, Zang J, Hu K, Zhou J G (2000) Numerically Simulating Seawall Overtopping. Proceedings ICCE2000 International Conference on Coastal Engineering Chassignet E P, Hurlburt E, Smedstad O M, Halliwell G R, Wallcraft A J, Metzger E, Lanton B O, Lozano C, Rao D B, Hogan P J, Srinivasan A (2006) Generalised vertical coordinates for eddy resolving global and coastal ocean forecasts. Oceanography 19(1): 20–31. http://coaps.fsu. edu/people/chassignet/allpubs.shtml. Accessed 15 January 2008 Cheung K F, Phadke A C, Wei Y, Rojas R, Douyere Y J M, Martino C D, Houston S H, Liu P L F, Lynett P J, Dodd N, Liao S, Nakazaki E (2003) Modeling of storm-induced coastal flooding for emergency management. Ocean Eng 30(2003): 1353–1386 Davidson N E, Entel M B, Hanstrum B N, Smith N R, Warren G R (2005) Enhancements to the Bureau of Meteorology’s ocean surge forecasting system using operational TC-LAPS. BMRC Research Report No. 114. http://www.bom.gov.au/bmrc/pubs/researchreports/researchreports. htm. Accessed 15 January 2008 Environment Agency (2004) Best Practice in Coastal Flood Forecasting. Defra/Environment Agency Flood and Coastal Defence R&D Programme, R&D Technical Report FD2206/TR1 Environment Agency (2005) Dependence Between Extreme Sea Surge, River Flow and Precipitation: A Study in South and West Britain. Defra/Environment Agency Flood and Coastal Defence R&D Programme. R&D Technical Report FD2308/TR3 Flather R A (2000) Existing operational oceanography. Coast Eng 41: 13–40 Hicks S D (2006) Understanding Tides. Centre for Operational Oceanographic Products and Services, NOAA. http://tidesandcurrents.noaa.gov/publications/Understanding_Tides_by_ Steacy_finalFINAL11_30.pdf. Accessed 15 January 2008 Holland G (Ed.) (2007) Global Guide to Tropical Cyclone Forecasting. Bureau of Meteorology Research Centre (Australia) WMO/TC-No. 560, Report No. TCP-31, World Meteorological Organization, Geneva, Switzerland http://www.bom.gov.au/bmrc/pubs/tcguide/globa_guide_intro.htm. Accessed 15 January 2008 Holthuijsen L H (2007) Waves in Oceanic and Coastal Waters. Cambridge University Press, Cambridge Horsburgh K J, Wilson C (2007) Tide-surge interaction and its role in the distribution of surge residuals in the North Sea. J Geophys Res 112: C08003, doi:10.1029/2006JC004033

References

289

Hu K, Wotherspoon C (2007) Numerical modelling in coastal flood forecasting and warning in England and Wales. In: Begum S, Stive M J F, Hall J W (Eds.), Flood Risk Management in Europe. Innovation in Policy and Practice. Springer, The Netherlands Huang W, Murray C (2003) Application of an Artificial Neural Network to Predict Tidal Currents in an Inlet. Technical Report ERDC/CHL CHETN-IV-58, US Army Corps of Engineers Hubbard M E, Dodd N (2002) A 2D numerical model of wave run-up and overtopping. Coast Eng 47: 1–26 JCOMM (2007) Guide to Storm Surge Forecasting (in preparation). IOC/WMO Joint Technical Commission for Oceanography and Marine Meteorology Jelesnianski C P, Chen J, Schaffer W A (1992) SLOSH: Sea, Lake and Overland Surges from Hurricanes. NOAA Technical Report NWS 48, NOAA Kirby J T, Dalrymple R A, Shi F (2005) Combined Refraction/Diffraction Model REF/DIF 1 Version 3.0 Documentation and User’s Manual. Center for Applied Coastal Research, Department of Civil and Environmental Engineering, University of Delaware, Newark, DE Knabb R D, Rhome J R, Brown D P (2006) Tropical Cyclone Report: Hurricane Katrina 23–30 August 2005. National Hurricane Center, 20 December 2005 (updated 10 August 2006 for tropical wave history, storm surge, tornadoes, surface observations, fatalities, and damage cost estimates) Komen G J, Cavaleri P, Donelan M, Hasselmann K, Hasselmann S, Janssen P (1994) Dynamics and Modelling of Ocean Waves. Cambridge University Press, Cambridge Lam H, Lam Q C C, Chu H C K (2005) Risk Management of Tropical Cyclone-Induced Flooding in Hong Kong. Workshop on Risk Management Towards Development Goals and Socioeconomic Impact Assessment of Typhoon-related Disaster, Kuala Lumpur, Malaysia, 5–9 September 2005 Lees M, Young P, Ferguson S, Beven K J, Burns J (1994) An adaptive flood warning system for the River Nith at Dumfries. In: White W R, Watts J (Eds.), 2nd International Conference on River Flood Hydraulics. Wiley, Chichester, UK Makarynskyy O, Makarynska D, Kuhn M, Featherstone W E (2004) Predicting sea level variations with artificial neural networks at Hillarys Boat Harbour, Western Australia. Estuarine, Coastal and Shelf Science 61(2): 351–360. http://espace.lis.curtin.edu.au/archive/00000025/01/ W291.pdf. Accessed 15 January 2008 Makarynskyy O, Pires-Silva A A, Makarynska D, Ventura-Soares C (2005) Artificial neural networks in wave predictions at the west coast of Portugal. Comput Geosci 31(4): 415–424. http://espace.lis.curtin.edu.au/archive/00000243/01/Comp&Geos04.PDF. Accessed 15 January 2008 Massey W G, Gangai J W, Drei-Horgan E, Slover K J (2007) History of Coastal Inundation Models. Mar Technol Soc J 41(1): 7–17 Patrick A R, Collins W G, Tissot P E, Drikitis A, Stearns J, Michaud P R (2002) Use of the NCEP MESOETA Data in a Water Level Predicting Neural Network. Proceedings of the 19th AMS Conference on Weather Analysis and Forecasting/15th AMS Conference on Numerical Weather Prediction, San Antonio, TX, pp. 369–372. http://wave.oregonstate.edu/Research/ Publications/Cox/pctdsmc-AMS-02.pdf. Accessed 15 January 2008 Prouty D B, Tissot P E, Anwar A A (2005) Using Artificial Neural Networks for Predicting Storm Surge Propagation in North Sea and the Thames Estuary. Fourth Conference on Artificial Intelligence Applications to Environmental Science, San Diego, CA. http://ams.confex.com/ ams/Annual2005/techprogram/programexpanded_264.htm. Accessed 15 January 2008 Randon N, Lawry J, Horsburgh K J, Cluckie I (2007) Fuzzy Bayesian Modelling of Sea-Level Along the East Coast of Britain. IEEE Transaction on Fuzzy Systems (in press) Sheng Y P, Paramygin V A, Alymov V, Davis J R (2005) A Real-Time Forecasting System for Hurricane Induced Storm Surge and Coastal Flooding. 9th International Conference on Estuarine and Coastal Modeling, Charleston, SC Shi F, Dalrymple R A, Kirby J T, Chen Q, Kennedy A (2001) A fully nonlinear Boussinesq model in generalized curvilinear coordinates. Coast Eng 42: 337–358 Solomatine D P, Velickov S, Wüst J C (2001) Predicting water levels and currents in the North Sea using chaos theory and neural networks. Proceedings of IAHR Congress, Beijing, September 2001

