D I S A S T E R R I S K M A N A G E M E N T S E R I E S N O . 5 34423 Natural Disaster Hotspots A Global Risk Analysis THE WORLD BANK Other Disaster Risk Management Series Titles 1 Managing Disaster Risk in Mexico: Market Incentives for Mitigation Investment 2 Managing Disaster Risk in Emerging Economies 3 Building Safer Cities: The Future of Disaster Risk 4 Understanding the Economic and Financial Impacts of Natural Disasters Disaster Risk Management Series Natural Disaster Hotspots A Global Risk Analysis by Maxx Dilley,1 Robert S. Chen,2 Uwe Deichmann,3 Arthur L. Lerner-Lam,4 and Margaret Arnold5 with Jonathan Agwe,5 Piet Buys,3 Oddvar Kjekstad,6 Bradfield Lyon,1 and Gregory Yetman2 The World Bank Hazard Management Unit 2005 Washington, D.C. 1International Research Institute for Climate Prediction (IRI), Columbia University 2Center for International Earth Science Information Network (CIESIN), Columbia University 3Development Economics Research Group (DECRG), The World Bank 4Center for Hazards and Risk Research (CHRR) and Lamont-Doherty Earth Observatory (LDEO), Columbia University 5Hazard Management Unit (HMU), The World Bank 6International Centre for Geohazards (ICG), Norwegian Geotechnical Institute (NGI) © 2005 The International Bank for Reconstruction and Development / The World Bank and Columbia University 1818 H Street, NW Washington, DC 20433 Telephone 202-473-1000 Internet www.worldbank.org E-mail feedback@worldbank.org All rights reserved. 1 2 3 4 08 07 06 05 Copyright 2005, International Bank for Reconstruction and Development/The World Bank and Columbia University. This material may be copied for research, education, or scholarly purposes. All materials are subject to revision. The views and interpretations in this document are those of the individual author(s) and should not be attributed to the World Bank or Columbia University. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of the World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this work is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of applicable law. The World Bank encourages dissemination of its work and will normally grant permission promptly. For permission to photocopy or reprint any part of this work, please send a request with complete information to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; telephone 978-750-8400; fax 978-750-4470; www.copyright.com. All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, World Bank, 1818 H Street NW; Washington, DC 20433, USA; fax 202-522-2422; e-mail pubrights@worldbank.org. ISBN 0-8213-5930-4 978-0-8213-5930-3 e-ISBN 0-8213-5931-2 Library of Congress Cataloging-in-Publication Data has been applied for. Contents Preface vii Acronyms and Abbreviations xi 1. Executive Summary 1 Project Approach 1 Key Findings of the Global Analyses 1 Key Findings of the Case Studies 12 Conclusions and the Way Forward 12 2. Project Objectives 19 3. Project Approach 23 Risk Assessment Framework 23 Selection of Natural Hazards 25 Units of Analysis 26 Summary of Data Sources and Data Preparation 27 Global Hotspots Classification 33 4. Single-Hazard Exposure Analysis 35 Cyclones 35 Drought 35 Floods 35 Earthquakes 43 Volcanoes 44 Landslides 44 Single-Hazard Analysis of Exposure 44 5. Multihazard Exposure Analysis 47 Simple Multihazard Index 47 Reclassification of Multihazard Areas by Population Density 52 6. Multihazard Risk Assessment 55 Derivation of Vulnerability Coefficients 55 Single-Hazard Risk Assessment Results 60 7. Multihazard Risk Assessment Results 81 8. Case Studies 93 Scale Issues 94 iii iv Natural Disaster Hotspots: A Global Risk Analysis Summary of Case Study Results 94 Linkages to and Lessons for Global Analysis 110 9. Conclusions and the Way Forward 113 The Costs of Disaster Risks 113 Implications for Decision Making 115 Information Development for Disaster Risk Management 117 Appendix A: Technical Appendix for Global Analysis 119 A.1 Derivation of Tropical Cyclone and GDP Surfaces 119 A.2 Reclassification of Hazardous Areas Weighted by Exposure 120 A.3 World Bank Country Income Classifications 127 References 130 Boxes Box 6.1 Risk Assessment Procedure for Both Mortality and Economic Losses, Illustrated by the Mortality Example 59 Tables Table 1.1 Countries Most Exposed to Multiple Hazards 4 Table 1.2 Countries at Relatively High Mortality Risk from Multiple Hazards 8 Table 3.1 Ranking of Major Natural Hazards by Number of Deaths Reported in EM-DAT 26 Table 3.2 Number of Input Units Used in the Gridded Population of the World (GPW) Data Sets, Versions 1-3 27 Table 3.3 Summary of Data Sources for Each Hazard 29 Table 3.4 Summary of Data Sources for Exposure 31 Table 3.5 Summary of Exposure Data for World and Unmasked Areas 32 Table 4.1 Characteristics of High-Hazard Areas by Hazard: Top Three Deciles 43 Table 5.1 Summary Statistics for the Simple Multihazard Index 48 Table 5.2 Hazard Profile for High-Cyclone Exposed Areas 52 Table 5.3 Summary Statistics for the Population-Weighted Multihazard Index 52 Table 6.1 Mortality-Related Vulnerability Coefficients 56 Table 6.2 Economic Loss-Related Vulnerability Coefficients 57 Table 6.3 Characteristics of High-Risk Areas by Hazard 64 Table 7.1 Characteristics of High-Risk Disaster Hotspots 88 Table 7.2 Countries at Relatively High Economic Risk from Multiple Hazards 89 Table 8.1 Summary of Case Studies 94 Table 8.2 An Expert Synthesis of Storm Surge Hotspots around the World 102 Table 8.3 Potential and Actual Hotspots Vulnerable to Flooding by Storm Surge 112 Table 9.1 Countries Receiving High Levels of International Disaster Assistance, 1992 through 2003 114 Table 9.2 Countries Receiving Emergency Loans and Reallocation of Existing Loans to Meet Disaster Reconstruction Needs, 1980 through 2003 115 Table 9.3 Direct and Indirect Losses for Six Major Disasters 116 Table A1.1 Available Tropical Cyclone Data by Region 119 Table A1.2 Subnational GDP Data 120 Table A3.1 World Bank Country Income Classifications: High Income 127 Table A3.2 World Bank Country Income Classifications: Low and Middle Income 128 Contents v Figures Figure 1.1 Global Distribution of Areas Highly Exposed to One or More Hazards, by Hazard Type 3 Figure 1.2 Global Distribution of Highest Risk Disaster Hotspots by Hazard Type 5 Figure 1.3 Proportion of National Population in Highest Risk Areas from Two or More Hazards (Mortality) 10 Figure 1.4 Proportion of National Population in Highest Risk Areas from One or More Hazards (Mortality) 11 Figure 1.5 Proportion of GDP in Highest Risk Areas from Two or More Hazards (Economic Losses) 13 Figure 1.6 Proportion of GDP in Highest Risk Areas from One or More Hazards (Economic Losses) 14 Figure 3.1 Mask Used to Eliminate Sparsely Populated, Nonagricultural Areas 28 Figure 4.1 Distribution of Hazardous Areas by Hazard Type 36 Figure 4.2 Exposure Measures by Hazard Decile 45 Figure 5.1 Global Distribution of Areas Significantly Exposed to One or More Hazards, by Number of Hazards 49 Figure 5.2 Detailed View of Multihazard Areas 50 Figure 5.3 Global Distribution of Multiple Hazards by Population Density Category 53 Figure 6.1 Global Distribution of Cyclone Risk 61 Figure 6.2 Global Distribution of Drought Risk 65 Figure 6.3 Global Distribution of Flood Risk 69 Figure 6.4 Global Distribution of Earthquake Risk 72 Figure 6.5 Global Distribution of Volcano Risk 75 Figure 6.6 Global Distribution of Landslide Risk 78 Figure 7.1 Global Distribution of Disaster Risk Hotspots for All Hazards 82 Figure 7.2 Global Distribution of Disaster Risk Hotspots by Number of Hazards 85 Figure 8.1 Frequency with Which Climatic Drought Hazard Events Were Accompanied by Drought Disasters or Not from 1979 through 2001 95 Figure 8.2 WASP Estimates of Climatic Drought and Drought Disasters for Central Southwest Asian Countries 96 Figure 8.3 WASP Estimates of Climatic Drought and Drought Disasters for Lao PDR and India 97 Figure 8.4 Modeled Landslide Zonation and GEORISK Landslide Inventory in Armenia 98 Figure 8.5 Landslide Hazard Map for Central America and Andean South America 99 Figure 8.6 Landslide Mortality Risks Calibrated with Historical Landslide-Related Mortality from the EM-DAT International Disaster Database 100 Figure 8.7 Multihazard Risk Map Constructed by Weighting Each Hazard Index by Incidence Frequency Data from EM-DAT Database 104 Figure 8.8 Multihazard Risk Map Constructed by Weighting Each Hazard Index by the Relief Expenditure Data for Each Hazard between 1948 and 1992 105 Figure 8.9 Multihazard Disaster Risk, Caracas 107 Figure 8.10 Location Map of Tana River and Garissa Districts with Coverage of Tana River Basin in Garissa District, Kenya 108 Figure 8.11 Livelihood Zones Overlaid on El Niño 1997­98 Flood Case 109 Figure A2.1 Single-Hazard Exposure Index Based on Top Three Population-Weighted Deciles 121 Preface As this volume goes to print, millions of people in Asia with the newly established Center for Hazards and Risk attempt to rebuild their lives and communities follow- Research (CHRR) at Columbia University to discuss the ing the devastating earthquake and tsunami that occurred possibility of a global-scale, multihazard risk analysis on December 26, 2004. The earthquake occurred off the focused on identifying key "hotspots" where the risks coast of Sumatra, registering 9.0 on the Richter scale, of natural disasters are particularly high. The project and causing tsunami waves that swept through the Indian would aim to provide information and methods to inform Ocean at a rate of 500-700 km per hour, devastating priorities for reducing disaster risk and making deci- coastal areas of countries across South and Southeast sions on development investment. Discussions culmi- Asia and East Africa. More than 220,000 people were nated in a jointly sponsored "brainstorming" workshop killed,thousandsmorewereinjured,andmillionsaffected. held at Columbia in September 2001 at which a small Damage to infrastructure, social systems, and the envi- group of experts examined in depth whether such an ronment has been substantial. At the time of this writ- analysis was feasible and worthwhile. A summary of ing, preliminary damage and needs assessments theworkshopandpresentationsisavailableontheProVen- undertaken by the World Bank and other partners esti- tion Consortium Web site at: http://www.provention- mate the damages at nearly $6 billion for Indonesia, the consortium.org/conferences/highriskhotspots.htm. Maldives, and Sri Lanka alone. Developed from that initial workshop, the Identifi- The tragic impacts and seeming enormity of this event cation of Global Natural Disaster Risk Hotspots (Hotspots) have thrown many around the world into a state of dis- project was implemented under the umbrella of the belief. As shocking as the tsunami disaster is, however, ProVention Consortium by World Bank staff from the it's important to remember that events of this magni- HMU and the Development Economics Research Group tude have happened in other places around the world, (DECRG) and Columbia University staff from the CHRR, and they will happen again. In 1984, persistent droughts the Center for International Earth Science Information in Ethiopia and Sudan killed 450,000. In Bangladesh in Network (CIESIN), the International Research Institute 1991, nearly 150,000 lives were taken by a cyclone. for Climate Prediction (IRI), and the Lamont-Doherty Hundreds of natural disasters, both large and small, occur Earth Observatory (LDEO). The project has also bene- each year. While the largest capture the attention of the fited greatly from close collaboration with the Norwe- global media, there are hundreds more events that we gian Geotechnical Institute (NGI), the United Nations don't hear about. The cumulative effect of these smaller Development Programme (UNDP), the United Nations and medium-sized disasters have equally devastating Environment Programme (UNEP), the United Nations impacts on developing countries: loss of development Office for the Coordination of Humanitarian Affairs gains, torn communities, and increased impoverishment. (OCHA), the United Nations World Food Programme The poor in these countries are consistently the most (WFP), the U.S. Geological Survey (USGS), the Inter- severely affected. national Strategy for Disaster Reduction (ISDR), and The Hotspots initiative began in 2001, when the World other individuals and groups. Bank's Disaster Management Facility (DMF), now the In November 2002, a second workshop was held at Hazard Management Unit (HMU), initiated discussions Columbia University involving experts on key natural vii viii Natural Disaster Hotspots: A Global Risk Analysis hazards as well as potential case study authors. (For more providing complementary funding of the project and information on this workshop, see http://www. their support of the Caracas case study. proventionconsortium.org/conferences/high- The Hotspots project benefited enormously from inter- riskhotspots2002.htm.) This workshop reviewed the ini- actions with the project on Reducing Disaster Risk, a col- tial plans and approaches under development by the laborative effort involving UNDP, UNEP, and others. core project staff, coordinated plans for the case stud- WeespeciallythankYasminAysan,PascalPeduzzi,Andrew ies, and obtained feedback from the World Bank and Maskrey, and Ron Witt for their willingness to exchange others, including the new director of the Earth Institute data, methods, and ideas. These two projects share a at Columbia University, Professor Jeffrey Sachs. This common approach with regard to analysis of disaster workshop led to the preparation of a revised work plan, risk and vulnerability. Pablo Recalde played a key role including the addition of several new case study activ- in organizing WFP participation in the project and case ities to the project. Intensive project work continued in studies. We also acknowledge the support of the U.S. 2003, culminating in a working meeting in December Agency for International Development (USAID) for the 2003 at which key results were reviewed and plans devel- Tana River case study. oped for the final project reports and dissemination of We thank Kathy Boyer for her extensive help with results. In March 2004, a review and synthesis meeting project management and implementation, especially with was held at the World Bank in Washington, D.C., regard to the case studies. We very much appreciate the where project results were presented to experts from tireless efforts of Piet Buys of DECRG and Greg Yetman the ISDR Working Group III on Vulnerability, Risk and and Kobi Abayomi of CIESIN to access, transform, and Impacts; the World Bank; and other interested organi- analyze the wide range of global data used in this proj- zations. ect. We gratefully acknowledge the extensive adminis- This report contains the results of the global hotspots trative and organizational support provided by Stacey analysis as well as summaries of the case studies, which Gander of the CHRR and Jennifer Mulvey, Ed Ortiz, are being published as a separate volume. The list of case and Hannia Smith of CIESIN. We also thank our col- studies and contributors is provided in Table 8.1. This leagues within the Earth Institute at Columbia Univer- publication does not examine tsunami hazard risk, as sity for their extensive inputs and guidance on a wide comprehensive data sets were not available during the range of issues, both organizational and technical. These course of the study. However, plans are being made to individuals include Deborah Balk, George Deodatis, include an analysis of tsunami-related risks in a subse- Klaus Jacob, Upmanu Lall, Marc Levy, Brad Lyon, Roberta quent phase of hotspots research. Balstad Miller, Chet Ropelewski, Jeffrey Sachs, Andrew The project team wishes to thank the HMU--espe- Smyth, Angeletti Taramelli, Jeff Weissel, and Lareef Zubair. cially its former manager, Alcira Kreimer--for her strong We are grateful to Matt Barlow, Klaus Jacob, Oddvar support, guidance, and encouragement throughout Kjekstad, and Sylvia Mosquera for their helpful reviews this challenging project. We thank Maryvonne Plessis- of the final draft. Of course, the opinions, conclusions, Fraissard, Director of the Transport and Urban Devel- and recommendations provided in this report are those opment Department, and Eleoterio Codato, Sector of the authors and not necessarily those of the World Manager for Urban Development, for their support of Bank, the Trustees of Columbia University in the City the initiative. We thank Maria Eugenia Quintero and Zoe of New York, our sponsors, partners, or colleagues. Trohanis at the HMU for their technical and organiza- Hotspots aims to provide a tool to get ahead of the tional contributions to the project. We especially thank disaster trend by highlighting areas that are most vul- theUnitedKingdom'sDepartmentforInternationalDevel- nerable to a number of hazards. We hope that develop- opment (DFID) and Norwegian Ministry of Foreign ment agencies and policymakers will use the information Affairs for their interest and financial support. We are to plan ahead for disasters and minimize their impacts. grateful to the CHRR, the Earth Institute, and the Lamont- This implies understanding the risk facing a particular Doherty Earth Observatory of Columbia University for community, city, or region, and integrating this under- Preface ix standing into development planning decisions. The knowledge and affordable technologies do exist to allow even low-income countries to significantly reduce the devastating social and economic impacts caused by such hazards as droughts, floods and earth- quakes that are part of the natural cycle of so many coun- tries. The triggers may be natural, but responsibility for the impacts of disasters belongs to all of us. Maxx Dilley, IRI Robert S. Chen, CIESIN Uwe Deichmann, DECRG, World Bank Art Lerner-Lam, CHRR/LDEO Margaret Arnold, HMU, World Bank Acronyms and Abbreviations CAS Country Assistance Strategy CHRR Center for Hazards and Risk Research CIESIN Center for International Earth Science Information Network CRED Centre for Research on the Epidemiology of Disasters DECRG Development Economics Research Group DFID UK Department for International Development DMF Disaster Management Facility (now HMU) DRI Disaster Risk Index ECLAC Economic Commission for Latin America and the Caribbean EM-DAT Emergency Events Database ENSO El Niño-Southern Oscillation ERL Emergency Reconstruction Loan FTS Financial Tracking System GDP Gross domestic product GIS Geographic Information System GPW Gridded Population of the World GSHAP Global Seismic Hazard Program HMU Hazard Management Unit ICG International Centre for Geohazards IFPRI International Food Policy Research Institute IFRC International Federation of the Red Cross IRI International Research Institute for Climate Prediction ISDR International Strategy for Disaster Reduction LDEO Lamont-Doherty Earth Observatory NGDC National Geophysical Data Center NGI Norwegian Geotechnical Institute NIMA National Imagery and Mapping Agency NRC National Research Council OCHA Office for the Coordination of Humanitarian Affairs pga Peak ground acceleration PNG Papua New Guinea PPP Purchasing power parity PreView Project of Risk Evaluation, Vulnerability, Information and Early Warning SRTM Shuttle Radar Topographic Mission UNDP United Nations Development Programme xi xii Natural Disaster Hotspots: A Global Risk Analysis UNEP United Nations Environment Programme USGS United States Geological Survey VEI Volcanic Explosivity Index VMAP(0) Vector Map Level 0 WASP Weighted Anomaly of Standardized Precipitation WFP World Food Programme WRI World Resources Institute Chapter 1 Executive Summary Earthquakes, floods, drought, and other natural haz- cyclones. By calculating relative risks for grid cells rather ards continue to cause tens of thousands of deaths, hun- than for countries as a whole, we are able to estimate dreds of thousands of injuries, and billions of dollars risk levels at subnational scales. in economic losses each year around the world. The The global analysis is limited by issues of scale as well Emergency Events Database (EM-DAT), a global disas- as by the availability and quality of data. For a number ter database maintained by the Centre for Research on of hazards, we had only 15- to 25-year records of events the Epidemiology of Disasters (CRED) in Brussels, records for the entire globe and relatively crude spatial infor- upwards of 600 disasters globally each year (http:// mation for locating these events. Data on historical dis- www.cred.be). Disaster frequency appears to be increas- aster losses, and particularly on economic losses, are ing. Disasters represent a major source of risk for the also limited. poor and wipe out development gains and accumulated While the data are inadequate for understanding the wealth in developing countries. absolute levels of risk posed by any specific hazard or As the recognition grows that natural disaster risk combination of hazards, they are adequate for identify- must be addressed as a development issue rather than ing areas that are at relatively higher single- or multi- one strictly of humanitarian assistance, so must our ple-hazard risk. In other words, we do not feel that the efforts to develop the tools to effectively mainstream data are sufficiently reliable to estimate, for example, disaster risk management into development activities. the total mortality risk from flooding, earthquakes, and This project has attempted to develop a global, synop- drought over a specified period. Nevertheless, we can tic view of the major natural hazards, assessing risks of identify those areas that are at higher risk of flood losses multiple disaster-related outcomes and focusing in par- than others and at higher risk of earthquake damage than ticular on the degree of overlap between areas exposed others, or at higher risk of both. We can also assess in to multiple hazards. The overall goal is to identify geo- general terms the exposure and potential magnitude of graphic areas of highest disaster risk potential in order losses to people and their assets in these areas. Such to better inform development efforts. information can inform a range of disaster prevention and preparedness measures, including prioritization of resources, targeting of more localized and detailed risk Project Approach assessments, implementation of risk-based disaster man- agement and emergency response strategies, and devel- In this report we assess the risks of two disaster-related opment of long-term land use plans and multihazard outcomes: mortality and economic losses. We estimate risk management strategies. risk levels by combining hazard exposure with histor- A set of case studies explores risks from particular ical vulnerability for two indicators of elements at risk-- hazards or for localized areas in more detail, using the gridded population and gross domestic product (GDP) same theoretical framework as the global analysis. We per unit area--for six major natural hazards: earth- hope that in addition to providing interesting and useful quakes, volcanoes, landslides, floods, drought, and results, the global analysis and case studies will stimu- 1 2 Natural Disaster Hotspots: A Global Risk Analysis late additional research, particularly at national and local The fact that some areas of the world are subject to levels, which will be increasingly linked to policy multiple hazards will not surprise many residents of making and practice in disaster risk reduction. those areas, but what this analysis reveals is the extent Within the constraints summarized above, we devel- to which, at global and regional scales, there is sub- oped three indexes of disaster risk: stantial overlap between different types of hazards and population concentrations. The world's geophysical 1. Mortality risks, assessed for global gridded popula- hazards--earthquakes and volcanoes--tend to cluster tion along fault boundaries characterized by mountainous 2. Risks of total economic losses, assessed for global terrain. Hazards