290

References

Tissot P E, Michaud P R, Cox D T (2003) Optimization and Performance of a Neural Network Model Forecasting Water Levels for the Corpus Christi, Texas, Estuary. 3rd Conference on Artificial Intelligence Applications to Environmental Science, Long Beach, CA. http://ams.confex.com/ams/annual2003/techprogram/programexpanded_148.htm. Accessed 15 January 2008 Tolman H L (1999) User Manual and System Documentation of WAVEWATCH-III Version 1.18. NOAA/NWS/NCEP/OMB Technical Note 166, National Weather Service, NOAA, U.S. Department of Commerce Tozer N, Saulter A, Hawkes P (2007) Coastal Flood Forecasting. 42nd Defra Flood and Coastal Management Conference, York, England Verlaan M, Zijderveld A, Vries H, Kroos J (2005) Operational storm surge forecasting in the Netherlands: developments in the last decade. Phil. Trans. R. Soc. Lond. 363(1831): 1441–1453 Wedge D, Ingram D, Mingham C, McLean D, Bandar Z (2005) On Neural Network Architectures and Wave Overtopping. Proc. ICE, Maritime Engineering, 158(MA3), 123–133 World Meteorological Organisation (WMO) (1998) Guide to Wave Analysis and Forecasting. 2nd edition. WMO No. 702, Geneva, Switzerland

Chapter 8

Selected Applications

ACTIF (2004) Some Research Needs for River Flood Forecasting in FP6. Achieving Technological Innovation in Flood Forecasting. European Commission Project EVK1-CT-2002-80014. http://www.actif-ec.net/documents/ACTIFResearchNeedsfor%20FP6V1.4.pdf. Accessed 15 January 2008 Bhattacharya B, Lobbrecht A H, Solomatine D P (2003) Neural Networks and Reinforcement Learning in Control of Water Systems. J. Water Resour. Plann. Manage. 129(6): 458–465 Bell V A, Moore R J, Brown V (2000) Snowmelt forecasting for flood warning in upland Britain. In Lees M, Walsh P (Eds.), Flood Forecasting: What does Current Research Offer the Practitioner. BHS Occasional Paper 12, British Hydrological Society, London Blackburn J, Hicks F (2003) Suitability of dynamic modeling for flood forecasting during ice jam release surge events. J. Cold Reg. Engrg. 17(1): 18–36 Bowles D S, Mathias J D, Chauhan S S, Countryman J D (2004) Reservoir Release Forecast Model for Flood Operation of the Folsom Project include Pre-releases. Proceedings of the 2004 USSD Annual Lecture, March 2004, St. Louis, MO. http://uwrl.usu.edu/www/faculty/ DSB/USSD2004.pdf. Accessed 15 January 2008 Castro D, Einfalt T, Hatzfeld F, Wagner A (2006) URBAS: Prediction and Management of Flash Floods in Urban Areas. International Workshop on Flash Floods in Urban Areas and Risk Management, Oman. http://www.rcuwm.org.ir/events/workshop/8/articles.html. Accessed 15 January 2008 Cluckie I D, Lane A, Griffith R J (1998) Real Time Control of Urban Drainage Systems (UDS) Using Weather Radar. Radar Hydrology for Real Time Flood Forecasting. Proceedings of an advanced study course, University of Bristol, 24 June–3 July 1998, pp. 305–321 Collischonn W, Allasia D G, Silva B C, Tucci C E M (2007a) The MGB-IPH model for large scale rainfall runoff modeling. Hydrol. Sci. J. 52(5): 878–895 Collischonn W, Tucci C E M, Clarke R T, Chou S C, Guilhon L G, Cataldi M, Allasia D (2007b) Medium-range reservoir inflow predictions based on quantitative precipitation forecasts. J. Hydrol. 344: 112–122 Daly S F (2003) A State–Space Model for River Ice Forecasting. Cold Regions Research and Engineering Laboratory Report ERDC/CRREL TR-03-9, New Hampshire. http://www. crrel.usace.army.mil/techpub/CRREL_Reports/html_files/Cat_D.html. Accessed 15 January 2008 Dunn S M, Colohan R J E (1999) Developing the snowmelt component of a distributed hydrological model: a stepwise approach based on multi objective analysis. J. Hydrol. 223(1–2): 1–16

References

291

Erikkson M (2006) Flash Floods in the Himalaya – Challenges and Opportunities for Regional Cooperation. Presentation at the First International Flash Flood Forecasting Workshop, World Meteorological Organisation, March 13–17 2006, Costa Rica, CA. http://www.nws.noaa.gov/ iao/iao_FFW.php. Accessed 15 January 2008 Garcia R R, Sáez A L, Salete E G, García L L (2005) Reservoir Management in Real Time Flood Forecasting and Decision Support in the Ebro river basin. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October 2005, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Georgakakos K P (2005) Modern Operational Flash Flood Warning Systems Based on Flash Flood Guidance Theory: Performance Evaluation. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October 2005, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Göppert H, Ihringer J, Plate E J, Morgenschweis G (1998) Flood forecast models for improved reservoir management in the Lenne River catchment, Germany. Hydrol. Sci. J. 43(2): 215–242 Graham W J (2000) Floods caused by dam failure. In Parker D J (Ed.), Floods. Routledge, London Guo S, Zhang H, Chen H, Peng D, Liu P, Pang B (2004) A reservoir flood forecasting and control system for China. Hydrol. Sci. J. 49(6): 959–972 Henry E, Klepiszewski K, Fiorelli D, Solvi A-M, Weidenhaupt A (2005) Modelling of a Sewage Network: Contribution to the Management of Pollution Risks at the Haute-Sûre Drinking Water Reservoir. 10th International Conference on Urban Drainage, Copenhagen, Denmark, 21–26 August 2005 Hock R (2003) Temperature index melt modelling in mountain areas. J. Hydrol. 282(1–4): 104–115 Hom V T, Capone E J, Econopouly T W (2004) Forecasting Ice Affected Rivers at the Northeast River Forecast Centre. 61st Eastern Snow Conference, Portland, Maine. http://www.easternsnow.org/proceedings/2004/hom_et_al.pdf. Accessed 15 January 2008 Howard C D D (2004) Ensemble Optimization for Hydroelectric Operations. Workshop on the Hydrological Ensemble Prediction Experiment (HEPEX), ECMWF, 8–10 March 2004. http:// ecmwf.int/newsevents/meetings/workshops/2004/HEPEX/presentations.html. Accessed 15 January 2008 Howard, C D D (2007) Hydroelectric Operations – Ensemble Optimisation Procedures. Workshop on the Hydrological Ensemble Prediction Experiment (HEPEX), Stresa, Italy, 27–29 June, 2007 Huband M, Sene K J (2005) Integrated Catchment Modelling Issues for Flow Forecasting Applications. Scottish Hydraulics Study Group, Catchment Modelling for Flood Risk Management, 18 March 2005 IMPACT (2005) Investigation of Extreme Flood Processes and Uncertainty. Final Technical Report. http://www.samui.co.uk/impact-project/general_publications.htm. Accessed 15 January 2008 Koren V, Schaake J, Mitchell K, Duan Q Y, Chen F, Baker J M (1999) A parameterization of snowpack and frozen ground intended for NCEP weather and climate models. J. Geophys. Res. 104(D16): 19569–19585 Kubat I, Sayed M, Savage S, Carrieres T (2005) Implementation and Testing of a Thickness Redistribution Model for Operational Ice Forecasting. Proceedings 18th International Conference on Port and Ocean Engineering under Arctic Conditions, POAC’05. Volume 2, pp. 781–791, Potsdam. ftp://ftp2.chc.nrc.ca/CRTreports/CIS/POAC_05_Forecasting.pdf. Accessed 15 January 2008 Kuchment L S, Gelfan A N (2005) Long Term Ensemble Forecasting Spring Runoff Volume and Hydrograph Peak. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October 2005, Tromsø, Norway. http://www.actifec.net/conference2005/proceedings/index.html. Accessed 15 January 2008

292

References

Lobbrecht A H, Dibike Y B, Solomatine D P (2005) Networks and Fuzzy Systems in Model Based Control of the Overwaard Polder. J. Water Resour. Plan. Manage. 131(2): 135–145 Mahabir C, Hicks F E, Robichaud C, Fayek A R (2006) Forecasting breakup water levels at Fort McMurray, Alberta, using multiple linear regression. Can. J. Civ. Eng. 33(9): 1227–1238 Meon G (2006) Past and Present Challenges in Flash Flood Forecasting. WMO International Workshop on Flash Flood Forecasting, 13–17 March 2006, Costa Rica, CA. http://www.nws. noaa.gov/iao/iao_FFW.php. Accessed 15 January 2008 Mirtskhoulava Ts E (2000) Mud floods and their prediction in the Caucasus and Central Asia. In Parker D J (Ed.), Floods. Routledge, London Morse B, Hicks F (2005) Advances in river ice hydrology 1999–2003. Hydrol. Process. 19(1): 247–263 Nandalal K D W, Bogardi J J (2007) Dynamic Programming Based Operation of Reservoirs Applicability and Limits. Cambridge University Press, Cambridge NOAA-USGS (2005) NOAA-USGS Debris-Flow Warning System. Final Report: U.S. Geological Survey Circular 1283. http://pubs.usgs.gov/circ/2005/1283/. Accessed 15 January 2008 Rickenmann D, Chen C L (2003) Debris-Flow Hazards Mitigation: Mechanics, Prediction, and Assessment. Millpress, Rotterdam Rutter N, Essery R (2006) Evaluation of Forest Snow Process Models (SnowMIP2). Poster presentation, European Geosciences Union, Vienna, April 2006 Satake K (2000) Characteristics, causes and predictability of tsunamis. In Parker D J (Ed.), Floods. Routledge, London Schilling W (Ed.) (1989) Real Time Control of Urban Drainage Systems. The State of the Art. IAWPRC Task Group on Real Time Control of Urban Drainage Systems, London Schütze M, Campisano A, Colas H, Schilling W, Vanrolleghem P A (2004) Real time control of urban wastewater systems—where do we stand today? J. Hydrol. 299: 335–348 Snorrason Á, Björnsson H, Jóhannesson H (2000) Causes, characteristics and predictability of floods in regions with cold climates. In Parker D J (Ed.), Floods. Routledge, London Sperfslage J A, Georgakakos K P, Carpenter T M, Shamir E, Graham N E (2005) Design and Implementation of a Real-Time Regional Flash Flood Guidance System for Central America Driven by Adjusted GOES Satellite Rainfall. 21st International Conference on Interactive Information Processing Systems (IIPS) for Meteorology, Oceanography, and Hydrology. http://ams.confex.com/ams/Annual2005/techprogram/session_17767.htm. Accessed 15 January 2008 Todini E, Codeluppi R (1998) Operational Use of Real-Time Flood Forecasting; the Case of Lake Como. EWC98 International Conference on Early Warning Systems for Natural Disaster Reduction, Potsdam Vieux B E, Bedient P E, Mazroi E (2005) Real-Time Urban Runoff Simulation Using Radar Rainfall and Physics-based Distributed Modeling for Site-Specific Forecasts. 10th International Conference on Urban Drainage, 21–26 August 2005, Copenhagen/Denmark Villars M, Gerritsen H, Heynert K, Kleiss B, Koster K (2007) A Demonstrator for Management of the St Petersburg Flood Protection Barrier. 32nd Congress of the International Association for Hydraulic Research, Venice, Italy Vitasovic Z C (2006) Real Time Control of Urban Drainage Networks. U.S. Environmental Protection Agency (EPA) Report EPA/600/R-06/120, Office of Research and Development, Washington, DC. http://www.epa.gov/nrmrl/pubs/600r06120/600r06120.pdf. Accessed 15 January 2008 White K D (2003) Review of prediction methods for breakup ice jams. Can. J. Civ. Eng. 30: 89–100 World Meteorological Organisation (1982) Flash Flood Forecasting. Operational Hydrology Report No. 18, WMO No. 577, Geneva World Meteorological Organisation (1994) Guide to Hydrological Practices. 5th Edition. WMO No. 168, Geneva World Meteorological Organisation (2007) Prospectus for Implementation of a Flash Flood Guidance System with Global Coverage. Geneva