driven mainly by hydro-meteorological gridded GDP per unit area processes--floods, cyclones, and landslides--strongly 3. Risks of economic losses expressed as a proportion affect the eastern coastal regions of the major continents of the GDP per unit area for each grid cell as well as some interior regions of North and South Risks of both mortality and economic losses are cal- America, Europe, and Asia. Drought is more widely dis- culated as a function of the expected hazard frequency persed across the semiarid tropics. The areas subject to and expected losses per hazard event. We obtained global both geophysically- and hydro-meteorologically-driven hazard data on cyclones, drought, earthquakes, floods, hazards fall primarily in East and South Asia and in Cen- landslides, and volcanoes from a variety of sources. tral America and western South America. Many of these The global hazard data sets were improved upon or, in areas are also more densely populated and developed the case of droughts and landslides, created specifi- than average, leading to high potential for casualties and cally for the analysis. Vulnerability was estimated by economic losses. Of particular concern in these areas obtaining hazard-specific mortality and economic loss are possible interactions between different hazards, for rates for World Bank regions and country wealth classes example, landslides triggered by cyclones and flooding, within them based on 20 years of historical loss data or earthquakes that damage dams and reservoirs needed from the EM-DAT database. for drought and flood protection. We masked out low-population and nonagricultural The global analysis supports the view that disaster areas where risks of losses are negligible. After calculat- risk management is a core issue of development. Com- ing the expected losses for each remaining grid cell, we paring Figures 1.1 and 1.2a illustrates the degree to ranked the grid cells and classified them into deciles (10 which exposure to hazards in developed countries has classes composed of roughly equal numbers of cells). not led to relatively high mortality in the past two decades Cells falling into the highest three deciles for either mor- in these areas. Areas of Europe and North America that talityoreconomiclossesareconsidereddisasterriskhotspots. are highly exposed to natural hazards as shown in Figure 1.1, for example, have not experienced correspondingly high mortality from these hazards over the past two Key Findings of the Global Analysis decades. The United States is noteworthy in that more than one-third of its population lives in hazard-prone Among the findings are that on the order of 25 million areas but only 1 percent of its land area ranks high in square kilometers (km2) (about 19 percent of the Earth's mortality risk. land area) and 3.4 billion people (more than half of the Figure 1.2 shows the types of hazards for which world's population) are relatively highly exposed to at each grid cell appeared in the top three deciles of the least one hazard. Some 3.8 million square kilometers global risk distribution for mortality (a) and economic and 790 million people are relatively highly exposed losses (b and c). Figure 1.2b shows that areas at high to at least two hazards. About 0.5 million square kilo- risk of economic losses are more widely distributed in meters and 105 million people are relatively highly industrial and lower-middle-income countries than areas exposed to three or more hazards (Figure 1.1). In some of high mortality risk. In addition to portions of Cen- countries, large percentages of the population reside in tral America and East and South Asia, large areas of the hazard-prone areas (Table 1.1). eastern Mediterranean and Middle East appear at high Executive Summary Figure 1.1. Global Distribution of Areas Highly Exposed to One or More Hazards, by Hazard Type Hazard Groups Top 3 Deciles Exposed to: Drought Only Geophysical Only Hydro Only Drought and Hydro Geophysical and Hydro Drought and Geophysical Drought, Hydro, and Geophysical Note: Geophysical hazards include earthquakes and volcanoes; hydrological hazards include floods, cyclones, and landslides. 3 4 Natural Disaster Hotspots: A Global Risk Analysis Table 1.1. Countries Most Exposed to Multiple Hazards a) Three or more hazards (top 15 based on land area) Country Percent of Percent of Max. Number Country Percent of Percent of Max. Number Total Area Population of Hazards Total Area Population of Hazards Exposed Exposed Exposed Exposed Taiwan, China 73.1 73.1 4 Vietnam 8.2 5.1 3 Costa Rica 36.8 41.1 4 Solomon Islands 7.0 4.9 3 Vanuatu 28.8 20.5 3 Nepal 5.3 2.6 3 Philippines 22.3 36.4 5 El Salvador 5.1 5.2 3 Guatemala 21.3 40.8 5 Tajikistan 5.0 1.0 3 Ecuador 13.9 23.9 5 Panama 4.4 2.9 3 Chile 12.9 54.0 4 Nicaragua 3.0 22.2 3 Japan 10.5 15.3 4 b) Two or more hazards (top 60 based on land area) Country Percent of Percent of Max. Number Country Percent of Percent of Max. Number Total Area Population of Hazards Total Area Population of Hazards Exposed Exposed Exposed Exposed St. Kitts and Nevis 100.0 100.0 2 Mexico 16.5 9.6 4 Macao, China 100.0 100.0 2 Korea, Dem. 16.4 13.5 3 Antigua and Barbuda100.0 100.0 2 People's Rep. of Hong Kong, China 100.0 100.0 2 Lao People's 15.2 12.6 3 Taiwan, China 99.1 98.9 4 Dem. Rep. of Vanuatu 80.8 75.6 3 Turkey 15.1 11.3 3 Costa Rica 80.4 69.2 4 Panama 15.0 12.6 3 Philippines 62.2 73.8 5 Swaziland 14.3 14.2 2 Nepal 60.5 51.6 3 Nicaragua 12.4 49.8 3 Guatemala 56.6 83.4 5 Afghanistan 11.1 29.5 3 Korea, Rep. of 53.0 53.6 2 Myanmar 10.7 10.4 4 Ecuador 47.6 74.6 5 India 10.5 10.9 4 Réunion 45.7 45.7 2 Lesotho 10.3 3.7 2 Vietnam 45.1 38.7 3 Iceland 9.4 4.8 2 Somalia 43.1 53.8 2 Colombia 8.9 7.5 3 South Africa 43.1 46.9 2 China 8.4 15.7 3 Japan 38.1 48.4 4 Kyrgyz Rep. 8.3 5.8 2 Cayman Islands 36.8 45.6 2 Dominica 8.1 6.2 2 Bangladesh 35.6 32.9 4 Peru 7.4 26.3 3 El Salvador 32.4 39.7 3 Iraq 7.3 9.6 3 Cambodia 27.9 4.4 3 Cuba 6.6 4.3 2 Chile 26.2 62.6 4 Papua New Guinea 5.9 6.4 3 Thailand 25.2 17.7 2 Jamaica 5.7 7.2 2 Fiji 23.2 29.0 2 Pakistan 5.6 18.2 2 Tajikistan 23.2 9.5 3 Indonesia 4.5 14.1 3 Solomon Islands 22.8 16.6 3 New Zealand 4.3 1.7 3 Madagascar 20.2 9.9 2 United Arab Emirates 4.1 6.8 2 Bhutan 20.1 29.2 4 Armenia 3.1 1.5 3 Georgia 17.4 5.9 3 Mongolia 2.8 0.7 2 Iran, Islamic Rep. of 17.1 22.2 4 Nigeria 2.7 6.7 2 Kenya 16.9 8.8 2 United States 2.6 11.2 4 Executive Summary Figure 1.2. Global Distribution of Highest Risk Disaster Hotspots by Hazard Type a) Mortality Risks High Mortality Risk Top 3 Deciles at Risk from: Drought Only Geophysical Only Hydro Only Drought and Hydro Geophysical and Hydro Drought and Geophysical Drought, Hydro, and Geophysical Note: Geophysical hazards include earthquakes and volcanoes; hydrological hazards include floods, cyclones, and landslides. 5 6 Figure 1.2. Global Distribution of Highest Risk Disaster Hotspots by Hazard Type b) Total Economic Loss Risks High Total Economic Loss Risk Top 3 Deciles at Risk from: Natural Drought Only Geophysical Only Disaster Hydro Only Drought and Hydro Geophysical and Hydro Hotspots: Drought and Geophysical Drought, Hydro, and Geophysical A Global Note: Geophysical hazards include earthquakes and volcanoes; hydrological hazards include floods, cyclones, and landslides. Risk Analysis Executive Summary Figure 1.2. Global Distribution of Highest Risk Disaster Hotspots by Hazard Type c) Economic Loss Risks as a Proportion of GDP Per Unit Area High Proportional Economic Loss Risk Top 3 Deciles at Risk from: Drought Only Geophysical Only Hydro Only Drought and Hydro Geophysical and Hydro Drought and Geophysical Drought, Hydro, and Geophysical Note: Geophysical hazards include earthquakes and volcanoes; hydrological hazards include floods, cyclones, and landslides. 7 8 Natural Disaster Hotspots: A Global Risk Analysis risk of loss from multiple hazards. These regions still or GDP in hotspots are especially likely to incur repeated rank high when the risk is recalculated by dividing the disaster-related losses and costs. Comparison of these losses per grid cell by each grid cell's GDP estimate maps with data on relief and reconstruction costs is (Figure 1.2c). In contrast, much of Europe and the instructive in this regard. Data on relief costs associ- United States no longer rank among the highest risk ated with natural disasters from 1992 to 2003 are areas when grid cells are ranked according to losses as available from the Financial Tracking System (FTS) of a proportion of GDP. the United Nations Office for the Coordination of Human- The statistics also suggest that future disasters will itarian Affairs (OCHA) (http://www.reliefweb.int/fts/). continue to impose high costs on human and eco- Total relief costs over this period are US$2.5 billion. Of nomic development. In 35 countries, more than 1 in this, US$2 billion went to just 20 countries, primarily 20 residents lives in an area identified as relatively high for disasters involving the following hazards (listed in in mortality risk from three or more hazards (Table 1.2a). order of magnitude of the relief amount allocated): China More than 90 countries have more than 10 percent of (earthquakes and floods); India (earthquakes, floods, their total population in areas at relatively high mor- and storms); Bangladesh (floods); the Arab Republic of tality risk from two or more hazards (Table 1.2b and Egypt (earthquakes); Mozambique (floods); Turkey Figure 1.3). And 160 countries have more than one- (earthquakes); Afghanistan (drought and earthquakes); fourth of their total population in areas at relatively high El Salvador (earthquakes); Kenya (drought and floods); mortality risk from one or more hazards (Figure 1.4). the Islamic Republic of Iran (earthquakes); Pakistan Similarly, many of the areas at higher risk of loss from (drought and floods); Indonesia (drought, earthquakes, multiple hazards are associated with higher-than-aver- and floods); Peru (earthquakes and floods); Democra- age densities of GDP, leading to a relatively high degree tic Republic of Congo (volcanoes); Poland (floods); Viet- of exposure of economically productive areas (Figures nam (floods and storms); Colombia (earthquakes); 1.5 and 1.6). Venezuela (floods); Tajikistan (droughts and floods); Until vulnerability, and consequently risks, are and Cambodia (floods). All of these countries except reduced, countries with high proportions of population Egypt have more than half of their population in areas Table 1.2. Countries at Relatively High Mortality Risk from Multiple Hazards a) Three or more hazards (top 35 based on population) Country Percent of Percent of Country Percent of Percent of Total Area Population in Total Area Population at Risk Areas at Risk at Risk at Risk Taiwan, China 90.2 95.1 Madagascar 6.3 24.8 El Salvador 51.7 77.7 Trinidad and Tobago 10.0 23.5 Costa Rica 38.2 77.1 Ecuador 3.6 21.4 Philippines 45.6 72.6 Bhutan 10.5 18.8 Dominica 70.8 71.1 Chile 1.0 18.7 Antigua and Barbuda 46.2 69.5 Malawi 5.5 12.9 Guatemala 28.8 69.4 Solomon Islands 0.1 12.0 Japan 23.2 69.4 Mexico 4.4 10.8 Dominican Rep. 33.7 66.0 Fiji 4.1 9.4 Jamaica 40.5 58.8 Albania 4.0 8.6 Nicaragua 4.4 42.7 Cuba 3.5 8.5 Indonesia 4.4 40.1 Samoa 0.7 8.3 Comoros 39.6 32.0 Afghanistan 0.8 8.1 Honduras 18.1 31.8 Pakistan 1.4 5.9 Nepal 31.9 28.0 Venezuela 0.9 5.6 Bangladesh 30.0 26.2 Cameroon 1.1 5.5 Colombia 1.8 25.9 Panama 2.6 5.1 Mozambique 4.7 25.5 Executive Summary 9 Table 1.2. Countries at Relatively High Mortality Risk from Multiple Hazards b) Two or more hazards (top 96 based on population) Country Percent of Percent of Country Percent of Percent of Total Area Population in Total Area Population at Risk Areas at Risk at Risk at Risk Bangladesh 97.1 97.7 Afghanistan 7.2 46.0 Nepal 80.2 97.4 Georgia 19.2 44.0 Dominican Rep. 97.3 96.8 Cameroon 9.2 42.0 Burundi 96.3 96.6 Fiji 20.0 42.0 Haiti 93.4 96.5 St. Vincent and Grenadines 41.6 41.6 Taiwan, China 92.5 95.5 Mexico 15.1 41.3 Malawi 70.8 95.3 Togo 61.2 39.3 El Salvador 83.0 92.6 St. Kitts and Nevis 31.8 39.1 Honduras 64.5 91.5 Zimbabwe 10.1 39.0 Guatemala 54.9 89.5 Congo,Rep. Of 1.9 38.8 Philippines 76.6 88.6 Benin 37.2 38.6 Costa Rica 53.6 86.1 Belize 19.8 38.2 Trinidad and Tobago 63.4 85.1 Sierra Leone 13.0 35.7 Japan 34.7 84.0 United States 1.1 35.1 Antigua and Barbuda 54.5 82.0 China 10.6 33.4 Dominica 84.7 82.0 Romania 14.4 33.3 Nicaragua 38.1 81.9 Uzbekistan 2.5 30.6 South Africa 12.1 78.7 Mali 2.9 29.6 Cuba 87.0 77.5 Lebanon 19.2 29.2 Niger 14.0 76.4 Sudan 5.0 28.8 Korea, Dem. People's Rep. of 58.5 72.8 Tajikistan 5.8 28.2 Vietnam 59.3 71.4 India 21.9 27.2 Ethiopia 29.9 69.3 United Kingdom 7.9 27.0 Nigeria 47.5 68.8 Liechtenstein 23.1 26.6 Chile 5.3 68.3 Uganda 27.5 26.6 Ecuador 20.3 67.2 Canada 0.04 25.3 Korea, Rep. of 25.2 66.7 Syrian Arab Rep. 8.0 24.9 Colombia 12.8 66.3 Turkey 12.6 24.7 Kenya 29.0 63.4 Bolivia 0.6 24.7 Burkina Faso 35.1 61.7 Lao People's Dem. Rep. 9.1 22.4 Bhutan 31.2 60.8 New Zealand 0.8 22.4 Venezuela 6.7 60.1 Ireland 0.6 21.9 Indonesia 10.6 59.3 Congo, Dem. Rep. of 2.5 21.6 Mozambique 16.9 58.9 Chad 2.7 20.5 Jamaica 40.5 58.8 Central African Rep. 0.5 19.7 Guam 23.8 58.5 Jordan 3.0 17.7 Peru 5.7 57.5 Yugoslavia Fed. Rep. 17.1 17.5 Albania 33.4 56.7 (Serbia/Montenegro) Madagascar 15.7 56.0 Myanmar 4.5 16.8 Barbados 54.9 54.9 Angola 0.2 14.8 Comoros 59.0 54.2 Rwanda 13.3 14.2 Tanzania 27.7 53.7 Panama 9.3 14.1 Somalia 15.4 53.3 Samoa 1.4 13.9 Senegal 10.1 52.9 Macedonia, FYR 22.4 13.7 Grenada 52.1 52.1 Kyrgyz Rep. 2.3 13.2 Lesotho 52.4 50.5 Solomon Islands 0.1 12.0 Montserrat 50.3 50.3 Ghana 15.2 11.6 Pakistan 22.8 49.6 Thailand 2.6 10.7 Iran, Islamic Rep. of 14.3 46.6 10 Figure 1.3. Proportion of National Population In Highest Risk Areas from Two or More Hazards (Mortality) Population at Risk from 2+ Natural Hazards (Mortality Weighted) Proportion of National Population Disaster 0 ­ 0.25 0.26 ­ 0.50 0.51 ­ 0.75 Hotspots: 0.79 ­ 0.98 A Global Risk Analysis Executive Summary Figure 1.4. Proportion of National Population In Highest Risk Areas from One or More Hazards (Mortality) Population at Risk from 1+ Hazards (Mortality Weighted) Proportion of National Population 0.00 ­ 0.25 0.26 ­ 0.50 0.51 ­ 0.75 0.76 ­ 1.00 11 12 Natural Disaster Hotspots: A Global Risk Analysis at relatively high risk from one or more hazards 1. Scale matters. Geographic areas that are identified as (Figure 1.4). The countries subject to multiple hazards hotspots at the global scale may have a highly vari- in this list also are among those countries with at least able spatial distribution of risk at finer scales. one-fourth of their populations in areas at risk from two 2. Scale affects data availability and quality. Hazard, expo- or more hazards (Figure 1.3). The correspondence with sure, and vulnerability data are available at sub- economic losses is not quite as strong (Figure 1.6). national resolutions for individual countries and even Total World Bank emergency lending from 1980 to cities, as the analyses for Sri Lanka and Caracas show. 2003 was US$14.4 billion (http://www.worldbank.org/ More comprehensive, finer resolution, and better hazards). Of this, US$12 billion went to 20 countries, quality data permit more complete, accurate, and primarily for the following hazards (listed in order of reliable identification of multihazard hotspots. highest loan amount): India (drought, earthquakes, and 3. Scale affects the utility of the results. Better data reso- storms); Turkey (earthquakes and floods); Bangladesh lution and a richer set of variables contribute to results (floods and storms); Mexico (earthquakes and floods); that are more relevant for risk management planning Argentina (floods); Brazil (floods); Poland (floods); at the national to local scale, as illustrated in the Colombia (earthquakes and floods); the Islamic Repub- case study from Caracas. This is highly important, lic of Iran (earthquakes); Honduras (floods and storms); as decisions made at the local and national scales China (earthquakes and floods); Chile (earthquakes); have perhaps the greatest potential to affect risk levels Zimbabwe (drought); the Dominican Republic (storms); directly, whether positively or negatively. El Salvador (earthquakes); Algeria (earthquakes and 4. The global- and local-scale analyses are complemen- floods); Ecuador (earthquakes and floods; Mozambique tary. In some instances, national-to-local level risk (drought and floods); the Philippines (earthquakes); assessors and planners may be able to "downscale" and Vietnam (floods). All of these countries except global data for finer scale risk assessment to com- Poland have half of their population in areas at rela- pensate for a lack of local data. Ideally, however, global tively high mortality risk from one or more hazards analyses would be scaled up--generalized from more (Figure 1.4), and all of them have at least half of their detailed, finer scale data. In practice, many barriers GDP in areas of relatively high economic risk from one still remain. The global infrastructure for systemat- or more hazards (Figure 1.6). ically assembling and integrating relevant data sets for disaster risk assessment at multiple scales remains inadequate. Nonetheless, the fact that relevant data Key Findings of the Case Studies sets can be obtained and integrated at various scales creates the hope that one day data can be collected Recognizing the limitations of the global analysis, we and shared routinely to improve disaster risk assess- undertook a number of case studies designed to inves- ment both globally and locally. tigate the potential of the hotspots approach at regional, national, and subnational scales, drawing on more detailed and reliable data sources as well as on expert Conclusions and the Way Forward knowledge concerning specific hazards and regions. Three case studies addressed specific hazards: storm The Hotspots project has created an initial picture of surges, landslides, and drought. Three case studies the location and characteristics of disaster hotspots: addressed regional multihazard situations: Sri Lanka, areas at relatively high risk from one or more natural the Tana River basin in Kenya, and the city of Caracas, hazards. The findings of the analysis support the view Venezuela. that disasters will continue to impose high costs on The following are the key findings from the case human and economic development, and that disaster studies: risk should be managed as an integral part of develop- ment planning rather than thought of strictly as a human- Executive Summary Figure 1.5. Proportion of GDP In Highest Risk Areas from Two or More Hazards (Economic Losses) GDP at Risk from 2+ Hazards (Economic Loss Weighted) Proportion of National GDP 0.0 ­ 0.25 0.26 ­ 0.50 0.51 ­ 0.75 0.76 ­ 0.98 13 14 Figure 1.6. Proportion of GDP In Highest Risk Areas from One or More Hazards (Economic Losses) GDP at Risk from 1+ Hazards Natural (Economic Loss Weighted) Proportion of National GDP Disaster 0.00 ­ 0.25 0.26 ­ 0.50 Hotspots: 0.51 ­ 0.75 0.76 ­ 1.00 A Global Risk Analysis Executive Summary 15 itarian issue. The following paragraphs detail how dis- resources from productive investments to support con- aster risk information can be useful for development sumption over short periods. Emergency loans have policy and decision makers, and how it can be further questionable value as vehicles for long-term investment developed in order to increase its usefulness. and contribute to country indebtedness without nec- essarily improving economic growth or reducing poverty. As disasters continue to occur, high-risk countries The Costs of Disaster Risks will continue to need high levels of humanitarian The combination of human and economic losses, plus relief and recovery lending unless their vulnerability the additional costs of relief, rehabilitation, and recon- is reduced. struction, make disasters an economic as well as a humanitarian issue. Until vulnerability, and conse- Implications for Decision Making quently risks, are reduced, countries with high pro- portions of population or GDP in hotspots are The Hotspots analysis has implications for develop- especially likely to incur repeated disaster-related ment investment planning, disaster preparedness, losses and costs. Disaster risks, therefore, deserve and loss prevention. The highest risk areas are serious consideration as an issue for sustainable those in which disasters are expected to occur most development in high-risk areas. frequently and losses are expected to be highest. This provides a rational basis for prioritizing risk-reduc- The significance of high mortality and economic loss tion efforts and highlights areas where risk man- risks for socioeconomic development extends well agement is most needed. beyond the initial direct losses to the population and economy during disasters. Covariate losses accompa- International development organizations are key nying mortality, for example, include partial or total loss stakeholders with respect to the global analysis. The of household assets, lost income, and lost productivity. analysis provides a scientific basis for understanding Widespread disaster-related mortality can affect house- where risks are highest and why, as well as a method- holds and communities for years, decades, and even ological framework for regional- and local-scale analy- generations. sis. The identified risks then can be evaluated further In addition to mortality and its long-term conse- using more detailed data in the context of a region's or quences, both direct and indirect economic losses must country's overall development strategy and priorities. be considered (ECLAC and the World Bank 2003). This would serve development institutions and the coun- Direct losses are losses to assets, whereas indirect losses tries in several ways to facilitate the development of are the losses that accrue while productive assets remain better-informed investment strategies and activities. damaged or destroyed. During disasters, both direct and indirect losses accumulate across the social, pro- Assistance Strategies. A development institution such ductive, and infrastructure sectors. The pattern of losses as the World Bank may use the analysis at the global depends on the type of hazard and the affected sec- and/or regional level to identify countries that are at tors' vulnerabilities to the hazard. In large disasters, higher risk of disasters and "flag" them as priorities to cumulative losses across sectors can have macro- ensure that disaster risk management is addressed in economic impacts. the development of a Country Assistance Strategy (CAS). Disasters impose costs in addition to human and While in some countries there can be a seemingly long economic losses. Costs include expenditures for dis- list of urgent priorities to address in a CAS--e.g., reduc- aster relief and recovery and for rehabilitation and ing extreme poverty, fighting HIV/AIDS, promoting edu- reconstruction of damaged and destroyed assets. In cation, achieving macroeconomic stability--managing major disasters, meeting these additional costs can disaster risk should be considered an integral part of require external financing or international humanitar- the development planning to protect the investments ian assistance. Disaster relief costs drain development made rather than as a stand-alone agenda. The CAS 16 Natural Disaster Hotspots: A Global Risk Analysis should consider the consequences of unmitigated dis- resources toward investments that would restore eco- aster risk in terms of possible tradeoffs with long-term nomic activity quickly and relieve human suffering. socioeconomic goals. This report's global disaster risk analysis provides a basis for identifying situations in which future emer- Sector Investment Operations. In high-risk regions gency recovery loans are likely to be needed. This cre- and countries, it is particularly important to protect ates an opportunity for "preappraising" emergency investments from damage or loss, either by limiting loans, that is, designing a risk management strategy to hazard exposure or