References

Chapter 9

293

Preparedness

Direction de la Défense et de la Sécurité Civiles (2006) Guide ORSEC departemental methode generale Organisation de la Réponse de Securité Civile. Tome G.1. http://www.cedre.fr/fr/ lutte/guide-elu/guide-orsec06.pdf. Accessed 15 January 2008 Emergency Management Australia (1999) Guide 4 – Flood Preparedness. Australian Emergency Manuals Series, Part III Emergency Management Practice. Volume 3 - Guidelines. http://www. ema.gov.au/. Accessed 15 January 2008 Environment Agency (2007) Risk Assessment for Flood Incident Management: Framework and Tools. Joint Defra/Environment Agency Flood and Coastal Erosion Risk Management R&D Programme. R&D Technical Report SC050028/SR1 Federal Aviation Authority (2000) FAA System Safety Handbook. Chapter 9: Analysis Techniques. FAA, Washington, DC. http://www.faa.gov/library/manuals/aviation/risk_ management/ss_handbook/. Accessed 15 January 2008 FEMA (1997) Your Family Disaster Plan, Your Family Disaster Supplies Kit, Food and Water in an Emergency, and Emergency Preparedness Checklist. Federal Emergency Management Agency, Washington, DC. http://www.fema.gov/plan/index.shtm. Accessed 15 January 2008 FEMA (2005) Reducing Damage from Localized Flooding: A Guide for Communities. Chapter 6: Warning and Emergency Services, Report 511–07. http://www.fema.gov/hazard/flood/pubs/ flood-damage.shtm. Accessed 15 January 2008 FEMA (2007) National Incident Management System Incident Resource Inventory System (NIMS-IRIS). Factsheet NIMS RM03, 202–646–3850 Gyorfi J S, Buhrke E R, Tarlton M A, Lopez J M, Valliath G T (2006) VICC: Virtual Incident Command Center. Presence 2006 Handmer J (2002) Flood warning reviews in North America and Europe: statements and silence. Aust. J Emerg. Manage. 17(3): 17–24. http://www.ema.gov.au/agd/EMA/emaInternet.nsf/ Page/Publications. Accessed 15 January 2008 Holland G (Ed.) (2007) Global Guide to Tropical Cyclone Forecasting. Bureau of Meteorology Research Centre (Australia) WMO/TC-No. 560, Report No. TCP-31, World Meteorological Organization. Geneva, Switzerland. http://www.bom.gov.au/bmrc/pubs/tcguide/globa_guide_ intro.htm. Accessed 15 January 2008 HM Government (2005) Emergency Preparedness Guidance on Part 1 of the Civil Contingencies Act 2004, Its Associated Regulations and Non-statutory Arrangements. Emergency Planning College ISDR (2006) Developing Early Warning Systems: A Checklist. EWC III Third International Conference on Early Warning from Concept to Action, 27–29 March 2006, Bonn, Germany. UN/ISDR Platform for the Promotion of Early Warning. http://www.unisdr.org/ppew/. Accessed 15 January 2008 Ishigaki T, Toda K, Baba Y, Nakagawa H, Shimada H (2006) Difficulty of Evacuation from Underground Space in Urban Flood. 7th International Conference on Hydroinformatics, HIC 2006, Nice, France MacFarlane R (2005) A Guide to GIS Applications in Integrated Emergency Management. Emergency Planning College, Cabinet Office NOAA/NWS (1997) Automated Local Flood Warning Systems Handbook, Weather Service Hydrology Handbook No. 2. National Oceanic and Atmospheric Administration, National Weather Service, Office of Hydrology, Washington, DC. http://www.weather.gov/oh/docs/ alfws-handbook/. Accessed 15 January 2008 Pajorova E, Hluchy E, Astalos J (2007) 3D Geovisualization Service for Grid-Oriented Applications of Natural Disasters. Cracow 07 Grid Workshop, Krakow, Poland Smith K, Ward R (1998) Floods: Physical Processes and Human Impacts. Wiley, Chichester, UK United Nations (2006) Guidelines for Reducing Flood Losses. A Contribution to the International Strategy for Disaster Reduction. United Nations. http://www.un.org/esa/sustdev/publications/ flood_guidelines.pdf. Accessed 15 January 2008

294

References

U.S. Army Corps of Engineers (USACE) (1995) Technical Guidelines for Hurricane Evacuation Studies. September 1995 U.S. Army Corps of Engineers (USACE) (1996) Hydrologic Aspects of Flood WarningPreparedness Programs. Report ETL 1110–2–540. Office of Chief of Engineers, Washington, DC. http://www.tpub.com/content/USACETechnicalletters/ETL-1110–2–540/index.htm. Accessed 15 January 2008 U.S. House of Representatives (2006) A Failure of Initiative. Final Report of the Select Bipartisan Committee to Investigate the Preparation for and Response to Hurricane Katrina. US Government Printing Office, Washington, DC. http://www.gpoaccess.gov/katrinareport/mainreport.pdf. Accessed 15 January 2008 Van Oosterom P, Zlatanova S, Fendel E M (Eds.) (2005) Geo-information for Disaster Management. Springer, Berlin Wolshon B, Urbina E, Levitan M (2001) National Review of Hurricane Evacuation Plans and Policies. LSU Hurricane Centre, Baton Rouge, LA Wolshon B, Urbina E, Wilmot C, Levitan M (2005a) Review of policies and practices for hurricane evacuation. I: transportation planning, preparedness, and response. Nat. Hazards Rev. 6(3): 129–142 Wolshon B, Urbina E, Levitan M, Wilmot C (2005b) Review of policies and practices for hurricane evacuation. II: traffic operations, management, and control. Nat. Hazards Rev. 6(3): 143–161 World Meteorological Organisation (2006a) Preventing and Mitigating Natural Disasters. WMO No. 993, Geneva World Meteorological Organisation (2006b) Social Aspects and Stakeholder Involvement in Integrated Flood Management. WMO No. 1008, Geneva

Chapter 10

Response

Collier C G, Cross R, Khatibi R, Levizzani V, Solheim I, Todini E (2005) ACTIF Best Practice Paper – The Requirements of Flood Forecasters for the Preparation of Specific Types of Warnings. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October 2005, Tromsø, Norway. http://www.actifec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Costes M, Abadie J P, Ducuing J L, Denier J S (2002) Tangible Results and Progress in Flood Risks Management with the PACTES Initiative. 34th COSPAR Scientific Assembly, The Second World Space Congress, 10–19 October 2002, Houston, TX Demeritt, D, Cloke H, Pappenberger F, Thielen J, Bartholmes J, Ramos M (2007) Ensemble predictions and perceptions of risk, uncertainty, and error in flood forecasting. Environ. Hazards 7: 115–127 Demuth J., Morss R E, Lazo J K, Stewart A E (2007) Assessing How the US Public Understands and Uses Weather Forecast Uncertainty Information. http://ams.confex.com/ams/pdfpapers/119820. pdf. Accessed 15 January 2008 Emergency Management Australia (1999) Guide 5 – Flood Warning. Australian Emergency Manuals Series, Part III Emergency Management Practice. Volume 3, 2nd edition. http://www. ema.gov.au/. Accessed 15 January 2008 Environment Agency (2007) Probabilistic Flood Forecasting Scoping Study. Joint Defra/EA Flood and Coastal Erosion Risk Management R&D Programme, R&D Technical Report FD2901/TR Erlich M (2007) OSIRIS - an example of citizen-oriented technology development in the area of dissemination of information on flood risk management. In Begum S, Stive M J F, Hall J W (Eds.), Flood Risk Management in Europe. Innovation in Policy and Practice, Springer, Dordrecht, The Netherlands Flikweert J J, Coremans C, De Gooijer K, Wentholt L (2007) Automation of flood contingency plans: benefits and implementation experiences. In Begum S, Stive M J F, Hall J W (Eds.),

References

295

Flood Risk Management in Europe. Innovation in Policy and Practice, Springer, Dordrecht, The Netherlands Fu H (2004) Development of Dynamic Travel Demand Models for Hurricane Evacuation. Ph.D. dissertation, Louisiana State University and Agriculture and Mechanical College Glassheim E (1997) Fear and loathing in North Dakota. Natural Hazards Observer XXI(6): 1–4 Huaimin G (2005) Flood Forecasting and Management in China. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October 2005, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 ISDR (2006) Guidelines for reducing flood losses. International Strategy for Disaster Reduction. http://www.unisdr.org/eng/library/isdr-publication/flood-guidelines/isdr-publication-floods. htm. Accessed 15 January 2008 Krzysztofowicz R (2001) The case for probabilistic forecasting in hydrology. J. Hydrol. 249: 2–9 Langkamp E J, Wentholt L R, Pengel B E, Gooijer C, de, Flikweert J J (2005) NOAH, the right information at the right time at the right place. Proceedings of Third International Flood Defence Symposium, Nijmegen, The Netherlands MacFarlane R (2005). A Guide to GIS Applications in Integrated Emergency Management. Emergency Planning College, Cabinet Office Martini F, De Roo A (Eds.) (2007) EXCIFF Guide: Good Practice for Delivering Flood Related Information to the General Public. European Commission/Joint Research Centre report EUR22760EN. http://exciff.jrc.it/downloads/exciff-related-documents/EXCIFF_guide.pdf. Accessed 15 January 2008 National Research Council (2003) Communicating Uncertainties in Weather and Climate Information: A Workshop Summary. National Academies Press, Washington. DC http://www. nap.edu/catalog.php?record_id = 10597. Accessed 15 January 2008 National Research Council (2006) Completing the Forecast: Characterizing and Communicating Uncertainty for Better Decisions Using Weather and Climate Forecasts. National Academies Press, Washington, DC. http://www.nap.edu/catalog.php?record_id = 11699. Accessed 15 January North Cornwall District Council (2004) Boscastle The Flood 16–08–04 Pierce C, Bowler N, Seed A, Jones D, Moore R (2005) Towards Stochastic Fluvial Flood Forecasting: Quantification of Uncertainty in Very Short Range QPF’s and Its Propagation Through Hydrological and Decision Making Models. Second ACTIF Workshop on Quantification, Reduction and Dissemination of Uncertainty in Flood Forecasting, Delft, 23–24 November 2004. http://www.actif-ec.net/Workshop2/ACTIF_WS2_Session1-cont. html. Accessed 15 January 2008 Prastacos P, Castaings W, Courtois N, Hluchy L, Houdry P, Tran V (2004) ANFAS: a decision support system for flood risk assessment. In Prastacos P, Cortes U, de Leon J L D, Murillo M (Eds.), e-Environment: Progress and Challenges. Instituto Politecnico Nacional, Mexico Prendergast S (2007) London’s Command Support System. Fire Magazine, July 2007: 26–29 PREVIEW (2007) User Link: Special Issue on Floods. PREVIEW and RISKEOS project newsletter Issue No. 3, http://www.preview-risk.com and http://www.risk-eos.com. Accessed 15 January 2008 Regnier E, Harr P A (2006) A dynamic decision model applied to hurricane landfall. Weather Forecast 21(5): 764–780 Ritzen M R (2005) The High Water Information System and Flood Management in the Netherlands. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October 2005, Tromsø, Norway. http://www.actifec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Rodrigues A S, Santos M A, Santos A D, Rocha F (2006) Dam-break flood emergency management system. Water Resour. Manage. 16(6): 1573–1650 Roulin E (2007) Skill and relative economic value of medium-range hydrological ensemble predictions. Hydrol. Earth Syst. Sci. 11: 725–737. http://www.hydrol-earth-syst-sci. net/11/725/2007/hess-11-725-2007.html. Accessed 15 January 2008 Simonovic S P, Ahmad S (2005) Computer-based model for flood evacuation emergency planning. Nat. Hazards 34: 25–51

296

References

Van der Vaat M, Erlich M, Lumbroso D, Mens M (2007) Methodology for Decision Support Systems for Flood Event Management. Floodsite Report T19–07–02. http://www.floodsite. net/. Accessed 15 January 2008 Wolshon B, Urbina E, Wilmot C, Levitan M (2005) Review of policies and practices for hurricane evacuation. I: transportation planning, preparedness, and response. Nat. Hazards Rev. 6(3): 129–142 World Meteorological Organisation (1994) Guide to Hydrological Practices. 5th edition. WMO No. 168, Geneva Zhou H, He B, Zhu L, Liang G (2004) The Application of Web GIS in Emergency Flood Control System. IAHS Publication, 289, ICGRHWF Conference, Three Gorges Dam, China, September 2003

Chapter 11

Review

Andryszewski A, Evans K, Haggett C, Mitchell B, Whitfield D, Harrison T (2005) Levels of Service Approach to Flood Forecasting and Warning. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October 2005, Tromsø, Norway. http://www.actif-ec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Andryszewski A, Evans K, Haggett C, Mitchell B, Whitfield D (2005) Flood Warning – The Next Step Change. 40th Defra Flood and Coastal Management conference, York, England Ash J, Postle M, Fenn T, Crossman M (2005) Evaluating a Multi-Criteria Analysis Methodology for Application to Flood Management and Coastal Defence Appraisals. 40th Defra/ Environment Agency Flood and Coastal Management Conference, July 2005, York, England Barnes L R, Gruntfest E C, Hayden M H, Schultz D M, Benight C (2007) False alarms and close calls: a conceptual model of warning accuracy. Weather Forecast. 22(5): 1140–1147 Basher R (2006) Global early warning systems for natural hazards: systematic and people-centred. Phil. Trans. R. Soc. A 364: 2167–2182. doi:10.1098/rsta.2006.1819 Betts R (2003) The missing links in community warning systems: findings from two Victorian community warning system projects. Aust. J. Emerg. Manage. 18(3): 37–45 Carsell K M, Pingel N D, Ford D T (2004) Quantifying the benefit of a flood warning system. Nat. Hazards Rev., ASCE, August 2004: 131–140 CNS Scientific and Engineering Services (1991) The Benefit Cost of Hydrometric Data – River Flow Gauging. Report prepared for Department of Environment, Reading Defra (2004) Making space for water. Developing a new Government strategy for flood and coastal erosion risk management in England. Flood Warning and Forecasting. One of a series of background papers and technical documents to accompany the consultation exercise. Flood Management Division, July 2004 Defra (2005) Evaluating a Multi-Criteria Analysis (MCA) methodology for Application to Flood Management and Coastal Defence Appraisals. Joint Defra/Environment Agency Flood and Coastal Erosion Risk Management R&D Programme, R&D Technical Report FD2013 Dixon R, Boissonnade A, Krebs T, O’Shay A (2006) Event Response Activities for the Near-Real Time Assessment of Financial Losses in Landfalling Hurricanes. AMS 27th Conference on Hurricanes and Tropical Meteorology. http://ams.confex.com/ams/27Hurricanes/techprogram/ programexpanded_339.htm. Accessed 15 January 2008 Drabek T E (2000) The social factors that constrain human responses to flood warnings. In Parker D J (Ed.), Floods. Routledge, London Elliot J F, Stewart B J (2000) Early warning for flood hazards. In Parker D J (Ed.), Floods. Routledge, London Elliott J, Handmer J, Keys C, Tarrant M (2003) Improving Flood Warning – Which Way Forward ? Paper presented at the Australian Disaster Conference, Canberra. http://www.ses.nsw.gov.au/ infopages/2337.html. Accessed 15 January 2008