by reducing vulnerability. Risks of guide the allocation of emergency reconstruction damage and loss should also be taken into account when resources should such resources become necessary, or estimating economic returns during project prepara- to arrange for other types of contingency financing with tion. Investment project preparation, particularly in the development banks. high-risk areas identified in the global analysis, would benefit from including a risk assessment as a standard Improved Information for Disaster Risk Management practice. This report's theory and methods can be trans- lated easily into terms of reference for such assessments. The Hotspots project provides a common framework Such assessments should identify probable hazards, as for improving risk identification and promoting risk well as their spatial distribution and temporal charac- management through a dialogue between organiza- teristics (including return periods), and should evalu- tions and individuals operating at various geographic ate vulnerabilities to the identified hazards that should scales. The methods and results provide useful tools be addressed in the project design. for integrating disaster risk management into devel- opment efforts and should be developed further. Risk Reduction Operations. In high-risk countries and As a global analysis conducted with very limited local- areas within countries, repeated, large-scale loss events level participation and based on incomplete data, the can harm economic performance (Benson and Clay results presented here should not provide the sole 2004). It may be impossible to achieve development basis for designing risk management activities. The goals such as poverty alleviation in these areas without analysis does, however, provide a scientific basis for concerted efforts to reduce recurrent losses. Increas- understanding where risks are highest and why, as well ingly, risk and loss reduction are being seen as invest- as a methodological framework for regional- and local- ments in themselves, and disaster-prone countries are scale analysis. The identified risks then can be evalu- demonstrating a willingness to undertake projects in ated further using more detailed data in the context of which disaster and loss reduction are the principal aims. a region's or country's overall development strategy Such projects can include both hard and soft compo- and priorities. nents: measures to reduce the vulnerability and expo- We have designed the Hotspots approach to be open- sure of infrastructure, as well as emergency funds and ended to allow additional studies to be incorporated institutional, policy and capacity-building measures on an ongoing basis. It provides a common framework designed to increase the abilities of countries to manage for improving risk identification and promoting risk disaster risks. management through a dialogue between organizations and individuals operating at various geographic scales. Contingency Financing. Emergency recovery and The Hotspots analysis can be improved upon as a tool reconstruction needs after a major disaster may create and developed in several directions. a high demand for emergency financing. While such loans are usually appraised and approved relatively Improve Underlying Databases. The first direction is quickly, at times there can be delays in disbursing the to pursue the many opportunities in both the short funds, which increase the social and economic impacts and long term to improve the underlying databases for of the disaster. Advance planning for recovery and assessing disaster risks and losses. A range of new global- resource allocation would allow for better targeting of scale data sets is currently under development, includ- Executive Summary 17 ing a new global urban-extent database being devel- Explore Long-term Trends. A third direction is to oped by CIESIN in support of the Millennium Ecosys- explore a key long-term issue: the potential effect of tem Assessment. A joint project between the Earth underlying changes in hazard frequency (for example, Institute, the World Bank, and the Millennium Project due to human-induced climatic change) coupled with will develop a much more detailed and complete data- long-term trends in human development and settlement base on subnational poverty and hunger. Much more patterns. To what degree could changes in tropical storm comprehensive regional data sets will become avail- frequency, intensity, and position interact with contin- able in specific areas of interest. On a regional scale, ued coastal development (both urban and rural) to there are also much longer records of hazard events for increase risks of death and destruction in these regions? specific hazards that could be harnessed to improve esti- Are agricultural areas, already under pressure from mates of hazard frequency and intensity in high-risk urbanization and other land use changes, likely to become areas (for example, O'Loughlin and Lander 2003). Sig- more or less susceptible to drought, severe weather, or nificant improvements could be made in characterizing floods? Could other hazards such as wildfires poten- flood, drought and landslide hazards in particular. Exist- tially interact with changing patterns of drought, land- ing data on disaster-related losses is being compiled into slides, deforestation, and land use to create new types a multi-tiered system through which regularly updated of hotspots? Although some aspects of these questions historical data from multiple sources can be accessed. have been addressed in the general context of research Additional work to link and cross-check existing data on climate change impacts, the interactions between is needed, however, as is improvement in the assess- climate change, the full range of hazards, and evolving ment and documentation of global economic losses. human hazard vulnerability have not been fully explored (for example, Brooks and Adger 2003; Chen 1994). Undertake Case Studies. A second direction is to Pursuing work in these directions will necessarily explore more fully the applicability and utility of the involve a wide range of institutions--national, regional Hotspots approach to analysis and decision making at and international, public and private sector, academic regional, national, and local scales. The initial case stud- and operational. We hope that the Hotspots project ies are promising, but are certainly not on their own has contributed a building block in the foundation of sufficient to demonstrate the value of the overall approach a global effort to reduce disaster-related losses by man- or the specific data and methods under different con- aging risks rather than by managing emergencies. We ditions. More direct involvement of potential stake- look forward to continuing collaboration with part- holders would be valuable in extending the approach ners at all levels to put in place a global disaster risk to finer scales of analysis and decision making. To be management support system in order to mobilize the effective, efforts to improve risk identification in hotspot knowledge and resources necessary to achieve this goal. areas should be part of a complete package of techni- cal and financial support for the full range of measures needed to manage disaster risks, including risk reduc- tion and transfer. Chapter 2 Project Objectives Hundreds of disasters occur worldwide each year in For the most part, both scientists and decision makers locations without sufficient local capacity or resources tend to deal with different hazards separately. For exam- to prevent death and destruction and to support rapid ple, seismologists, structural engineers, and urban plan- recovery. Continuing rapid urbanization and coastal ners typically focus on mitigating earthquake risks development in hazard-prone regions and the poten- through such efforts as strengthening building codes tial for long-term changes in the intensity and frequency and structures, whereas climatologists, agronomists, of some hazards pose a serious challenge to sustainable and water resource managers address flood and drought development in both the developing and industrial risks through the development and maintenance of worlds. Decision makers at all levels of governance, from dams, reservoirs, and other water resource systems or the international to community levels, will face difficult through demand management. Although this approach choices about priorities for mitigating the risks of, for is appropriate to some degree, given the differences in example, frequent, smaller hazards such as floods and hazards and vulnerabilities, it is also important to con- landslides versus the risks of less frequent, more sider and manage the combined risks of all hazards and uncertain, but potentially much more deadly hazards vulnerabilities. such as earthquakes and tsunamis. Disaster response is often handled by a variety of Natural disasters occur when large numbers of people organizations at different levels of government and soci- or economic assets are damaged or destroyed during a ety, ranging from local volunteer groups to national civil- natural hazard event. Disasters have two sets of causes. ian and military agencies to international relief agencies The first set is the natural hazards themselves, includ- and nongovernmental organizations--each with its own ing floods, drought, tropical storms, earthquakes, vol- areas of expertise with regard to particular disaster types canoes, and landslides. The second set comprises the and its own limitations in terms of jurisdiction and mode vulnerabilities of elements at risk--populations, infra- of operation. A more complete picture of multihazard structure, and economic activities--that make them risks can assist in developing coordinated strategies for more or less susceptible to being harmed or damaged total risk management. by a hazard event. The Hotspots project seeks to contribute to existing Disaster-prone countries can be identified readily knowledge on global natural-disaster risks in the fol- from existing databases of past disasters. Countries them- lowing ways: selves may be aware of disaster-prone areas, either 1. Development of a spatially uniform, first-order, global through local knowledge and experience or through disaster risk assessment through the use of global formal risk assessments and historical data. The role of data sets in which the spatial distributions of haz- vulnerability as a causal factor in disaster losses tends ards, elements at risk, and vulnerability factors, rather to be less well understood, however. The idea that dis- than national-level statistics, are the primary inde- asters can be managed by identifying and managing spe- pendent variables cific risk factors is only recently becoming widely recognized. 19 20 Natural Disaster Hotspots: A Global Risk Analysis 2. Rigorous and precise definition of specific social nerability, and risk? An international relief organization and economic disaster-related outcomes, the risks concerned with prepositioning disaster relief supplies of which can be quantitatively assessed globally might ask, What hazards are likely to be of concern in areas 3. Identification of the hazard- and vulnerability-related inhabited by vulnerable populations? How can limited sup- causal components of risk on a hazard-by-hazard plies be positioned optimally to address a range of possible basis, taking into account the damaging character- hazard scenarios? istics of each hazard and the contingent vulnerabil- In the long run, we also expect the Hotspots approach ity characteristics of potentially affected exposed to be useful at the national and subnational levels. A elements national government might ask, In areas that face risks from multiple hazards, which pose the most significant risks? 4. Assessment of overall, multihazard, global natural What measures would be most effective in reducing vulner- disaster risks, stated in terms of specific disaster ability to all hazards? How much will achieving an accept- outcomes (mortality and economic losses) for pop- able level of risk cost, and how should resources be allocated? ulations, infrastructure, and economic activities at A local government or community organization might risk ask, Should certain risk management measures be avoided 5. Verification of the global risk assessment through a because they increase risks from other hazards? Can simple limited number of case studies of limited geo- changes to development and mitigation plans result in graphic scope that allow risk factors to be charac- long-term risk reduction? Is it possible to combine mitiga- terized in greater detail through the use of larger scale tion measures for single hazards cost-effectively? data and involvement of national- to local-level stake- Both international institutions and the regions and holders countries they serve may seek a deeper understanding 6. Documentation of the hazard, vulnerability, and of potential barriers to disaster mitigation--not only risk assessment methods used or generated in the technical and economic, but also cultural and politi- analysis to extend the project's scope by enlisting cal. They may wish to understand the long-term con- others who wish to contribute to an ongoing, long- sequences of unmitigated disaster risk in terms of possible term, scientific effort to assess global risk tradeoffs with long-term socioeconomic goals. What are Disaster relief and recovery not only consume the lion's the opportunity costs and benefits of addressing disaster risk? share of resources available for disaster management, How would overall wealth and the distribution of wealth be but also drain resources away from other social and affected in the longer term? Could persistent impacts of dis- economic development priorities. Risk management asters alter a country's global position in terms of future lend- investments in high-risk areas can be cost-effective in ing opportunities, trade, public health, or military security? preventing disaster losses and increasing disaster There is growing recognition of the need for better preparation, leading to quicker, better planned recov- data and information on hazards and disasters at both ery. Currently, high-risk areas typically are identified on national and international levels. Within the United the basis of national-level data of historical disasters States, several recent reports by the U.S. National Research and unevenly applied local knowledge. This project seeks Council (NRC) and the U.S. government have high- to assess the geographic distribution of risks across lighted the importance of both historical and current national boundaries. Uniform data and methods pro- data on hazard events and their associated impacts (NRC vide comparability from one area to another. 1999a, 1999b; Subcommittee on Disaster Reduction Key stakeholders for the global analysis are interna- 2003). At the international level, there is strong inter- tional organizations that promote disaster risk man- est in improving disaster information systems and asso- agement. For example, a global or regional lending ciated decision support tools (for example, ISDR 2003). organization might ask, Where could a new lending pro- A welcome shift in emphasis appears to be under way gram have the greatest risk reduction impact over the next from managing disasters by managing emergencies to 10 years? To what extent can existing data provide an ade- managing disaster risks. This shift is evident in recent quate assessment of the degrees of hazard, exposure, vul- publications such as the 2002 World Disasters Report: Project Objectives 21 Focus on Reducing Risk (International Federation of Red Near-term applications of the analysis are expected Cross and Red Crescent Societies 2002), Living with Risk to include the following: (ISDR 2004), and Reducing Disaster Risk: A Challenge for 1. A basis for further focus on high-risk areas by inter- Development (UNDP 2004). Risk assessment, reduction, national institutions concerned with disaster risk and transfer are the major elements of risk management management (Kreimer and others 1999), offering a desirable alter- native to managing disasters through emergency response. 2. Promotion of global/local partnerships for additional Risk reduction requires risk assessment in order to deter- risk assessment and collaborative development and mine which areas are at highest risk of disaster and why, implementation of risk reduction plans in high-risk so that appropriate and cost-effective mitigation meas- areas ures can be identified, adapted, and implemented. 3. Stimulation of further research on hazard and vul- As a global analysis conducted with very limited local- nerability risk factors in high-risk areas and on appro- level participation and based on incomplete data, the priate and cost-effective risk reduction and transfer results presented here should not provide the sole measures basis for designing risk management activities. The 4. A model mode of analysis based on consistent dis- analysis does, however, provide a scientific basis for aster risk theory, assessment methods, and data that understanding where risks are highest and why, as well can be improved upon and applied globally and in as a methodological framework for regional- and local- particular locations scale analysis. The identified risks then can be evalu- ated further using more detailed data in the context of 5. A platform of static risks over which dynamic risks a region's or country's overall development strategy can be overlaid at varying time scales, capturing and priorities. seasonal-to-interannual fluctuations in hazard prob- We have designed the Hotspots approach to be open- abilities such as those associated with El Niño- ended to allow additional studies to be incorporated Southern Oscillation (ENSO) events or long-term cli- on an ongoing basis. It provides a common framework matic trends, as well as socioeconomic risk factors for improving risk identification and promoting risk and trends fluctuating on both short and long time management through a dialogue between organizations scales and individuals operating at various geographic scales. Chapter 3 Project Approach A wide range of natural hazards cause death, damage, 3. The vulnerability of the elements exposed to specific and other types of losses in both industrial and devel- hazards. oping countries. Small-scale hazard events such as a Disaster losses are caused by interactions between hazard small flood, tornado, landslide, lightning strike, or earth events and the characteristics of exposed elements that tremor may cause very localized damage, injuring or make them susceptible to damage. A hazard's destruc- killing a few individuals and destroying or damaging a tive potential is a function of the magnitude, duration, limited number of structures. In contrast, large-scale location, and timing of the event (Burton and others events such as hurricanes and tropical cyclones, strong 1993). To be damaged, however, elements exposed to earthquakes, large volcanic eruptions, tsunamis, major a given type of hazard must also be vulnerable to that floods, and drought can kill tens of thousands of people hazard; that is, the elements must have intrinsic char- and injure many more; they can also cause significant acteristics that allow them to be damaged or destroyed economic and social disruption as a result of both direct (UNDRO 1979). Valuable but vulnerable elements damage and indirect economic losses. Often large- include people, infrastructure, and economically or envi- scale events such as storms, earthquakes, and droughts ronmentally important land uses. spawn ancillary hazards such as floods, landslides, and The destructive power of natural hazards combined wildfires that may add to casualties and economic losses. with vulnerabilities across a spectrum of exposed ele- The severity of such secondary events may depend in ments can lead to large-scale covariate losses during part on environmental conditions such as soil moisture, hazard events in areas where population and economic land cover, and topography as well as on the presence investment are concentrated. Aggregate losses start with and condition of protective works such as dams, dikes, losses to individual elements, reaching a point in dis- and drainage systems. aster situations where economic and social systems break down partly or completely, leading to higher net socioe- conomic impacts. Risk Assessment Framework Risks of individual element losses or of aggregate covariate losses can be stated as the probability of loss, General Framework or as the proportion of elements that will be damaged In this project, we use the commonly accepted risk or lost, over time (Coburn and others 1994). Disaster assessment framework for natural hazards (for exam- risk assessment examines the factors that cause losses ple, Coburn and others 1994; Mileti 1999). In essence, in order to estimate loss probabilities. Risk factors include we distinguish among three components that contribute the probability of destructive hazard events as well as to the overall risk of natural hazards: the contingent vulnerabilities of the exposed elements at risk. 1. The probability of occurrence of different kinds and The hazards research community has evolved a intensities of hazards dynamic paradigm for hazards analysis that includes a 2. The elements exposed to these hazards four-stage process of hazard preparedness, response, 23 24 Natural Disaster Hotspots: A Global Risk Analysis recovery, and mitigation (Mileti 1999). Within this 2. Multihazard hotspots. Some areas may be subject to a paradigm, assessment of vulnerability and risk is most variety of natural hazards and associated moderate useful at the stage of assessing hazard preparedness to high levels of risk of loss. In some cases, the haz- and designing hazard mitigation strategies. Indeed, Mileti ards themselves may be largely independent of each and colleagues have recommended adoption of a "global other; that is, the occurrence of one hazard does not systems perspective" that recognizes the complex "inter- significantly affect the probability that other hazards actions between earth and social systems, within and will occur. However, even if this is the case, the occur- across the global-to-local levels of human aggregation" rence of one hazard might significantly affect the over- (Mileti 1999: 27). The Hotspots approach is consistent all impacts of other hazards. For example, after a with this perspective. major tsunami hit Papua New Guinea (PNG) in July 1998, the PNG embassy issued an appeal in which it noted, "The tsunami is the latest of a series of nat- Terminology ural disasters striking Papua New Guinea in the last In its simplest terms, we define a natural disaster "hotspot" three and a half years. The volcano eruption in Rabaul, as a specific area or region that may be at relatively high cyclone Justin's destruction in the Milne Bay area, risk of adverse impacts from one or more natural hazard and the El Niño-induced drought in most parts of events. Use of the term "adverse" implies a normative the country, have caused a horrendous burden on judgment that at least some of the major consequences the Government and the people of Papua New Guinea" of a hazardous event are considered undesirable by those (International Disaster Situation Reports, 23 July 1998; affected: for example, the death or injury of people, see http://www.cidi.org/disaster/98b/0021.html). damage to, or loss of, economically valuable assets, or For both types of hotspot, exposure and vulnerabil- lost income and employment. Impacts on natural eco- ity must be high before risks are considered signifi- systems may also be of concern but are not explicitly cant. Such exposure and vulnerability could be in the addressed in this project. However, it is important to form of important economic assets, such as agricul- recognize that, for example, tropical storms may have tural areas that are vulnerable to drought or flood haz- adverse impacts on coastal populations in their imme- ards. In areas of relatively low population density, diate path but beneficial effects on agriculture and water some hazards could still pose high mortality (and mor- resources over much larger areas. The focus of disaster bidity) risks if vulnerability is high because of fragile management is to reduce or ameliorate the adverse infrastructure or other factors. In very high-density areas, impacts, generally in the context of other societal efforts even low vulnerability (low casualty rates) could result to take advantage of beneficial effects. in substantial losses in absolute terms (many deaths), Given the variety of natural hazards that continue to especially among those who may have higher-than-aver- cause significant adverse impacts in both industrial and age vulnerability (for example, slum dwellers living on developing countries, we categorize hotspots into two steep slopes). major types: Throughout this report, we use the term hazard 1. Single-hazard hotspots. Some areas or regions may be to represent a specific family of natural phenomena at relatively high risk of adverse impacts associated and degree of hazard to signify a particular hazard- with one major natural hazard. For example, seis- dependent measure of severity. Exposure represents the mologists have predicted that there is a 47­77 per- overlap of time and spatial distribution of human cent probability that the city of Istanbul, Turkey--with assets and the time and spatial distribution of hazard a population estimated at 8.7 million in 2000 (U.N. events. We use the term vulnerability to represent the Population Division 2004)--will experience strong apparent weaknesses of physical and social systems to shaking during the first 30 years of this century, particular hazards. Physical system vulnerability is usu- with great potential for death, injury, damage, and ally defined (especially in the engineering community) economic disruption (Hubert-Ferrari and others 2000; in terms of fragility curves, in which the weaknesses of Parsons and others 2000). physical systems (buildings and infrastructure, for exam- Project Approach 25 ple) are quantified as a function of hazard severity. fragility, and loss, especially in the time frames required Similar fragility curves for social systems--that is, a for policy decisions and mitigation investments. Instead, quantification of social vulnerability--are complex func- we propose that this analysis be a basis for developing tions of social, economic, political, and cultural vari- scenarios and counter-factual analyses of mitigation ables and are addressed in this report through the use alternatives to give policymakers a framework for their of proxies. In general terms, risk is a multiplicative func- investment decisions. The role of uncertainties can be tion of hazard severity, exposure, and fragility. included in such scenarios, as they relate to decision support, but the actual degrees of uncertainty are unlikely Limitations and Uncertainty to be useful in the near future. However, this lack of certainty should not be taken as an excuse for inaction. In designing the methodology for this report, we have been forced to accommodate the inherent hetero- geneity that characterizes risk assessments across mul- Selection of Natural Hazards tiple natural hazards. Although some attempts have been made (most notably by the insurance industry) to develop Data on natural hazards have been collected by differ- common risk metrics (such as average annualized loss: ent groups for different purposes in different ways. see Risk Management Solutions 2004), such methods The most comprehensive, publicly available global data- are themselves based on highly variable data quality, base on natural hazards and their impacts is the EM- incomplete fragility analysis, and insufficient historical DAT data set maintained by the Centre for Research on records. Where such data and analysis exist, as they do the Epidemiology of Disasters (CRED) in Brussels (Sapir for some regions, more comprehensive risk assessment and Misson 1992; see www.cred.be). This database con- is possible (as we point out in the case studies). tains more than 12,000 records of disasters from 1900 Our goal in this report is to estimate the relative multi- to the present, compiled from multiple sources. It includes hazard risk countries face using defensible measures of estimates of numbers of people killed and affected as degree of hazard and defensible proxies for physical and well as estimates of economic losses, derived from social vulnerabilities. Our metrics for degree of hazard documented sources. In many cases, these loss estimates or hazard severity vary according to the hazard. In our include direct losses not only from the primary event view, the science of hazard occurrence and magnitude (for example, a cyclone or earthquake) but also from has not developed enough to permit a globally consis- subsequent related events such as landslides and tent single metric for multihazard severity. Such met- floods. The database generally does not include geo- rics are currently the subject of basic research programs. physical or hydro-meteorological events that were not Lacking widely applicable measures of physical fragility reported as causing heavy losses, either because the events and social vulnerability, and lacking even uniform occurred in areas that were thinly populated at the standards for collecting the loss data needed to calibrate time, or because the losses were not reported in English- fragility, we have chosen to use broadly accepted and or French-language periodicals. relatively uniform proxies for vulnerability in the form In addition to the EM-DAT database, this project of masked population density, GDP, and transporta- has taken advantage of data sets developed by different tion network density, as normalized by total losses in groups around the world focused on specific hazard the Emergency Events Database (EM-DAT). As we explain probabilities, occurrences, or extents. This approach in the next section, we use a geographic mask designed has permitted us to identify areas that will be at rela- to identify agricultural land use and high population tively high risk of particular types of hazard events in density as first-order selection criteria to quantify the the future, regardless of their past levels of exposure or geographic distribution of exposure. actual losses. Our approach assumes that existing An analysis of this sort is not amenable to a quanti- databases are more likely to underreport smaller tative estimation of either aleatoric or epistemic uncer- events than large events. Areas at higher risk from large tainties. A meaningful error analysis may not be possible events therefore probably will be more accurately given the state of knowledge about hazard occurrence, identified than areas that suffer from smaller, more fre- 26 Natural Disaster Hotspots: A Global Risk Analysis quent events. However, the short record periods for 1975), these efforts were severely constrained by the some large but infrequent hazard events (for example, lack of detailed data, especially at the global level, as volcanic eruptions) suggest that efforts to assess absolute well as by limitations in computational capabilities levels of risk or to compare risk levels across hazards and data integration methods. would be premature. In 1994­95, the first global-scale gridded popula- Table 3.1 lists the major natural hazards reported in tion data set, known as the Gridded Population of the EM-DAT, ranked by the total number of deaths reported. World (GPW), version 1 data set, was developed with For this analysis, we selected six major disaster types primary support from CIESIN (Tobler and others 1995). for analysis: drought, tropical storms, floods, earth- This data set transformed population census data, which quakes, volcanoes, and landslides. We did not attempt most countries collected for subnational administrative to assess extreme temperature events (heat and cold units, into a regular "grid" of "spherical quadrilaterals" waves), wildfires, and wave/surge events such as tsunamis, with the dimensions of 5 minutes (5') of latitude and 5 owing to data and resource limitations. Nor did we assess minutes (5') of longitude and an average area of about some hazards that are primarily of regional or economic 55 square kilometers each (85 square kilometers at the concern, such as tornadoes, hail, and lightning. How- equator). Each cell contained an estimate of total pop- ever, in principle these hazards could be included in ulation and population density (on land) for 1994, based future efforts to improve and expand the hotspots on the overlap between the irregular boundaries of the approach. administrative units and the regular boundaries of the grid. Version 2 of GPW was developed by CIESIN in collaboration with the International Food Policy Research Units of Analysis Institute (IFPRI) and the World Resources Institute (WRI). Its cells have a nominal resolution of 2.5' lati- Most efforts to assess the impacts of natural hazards have tude by 2.5' longitude and contain population estimates used either events or countries as the basic unit of for 1990 and 1995 (CIESIN and others 2000). A beta analysis. That is, they have examined known occurrences test version of Version 3 is currently available with of hazards and associated impacts either on an event- population estimates for 1990, 1995, and 2000 with by-event basis or as aggregated to the national level. the same nominal resolution as GPW Version 2 (CIESIN This project takes advantage of new methods and and others 2004). With each new version, the number data that make possible a more detailed geospatial analy- of subnational administrative units used to create sis across multiple hazards. Although hazard mapping these gridded population estimates has increased, efforts began in the 1970s (for example, White and Haas from about 19,000 units in Version 1 to 127,000 in Ver- sion 2 to about 375,000 in Version 3. The underlying Table 3.1. Ranking of Major Natural Hazards by detail of the spatial distributions has therefore increased Number of Deaths Reported in EM-DAT dramatically. The improvement in resolution is sum- Rank Disaster Type All Deaths Deaths marized in Table 3.2. 1980­2000* 1992­2001** Using the 2.5' x 2.5' grid as a base, it is possible to 1 Drought 563,701 277,574 make a variety of estimates of hazard probability, occur- 2 Storms 251,384 60,447 rence, and extent on a common geospatial frame of ref- 3 Floods 170,010 96,507 erence. It is also possible to add supplementary measures 4 Earthquakes 158,551 77,756 of exposure such as the density of roads and railroads, 5 Volcanoes 25,050 259 6 Extreme temperature 19,249 10,130 the amount of agricultural land, and the economic value- 7 Landslides 18,200 9,461 added to the same framework. The result is a grid of 8 Wave/surge 3,068 2,708 approximately 8.7 million cells covering most of the 9 Wildfires 1,046 574 Total 1,211,159 535,416 occupied land area of the Earth within latitudes 85°N to 58°S. Each grid cell contains estimates of land area, * Compiled by O. Kjekstad, personal communication ** 2002 IFRC World Disaster Report (http://www.cred.be/emdat/intro. population, population density, various hazard proba- htm) Project Approach 27 Table 3.2. Number of Input Units Used in the Gridded Summary of Data Sources and Data Preparation Population of the World (GPW) Data Sets, Versions 1­3 Version Year Released Estimates for Input Units Hazard Data GPW v1 1995 1994 19,000 GPW v2 2000 1990, 1995 127,000 The first step in the hotspots analysis was to examine GPW v3 2003/04 1990, 1995, 2000 ~ 375,000 each hazard individually in terms of available spatial data on probability, occurrence, or extent. The most desirable input data would be complete probability den- bilities, and associated exposure and vulnerability char- sity functions for each hazard, that is, the probabilities acteristics. These grid cells may be aggregated, either of occurrence of a specific hazard for a range of sever- to a larger grid (for example, a 1° x 1° latitude/longi- ities or intensities in a specific future time period. Unfor- tude grid) or to national boundaries (making simple tunately, detailed probabilistic data of this type do not assumptions about grid cells along borders). exist for any hazards at the global level. A more limited Sincetheobjectiveofthisanalysisistoidentifyhotspots probabilistic estimate is available for earthquakes: the where natural hazard impacts may be large, it need not Global Seismic Hazard Program (GSHAP) has used both include the large proportion of the Earth's surface that is historic data and expert judgment to derive a global sparsely populated and not intensively used. We have map of the peak ground acceleration (pga) for which therefore chosen to mask out grid cells with population there is a 10 percent chance of exceedance in the next densities less than five persons per square kilometer (cells 50 years. with less than about 105 residents) and without signifi- Even without detailed probabilistic data, however, cant agriculture. Even if all residents of such cells were it is still possible to distinguish between areas of higher exposed and highly vulnerable to a hazard, total casual- and lower risk using occurrence data, that is, data on ties would still be relatively small in absolute terms, and specific events that took place during a given histori- the potential agricultural impact would be limited.1 cal period. The area affected by the events must be deter- Masking these cells reduces data processing require- mined by analysis or modeling of available data. ments and ensures that the large number of very low The data identified and used for each hazard are sum- risk cells do not dominate the results. In addition, hazard marized in Table 3.3. More detailed descriptions of the reporting and frequency data are likely to be poorest in individual data sets acquired and the transformations rural, sparsely populated areas, so masking could reduce applied are given in Appendix A.1. A brief summary anomalies caused by poor data. A total of approximately for each hazard follows: 4.1 million grid cells remain after applying the mask (Figure 3.1). These cells (colored orange, blue, or 1. Cyclones. For cyclones, we used storm track data green in the figure) include slightly more than half of collected from multiple sources and assembled into the world's estimated land area (about 72 million square geographic information system (GIS) coverages by kilometers, or about 55 percent of the total), but most the UNEP/GRID (Global and Regional Integrated of the world's population (6 billion people, or about Data)-Geneva Project of Risk Evaluation, Vulnera- 99.2 percent of the population estimate in GPW for bility, Information and Early Warning (PreView). This the year 2000). data set includes more than 1,600 storm tracks for the period 1 January 1980 through 31 December 2000 for the Atlantic, Pacific, and Indian Oceans.2 As described in detail in Appendix A.1, we modeled the wind speeds around the storm tracks in order to 1To determine agricultural land use, we used the U.S. Geologi- cal Survey (USGS) Global Land Cover Classification database at assess the grid cells likely to have been exposed to 30" resolution and dropped from the mask any cells with any high wind levels. one of three land covers typically associated with agriculture (Sebastian, personal communication, 2003). If any of the 25 2The record for the 1980s for some parts of the Indian and Pacific 30" cells in a 2.5' cell included an agricultural land cover, we Oceans are incomplete in some cases. See: http://www.grid. dropped the entire 2.5' cell from the mask. unep.ch/data/grid/gnv199.php. 28 Figure 3.1. Mask Used to Eliminate Sparsely Populated, Nonagricultural Areas Natural Analysis Mask Disaster Agriculture Only Inhabited Only Hotspots: Agriculture and Inhabited Note: Colored cells are those retained. A Global Risk Analysis Project Approach 29 2. Drought. For drought, we used the Weighted Anom- through 2002 (Advanced National Seismic System aly of Standardized Precipitation (WASP) devel- 1997). The GSHAP data were sampled at 1' inter- oped by IRI, computed on a 2.5° x 2.5° grid from vals, with a minimum peak ground acceleration of monthly average precipitation data for 1980 through 2 meters per second per second (m/s2), or approxi- 2000. The WASP assesses the precipitation deficit mately one-fifth of surface gravitational acceleration. or surplus over a specified number of months, 5. Volcanoes. For volcanoes, we used a spatial coverage weighted by the magnitude of the seasonal cyclic of volcanic activity (79 A.D. through 2000 A.D.) variation in precipitation. A three-month running developed by UNEP-GRID Geneva based on the average was applied to the precipitation data and Worldwide Volcano Database and available at the the median rainfall for the 21-year period calculated National Geophysical Data Center (http://www. for each grid point. A mask was applied to eliminate ngdc.noaa.gov/seg/hazard/vol_srch.shtml). This data- grid points where the three-month running average base includes nearly 4,000 events categorized as mod- precipitation was less than 1 millimeter per day. erate or above (values 2­8) according to the Volcano This excluded both desert regions and dry seasons Explosivity Index (VEI) developed by Simkin and from the analysis. For the remaining points, a drought Seibert (1994). Some volcanoes are located to the event was identified when the magnitude of a monthly nearest thousandth of a degree, but most have been precipitation deficit was less than or equal to 50 georeferenced to the nearest tenth or hundredth of a percent of its long-term median value for three or degree. more consecutive months. 6. Landslides. The NGI, working with UNEP GRID- 3. Floods. The Dartmouth Flood Observatory has com- Geneva and this project, has developed a global land- piled a global listing of extreme flood events from slide and snow avalanche hazard map that has been diverse sources and georeferenced to the nearest used for global analysis of these hazards. The map degree for 1985 through 2003. Flood extent data is based on a range of data including slope, soil and are based on regions affected by floods, not neces- soil moisture conditions, precipitation, seismicity, sarily on flooded areas. Data are poor or missing in and temperature (NGI 2004). This index takes advan- the early-mid 1990s. tage of more detailed elevation data that recently 4. Earthquakes. For earthquakes, we used both the became available from the Shuttle Radar Topographic GSHAP data and a database of actual earthquake Mission (SRTM) at 30" resolution, compiled and cor- events greater than 4.5 on the Richter scale for 1976 rected by Isciences, L.L.C. (http://www.isciences.com). Table 3.3. Summary of Data Sources for Each Hazard Hazard Parameter Period Resolution Source(s) Cyclones Frequency by wind strength 1980­2000 30" UNEP/GRID-Geneva PreView Drought Weighted Anomaly of Standardized 1980­2000 2.5° IRI Climate Data Library Precipitation (50% below normal precip. for a 3-month period) Floods Counts of extreme flood events 1985­2003* 1° Dartmouth Flood Observatory, World Atlas of Large Flood Events Earthquakes Expected pga > 2 m/s2 (10% n/a sampled at 1' Global Seismic Hazard Program probability of exceedance in 50 years) Frequency of earthquakes > 1976­2002 sampled at 2.5' Advanced National Seismic System 4.5 on Richter Scale Earthquake Catalog Volcanoes Counts of volcanic activity 79­2000 Sampled at 2.5' UNEP/GRID-Geneva and NGDC Landslides Index of landslide and snow n/a 30" NGI avalanche hazard *Missing data for 1989, 1992, 1996, and 1997; quality of spatial data for 1990­91 and 1993­95 limited. n/a = not available. 30 Natural Disaster Hotspots: A Global Risk Analysis Selection of Severity Metrics We also choose a lower cutoff of 2 m/s2 for ground acceleration. The choice of cutoff determines to a large As we assert above, globally uniform data do not yet extent (for fixed repeat time) the geographic overlap exist to produce a justifiable measure of hazard sever- with assets, based on the GSHAP calculation alone. ity that can be applied consistently across multiple However, the geographic area susceptible to destruc- hazards. Instead we have chosen to use severity met- tive ground shaking from a particular event is also affected rics appropriate for each of the hazards studied in this by such unmodeled variables as soil quality, attenua- report. This necessarily introduces a degree of hetero- tion of earthquake energy in the earth's crust, and patho- geneity in our analysis. But at this stage in natural hazard logical characteristics of the earthquake source itself. studies, it is more important to develop a homogeneous Although weak buildings can be damaged severely by representation of normalized loss and risk, which we ground shaking as low as 1 m/s2 (or even lower), the have attempted to do through the use of well-accepted uncertainties associated with these other factors proxies. We argue that the choice of different hazard render arguments over the lower cutoff moot. Instead, severity metrics for different hazards should be informed we have chosen the cutoff to represent subjectively the by the known relative losses for historical events, so that major attributes of the spatial distribution of damaging we can achieve relative parity in the treatment of mul- earthquakes, namely, that the most significant damage tiple hazards on an expected loss and geographical basis. occurs near the known geologic expression of active Such an approach would tend to underemphasize tectonic boundaries. Some events, such as the Mexico extreme events, that is, infrequent or unexpected high- City earthquake in 1985, have pathological damage dis- impact events. (The question remains how extreme tribution patterns and are not well modeled by this events should be treated in a global relative analysis. choice. On the other hand, the choice of a lower cutoff By definition, their occurrences are highly uncertain would enlarge the geographic overlap to such an and their impacts highly specific. However, singular extent that comparisons with other hazards would be events may dominate a country's total loss profile. This vitiated and unrealistic. is an area for further research.) We emphasize this last point by showing actual earth- quake locations (seismicity) as well as the pga maps. Earthquakes This is somewhat redundant, as the GSHAP calcula- We rely on the exceedance probabilities presented in tion uses essentially the same data. However, the seis- the GSHAP maps. This calculation uses the empirical micity maps support the choice of lower pga cutoff. space-time distribution of earthquake occurrence to develop a probabilistic estimate of maximum ground Drought shaking at each grid point. To use these maps in our The definition of drought hazard events as rainfall at 50 analysis, we must therefore choose both a lower cutoff percent or less of the median for three months was for shaking amplitude (ground acceleration) and a char- based on a number of factors. A modified version of this acteristic repeat time for that cutoff to be achieved. We definition was found to maximize the correlation between have chosen to use a relatively short repeat time (chance drought hazard events and EM-DAT mortality in a com- of exceeding lower ground motion cutoff is 10 percent panion study (UNDP 2004). In the semiarid tropics, in 50 years, or once in 500 years) in order to empha- where drought-related risks are highest, three- to four- size more common events and repetitive disasters. month rainfall seasons are typical. Rainfall at 50 percent This deemphasizes rare extreme events, but losses or less of the median for three months therefore poses a from such events are not well calibrated and thus have significant