References

297

Emergency Management Australia (1999) Guide 5 – Flood Warning. Australian Emergency Manuals Series, Part III Emergency Management Practice. Volume 3, 2nd edition. http://www. ema.gov.au/. Accessed 15 January 2008 Environment Agency (2002) Fluvial Flood Forecasting for Flood Warning - Real Time Modelling. Defra/Environment Agency Flood and Coastal Defence R&D Programme. R&D Technical Report W5C-013/5/TR Environment Agency (2003) Flood risks to people Phase 1. Defra/Environment Agency Flood and Coastal Defence R&D Programme, R&D Technical Report FD2317/TR1 Environment Agency (2004) Best Practice in Coastal Flood Forecasting. Defra/Environment Agency Flood and Coastal Defence R&D Programme, R&D Technical Report FD2206/TR1 Environment Agency (2007) Risk assessment for flood incident management. Joint Defra/ Environment Agency Flood and Coastal Erosion Risk Management R&D Programme. R&D Technical Report SC050028/SR1 Handmer J, Henson R, Sneeringer P, Konieczny R, Madej P (2001) Warning systems for flash floods: research needs, opportunities and trends. In Gruntfest E, Handmer J (Eds.), Coping with Flash Floods. Kluwer, Dordrecht, The Netherlands, pp. 77–89 Handmer J (2001) Are flood warnings futile? Risk communication in emergencies. Aust. J. Disaster Trauma Stud. 2000(2), New Zealand Henson R (2001) U.S. Flash flood warning dissemination via radio and television. In Gruntfest E, Handmer J (Eds.), Coping with Flash Floods. Kluwer, Dordrecht, The Netherlands, pp. 243–252 ISDR (2006) Lessons for a Safer Future: Drawing on the Experience of the Indian Ocean tsunami disaster. International Strategy for Disaster Reduction. http://www.unisdr.org/ppew/tsunami/ pdf/lesson-for-a-safer-future.pdf. Accessed 15 January 2008 Jonkman S N, Kelman I (2005) An analysis of the causes and circumstances of flood disaster deaths. Disasters 29(1): 75–97 Krzysztofowicz R, Kelly K S, Long D (1994) Reliability of flood warning systems. ASCE J. Water Res. Plan. Manage. 120(6): 906–926 Martini F, De Roo A (Eds.) (2007) EXCIFF Guide: Good Practice for Delivering Flood Related Information to the General Public. European Commission/Joint Research Centre report EUR22760EN. http://exciff.jrc.it/downloads/exciff-related-documents/EXCIFF_guide.pdf. Accessed 15 January 2008 Merz B, Kreibich H, Thieken A, Schmidtke R (2004) Estimation uncertainty of direct monetary damage to buildings. Nat. Hazards Earth Syst. Sci. 4: 153–163 National Hydrologic Warning Council (2002) Use and Benefits of the National Weather Service River and Flood Forecasts. http://www.nws.noaa.gov/oh/ahps/AHPS%20Benefits.pdf. Accessed 15 January 2008 Norouzi A, Ghavasieh A, Attari J (2007). Uncertainty Evaluation of Flood Warning System; Reliability and Trade-off. 32nd Congress of IAHR, the International Association of Hydraulic Engineering & Research, July 1–6 2007, Venice, Italy Parker D J, Fordham M, Torterotot J P (1994) Real time hazard management: flood forecasting, warning and response. In Penning-Rowsell E, Fordham M (Eds.), Floods Across Europe: Hazard Assessment, Modeling and Management. Middlesex University Press, UK Parker D J (2003) Designing Flood Forecasting, Warning and Response Systems from a Societal Perspective. Paper presented at the International Conference on Alpine Meteorology and Meso-Alpine Programme, 21 May 2003, Brig, Switzerland. http://www.map.meteoswiss.ch/ map-doc/icam2003/Presentation/10.1/Brig_document.pdf. Accessed 15 January 2008 Parker D, Tunstall S, Wilson T (2005) Socio-Economic Benefits of Flood Forecasting and Warning. ACTIF International Conference on Innovation Advances and Implementation of Flood Forecasting Technology, 17–19 October 2005, Tromsø, Norway. http://www.actifec.net/conference2005/proceedings/index.html. Accessed 15 January 2008 Pfister N (2002) Community response to flood warnings: the case of an evacuation from Grafton, March 2001. Aust. J. Emerg. Manage., Autumn 2002. http://www.ema.gov.au. Accessed 15 January 2008

298

References

Sene K J, Ainsworth E J, Graham G (2006) Risk Based Design of Telemetry Networks for Flood Warning Applications. 41st Defra/Environment Agency Flood and Coastal Management Conference, York, England Simonovic S P, Ahmad S (2005) Computer-based model for flood evacuation emergency planning. Nat. Hazards 34: 25–51 Tilford K A, Sene K J, Khatibi R (2007) Flood forecasting model selection: a structured approach. In Begum S, Stive M J F, Hall J W (Eds.), Flood Risk Management in Europe, Springer, Dordrecht, The Netherlands United Nations (2006a) Global Survey of Early Warning Systems. An Assessment of Capacities, Gaps and Opportunities Towards Building a Global Early Warning System for All Natural Hazards. United Nations, Geneva. http://www.unisdr.org/ppew/. Accessed 15 January 2008 United Nations (2006b) Guidelines for Reducing Flood Losses. International Strategy for Disaster Reduction. http://www.unisdr.org/eng/library/isdr-publication/flood-guidelines/isdr-publication-floods.htm. Accessed 15 January 2008 U.S Army Corps of Engineers (USACE) (1994) Framework for Estimating National Economic Development Benefits and Other Beneficial Effects of Flood Warning and Preparedness Systems. Institute for Water Resources Report IWR-94–3, Alexandria, Virginia U.S. Army Corps of Engineers (USACE) (1996) Hydrologic Aspects of Flood WarningPreparedness Programs. Report ETL 1110-2-540, Office of Chief of Engineers, Washington, DC. http://www.tpub.com/content/USACETechnicalletters/ETL-1110-2-540/index.htm. Accessed 15 January 2008 U.S. House of Representatives (2006) A Failure of Initiative. Final Report of the Select Bipartisan Committee to Investigate the Preparation for and Response to Hurricane Katrina. US Government Printing Office, Washington, DC. http://www.gpoaccess.gov/katrinareport/mainreport.pdf. Accessed 15 January 2008 World Meteorological Organisation (1973) Benefit and Cost Analysis of Hydrological Forecasts: A State of the Art Report. Operational Hydrology Report No. 3, WMO No. 341, Geneva World Meteorological Organisation (1989) Natural Disaster Reduction: How Meteorological and Hydrological Services Can Help. WMO No. 722, Geneva World Meteorological Organisation (1994) Guide to Hydrological Practices. 5th edition. WMO No. 168, Geneva World Meteorological Organisation (2007) Economic Aspects of Integrated Flood Management. WMO/GWP Associated Programme on Flood Management. WMO No. 1010, Geneva