threat. Finally, experimentation with various relatively little predictive value. At the same time, basic cutoffs--longer or shorter periods or alternative per- research into the occurrence of such large events is centages of the median--resulted in spatial patterns in continuing; such research holds the promise that the which droughts were either too pervasive or too insignif- occurrence of large earthquakes will be better under- icant to explain observed losses. Although the resulting stood in the near future. definition produces a spatial pattern of drought frequency Project Approach 31 that is not totally satisfactory in all regions, it was judged tality rates, we have little basis for projecting these the best overall. The IRI and CRED are currently design- changes into the future (Gaffin and others 2004). ing a study to systematically examine climate-loss rela- To capture the hazard exposure of human economic tionships over 800 historical drought disasters to arrive infrastructure and activity, it would be ideal to have at a more rigorous drought hazard definition. detailed measures of the extent and quality of infra- structure and the economic value of the exposed land and resources. Unfortunately, consistent, spatially dis- Exposure Data aggregated data on these parameters are very limited. To understand the risks posed by a range of hazards, it In the United States, for example, although most local is also essential to characterize the exposure of people jurisdictions assess property values for the purpose of and their economic activities to the different hazards. tax assessment, the ratio between actual market values Ideally, we would have a complete probability density and assessed values varies greatly. Detailed sample sur- function for population exposure to specific types of veys of this ratio were conducted at the county level in events; that is, we would know the probabilities that the early 1970s but were later discontinued (Schneider particular populations for a range of event sizes and and Chen 1980). We have also explored whether satel- characteristics would be present in the grid cells directly lite remote sensing data could be used to assess infra- affected by those events. Such estimates might vary structure consistently around the world, but the available depending on the time of day, day of week, or month techniques and data at this point appear insufficient of the year, as well as on local holiday schedules, given (Nghiem and others 2002). that individuals in many industrial and developing coun- We have therefore chosen to use several crude tries may travel across multiple grid cells in the course measures of economic activity and infrastructure on an of a day, week, or month. For longer term events such exploratory basis (Table 3.4). More detailed descrip- as drought, a time-averaged population estimate or an tions of the methods and data sources are provided in estimate of population involved in agriculture might Appendix A.1. be more appropriate for assessing exposure. The total level of economic activity at the national For this initial, global-scale analysis, however, we level is measured by the GDP, the annual market value believe that a consistent population estimate based on of final goods and services produced by a country. For reported residence is appropriate to characterize pop- about 50 countries, more than half of which are devel- ulation exposure across different hazard types. We use oping or transitional economies (including Bangladesh, an estimate of population for the year 2000, developed Brazil, China, India, Indonesia, and Mexico), GDP as part of GPW Version 3, to characterize the "current" data are available for subnational units. Following Sachs distribution of population. Although population dis- and coauthors (2001), we applied these subnational tribution is likely to change in the future because of estimates to population density, using the World Bank differential rates of population change, including urban estimates of GDP based on purchasing power parity and coastal migration and different fertility and mor- (PPP) for 2000. Table 3.4. Summary of Data Sources for Exposure Exposure Parameter Period Resolution Source(s) Land Land area 2000 2.5" GPW Version 3 (beta) Population Population counts/density 2000 2.5" GPW Version 3 (beta) Economic Activity National/subnational GDP 2000 2.5" World Bank DECRG based on Sachs and others (2001) Agricultural Activity National agricultural GDP 2000 2.5" World Bank DECRG based on Sachs allocated to agricultural land area and others (2001) Road Density Length of major roads c. 1993 2.5" VMAP(0) and railroads 32 Natural Disaster Hotspots: A Global Risk Analysis To provide an exposure measure relevant to floods For a given hazard, vulnerability will vary across sim- and drought, we allocated available national estimates ilar elements and from one element to the next. Irri- of agricultural GDP to grid cells based on the amount gated agricultural areas tend to experience lower losses of agricultural land. This assumes, very crudely, that during droughts than areas that depend on rainfall, for all agricultural areas contribute equally to the total agri- example. Buildings that are constructed to seismic safety cultural GDP for the country. standards are less likely to be damaged during an Finally, as a measure of infrastructure development, earthquake than those built of unreinforced masonry. we computed the total length of major roads and rail- Houses with raised platforms are better suited to with- roads for each grid cell based on the VMAP(0) data sets standing flood conditions that those without. People developed by the National Geospatial-Intelligence Agency and societies with resources and economic alternatives (formerly the National Imagery and Mapping Agency) tend to be better protected from harm and able to recover and made available by the USGS (http://erg.usgs.gov/ more quickly than people with fewer options and nimamaps/dodnima.html#Digital). resources. We recognize that these measures provide at best only The set of elements that may be damaged by a given first-order indicators of the level of exposure and, hazard is often quite large. Urban infrastructure, for given the lack of detailed spatial data, may signifi- example, consists of multiple sectors--transport, power, cantly under- or overstate exposure in any particular water and sanitation, housing, and communications-- grid cell. However, we believe that these indicators each of which in turn may encompass many separate may be useful in broadly classifying different levels of systems. Each system is made of subsystems and so on, exposure in order to better characterize the relative risks down to the level of individual components. associated with different hazards. A summary of the When a complex entity such as an urban area is totals of each measure for the world and the area retained subjected to a severe hazard event like a flood or vol- after applying the mask is provided in Table 3.5. canic eruption, widespread failures of vulnerable com- ponents can cause total or partial system failure, resulting in a disaster. Given the number of systems, sub- Vulnerability Data systems, and components, each of which responds dif- The stresses to which a given element at risk is sub- ferently when subjected to a given hazard, it is possi- jected during a hazard event depend on the hazard. ble to characterize vulnerability only generally (or perhaps These stresses include shaking in the case of earthquakes, stochastically) when operating at scales larger than indi- moisture stress in the case of drought, inundation during vidual installations or facilities. Similarly, when social floods, and so on. A given element may be severely chal- systems such as communities or households are the unit lenged by one hazard but completely unaffected by of analysis, vulnerability analysis requires detailed knowl- another. A building, for example, may collapse when edge of household or community characteristics. In a subjected to seismic shaking or incur damage due to global analysis such as the current one, therefore, vul- floods, but may suffer little or no impact during a drought. nerability assessment is at best possible only through Similarly, the fertility of agricultural land may benefit the use of general proxies. directly as a result of flooding, whereas exposed infra- This analysis assesses global disaster-related risks of structure may be severely damaged. mortality and economic losses. The elements at risk are Table 3.5. Summary of Exposure Data for World and Unmasked Areas Total* Within Mask Percentage Within Mask Land Area (million km2) 131 72 54.9 Population 2000 (millions) 6,054 6,008 99.2 GDP (PPP, billion US$, 2000) 44,198 43,544 98.5 Agricultural Value (billion US$, 2000) 1,361 1,359 99.8 Transportation (million km) 7.9 6.4 80.8 * All grid cells within GPW, Version 3 (beta). Project Approach 33 people in the first instance and an estimated value of tributions of the resulting hazard maps. The data may goods and services produced annually per unit area in be inadequate for assessing absolute risk levels or for the second. Ideally, we would have a complete proba- detailed comparisons of risk levels across hazards. For bility density function for the loss expected to result when a number of the available hazard data sets, such as those particular populations or economic assets are exposed based on media reports, we also expect that relatively to a range of hazards and hazard severities (that is, we small or modest events may be undercounted substan- would know the probabilities of different levels of losses tially, especially in developing countries where report- likely to be experienced by the exposed units in the grid ing is likely to be less complete. cells directly affected by different hazard events). Owing We therefore divided the total number of grid cells to data limitations, we used historical loss rates, using a into deciles, ten groups of approximately equal number methodology described in detail below. We calculated of cells, based on the value of each individual hazard loss rates for each hazard from historical losses over 20 indicator. Cells with the value of zero for an indicator years (1981 through 2000) obtained from EM-DAT. For were excluded. When hazard indicators have large num- each hazard we calculated 28 loss rates, one for each com- bers of cells with the same values (cyclones, drought, bination of seven regions and four country wealth status floods, and earthquakes), deciles may be grouped groups based on World Bank classifications. together. For example, the result of dividing the flood Estimates of losses per disaster and the degree to data into deciles results in output values of 1, 4, and 7 which disaster events are consistently captured vary through 10. Since many grid cells have only one or from one data source to the next (Sapir and Misson two flood events, the first through third deciles are com- 1992). For the purpose of estimating loss rates, how- bined and given the output value 1, and the fourth ever, it is not necessary to assume that EM-DAT con- through sixth deciles are combined and given the output tains a complete inventory of all deaths and economic value 4. In all cases, the combined deciles are at the losses over the 20-year period. Rather, in this analysis, low end of the scale (sixth decile or less). it is only necessary that the deaths and economic dam- Results for each hazard are discussed in detail in Sec- ages recorded in EM-DAT capture relative differences tion 4. In general, at least the top three deciles of cells in mortality and economic losses between hazards, were needed to identify areas of known hazard around regions, and country wealth groupings. Improvements the world. As an initial arbitrary cutoff, we therefore in mortality and economic loss data by event in data chose the top three deciles as our first-order definition sets such as EM-DAT would make loss rate calculation of "relative significance" in terms of hazard frequency more precise. For example, the insurance industry has or probability, exposure, and overall risk. Some cells are been developing more consistent loss databases for classified as relatively high in significance according to selected regions and in at least one case has developed more than one hazard, that is, they fall within the top a multihazard index of average annual loss based on three deciles of more than one hazard indicator. We modeled exposure to hazard events (Risk Management therefore built an index that simply sums the decile Solutions 2004). values for each hazard, with 8 representing the third highest decile, 9 representing the second highest decile, and 10 the highest decile. Thus, a cell in the third Global Hotspots Classification decile for just one hazard would have an index value of 8, and a cell in the third highest decile for just two Classification of hotspots on a global basis addresses hazards would have an index value of 16. A cell in the the central concern of the project--the identification top three deciles for three hazards would have an and characterization of high-risk natural disaster hotspots. index value between 24 and 30. Results are presented Because of the limited time period and quality of the in Chapter 5. input data, we believe that it is appropriate to use the Using the same cutoff of the top three deciles for each data to identify areas at relatively high risk of a partic- natural hazard identifies those cells that are at higher ular natural hazard, and then to compare the spatial dis- relative probability compared with other cells for each 34 Natural Disaster Hotspots: A Global Risk Analysis hazard, but does not necessarily result in comparable ble absolute levels of probability. Moreover, the poten- levels of absolute probability across hazards. That is, a tial exposure of land, population, and other features of cell in the top three deciles for both flood and drought each cell varies greatly both across cells and over time, hazards does not necessarily face the same probability so that the overall level of risk faced in a multihazard of hazard occurrence in terms of drought and flood hotspot will be determined by a range of highly uncer- frequency and intensity. Moreover, hazards such as tain factors. floods, earthquakes, and volcanoes have very different We have experimented with alternative approaches patterns of occurrence in terms of their spatial distri- to index construction, weighting the decile values by butions, temporal recurrence, and event characteristics, measures of exposure, population density, and GDP making absolute comparisons difficult. Given the very density, and then by regional measures of vulnerabil- limited records available at the global scale, we think ity. We present and compare these results in Chapters that it is currently impossible to determine compara- 5 and 6. Chapter 4 Single-Hazard Exposure Analysis Figures 4.1a­g presents hazard maps for the six natu- that many of the storm tracks in the database do not ral hazards, including two different indicators of earth- actually reach land. Since our indicator is based only quake hazard. All of these figures are based on deciles, on wind speed and not on storm surge potential, it with the top three deciles indicated in red, the next may underestimate the potential hazard. three deciles in yellow, and the bottom four deciles in blue. Areas with very low population and minimal agri- cultural production are masked as described in Chap- Drought ter 3. Drought exhibits a more dispersed pattern around the globe, with 2.5° x 2.5° grid cells in the top three deciles Cyclones (red areas of Figure 4.1b) appearing in both interior and coastal regions of most continents. These drought- At least 6.7 percent of the world's land area was sub- prone areas include parts of the western and midwest- ject to at least one instance of high wind speeds asso- ern United States, Central America, northeastern ciated with a tropical storm or cyclone during the 21-year Brazil, the sub-Saharan belt, the Horn of Africa, south- period of record. Notably, these coastal areas are more ern and central Africa, Madagascar, southern Spain and densely populated than average, so that approximately Portugal, central Asia, northwest India, northeast China, 24 percent of the world's population, more than 1.4 Southeast Asia, Indonesia, and southern Australia. billion people, lives in the affected areas. Similarly, GDP, About 38 percent of the world's land area has some agricultural production, and transportation infrastruc- level of drought exposure. This 38 percent contains ture also appear more concentrated in these areas than about 70 percent of both the total population and the average (33 percent, 19 percent, and 13 percent of agricultural value produced. In the top three deciles of total land area, respectively). drought-prone grid cells, about 1.1 billion people (18 The top three deciles of grid cells (red areas of percent) live in about 12.9 million square kilometers Figure 4.1a) include about 2.5 million square kilome- of land (10 percent). The highest decile alone contains ters (1.9 percent) and more than 550 million residents about 419 million people because of a somewhat higher (9.1 percent) (Table 4.1). GDP density is more than than average population density. eight times greater than the world average; these exposed areas represent about one-sixth of total world GDP. The most frequently hit areas are in the western Pacific, Floods southern Africa, the Caribbean, and southeastern United States. Surprisingly, Bangladesh and neighboring areas Some flooding is evident in more than one-third of the do not show up in the highest deciles, even though they world's land area, in which some 82 percent of the world's have been impacted significantly in the past by severe population resides (red, green, and blue areas of storm surge. This may be due to the somewhat more Figure 4.1c). The most flood-prone areas (indicated in limited record of Indian Ocean storms. It also appears red) encompass about 9 percent of the land area and 35 36 Figure 4.1. Distribution of Hazardous Areas by Hazard Type a) Cyclones Natural Cyclone Hazard Disaster Deciles 1st ­ 4th Hotspots: 5th­ 7th 8th­ 10th A Global Risk Analysis Single-Hazar d Exposur Figure 4.1. Distribution of Hazardous Areas by Hazard Type eAnalysis b) Drought Drought Hazard Deciles 1st ­ 4th 5th­ 7th 8th­ 10th 37 38 Figure 4.1. Distribution of Hazardous Areas by Hazard Type c) Floods Natural Flood Hazard Disaster Deciles 1st ­ 4th Hotspots: 5th­ 7th 8th­ 10th A Global Risk Analysis Single-Hazar d Exposur Figure 4.1. Distribution of Hazardous Areas by Hazard Type eAnalysis d) Earthquakes (pga) Earthquake Hazard PGA Deciles 1st ­ 4th 5th­ 7th 8th­ 10th 39 40 Figure 4.1. Distribution of Hazardous Areas by Hazard Type e) Earthquakes (count) Natural Earthquake Hazard Disaster Frequency Deciles 1st ­ 4th 5th­ 7th Hotspots: 8th­ 10th A Global Risk Analysis Single-Hazar d Exposur Figure 4.1. Distribution of Hazardous Areas by Hazard Type f) Volcanoes eAnalysis Volcano Hazard Deciles 1st ­ 4th 5th­ 7th 8th­ 10th 41 42 Figure 4.1. Distribution of Hazardous Areas by Hazard Type g) Landslides Natural Disaster Landslide Hazard Deciles 6th­ 7th Hotspots: 8th­ 10th A Global Risk Analysis Single-Hazard Exposure Analysis 43 Table 4.1. Characteristics of High-Hazard Areas by Hazard: Top Three Deciles Hazard Land Area (106 km2) Population (106) GDP (109 $) Agricultural GDP (109 $) Road/Rail Length (103 km) Cyclones 2.5 553 7,053 81 275 Drought 12.9 1,094 5,319 252 1,078 Floods 11.5 2,283 14,670 371 1,191 Earthquakes 2.9 328 3,425 50 242 Volcanoes 0.1 45 240 3 14 Landslides 0.8 66 782 10 45 Percent of World Cyclones 1.9% 9.1% 16.0% 5.9% 3.5% Drought 9.8% 18.1% 12.0% 18.5% 13.6% Floods 8.8% 37.7% 33.2% 27.2% 15.0% Earthquakes 2.2% 5.4% 7.8% 3.7% 3.1% Volcanoes 0.1% 0.8% 0.5% 0.2% 0.2% Landslides 0.6% 1.1% 1.8% 0.8% 0.6% more than 2 billion people (38 percent). These flood- set and the second on counts of reported earthquake prone regions include large areas of the midwestern activity. The latter is based on a relatively limited 27- United States, Central America, coastal South Amer- year record, so the overall area affected is more limited ica, Europe, eastern Africa, northeast India and than that captured by the GSHAP data set. The GSHAP Bangladesh, China, the Korean peninsula, Southeast data set reflects expert judgment on the potential Asia, Indonesia, and the Philippines. The high frequency severity of earthquakes at a fixed probability level (10 of flooding in Bangladesh and surrounding areas pre- percent in 50 years), taking into account scientific under- sumably reflects the influence of tropical storms, and standing of earthquake processes (as of about 1999, appears to compensate to some degree for the weak the end of the GSHAP activity) as well as longer peri- identification of storm-related hazard in the cyclone ods of record in many areas of the world. data noted previously. The very large areas of China and Approximately 10 million square kilometers of other parts of Asia highlighted in red may stem in part land, about 7.5 percent of the total world land area, is from the crude georeferencing of flood reports in these estimated to have a 10 percent probability of pga of at areas, which in turn may overemphasize these densely least 2 m/s2 in a 50-year period. An estimated 1.2 bil- populated areas. On the other hand, it is certainly clear lion people, or about 20 percent of world population, that large areas of China, such as the Yangtze River basin, live in these areas around the year 2000. are subject to significant flood risk affecting large areas Cells in the top three deciles of pga values had a and large populations. land area of nearly three million square kilometers and These areas also represent relatively high concen- a total population of more than 300 million people, trations of GDP and agricultural value added, more than about 5 percent of total world population, and an asso- triple the average for the world. This is consistent with ciated GDP of nearly 8 percent. Areas of relatively high the idea that flood-prone areas are also areas of more hazard include the western coast of the United States, intensive agricultural production and development. Central America, and the western coast of South America, as well as much of southern Europe, Turkey, the Islamic Republic of Iran, central Asia, southwest Earthquakes China, Nepal, Taiwan (China), Japan, the Philippines, and New Zealand. Road and rail lengths in these areas We have examined two different data sets for assessing are about average, roughly 240,000 kilometers or about earthquake hazards, the first based on the GSHAP data 3 percent of the total world transportation length. 44 Natural Disaster Hotspots: A Global Risk Analysis Volcanoes often overlapping areas, especially in parts of eastern North America, Central America, and Asia. Along with Volcanoes are the most spatially concentrated hazard landslides, they represent relatively frequent events with of the six considered here, affecting only 400,000 square a range of intensities. kilometers and 93 million people in all nonzero cells, The earthquake hazard is also dispersed, but the prob- mainly in Japan, the Philippines, Indonesia, the ability of a major earthquake is comparatively low over United States, Mexico, Central America, Colombia, a longer period. On the other hand, when exposure and Ecuador, and Chile. For the top three deciles (red areas vulnerability are high, significant numbers of deaths of Figure 4.1f), only 114,000 square kilometers and and considerable damage in a concentrated area can 45 million people are included. These areas are much result, as evidenced by the December 2003 earthquake more densely populated than average and have high in the Islamic Republic of Iran. The other geophysical GDP densities, about six times the world average, though hazard, volcanoes, is confined to much smaller and this is partly because of Japan's inclusion. The agricul- clearly defined regions. Large-scale volcanic events are tural value associated with these volcanic areas is relatively rare, but sometimes have substantial impacts about 2.7 times the world average density. No attempt on nearby populations through lava flows, ash deposi- has been made here to identify buffers around the vol- tion, pyroclastic events, and other phenomena. canoes or to address possible impacts from a larger scale As noted above, the highest hazard areas often have eruption (for example, ash deposits over a large area). higher-than-average densities of population, GDP, agricultural GDP, and transportation length. As shown in Figures 4.2a­d, this is generally the case for areas Landslides affected by cyclones, volcanoes, and floods. The den- sities of these socioeconomic variables do not vary sig- The NGI index of landslides and snow avalanches ranges nificantly with drought and earthquake deciles, though in value from 1 to 9 (Nadim and Kjekstad, in process; the slight increase in GDP density and decrease in agri- NGI 2004). However, values of 4 or less are considered culture GDP and transportation densities for the high- very low in frequency. We have mapped the cells using est earthquake decile is of interest. GDP density increases deciles, but the overall map of NGI classes 5­9 corre- but transportation density decreases at higher levels of sponds closely with the decile groupings shown in Figure landslide hazard. 4.1g. Note that this map differs from the map shown In Appendix A.2, we weight each hazard distribu- in Figure 7.1 of NGI (2004), primarily because of the tion by population density to create spatial distributions masking of unpopulated, nonagricultural areas. of population hazard exposure. Since population den- The total land area subject to landslides is about 3.7 sity varies significantly across the grid cells included in million square kilometers with a population of nearly the analysis--from the minimum value of five people 300 million, or 5 percent of total world population. The per square kilometer to more than 30,000 people per relatively high-risk areas (top three deciles) cover about square kilometer--simple weighting of the hazard values 820,000 square kilometers with an estimated popula- by population density would result in the highest expo- tion of 66 million. GDP density is higher than average, sure deciles being simply the high population density but agricultural value added and road and rail length areas, including those exposed to only moderate levels are about average. of hazard. The alternative approach presented in Appendix A.2 is therefore to divide grid cells into deciles based Single-Hazard Analysis of Exposure on population density alone and to use the resulting index (1­10) to weight each hazard distribution. A The individual hazard maps demonstrate considerable grid cell with a drought decile value of 8 might there- diversity in the distribution of relatively high hazard fore have a drought population-exposure index rang- areas. Cyclones, droughts, and floods cover large, ing from 8 to 80. Because of the average increase in Single-Hazard Exposure Analysis 45 population density at higher hazard levels noted hazard distributions. Population totals and densities above, we expect that most high hazard cells will gen- increase to some degree in the most hazardous areas erally be classified as high on the hazard population- over the population totals and densities in hazardous exposure index as well. areas as defined by the unweighted, hazard-only In brief, the patterns for population hazard expo- index. sure remain very similar to those for the unweighted Figure 4.2. Exposure Measures by Hazard Decile a) Population Density 600 500 ) 2 400 Density 300 (people/km 200 Population 100 0 1 2 3 4 5 6 7 8 9 10 Hazard Decile Cyclones Drought Flood Earthquake Volcanoes Landslides Figure 4.2. Exposure Measures by Hazard Decile b) GDP Density 6000 5000 ) 2 4000 $/km Density 3000 GDP 2000 (thousands 1000 0 1 2 3 4 5 6 7 8 9 10 Hazard Decile Cyclones Drought Flood Earthquake Volcanoes Landslides 46 Natural Disaster Hotspots: A Global Risk Analysis Figure 4.2. Exposure Measures by Hazard Decile c) Agriculture GDP Density 45 40 35 ) 2 Density 30 $/km 25 GDP e 20 15 (thousands 10 Agricultur 5 0 1 2 3 4 5 6 7 8 9 10 Hazard Decile Cyclones Drought Flood Earthquake Volcanoes Landslides Figure 4.2. Exposure Measures by Hazard Decile d) Transportation Length Density 0.180 0.160 0.140 Density ) 0.120 2 0.100 (km/km 0.080 0.060 ransportationT 0.040 0.020 0.000 1 2 3 4 5 6 7 8 9 10 Hazard Decile Cyclones Drought Flood Earthquake Volcanoes Landslides Chapter 5 Multihazard Exposure Analysis Under ideal circumstances, it would be possible to deter- fits (for example, living along coasts), or risks that lead mine precisely the spatial and temporal distributions to long-term irreversible impacts. of risk for specific locations and time periods by com- Despite these problems, it is still important to char- paring, for different natural hazards, the estimated levels acterize risks of specific types of losses associated with of particular hazards, and exposure and vulnerability natural hazards as objectively as possible, making clear to those hazards. Unfortunately, many factors make this alternative assumptions that may lead to different quan- difficult. titative or qualitative results. Given the data limitations, First, natural hazards differ greatly in their tempo- it is important to proceed systematically from simpler ral and spatial patterns of occurrence. To estimate the to more complex methods for multihazard analysis. In risk of volcanic hazards, for example, one would need this section, we develop simple multihazard indexes records over many centuries or even millennia to based solely on hazard probability and exposure data. ascertain the frequency of events with any confidence. In Chapter 6, we address the more difficult problem of Over comparable time periods, higher frequency events incorporating vulnerability in order to compare risk such as droughts and floods might change significantly levels from the perspective of both mortality and eco- because of climate trends. During these periods, pop- nomic loss. ulation exposure and vulnerability also change, which makes prediction of expected losses over time difficult. Second, as noted previously, we often lack compa- Simple Multihazard Index rable, detailed data about the spatial location and extent of hazards, their intensity and duration, and other char- In this section, we construct a simple multihazard index acteristics that can interact with exposure and vulner- by summing category values between 8 and 10 across ability. Such data limitations make probability and risk all six natural hazards. This results in a multihazard estimates less certain and comparisons between haz- index that reflects the number of hazards considered ards more difficult. relatively significant in a particular grid cell. Cells that Finally, there are normative aspects to comparing are in the highest decile for multiple hazards will also risks. For example, some individuals or groups may rank slightly higher than those composed of slightly value future potential losses differently than present lower single-hazard decile values. potential losses, depending on their personal or social Total area, population, and other exposure charac- "discount rates" (Schneider and Chen 1980). Similarly, teristics by the combined hazards are summarized in individuals or groups may have disparate views on mor- Table 5.1. The overall global map is shown in Figure 5.1, tality, morbidity, economic losses, and social impacts, and a version based on type of hazard is given in Figure and may disagree, for example, on the relative costs of 1.1. Detailed regional maps are provided in Figure 5.2. loss of life as opposed to reduction in economic well- Areas exposed to three to five hazards fall mostly being. Some may also have different preferences regard- along the west coasts of North, Central, and South Amer- ing different types of risks, such as large-scale catastrophic ica, mountain regions of Central and South Asia, and risks, risks that come with significant perceived bene- western Pacific coastlines. These are all areas charac- 47 48 Natural Disaster Hotspots: A Global Risk Analysis Table 5.1. Summary Statistics for the Simple Multihazard Index No. of Hazards Index Values Land Area Population GDP Agricultural GDP Road/Rail Length (106 km2) (106) (109 $) (109 $) (103 km) 0 0 43.6 2,546.0 19,702 693 3,840 1 8­10 21.4 2,645.2 17,424 522 2,048 2 16­20 3.4 687.0 4,825 97 297 3­5 24­50 0.5 105.4 1,312 11 41 Percent of world 0 33.4% 42.1% 44.6% 50.9% 48.5% 1 16.4% 43.7% 39.4% 38.4% 25.9% 2 2.6% 11.3% 10.9% 7.2% 3.8% 3­5 0.4% 1.7% 3.0% 0.8% 0.5% terized by high relative susceptibility to both geophys- overlap with cyclones: some one-third of the area affected ical and hydro-meteorological hazards (yellow and red by cyclones is also affected by floods. These areas include areas of Figure 1.1). These areas encompass, or are in some 60 percent of the total population exposed to close proximity to, major cities such as San Francisco, cyclones, which totals over 300 million people. About Guatemala, Managua, Quito, San Jose, Santiago, Manila, 9 percent of the cyclone-prone area overlaps with areas Taipei, and Tokyo. Although the total land area affected of relatively high drought, and another 9 percent over- is relatively small, under 500,000 square kilometers, laps with relatively high earthquake hazards. It is strik- more than 100 million people live in these areas, asso- ing that the former areas are much less densely populated ciated with about 3 percent of total GDP. than the latter. This likely stems from the high expo- Much larger areas and populations are exposed to sure of urban areas in Japan, the Philippines, and Taiwan two hazards. Nearly 800 million people, or 13 percent (China) to both cyclones and earthquakes, as compared of the world's population, live in grid cells that have with more rural overlapping cyclone and drought regions relatively high exposure to two or more different haz- in Madagascar, Vietnam, and eastern Mexico. Landslides ards. Areas affected by two hazards cover much of the are also an issue for 5 percent of the land area and California coast and portions of the Gulf Coast and about 4 percent of the population. Volcanoes remain a Caribbean, areas of the Horn of Africa and Madagas- localized problem. Note that the land area and popu- car, and much of the Chinese coast and the Korean lation estimates for the five hazards include areas and Peninsula (yellow areas in Figure 1.1). population exposed to more than two hazards. Overall, more than half of the world's population lives Similar "bi-hazard" profiles may be generated for each in areas subject to at least one hazard at a significant hazard, globally and regionally, providing a simple level, on just under 20 percent of the world's land area. way to inform analysts and decision makers about the This finding is driven primarily by the wider extent of potential importance of multihazard analysis from their drought and flood in the database and therefore the regional and sectoral perspectives. larger number of cells classified in the top three Another application of these data is to identify areas deciles. As for most of the single hazards, population, that have comparable hazard profiles on multiple dimen- GDP, and transportation length are concentrated in the sions of hazard probability and exposure. For example, more hazardous areas. However, the effect is less pro- an analyst or decision maker might wish to identify com- nounced for the transportation measure. parable areas with a particular combination of exposures These multihazard distributions may help hazard to cyclone and drought within a particular range of managers primarily concerned with one hazard to under- population or GDP density. Such an analyst might find stand how much their area is also susceptible to other it useful to compare hazard impacts and response for the hazards. In Table 5.2, for example, the total area exposed three areas on the three different continents noted above to cyclones is assessed in terms of overlapping haz- that face both cyclone and drought. ards. This table indicates that floods have the strongest Multihazar d Exposur eAnalysis Figure 5.1. Global Distribution of Areas Significantly Exposed to One or More Hazards, by Number of Hazards Exposed Areas Top 3 Deciles Exposed to: 1 Hazard 2 Hazards 3 ­ 5 Hazards 49 50 Natural Disaster Hotspots: A Global Risk Analysis Figure 5.2. Detailed View of Multihazard Areas a) Western Hemisphere Exposed Areas Top 3 Deciles Exposed to: 1 Hazard 2 Hazards 3 ­ 5 Hazards Multihazard Exposure Analysis 51 Figure 5.2. Detailed View of Multihazard Areas b) Asia/Pacific Exposed Areas Top 3 Deciles Exposed to: 1 Hazard 2 Hazards 3 ­ 5 Hazards 52 Natural Disaster Hotspots: A Global Risk Analysis Reclassification of Multihazard Areas by Table 5.3 indicates that high population density areas Population Density that coincide with the highest three deciles of two or more hazards include more than 650 million people As in the case of the single-hazard assessment, it is impor- living on 1.75 million square kilometers. This under- tant to modify the wide range of population density to scores the need for using a multihazard management avoid giving too much weight to the spatial distribu- approach in densely populated regions, where interac- tion of population. In this case, we divide the popula- tions among urban development, social displacement, tion density into three categories: low (between 5 and and overlapping hazards could lead to areas of enhanced 14.49 people per square kilometer); medium (14.5 to risk at finer scales of resolution. 51.49 people per square kilometer); and high (51.5 or more people per square kilometer)(Figure 5.3). Table 5.2. Hazard Profile for High-Cyclone Exposed Areas Cyclones Drought Floods Earthquakes Volcanoes Landslides Land Area (103 km2) 2,452 229 822 220 23 117 Population (106) 552.5 11.6 331.8 91.3 5.9 21.7 Population Density 225 51 403 416 260 185 Percent of Total Cyclone Area/Population Land Area 100% 9.3% 33.5% 9.0% 0.9% 4.8% Population 100% 2.1% 60.1% 16.5% 1.1% 3.9% Table 5.3. Summary Statistics for the Population-Weighted Multihazard Index Population Density No. of Hazards Land Area Population GDP Agricultural GDP Class (millions of km2) (millions) (billions of $) (billions of $) High 0 8.8 2,078.8 15,351 338 1 8.1 2,440.6 15,285 337 2 1.8 653.1 4,394 77 3­5 0.3 100.7 1,177 8 Medium 0 11.7 324.3 2,870 159 1 5.3 152.4 1,464 79 2 0.8 25.9 323 13 3­5 0.1 4.1 103 2 Low 0 14.1 123.3 1,137 93 1 5.1 46.0 540 54 2 0.6 5.6 98 5 3­5 0.1 0.6 31 1 Multihazar d Exposur Figure 5.3. Global Distribution of Multiple Hazards by Population Density Category eAnalysis Exposure to All Hazards Top 3 Deciles With: Low Population Density Medium Population Density High Population Density 53 Chapter 6 Multihazard Risk Assessment Disaster risks are a function of hazard exposure and vul- risks. In each case, we calculate weights for each hazard, nerability. For a given level of hazard, risks of death stratified by region and the wealth of the country in and losses can differ greatly because of differences in which the losses occurred. To assign the wealth status exposure and vulnerability. For example, drought haz- for each country, we used standard World Bank classi- ards of the same apparent magnitude and affecting the fications based on GDP in 2000 (Appendix A.3). same numbers of people may be associated with high We used historical losses as recorded in EM-DAT mortality and small absolute economic losses in devel- across all events from 1981 through 2000 for each hazard oping countries, but with low mortality and large absolute type to obtain mortality and economic loss weights for economic losses in industrialized countries. In the absence each hazard across each region for four wealth classes of vulnerability information, risk indexes based solely within regions (Tables 6.1 and 6.2). The weights are an on relative measures of hazard and exposure could fail aggregate index of relative losses within each region and to identify relatively modest risks posed by some nat- country wealth class for each hazard over the 20-year ural hazards compared with much more severe risks period. posed by others. A high value in Table 6.1 or 6.2 indicates relatively In this section, therefore, we assess global risks of high historical losses for a given combination of hazard mortality and economic losses by incorporating esti- and region/wealth class combination; a low value indi- mated vulnerability by hazard, region, and country cates relatively low rates of historical losses. Interest- wealth status. Although data on vulnerability are ingly, the highest historical loss rates are not always in aggregated and limited, incorporating such data into the lowest income countries. This suggests that the the analysis allows us to identify single- and multi- truism that the poor are always the most vulnerable may hazard-disaster risk hotspots, incorporating all sources be an oversimplification. Although the data for drought of disaster causality. in Africa in Table 6.1 (mortality) certainly reinforce this perception, in other instances the middle- or even upper-income countries have historically experienced Derivation of Vulnerability Coefficients the highest rates of losses. Some middle-income coun- tries, for example, might have a relatively high value of In the following analysis, we weight the value of pop- economic assets at risk, without having instituted ade- ulation or GDP exposure to each hazard for each grid quate measures to reduce the vulnerability of those assets cell by a vulnerability coefficient to obtain an estimate during hazard events. of risk. The vulnerability weights are based on histori- The vulnerability coefficients given in Tables 6.1 and cal losses in previous disasters. There are two sets of 6.2 not only affect the relative significance of each hazard weights: one derived from historical mortality and the across the regions and country groupings, but also char- other from historical economic losses. The mortality acterize the relative significance of the six hazards within weights are applied to population exposure to obtain each group. Thus, for example, mortality rates associ- mortality risks; the economic loss weights are applied ated with cyclones are generally 3 to 20 times larger to GDP per unit area exposure to obtain economic loss than those associated with floods in most low-income 55 56 Natural Disaster Hotspots: A Global Risk Analysis Table 6.1. Mortality-Related Vulnerability Coefficients Region and Wealth Status Cyclones Drought Earthquakes Floods Landslides Volcanoes Africa Low 5.06 118.97 1.51 0.95 79.10 Lower middle 59.35 1.10 3.10 0.00 0.00 Upper middle 0.57 0.00 2.18 High 5.10 0.00 0.00 East Asia and the Pacific Low 10.17 0.42 2.60 2.24 2.08 0.79 Lower middle 5.03 0.15 3.17 2.22 4.74 13.20 Upper middle 39.22 0.00 0.51 23.31 High 1.33 0.00 5.48 1.10 1.20 0.51 Europe and Central Asia Low 0.00 0.75 2.82 5.69 Lower middle 2.50 0.00 62.16 0.67 1.46 0.00 Upper middle 0.00 0.00 0.33 0.00 High 1.65 0.00 1.77 0.25 2.67 0.00 Latin America and the Caribbean Low 39.72 0.00 4.22 2.36 0.00 0.12 Lower middle 44.16 0.00 3.24 4.44 8.53 231.68 Upper middle 4.27 0.01 13.86 11.21 4.24 1.62 High 3.26 0.00 0.00 0.00 0.00 0.00 Middle East and North Africa Low 0.00 5.81 0.00 Lower middle 0.00 271.25 5.11 2.54 Upper middle 0.00 0.00 0.54 1.91 0.00 High 0.00 0.00 0.00 0.19 North America High 1.01 0.00 0.39 0.19 0.00 0.00 South Asia Low 64.52 0.04 8.04 3.90 7.04 Lower middle 0.20 0.00 Upper middle High 0.00 Note: These coefficients are based on hazard-specific historical mortality rates (persons killed during 1981 through 2000 per 100,000 persons in 2000) and are used to weight population exposure to estimate mortality risk. Blank cells indicate insignificant recorded historical losses. The number of histori- cal events available to calculate each weight varies, with some weights based on as few as five to ten events. Multihazard Risk Assessment 57 Table 6.2. Economic Loss-Related Vulnerability Coefficients Region and Wealth Status Cyclones Drought Earthquakes Floods Landslides Volcanoes Africa Low 38.97 5.55 0.65 0.00 0.00 Lower middle 127.01 0.01 2.33 0.00 0.00 Upper middle 18.49 9.88 0.00 High 5.24 0.00 0.00 East Asia and the Pacific Low 59.47 0.66 0.92 25.97 0.07 7.58 Lower middle 8.62 0.54 10.72 17.45 0.08 12.02 Upper middle 953.20 0.00 0.07 0.00 High 4.02 8.54 47.97 1.53 0.17 0.00 Europe and Central Asia Low 4.52 16.34 5.56 3.80 Lower middle 0.00 0.76 82.12 24.96 4.23 0.00 Upper middle 4.13 0.00 10.13 0.00 High 24.04 3.29 19.23 4.23 4.65 0.31 Latin America and the Caribbean Low 71.65 7.50 2.23 0.36 0.00 0.17 Lower middle 48.84 2.74 8.82 7.04 3.97 22.94 Upper middle 14.48 1.28 11.72 5.88 1.04 0.37 High 104.27 0.00 0.00 0.00 0.00 0.00 Middle East and North Africa Low 0.00 168.87 0.00 Lower middle 9.35 38.98 5.90 0.00 Upper middle 0.00 0.00 0.00 10.60 0.00 0.00 High 1.03 0.00 0.00 North America High 13.00 0.97 30.82 2.84 0.00 0.00 South Asia Low 26.64 0.18 1.33 7.00 0.07 Lower middle 0.00 0.00 5.26 Upper middle High 0.00 Note: These coefficients are based on hazard-specific historical economic rates (economic losses per $100,000 GDP in 2000 during 1981 through 2000) and are used to weight GDP exposure to obtain economic loss risks. Blank cells indicate insignificant recorded historical losses. The number of historical events available to calculate each weight varies, with some weights based on as few as five to ten events. 