Glossary ACTIF (2004) Some research needs for river flood forecasting in FP6. Achieving Technological Innovation in Flood Forecasting. European Commission Project EVK1-CT-2002-80014. http://www.actif-ec.net/documents/ACTIFResearchNeedsfor%20FP6V1.4.pdf. Accessed 15 January 2008 Beven K J (2001) Rainfall Runoff Modelling: The Primer. Wiley, Chichester, UK Holland G (Ed.) (2007) Global Guide to Tropical Cyclone Forecasting. Bureau of Meteorology Research Centre (Australia) WMO/TC-No. 560, Report No. TCP-31, World Meteorological Organization, Geneva, Switzerland. http://www.bom.gov.au/bmrc/pubs/tcguide/globa_guide_ intro.htm. Accessed 15 January 2008 IDNDR (1992) Internationally agreed glossary of basic terms related to Disaster Management. IDNDR 1990-2000. DHA-Geneva, December 1992 UNESCO/WMO (2007) International Glossary of Hydrology. International Hydrological Programme. http://webworld.unesco.org/water/ihp/db/glossary/glu/aglu.htm. Accessed 15 January 2008 UN/ISDR (2004) Terminology: Basic terms of disaster risk reduction (http://www.unisdr.org/eng/library)

Index

A Action tables, 59, 73 ALERT systems, 45,75 Artificial neural networks, 54, 133, 141, 147, 169–171, 190, 194, 226, 239 Astronomical tides, 156, 157 Australia, 1, 82, 88, 102, 120, 158, 171, 217, 260 Automatic Weather Station, 24

B Bangladesh, 4, 82, 102, 183 Bayesian Techniques, 54, 135, 169, 170, 225, 253 Brazil, 102, 194 Business Continuity Management, 218, 224

C Canada, 17, 194, 201, 252 Catchment rainfall estimation, 26 Central America, 83, 102, 183–184 China, 4, 32, 102, 194, 241 Coastal Forecasting astronomical tides, 156–157 data based, 152, 169 hurricanes, 22, 149, 150, 151, 162, 163–165, 172 process based, 152, 156–169 surge, 66, 108, 120, 157–165, 169, 171, 172 transformation matrices, 171 tropical cyclones, see hurricanes wave overtopping, 167–169, 171, 172, 173 waves, 165–167, 171 Communication public awareness, 79, 82, 213, 251 risk, 9–12, 259

uncertainty, 17, 86, 120–122, 244–248 warning messages, 84–87, 250, 256 Conceptual Models flow routing, 145–146 rainfall runoff, 131, 137–139 Contingency planning, 72, 220–226 Contingency Table Cost Loss, 245 forecasting model performance, 111–112, 122 threshold performance, 68 Control Rooms, 77–79 Control Structures river, 142, 176, 190–196 Thames Barrier, 197–198 tidal barriers, 197–199 Cost Benefit Analysis, 261–265 Cost Loss polder operations, 55–56 rainfall alarms, 54 reservoir operations, 122, 194 Utility Function, 122, 246 Cyclones. See Tropical cyclones

D Dams. See Reservoirs Data Assimilation error prediction, 106–107 parameter updating, 107–108 state updating, 107–108, 153, 162, 186 Data Based Models artificial neural networks, 141, 147, 169–171, 190 coastal, 169–171 flow routing, 146–147 ice forecasting, 190 rainfall runoff, 131, 139–141 transfer function, 140, 147

299

300 Debris flows, 181, 205 Decision Support Systems, 122, 194, 237–244 Delta. See Estuaries Denmark, 30 Detection automatic weather station, 24 evaporation, 23 network design, 47–49 performance improvements, 254–255 rainfall, 24–36 remote sensing, 28–33 river levels and flows, 37–42 snow, 27–28, 33 stage-discharge relationship, 39–41 telemetry, 44–47 tidal levels, 37–39 wave monitoring, 42–43 weather radar, 28–32 Dikes. See Flood Defences Disdrometers, 25 Dissemination internet, 82–83 multimedia, 82–83 performance improvements, 256–257, 258–260 RANET, 85 techniques, 79–87 transient populations, 81, 83, 256 uncertainty, 17, 86, 120–122, 244–248 warning messages, 84–87, 250, 256

E Economic Analysis cost benefit, 261–265 flood warning benefits, 263–264 multi criteria analysis, 261, 265–266 stage-damage relationships, 264 Emergency Management Australia, 2, 4, 88, 209, 217, 250 Ensemble Forecasting coastal, 120, 159, 164 forecast products, 117–119, 120–122 meteorological, 35, 36 multi-model forecasts, 35, 116 performance measures, 114 rainfall thresholds, 54–56 river models, 120, 128, 183, 193–194 snowmelt, 188 Environment Agency, 64, 76–77, 99–100, 120, 171–172, 197–198, 264–265 Estuaries, 63, 64, 150, 158, 161, 169, 171, 198–199

Index Evacuation Decision Support Tools, 240 hurricanes, 216, 233, 240

F False alarms, 46, 67, 68, 69, 111, 233, 251–252, 255 Federal Emergency Management Agency (FEMA), 213, 214, 215–217, 220, 228, 240 Finland, 82, 102 Flash Flood Boscastle Event, 236–237 definitions, 181 Flash Flood Guidance, 53, 183–185 forecasting, 181–185 thresholds, 53–54., 56, 205 World Meteorological Organisation, 183–185 Flash Flood Guidance, 53, 183–185 Flood Defences breach, 12, 203–204, 240 emergency works, 232, 241 overtopping, 167–169, 173 temporary barriers, 15 Flood Emergency Plans General Principles, 71–73, 209–219 operational response, 59, 73 Table Top Exercises, 219–220 Validation and Testing, 73, 219–220 Flood Event Management Preparatory Actions, 231–234 Flood Forecasting data availability, 126–128 flow routing, 141–147, 190–199 ice, 188–190 integrated catchment models, 133, 175–180 model calibration, 108–112 model design, 93–97, 123–126, 149–153 performance improvements, 113–114, 257 rainfall runoff models, 132–141 simple forecasting techniques, 61–67 simple triggers, 61–67 surge, 157–165, 171–173 systems, 97–104 ungauged flows, 178–180 urban drainage, 199–202 waves, 165–169, 171–173