58 Natural Disaster Hotspots: A Global Risk Analysis countries in Africa, Asia, and Latin America and in high- scale (for example, frequency counts for droughts versus income areas of Europe and North America. Drought probability index values for landslides), the accumu- has minimal impact on mortality relative to other haz- lated mortality numbers are not easily comparable across ards, except in low-income Africa. hazards. Before combining the hazards into a multi- Aggregating across more than 6,000 entries in EM- hazard index that reflects total estimated impacts from DAT for this period helps compensate for missing data all disaster types, we apply a uniform adjustment to all and reporting inaccuracies. The aggregate indexes are values within a given region such that the total hazard- broadly reflective of patterns across hundreds of events specific mortality for all places in the region equals the rather than dependent on accurate loss estimations for actual number recorded in EM-DAT. The combined, individual events. This is particularly important in case mortality-weighted multihazard index is then simply of economic losses, since economic losses are unevenly the sum of the individual hazard mortality estimates recorded in EM-DAT. Only 30 percent of the entries in for a given place. EM-DAT from 1981 through 2000 contain data on eco- Reporting actual mortality numbers would give an nomic losses, and these economic loss data were assessed unrealistic impression of precision. Our more modest using nonstandardized methodologies. objective here is to provide a "relative" representation The procedure for assessing mortality and economic of disaster risk. For the purposes of cartographic loss risks for each grid cell was similar. In the case of output and interpretation, we therefore convert the result- mortality risks, the weights were based on historical ing numbers into index values from 1 to 10 that corre- mortality and applied to population exposure values spond to deciles of the distribution of place-specific at each grid point (Box 6.1). The derivation of eco- aggregate mortality. nomic loss risk is the same, with two exceptions: (1) The mortality-weighted multihazard index is strongly the unit of analysis is GDP per unit area rather than influenced by the choice of measure for the degree or population density, and (2) loss weights are based on severity of hazard. Ideally we would have sound rules historical economic losses rather than on historical for applying these measures and guiding the realloca- mortality. tion of mortality within regions. If we think of hazard The resulting regional differences in loss risks are in mortality in epidemiological terms, we can think of part due to regional differences in population density, measures of severity (frequency, duration, and magni- in the size of affected areas, and in the degree of tude, or combinations thereof) as the right-hand side hazard. But they also reflect differences in vulnerabil- term in a dose-response function that links the magni- ity. For instance, earthquakes in Japan tend to result in tude of an event to the resulting mortality. The form of lower mortality rates than in many developing coun- this function could be linear or exponential (for exam- tries thanks to better enforcement of building codes, ple, stronger storms cause proportionally higher damage), better emergency response, and the generally high level or it could be defined by some kind of threshold value of preparedness. (for example, serious damage occurs only beyond a cer- In the above series of steps (see Box 6.1 for more tain wind speed). Given a large enough set of records detail), we assume that mortality within a given region of hazard events and outcomes--combined with addi- is not uniformly distributed but rather is influenced by tional characteristics of the events and the exposed areas the frequency (and ideally, severity) of hazard events as controls--we could estimate a dose-response func- that have occurred in the region. We therefore allocate tion empirically. This would provide a sounder empir- more of the region's total mortality to places with a higher ical grounding of the proposed multihazard indicator apparent degree of hazard. and would also reduce the problem of including areas Rather than applying a constant mortality rate to a with relatively low disaster risk in the definition of region's population, we generate an accumulated mor- exposed areas. Clearly this represents a promising direc- tality by multiplying the mortality rate by the severity tion for future work. measure for each hazard. Since the degree of hazard To extend the mortality-weighted approach to eco- for each of the six hazards is measured on a different nomic loss risk assessment, we use the geographically Multihazard Risk Assessment 59 Box 6.1. Risk Assessment Procedure for Both Mortality and Economic Losses, Illustrated by the Mortality Example. 1. We extract the appropriate measure of total global losses from 1981­2000 from EM-DAT (in the mortality case, the number of fatalities) by hazard h: Mh. 2. Using the GIS data on the extent of each hazard, we compute the total population estimated to live in the area affected by that hazard in the year 2000: Ph. 3. We then compute a simple mortality rate for the hazard: rh = Mh / Ph. If we assume that the 1981­2000 period was representative, this rate is an estimate of the proportion of persons killed during a 20-year period in the area exposed to that hazard. Since the numbers are very small, they are expressed per 100,000 persons in 2000. Future revisions of the index could construct a mortality rate for the 20-year period based on annual rates which are computed using yearly mortality and population figures. As the results are intended only as an index of disaster risk, however, we believe that the computational simplification of using only end-of- period population is justified. 4. For each GIS grid cell i that falls into a hazard zone h, we compute the location-specific expected mortality by multiplying the global hazard-specific mortality rate by the population in that grid cell: Mhi = rh *Ph. We do this for all six hazards, then compute a mortality-weighted multihazard index value for each grid cell: 6 Yi = Phi. h = 1 This first estimate represents an unweighted index value that assumes that mortality rates are globally uni- form and that hazard severity has no influence on the relative distribution of mortality. In the following steps we relax these assumptions. 5. If we denote the various combinations of region and country wealth class (see Table 6.1) by j, then the esti- mated mortality in a given grid cell is now Mhij = rhj *Pi. 6. The global hazard data compiled for the analysis measures the degree of hazard in terms of frequency in most, although not all, cases (see Table 3.3). The various degree-of-hazard measures are used to redistribute the total regional mortality from EM-DAT across the grid cells in the area of the region exposed to each hazard. For example, if a grid cell were hit four times by a severe earthquake during the 20-year period, the regional mortality rate is multiplied by four to yield an accumulated mortality for that grid cell. More generally, if the degree of hazard measure is denoted by W, and assuming that the weighting scheme is identical across region/wealth-class combinations j, the accumulated mortality in the grid cell is: Mhij = rhj *Whi *Pi. Since the degree of hazard is not always measured on the same scale across hazards, simply adding up the resulting values would result in an index that could be unduly dominated by a hazard that happens to be measured on a scale with larger values. We therefore deflate the weighted hazard-specific mortality figures uniformly, so that the total mortality in each region adds up to the total recorded in EM-DAT. The resulting weighted mortality from hazard h in grid cell i and region/wealth-class combination j is thus: n M*hij = M'hij Mhj / M'hij, i = 1 where n is the number of grid cells in the area exposed to hazard h. Future revisions could be based on alter- native definitions of severity such as wind speed and duration for storms or earthquake and volcanic erup- tion magnitudes. 7. A mortality-weighted multihazard disaster risk hotspot index can be calculated as the sum of the adjusted single-hazard mortalities in the grid cell across the six hazard types: 6 Y*i = M*hij. h = 1 continued 60 Natural Disaster Hotspots: A Global Risk Analysis Box 6.1. Risk Assessment Procedure for Both Mortality and Economic Losses, Illustrated by the Mortality Example (continued) 8. To avoid literal interpretation of the multihazard disaster risk hotspot index as the number of persons expected to be killed in a 20-year period and in recognition of the many limitations of the underlying data, we convert the resulting measure into an index from 1 to 10, classifying the global distribution of unmasked grid cell values into deciles. referenced database of subnational GDP per unit area. Cyclones Although the global GDP surface is less detailed than Not surprisingly, mortality risk for cyclones is highest the population data set, it represents the best available along the Pacific and Indian coastlines and in the disaggregated information on economic output. Car- Caribbean and Central America (red areas of Figure rying out the same steps as described above for mor- 6.1a). Despite the low relative hazard shown for the Bay tality yields measures of economic losses per unit of of Bengal in Figure 4.1a, when weighted by mortality, GDP. Reallocation of economic losses within regions this area is ranked much higher in terms of risk. The and country wealth classes is again guided by hazard- picture changes somewhat in examining aggregate specific loss weights based on historical economic losses economic risk: the eastern United States and the United from EM-DAT. The resulting economic hotspots indi- Kingdom show relatively high risk, and poor areas of cator for damage-weighted multihazard disaster risk Africa no longer rank in the top three deciles (Figure reflects that although mortality impacts are lower in 6.1b). However, large portions of India and Asia remain richer countries, economic losses for a given event are relatively high in terms of aggregate economic risk. higher. For instance, a hurricane in southern Florida When normalized for GDP, Madagascar and neighbor- causes considerably more economic damage than a sim- ing areas reappear in the top three deciles, but China, ilar hurricane in poorer countries, since the value of real Japan, and the Republic of Korea drop out (Figure 6.1c). estate, infrastructure, and other economically produc- The areas of significant risk in North and Central Amer- tive assets is much higher in the United States. Of course, ica and the Caribbean appear to shift slightly south. such damage is usually a higher proportion of regional Overall, one billion people, or about 18 percent of and national income in developing countries than in the world's population, live in areas at high risk of cyclone industrial countries and is also higher relative to avail- mortality (Table 6.3a). More than one-fourth of the able resources for relief and reconstruction. world's GDP is at high risk of economic losses from cyclones (Table 6.3b). Single-Hazard Risk Assessment Results Drought Global results below are for risks of mortality- and eco- The drought vulnerability coefficients for mortality nomic loss-related outcomes by hazard type. Economic shown in Table 6.1 give high weight to low-income losses are presented both in aggregate terms and nor- African countries and very limited weight to other regions. malized relative to GDP density. Note that this normal- This is reflected clearly in Figure 6.2a, in which all of ization essentially removes the effect of GDP density, the grid cells in the highest three deciles fall in Sub- leaving the economic loss-related vulnerability coeffi- Saharan Africa. In contrast, economic loss risks related cients applied to the underlying hazard distribution. to drought are much more dispersed (Figure 6.2b), with Multihazar d Risk Assessment Figure 6.1. Global Distribution of Cyclone Risk a) Mortality Cyclone Mortality Risk Deciles 1st ­ 4th 5th­ 7th 8th­ 10th 61 62 Figure 6.1. Global Distribution of Cyclone Risk b) Total Economic Loss Natural Cyclone Total Economic Loss Disaster Risk Deciles 1st ­ 4th Hotspots: 5th­ 7th 8th­ 10th A Global Risk Analysis Multihazar d Risk Assessment Figure 6.1. Global Distribution of Cyclone Risk c) Economic Loss as a Proportion of GDP Density Cyclone Proportional Economic Loss Risk Deciles 1st ­ 4th 5th­ 7th 8th­ 10th 63 64 Natural Disaster Hotspots: A Global Risk Analysis relatively significant risks on all continents. As indicated Asia and China, despite their high reliance on agricul- in Table 6.3b, these areas are heavily populated and ture. This could reflect underreporting of historical losses important in terms of both economic and agricultural or low vulnerability to drought thanks to irrigation. activity. Areas of Europe, Central America, and west- ern Africa that seem to have only modest drought expo- Floods sure (Figure 4.1b) appear to have higher overall risk. The distribution of drought-related losses normalized Application of the flood mortality vulnerability coeffi- by GDP density (Figure 6.2c) reflects a middle ground cients from Table 6.1 to global population flood expo- between these extremes, with high relative losses in Sub- sure yields roughly the same spatial pattern as that of Saharan Africa, but also noticeable areas of high risk in flood hazard frequency shown in Figure 4.1c. Minor Central and South America, southern Europe, the Middle differences include lower relative risk in the eastern East, the Republic of Korea, and southern Australia. United States, high relative risk in North Africa, and Surprisingly, drought risk normalized by GDP density somewhat greater areas of high relative risk in India, does not seem to be significant in South and Southeast China, and other parts of Asia (Figure 6.3a). Much of Table 6.3. Characteristics of High-Risk Areas by Hazard a) Top three deciles based on mortality Hazard Land Area Population GDP Agricultural GDP Road/Rail Length (106 km2) (106) (109 $) (109 $) (103 km) Cyclones 2.7 1,096 8,909 119 332 Drought 9.7 573 1,086 49 619 Floods 14.4 3,936 22,859 528 1,507 Earthquakes 2.9 865 5,282 89 334 Volcanoes 0.1 56 166 3 12 Landslides 1.1 215 1,431 23 87 Percent of World Cyclones 2.1% 18.1% 20.2% 8.8% 4.2% Drought 7.5% 9.5% 2.5% 3.6% 7.8% Floods 11.0% 65.0% 51.7% 38.8% 19.0% Earthquakes 2.2% 14.3% 12.0% 6.5% 4.2% Volcanoes 0.1% 0.9% 0.4% 0.2% 0.2% Landslides 0.8% 3.6% 3.2% 1.7% 1.1% b) Top three deciles based on economic losses Hazard Land Area Population GDP Agricultural GDP Road/Rail Length (106 km2) (106) (109 $) (109 $) (103 km) Cyclones 2.4 940 11,723 94 401 Drought 14.8 2,790 17,556 446 1,697 Floods 13.2 3,776 31,216 598 1,751 Earthquakes 2.8 614 7,032 86 381 Volcanoes 0.1 59 201 2 11 Landslides 0.9 124 2,077 20 110 Percent of World Cyclones 1.8% 15.5% 26.5% 6.9% 5.1% Drought 11.3% 46.1% 39.7% 32.8% 21.4% Floods 10.1% 62.4% 70.6% 43.9% 22.1% Earthquakes 2.1% 10.1% 15.9% 6.3% 4.8% Volcanoes 0.1% 1.0% 0.5% 0.2% 0.1% Landslides 0.7% 2.1% 4.7% 1.5% 1.4% Multihazar d Risk Assessment Figure 6.2. Global Distribution of Drought Risk a) Mortality Drought Mortality Risk Deciles 1st ­ 4th 5th­ 7th 8th­ 10th 65 66 Figure 6.2. Global Distribution of Drought Risk b) Total Economic Loss Natural Disaster Drought Total Economic Loss Risk Deciles 1st ­ 4th Hotspots: 5th­ 7th 8th­ 10th A Global Risk Analysis Multihazar d Risk Assessment Figure 6.2. Global Distribution of Drought Risk c) Economic Loss as a Proportion of GDP Density Drought Proportional Economic Loss Risk Deciles 1st ­ 4th 5th­ 7th 8th­ 10th 67 68 Natural Disaster Hotspots: A Global Risk Analysis Asia and parts of Central and South America appear average population and GDP densities, comparable to high in risk in terms of both mortality and economic those observed for cyclones and floods (Table 6.3). loss (Figures 6.3a, b). This is true even after normaliz- ing for GDP density (Figure 6.3c). Flood losses in the Volcanoes eastern United States appear relatively high in absolute terms but are relatively small when normalized against Less than one-third of the categories in Tables 6.1 and GDP. Economic risks in Africa are small in both absolute 6.2 have nonzero values for volcanoes, owing in part to and relative terms. Collectively, very high proportions the limited period of record for this relatively infre- of the world appear to live in areas of high relative quent event. The result is that the areas of high relative flood risk, and these areas have high total and agricul- risk from volcanoes are even more restricted than the tural GDP density (Table 6.3). volcano hazard shown in Figure 4.1f. In particular, the mortality and economic risks of volcanoes in Japan are not ranked in the top three deciles (Figure 6.5a­c). Risks Earthquakes are high in localized areas around volcanoes, with recent The risks associated with earthquakes are high with activity mainly in Central and South America, East Africa, respect to mortality and economic losses in most but and Indonesia. Only about 1 percent of the world's not all areas where the hazard exists (Figure 4.1d). These population lives in the top three deciles for volcanoes areas include Central America, Venezuela, southern (Table 6.3). Europe, the Caucasus and Zagros mountain regions, Japan, and the Philippines (Figure 6.4a­c). The west- Landslides ern United States is a region where economic risks are relatively high but mortality risks are low. Conversely, Landslide risks are significant in terms of both mortal- the Himalayan region stands out as an area of high mor- ity and economic loss in Central America, northwest- tality risk but minimal absolute and relative economic ern South America, the Caucasus region, and Taiwan risk. Despite high earthquake hazard in Peru, Chile, and (China) (Figure 6.6a,b). Mortality-weighted risks are New Zealand, these areas do not appear especially at highintheHimalayanregion,thePhilippines,andIndone- risk in terms of mortality and economic loss. In gen- sia, but low in southern Europe and Japan (Figure 6.6a), eral, areas of high earthquake risk also have higher-than- where economic risks appear high (Figure 6.6b). Multihazar d Risk Assessment Figure 6.3. Global Distribution of Flood Risk a) Mortality Flood Mortality Risk Deciles 1st ­ 4th 5th­ 7th 8th­ 10th 69 70 Figure 6.3. Global Distribution of Flood Risk b) Total Economic Loss Natural Flood Total Economic Loss Disaster Risk Deciles 1st ­ 4th Hotspots: 5th­ 7th 8th­ 10th A Global Risk Analysis Multihazar d Risk Assessment Figure 6.3. Global Distribution of Flood Risk c) Economic Loss as a Proportion of GDP Density Flood Proportional Economic Loss Risk Deciles 1st ­ 4th 5th­ 7th 8th­ 10th 71 72 Figure 6.4. Global Distribution of Earthquake Risk a) Mortality Natural Earthquake (PGA) Mortality Disaster Risk Deciles 1st ­ 4th Hotspots: 5th­ 7th 8th­ 10th A Global Risk Analysis Multihazar d Risk Assessment Figure 6.4. Global Distribution of Earthquake Risk b) Total Economic Loss Earthquake (PGA) Total Economic Loss Risk Deciles 1st ­ 4th 5th­ 7th 8th­ 10th 73 74 Figure 6.4. Global Distribution of Earthquake Risk c) Economic Loss as a Proportion of GDP Density Natural Earthquake (PGA) Proportional Economic Loss Disaster Risk Deciles 1st ­ 4th Hotspots: 5th­ 7th 8th­ 10th A Global Risk Analysis Multihazar d Risk Assessment Figure 6.5. Global Distribution of Volcano Risk a) Mortality Volcano Mortality Risk Deciles 1st ­ 4th 5th­ 7th 8th­ 10th 75 76 Figure 6.5. Global Distribution of Volcano Risk b) Total Economic Loss Natural Volcano Total Economic Loss Disaster Risk Deciles 1st ­ 4th Hotspots: 5th­ 7th 8th­ 10th A Global Risk Analysis Multihazar d Risk Assessment Figure 6.5. Global Distribution of Volcano Risk c) Economic Loss as a Proportion of GDP Density Volcano Proportional Economic Loss Risk Deciles 1st ­ 4th 5th­ 7th 8th­ 10th 77 78 Figure 6.6. Global Distribution of Landslide Risk a) Mortality Natural Landslide Mortality Disaster Risk Deciles 1st ­ 4th Hotspots: 5th­ 7th 8th­ 10th A Global Risk Analysis Multihazar d Risk Assessment Figure 6.6. Global Distribution of Landslide Risk b) Total Economic Loss Landslide Total Economic Loss Risk Deciles 1st ­ 4th 5th­ 7th 8th­ 10th 79 80 Figure 6.6. Global Distribution of Landslide Risk c) Economic Loss as a Proportion of GDP Density Natural Landslide Proportional Economic Loss Disaster Risk Deciles 1st ­ 4th Hotspots: 5th­ 7th 8th­ 10th A Global Risk Analysis Chapter 7 Multihazard Risk Assessment Results Multihazard risks were calculated by summing the Another way to look at multihazard mortality risk is vulnerability-weighted single-hazard mortality and eco- to show for how many hazards each mortality-risk grid nomic loss risk values for each grid cell across the six cell value falls into the top three deciles (Figure 7.2a). hazard types (Figure 7.1a­c). The strong drought- This presentation makes it easier to discriminate mortality signal in Sub-Saharan Africa shown in Figure within large regions where risks are evaluated as high 6.2a is strongly reflected in the multihazard map weighted across all hazards versus areas that are high risk for each by mortality (Figure 7.1a). So are the high-risk areas of hazard individually. Figures 1.2a­c show which types Central America, the Caribbean, the Bay of Bengal, of hazards are prevalent at each location. As noted above, China, and the Philippines because of cyclones and drought and combinations of drought and hydro- floods. Areas of high earthquake and landslide risk are meteorological hazards dominate both mortality and evident in Central America and Venezuela, Central Asia, economic losses in Sub-Saharan Africa. Geophysical the Himalayas, Japan, the Philippines, and Indonesia. hazards drive high mortality risk in western Asia, and Because of the relatively short period on which these combine with hydro-meteorological hazards in moun- rates are based, the vulnerability coefficients appear to tainous areas of Central America and Asia (Figure 1.2c). give greater weight to the higher frequency hazards such Drought is also an important driver of economic losses as cyclones, drought, landslides, and floods. Mortality in many other countries, including Mexico, Spain, the and economic losses are high in specific regions for Republic of Korea, and Australia. earthquakes and volcanoes where they have occurred The concentration of population and economic activ- in 1981 through 2000, but may underestimate poten- ity in areas at high relative risk from one or more haz- tial risks in other regions that face some degree of hazard. ards is of great interest from the viewpoint of national-level Based on the mortality-weighted index, nearly one- vulnerability and response capacity. As indicated in Table fourth of total land area and more than three-fourths of 1.2a, 35 countries have more than 5 percent of their the world's population are subject to a relatively high population living in areas identified as relatively high risk of mortality from one or more hazards. This reflects in mortality risk from three or more hazards. Ninety- the higher population densities of areas that have expe- six countries have more than 10 percent of their pop- rienced relatively high mortality during the past two ulation in areas at risk from two or more hazards decades according to EM-DAT. Only one-twentieth of (Table 1.2b and Figure 1.3). And 160 countries have the total land area (but about one-third of the popula- more than one-fourth of their population in areas at rel- tion) is subject to higher mortality risk from two or more atively high mortality risk from one or more hazards hazards. About 7 percent of the total population lives in (Figure 1.6). Similarly, many of the areas at higher risk areas at high mortality risk from three or more hazards of economic loss from multiple hazards are associated (Table 7.1a). More than four-fifths of GDP is located in with higher-than-average densities of GDP, leading to a areas of relatively high economic risk subject to one or relatively high degree of exposure of economically pro- more hazards and more than half in high-risk areas ductive areas (Table 7.2 and Figures 1.5 and 1.6). subject to two or more hazards (Table 7.1b). 