Index Flood Risk causes of flooding, 8–9 Flood Risk Assessment, 9–13, 217 Flood Warning Areas, 73–75 hydraulic modelling, 11–12 transient populations, 12 Flood Warning Procedures Flood Warning Areas, 73–75 Flow Control, 14, 73, 190–202, 232 Flow Routing Models conceptual, 145–146 data based, 146–147 process based, 142–145 Forecasting Points, 94–95, 124–126, 151 Forecasting Systems data hierarchy, 104 France, 54, 82, 217, 241, 242

G Geographical Information Systems, 227–228, 237–239 geotechnical risks, 202–206 Germany, 82, 147, 183, 194, 201, 241, 242 Glacial Lake Outburst Floods, 84, 181, 202, 204 Groundwater flooding, 203

H Hong Kong, 158, 205 Hurricanes, 9, 150, 162–165, 172, 215–217, 238 Hydraulic Models coastal, 157–165 ice forecasting, 189 river modelling, 142–145 urban, 200–201

I Ice forecasting, 188–190 ice jams, 189 stage discharge relationships, 189 India, 85, 183 International Strategy for Disaster Reduction (ISDR), 88, 212 Ireland, 241–242 Italy, 194

J Japan, 11, 32, 82

301 K Kalman filter, 108, 120, 190 KNMI, 55–56, 158

L Lead time evacuation, 216, 233 flash flooding, 181, 232, 236 flood warning, 67–68, 86 forecast, 95–96, 181 forecasting models, 127, 171 targets, 251 telemetry network design, 48 thresholds, 51, 57, 65–66, 255 time delays, 57, 95–96 tropical cyclones, 216 Lessons Learned Reports, 250 Levees. See Flood Defences Levels of service, 251–252, 265 Luxembourg, 201

M Mesoscale, 34, 162 Meteorburst telemetry, 28, 44, 45 Multi Criteria Analysis, 261, 265–266

N Nepal, 84, 183 Netherlands, 35, 55–56, 108, 158. 194, 199, 241, 242–244 Norway, 28, 102 Nowcasting, 35–36 Numerical Weather Prediction, 34–35, 162

P Performance Monitoring dissemination systems, 83 flood warning systems, 83, 249–253 forecasting models, 113–114 thresholds, 67–70, 255 Polders, 55–56, 192 Portugal, 171 Preparedness All-Hazard Approaches, 217 Flood Emergency Plans, 209–219 Resilience, 220–226 Probabilistic also. See Ensemble Forecasting flood warnings, 74, 244–248 forecasts, 16–17, 114–122

302 Probabilistic (cont.) Numerical Weather Prediction, 35 rainfall thresholds, 55–56 Risk Assessment, 9–13, 224–225 Process Based Models coastal, 156–169 flow routing, 142–145 rainfall runoff, 131, 135–137 Proudman Oceanographic Laboratory, 158–160 Public awareness, 79, 82, 213, 251

Q Quantitative Precipitation Forecast. See Numerical Weather Prediction

R Rainfall alarms, 51–56 catchment rainfall estimation, 26–27 depth duration, 52 disdrometers, 25 forecasts, 33–36 microwave attenuation, 33 nowcasting, 35–36 raingauges, 24–25 satellite observations, 32–33 thresholds, 51–56 weather radar, 28–32 Rainfall Runoff Models conceptual, 137–139 data based, 139–141 process based, 135–137 Rating curve. See stage discharge relationship Real time updating. See Data Assimilation Red River, 17, 247 Reservoirs dam break, 203–204, 240 decision support systems, 193–194 flood forecasting, 190–194 probabilistic forecasting, 122, 193–194 Resilience control rooms, 78 dissemination systems, 80, 88 flood warning systems, 220–226 forecasting systems, 100–101, 103–104 instrumentation, 38 telemetry networks, 46, 48, 104, 128 thresholds, 58, 62 Risk to life, 4, 10, 259, 262, 265 River Gauging Stations level monitoring, 37–42 Russia, 32

Index S Satellite altimetry, 38 rainfall measurements, 32–33 soil moisture measurement, 32–33 telemetry, 44, 45 wave monitoring, 43 SCADA, 46 SLOSH, 163–165 Snow degree-day method, 186 monitoring, 27–28 snowmelt forecasting, 185–188 Somalia, 18 Spain, 194 Stage-discharge relationship, 39–41 STOWA, 242 Surge forecasting, 66, 108, 120, 157–165, 169, 171, 172

T Taiwan, 194 Telemetry meteorburst, 28, 44, 45 networks, 48 radio, 44, 45 satellite, 44, 45 telephone, 44, 45 Thames Barrier, 197–198 Thresholds alarm handling, 46 meteorological indicators, 54, 182 performance, 67–70, 255 rainfall, 51–56, 182 risk based, 54 river level, 56–61, 182 tidal level, 56–61 Tidal barriers, 197–199 Timeline, 234–237 Transfer function, 120, 139–141, 147 Transient populations, 10, 81, 83, 256 Triggers. See Thresholds Tropical Cyclone Programme, 154–156 Tropical cyclones, 2, 4, 9, 88, 150, 151, 153–156, 162, 172, 210, 215, 216 Tsunami, 9, 205–206, 250

U Uncertainty Emergency Response, 244–248 forecasting models, 114–122 reservoir forecasting, 193–194 thresholds, 54–56, 58

Index Ungauged catchment, 136, 178–180 United Kingdom, 30, 76–77, 87, 99–100, 114, 120, 158–160, 177, 197–198, 218–219, 252, 264–265 Urban drainage, 199–202 Urban Drainage and Flood Control District, 83 US Army Corps of Engineers, 2, 88, 209, 240, 263 US National Weather Service, 28, 30, 45, 53, 75, 82, 88, 102, 120, 158, 164, 172, 184, 188, 240 USA, 11, 28, 30, 45, 53, 82, 88, 102, 120, 158, 163, 171, 172, 194, 205, 213, 215–217, 240, 260 Utility function, 54, 122, 194, 246

303 V Virtual Emergency Operations Centres, 238 Visualisation and simulation, 228–229 Vulnerability, 10, 12, 214

W Waves forecasting, 165–167 monitoring, 42–43 overtopping, 167–169 types, 150, 166 Weather radar, 28–32 World Meteorological Organisation, 5–7, 32, 45, 88, 134, 153–156, 183–185