81 82 Figure 7.1. Global Distribution of Disaster Risk Hotspots for All Hazards a) Mortality Natural All-Hazard Mortality Disaster Risk Deciles 1st ­ 4th Hotspots: 5th­ 7th 8th­ 10th A Global Risk Analysis Mulithazar d Risk Assessment Figure 7.1 Global Distribution of Disaster Risk Hotspots for All Hazards b) Total Economic Loss Results All-Hazard Total Economic Loss Risk Deciles 1st ­ 4th 5th­ 7th 8th­ 10th 83 84 Figure 7.1 Global Distribution of Disaster Risk Hotspots for All Hazards c) Economic Loss as a Proportion of GDP Density Natural All-Hazard Proportional Economic Loss Disaster Risk Deciles 1st ­ 4th Hotspots: 5th­ 7th 8th­ 10th A Global Risk Analysis Mulithazar d Risk Assessment Figure 7.2 Global Distribution of Disaster Risk Hotspots by Number of Hazards a) Mortality Results Mortality Risk Top 3 Deciles for: 1 Hazard 2 Hazards 3+ Hazards 85 86 Figure 7.2 Global Distribution of Disaster Risk Hotspots by Number of Hazards b) Total Economic Loss Natural Total Economic Loss Risk Disaster Top 3 Deciles for: 1 Hazard Hotspots: 2 Hazards 3+ Hazards A Global Risk Analysis Mulithazar d Risk Assessment Figure 7,2 Global Distribution of Disaster Risk Hotspots by Number of Hazards c) Economic Loss as a Proportion of GDP Density Results Proportional Economic Loss Risk Top 3 Deciles for: 1 Hazard 2 Hazards 3+ Hazards 87 88 Natural Disaster Hotspots: A Global Risk Analysis Table 7.1. Characteristics of High-Risk Disaster Hotspots a) Top three deciles based on mortality No. of Hazards Land Area (106 km2) Population (106) GDP (109 $) Agricultural GDP (109 $) Road/Rail Length (103 km) 1 22.8 2,602 15,648 470 1,890 2 6.3 1,629 9,276 184 614 3­5 0.9 432 3,219 28 109 Total 30.1 4,664 28,143 683 2,613 Percent of World 1 17.4% 43.0% 35.4% 34.6% 23.9% 2 4.8% 26.9% 21.0% 13.5% 7.8% 3­5 0.7% 7.1% 7.3% 2.1% 1.4% Total 23.0% 77.0% 63.7% 50.1% 33.0% b) Top three deciles based on economic losses No. of Hazards Land Area (106 km2) Population (106) GDP (109 $) Agricultural GDP (109 $) Road/Rail Length (103 km) 1 16.1 2,037 13,784 465 1,813 2 5.7 1,841 14,732 271 803 3-5 2.1 808 8,208 76 289 Total 23.9 4,686 36,724 812 2,905 Percent of World 1 12.3% 33.7% 31.2% 34.1% 22.9% 2 4.3% 30.4% 33.3% 19.9% 10.1% 3-5 1.6% 13.3% 18.6% 5.6% 3.7% Total 18.3% 77.4% 83.1% 59.6% 36.7% Mulithazard Risk Assessment Results 89 Table 7.2. Countries at Relatively High Economic Risk from Multiple Hazards a) Three or more hazards (top 33 based on GDP) Country Percent of Total Area at Percent of Population Percent of GDP in Areas at Risk in Areas at Risk at Risk Taiwan, China 97.0 96.6 96.5 Dominican Rep. 77.0 90.7 92.0 Jamaica 79.6 87.7 87.8 El Salvador 64.5 84.1 85.4 Guatemala 39.5 83.4 83.3 Antigua and Barbuda 53.4 80.4 80.4 Japan 51.3 75.8 80.2 Costa Rica 38.6 77.9 80.1 Philippines 35.8 72.5 78.7 Colombia 10.0 69.0 73.0 Bangladesh 41.9 55.6 62.7 Chile 2.9 58.4 62.6 Korea, Rep. of 24.7 61.6 61.6 Turkey 37.7 50.4 55.6 Barbados 53.4 53.4 53.4 Guam 53.2 59.7 51.6 Uzbekistan 5.0 51.4 51.4 Ecuador 10.0 50.5 50.0 Venezuela 3.1 40.6 47.6 Peru 1.4 30.4 43.9 St. Kitts and Nevis 33.8 41.6 41.6 Iran, Islamic Republic of 15.4 45.2 39.8 Indonesia 3.0 30.5 34.2 Honduras 7.9 30.9 33.2 Greece 19.9 18.6 32.0 Albania 35.9 27.6 29.6 Mexico 6.9 31.1 29.2 Hong Kong, China 51.4 29.5 28.2 Tajikistan 1.2 27.1 27.1 Mozambique 0.0 1.3 23.7 Syrian Arab Rep. 7.5 24.3 21.2 Bolivia 0.4 20.3 20.8 United States 1.6 20.6 20.8 90 Natural Disaster Hotspots: A Global Risk Analysis Table 7.2. Countries at Relatively High Economic Risk from Multiple Hazards b) Two or more hazards (top 75 based on GDP) Country Percent of Total Area Percent of Population Percent of GDP in Areas at Risk in Areas at Risk at Risk Taiwan, China 97.5 97.0 96.9 El Salvador 88.7 95.4 96.4 Jamaica 94.9 96.3 96.3 Dominican Rep. 87.2 94.7 95.6 Guatemala 52.7 92.1 92.2 Korea, Rep. of 82.8 92.2 91.5 Vietnam 33.2 75.7 89.4 Japan 65.6 86.5 89.0 Albania 86.4 88.6 88.5 Costa Rica 51.9 84.8 86.6 Colombia 21.2 84.7 86.6 Bangladesh 71.4 83.6 86.5 Philippines 50.3 81.3 85.2 Turkey 73.0 80.9 83.3 Trinidad and Tobago 66.7 82.4 83.1 Guam 83.6 84.5 82.6 Thailand 47.8 70.1 81.2 Antigua and Barbuda 53.4 80.4 80.4 Barbados 79.9 79.9 79.9 San Marino 66.7 55.3 73.1 Greece 65.5 73.7 72.6 Ecuador 24.4 73.6 72.2 Mexico 15.9 68.2 71.1 United States 9.3 67.5 68.7 Dominica 68.3 67.0 68.3 Nicaragua 21.6 68.7 67.9 Chile 5.2 64.9 67.7 Iran, Islamic Republic of 31.7 69.8 66.5 United Kingdom 38.6 64.5 66.0 Venezuela 4.9 61.2 65.9 Uzbekistan 9.3 65.6 65.5 St. Kitts and Nevis 0.0 52.8 64.9 Jordan 13.7 64.9 64.7 Argentina 1.8 57.4 63.2 South Africa 8.6 56.3 62.4 Tunisia 30.4 64.1 62.4 Indonesia 11.5 67.4 62.3 Cuba 22.5 56.7 57.9 China 13.1 49.8 56.6 Honduras 19.0 56.0 56.5 Haiti 44.4 47.9 56.0 Uruguay 3.0 55.0 55.0 Hong Kong, China 73.0 58.9 54.2 Netherlands 63.2 55.0 53.9 Peru 4.0 41.5 53.7 Liechtenstein 53.9 45.9 53.6 Kyrgyz Rep. 8.3 51.3 53.4 Montserrat 50.3 50.3 50.3 Romania 37.4 45.8 50.3 India 22.1 47.7 49.6 Algeria 3.1 49.3 48.3 Niue 48.1 48.1 48.1 continued Mulithazard Risk Assessment Results 91 Table 7.2. Countries at Relatively High Economic Risk from Multiple Hazards, continued b) Two or more hazards (top 75 based on GDP) Country Percent of Total Area Percent of Population Percent of GDP in Areas at Risk in Areas at Risk at Risk Cyprus 50.4 60.5 47.4 Korea, Dem. People's Rep. of 27.6 44.5 46.3 Andorra 43.5 19.4 45.0 Australia 0.3 44.0 44.7 Paraguay 2.0 45.6 42.9 Azerbaijan 15.6 42.3 42.4 Pakistan 9.0 40.1 41.6 St. Vincent and the Grenadines 41.6 41.6 41.6 Georgia 4.4 40.5 41.0 Germany 26.8 38.9 40.3 Ireland 9.6 32.7 39.8 Italy 42.2 35.4 38.8 Macedonia, FYR 38.8 29.6 38.7 Tajikistan 4.1 38.2 38.3 Bolivia 1.0 36.6 37.7 Mozambique 0.0 1.9 37.3 Syrian Arab Rep. 14.9 34.4 36.8 Djibouti 1.9 31.7 35.3 Cambodia 9.1 31.3 34.5 New Zealand 1.6 33.9 33.7 Morocco 3.4 30.4 33.4 Canada 0.2 36.0 32.1 Bulgaria 29.3 31.6 30.0 Chapter 8 Case Studies The Hotspots project is an effort to deepen understanding exposure to natural hazards and their vulnerability to of the risks posed by multiple natural hazards and the those hazards. potential for mitigation and response approaches that Two types of case studies were undertaken. One type take into account the interactions among different haz- examined the impacts of a particular hazard on a regional ards and hazard vulnerabilities. It identifies risks in or global scale. Three such studies were performed, deal- two ways: in a global multihazard analysis and in a set ing with drought and disaster in Asia, global landslide of hazard- or place-specific case studies. risks, and storm surges in coastal areas. Limitations of the global analysis include the fol- The other type of study was geographically limited lowing: and identified risks associated with a particular hazard or combination of hazards using a richer set of location- 1. Global spatial data sets do not exist for the vulnera- specific data. These geographically limited case studies bility characteristics of the major sets of elements at were designed to risk from each hazard, although in some cases vul- nerability may be inferred from existing data on a 1. "Ground truth" particular regions identified as poten- limited basis. tial hotspots; 2. Existing global spatial data sets on major hazards and 2. Explore specific cases where more detailed loss prob- elements at risk are of a coarse resolution, sufficient ability data and models exist than are available for resolving only relatively broad spatial patterns of globally; risk. 3. Ascertain what finer scale data may exist locally, for 3. Global data on socioeconomic "outcome" variables-- example, on vulnerability, response capacity, and such as mortality, morbidity, economic losses, and poverty; impoverishment--are universally available only at 4. Identify cross-hazard dependencies and interac- the country level, in the form of national statistics. tions among hazards, exposure, vulnerability, and However, such data are needed to verify the global multihazard risk management; risk assessment (that is, assessed spatial patterns of disaster risk hotspots should correspond to histori- 5. Examine the policy context for risk management and cal patterns of actual human and economic losses to the degree to which multiple hazards are recognized some degree). and addressed in an integrated manner; To partially address these limitations, case studies 6. Engage national- to local-level stakeholders; and were undertaken to complement the global-scale analy- 7. Demonstrate that the theory and methods that sis. The case studies use the same theory of disaster guide the global analysis can be applied on more causality as the global analysis: that over a given time regional or local geographic scales. period, the risks of a specified type of disaster-related loss to a set of elements are a function of the elements' 93 94 Natural Disaster Hotspots: A Global Risk Analysis The three case studies in this category deal with mul- up--generalized from more detailed local data. In prac- tihazard risks in Sri Lanka; multihazard risks in Cara- tice, many barriers remain. Data sets may contain cas, Venezuela; and flood risks in the Tana River Basin gaps, data for one country may not be available for a in Kenya. bordering country, and data on vulnerability charac- Table 8.1 lists the six case studies and their authors. teristics for each hazard remain scanty. Vulnerability often must be inferred from proxies at best. The global infrastructure for systematically assembling and inte- Scale Issues grating relevant data sets for disaster risk assessment at multiple scales remains inadequate. Nonetheless, the The place-based case studies demonstrate that scale mat- fact that relevant data sets can be obtained at various ters. Geographic areas that are subsumed into a single scales and integrated for the case studies below creates hotspot at the global scale are shown to have a highly the hope that one day data can be collected and shared variable spatial distribution of risk at a more localized routinely to improve disaster risk assessment both glob- scale. ally and locally. Scale also affects data availability and quality. Hazard, exposure, and vulnerability data are available at sub- national resolutions for individual countries and even Summary of Case Study Results cities, as the analyses for Sri Lanka and Caracas show. More comprehensive, finer resolution, and better Hazard-focused, geographically extensive case studies quality data contribute to more complete, accurate, and reliable risk assessments. Drought Disaster in Asia Better data resolution and a richer set of variables A drought disaster is caused by the combination of a also contribute to results that are more relevant for climate hazard event (deficits in rain or snow) and national- to local-scale risk management planning. This societal vulnerability (the economic, social, and polit- is highly important, as decisions made at this level may ical characteristics that render livelihoods susceptible have the greatest effect on risk levels, whether positively in affected areas). A pilot investigation for 27 countries or negatively. In some instances, risk assessors and plan- in Asia compared the incidence of drought disasters ners at the national and local levels may be able to "down- recorded in EM-DAT with climatically defined drought scale" global data for larger scale risk assessment to hazard events. compensate for a lack of locally collected data. In an Severe, persistent precipitation deficits as defined ideal world, however, global analyses would be scaled by the WASP index corresponded with reported drought Table 8.1. Summary of Case Studies Case Study Contributors Hazard-oriented Global Landslide and Avalanche Hotspots Farrokh Nadim, Anne Sophie Gregoire, Carlos Rodriguez, Pascal Peduzzi, Oddvar Kjekstad An Expert Assessment of Storm Surge "Hotspots" Robert J. Nicholls Toward Calculation of Global Drought Hazard: A Pilot Study for Asia Matthew Barlow, Heidi Cullen, Brad Lyon, Olga Wilhelmi Geographically focused Identification of Global Natural Disaster Risk Hotspots-- Vidhura Ralapanawe, Lareef Zubair Sri Lanka Case Study Disaster Resilient Caracas Kristina R. Czuchlewski, Klaus H. Jacob, Arthur L. Lerner-Lam, Kevin Vranes Reducing the Impacts of Floods through Early Warning and Hussein Gadain, Nicolas Bidault, Linda Stephen, Ben Watkins, Preparedness: A Pilot Study for Kenya Maxx Dilley, Nancy Mutunga Case Studies 95 disasters most frequently in Asian countries with low Varying degrees of correspondence between clima- average annual rainfall (Figure 8.1). In the 11 coun- tological drought events and drought disasters may be tries with annual precipitation less than 700 mm, drought explained by two groups of factors. First, differences disasters typically occurred 20 to 40 percent of the may be attributable to uneven reporting of disasters time in the three months following climatic drought throughout the study period, uncertain precipitation events (defined as country-average 12-month WASP data accuracy, and the choice of criteria for defining values of less than ­1) from 1979 through 2001. drought events. Second, differences may stem from the Widespread and prolonged precipitation deficits in types of land uses and economic activities that were the region from 1999 through 2001 were associated exposed to drought within the affected countries and with drought disaster events among the countries in their degree of vulnerability. Further research may central southwest Asia, but not during the weaker but clarify the combination of climatological and socio- more prolonged dry period of the 1980s (Figure 8.2). economic factors that results in disaster. Such research In other parts of the region, the WASP threshold cor- could enhance the prospects for using data for real-time responded with disaster occurrence in both the 1980s early warning of disasters. and 1990s (see Lao People's Democratic Republic in Figure 8.3), while some countries are so large that a Global Landslide Risks country average is not very meaningful (see India in This study performed a data-based, first-order identi- Figure 8.3). fication of the geographic areas that constitute hotspots Figure 8.1. Frequency with Which Climatic Drought Hazard Events Were Accompanied by Drought Disasters (Gold) or Not (Blue) from 1979 through 2001 Countries Have Been Ordered Left-To-Right Based on Annual Average Precipitation (Green Line, in millimeters). Matches (gold) and Misses (blue) for 12-Month WASP; Annual Precipitation (green line) 7 3000 6 2500 s n ssei 5 2000 oitati s/Me hcta 4 picerP 1500 M of 3 re verage b 1000 Al m u 2 N nnua A 500 1 0 0 lear na dr Is Jo natsike lia ai aq an na ai a a d o o ni ai la ai m ai Ir Ir gr p ae se d n na p La in hse g natsi antsi ne aise iasy In od n o na Syr stikji m Ch Ne Kor ilaa Ja h tnaei p dal n o b h Pak V ilip g mb laa M Ar Geo nkaLir d g Ta S T h n M Uz P Ca In Af Ba 96 Natural Disaster Hotspots: A Global Risk Analysis Figure 8.2. WASP Estimates of Climatic Drought (Shaded Brown Curve) and Drought Disasters (Red Bars) for Central Southwest Asian Ccountries Figure 8.3. WASP Estimates of Climatic Drought (Shaded Brown Curve) and Drought Disasters (Red Bars) for Lao PDR and India Case Studies 97 for global landslide disaster risk on a non-national scale, extreme water levels and flooding. Surges are most com- with an emphasis on developing countries. This iden- monly produced by the passage of atmospheric tropi- tification includes combining landslide hazards with cal or extratropical depressions. Positive surges can occur the vulnerability of people and infrastructure to obtain on any ocean coast, but they are best developed under risks of losses. extreme meteorological forcing and where the coastal The probability of landslides is estimated by model- morphology is favorable. Surges of two to three meters ing the physical processes combined with historical have been regularly observed in the southern North Sea, experience and statistics. Rapid mass movements such 6-meter surges are often associated with the landfall of as rockslides, debris flows, and snow avalanches are a category 5 hurricane, and the largest surge ever observed included, whether triggered by precipitation or earth- was 13 meters, during a tropical cyclone in Australia. quakes. The main input data for assessing the land- These surges are generally associated with strong wave slide hazard are topography and slope angles, activity, and the impacts of waves and surges need to precipitation, seismic activity, soil type, hydrological be considered together. condition, and vegetation. Snow avalanche probabili- Flooding caused by storm surges is a major hazard ties are calculated from slope and relief, precipitation for coastal residents. In the past 200 years, at least 2.6 values from the winter months and temperature. The million people may have drowned because of storm resulting combined global landslide and avalanche hazard surges, which also caused a range of other damage and map is an input to the global hotspots multihazard analy- disruption. Given that storms result in a number of haz- sis described previously. The calculated landslide hazard ards (surge/waves, winds, intense rainfall, tornadoes, corresponds well with historical data in selected case and waterspouts), it is often quite difficult to isolate study areas (Figure 8.4). the specific impacts of the surge and hence attribute Exposure and vulnerability derive from socio- specific damage to surges. However, drowning by surge economic factors such as population density, quality of is the major killer in most coastal storms with high fatal- infrastructure, and response capacity. These factors were ities, so it is generally possible to link fatalities to surge combined with the hazard estimates in an independ- events. This approach is followed in this case study. ent risk analysis calibrated with observed losses from The major impacts of surges are concentrated in a lim- the EM-DAT international disaster history database. The ited number of regions. Most fatalities in the past 200 combination of gridded hazard probabilities (Figure years have occurred in Asia, especially around the Bay 8.5) with exposure and vulnerability factors was used of Bengal, particularly in Bangladesh, where more than to calculate mortality risks (Figure 8.6). one million people may have died during the period Unlike the global risk analysis for landslides pre- (Murty 1984; Warrick and Ahmad 1996). This includes sented in Chapter 6, in this case vulnerability was both the 1970 surge, when 300,000 to 500,000 people characterized on the basis of national-level variables were killed and the 1991 surge, when about 140,000 identified during the construction of the Disaster Risk people were killed. It is noteworthy that most of the Index (DRI) that provided the basis for the global risk surge events that have killed substantial numbers of assessment presented in Reducing Disaster Risk: A Chal- people (over 10,000 deaths) have occurred where there lenge for Development (UNDP 2004). The case study have been substantial land claim and other human mod- companion volume to this report contains results of an ifications to the coastal zone, suggesting that the hazard analysis of global landslide risks prepared by the Nor- has coevolved with human modifications to the coastal wegian Geotechnical Institute. zone. These areas include Bangladesh, China, Japan, and the southern North Sea. All of these areas have signifi- Storm Surges in Coastal Areas cantly adapted to the threat posed by surges. This adap- Surges are positive or negative changes in sea level result- tation is best developed around the North Sea and Japan, ing from variations in atmospheric pressure and asso- where surges have had very limited impacts since the ciated winds. They are additive to normal tides. When 1960s, largely because of massive investment in flood positive surges are added to high tides, they can cause defense infrastructure. Even in Bangladesh, improved 98 Figure 8.4. Modeled Landslide Zonation (Shading) and GEORISK Landslide Inventory (Blue) in Armenia Natural Disaster Yerevan Landslides Hotspots: GEORISK Hazard Zones Landslide Zonation A No Hazard Global Low Hazard Risk Medium Hazard Analysis High Hazard Case Figure 8.5. Landslide Hazard Map for Central America and Andean South America Studies United States The Bahamas Cuba Turks and Caicos Islands Mexico Cayman Islands HaitiDominican Republic Jamaica Puerto Rico Anguilla Virgin Islands Belize Guatemala Honduras El Salvador Nicaragua Aruba Netherlands Antilles Costa Rica Panama Venezuela Global Landslide Hazard Zonation Hot-Spots Colombia Negligible to very low (Class 1­2) Low (Class 3) Low to moderate (Class 4) Moderate (Class 5) Ecuador Medium (Class 6) Medium to high (Class 7) Brazil High (Class 8) Peru Very high (Class 9) 0 250 500 1,000 1,500 2,000 Kilometers 99 100 Figure 8.6. Landslide Mortality Risks Calibrated with Historical Landslide-Related Mortality from the EM-DAT International Disaster Database Natural Expected Annual Deaths Disaster Per Pixel -8 -5 10 ­ 10 -5 -3 Hotspots: 10 ­ 10 0.001 ­ 1 A Global Risk Analysis Case Studies 101 flood warnings appear to be reducing the number of and landslides were weighted for their associated dis- fatalities significantly. However, there is no room for com- aster risk with proxies for economic losses to provide placency. Even in the United States, with its highly a risk map or a hotspots map. Principal findings include effective storm warning systems, the potential for large the following: numbers of fatalities remains (for example, a hurricane 1. Useful hazard and vulnerability analysis can be car- landfall on New Orleans or a major hurricane or north- ried out with the type of data that is available in- easter impacting New York City). country. The hazard estimates for droughts, floods, The storm surge hazard will also continue to evolve cyclones, and landslides show marked spatial vari- as a result of socioeconomic and climate change, as well ability. Vulnerability shows marked spatial variabil- as continuing efforts to mitigate this hazard. Increased ity as well. Thus, the resolution of analysis needs to exposure in surge-prone areas may be problematic: areas match the resolution of spatial variations in relief, such as eastern Africa that presently appear to have lim- climate, and other features. Analyses of disasters need ited surge problems may see problems emerge if there higher spatial and temporal resolution for planning is substantial population growth along the coast. and action at the local level. Given that populated areas are particularly exposed, the likelihood of storm surge needs to be considered as these 2. Multihazard analysis brought out regions of high risk areas are developed. such as the Kegalle and Ratnapura districts in the The available data are not sufficient to define precise southwest; the Ampara, Batticaloa, Trincomalee, Mul- surge hotspots. It is most realistic to define the regions laitivu, and Killinochchi districts in the northeast; where major surge impacts might occur (see Table and the districts of Nuwara Eliya, Badulla, Ampara, 8.2). Potential hotspots within these regions can then and Matale, which contain some of the sharpest hill be identified on the basis of coastal elevation, coastal slopes of the central mountain massifs. land use, and historical experience (as well as on the 3. There is a distinct seasonality to risks posed by drought, consideration of relevant scenarios). See Table 8.3. floods, landslides, and cyclones. Whereas the east- ern regions have hotspots during the boreal fall and Place-based case studies early winter, the western-slopes regions are risk prone Multihazard Risks in Sri Lanka in the summer and the early fall. Thus, attention is This case study exemplifies a high-resolution assess- warranted not only on hotspots but also on "hot ment of natural hazards, vulnerability to hazards, and seasons." disaster risk. Drought, flood, cyclone, and landslide haz- 4. Climate data were useful for estimating the degree ards, as well as vulnerability to those hazards, were iden- of hazard in the case of droughts, floods, and cyclones tified using data from Sri Lankan government agencies. and the risk of flood and landslide. The methodolo- Drought- and flood-prone areas were mapped using gies used here for hazard analysis of floods and rainfall data that were gridded at a resolution of 10 kilo- droughts present an explicit link between climate meters. Cyclone and landslide hazards were mapped and hazard. This link can be used in conjunction based on long-term historical incidence data. Indexes with seasonal climate prediction to provide predic- for regional industrial development, infrastructure devel- tive hazard risk estimates in the future. opment, and agricultural production were estimated on the basis of proxies. An assessment of regional food inse- 5. Climatic, environmental, and social changes such as curity from the World Food Programme was used in the deforestation, urbanization, and war affect hazard analysis. Records of emergency relief were used in esti- exposure and vulnerability. It is more difficult to quan- mating a spatial proxy for disaster risk. A multihazard tify such changes than the baseline conditions. How- map was developed for Sri Lanka based on the estimates ever, climate change is already making parts of the of regional drought, flood, cyclone, and landslide haz- island more prone to drought hazard. ards. The hazard estimates for drought, floods, cyclones, 102 Natural Disaster Hotspots: A Global Risk Analysis Table 8.2. An Expert Synthesis of Storm Surge Hotspots around the World Surge-Prone Regions Hotspots Commentary Fatalities* Other Damage Bay of Bengal High High Improved flood warnings may reduce fatalities (Bangladesh and Eastern India) Western India/Pakistan Unclear Unclear Cyclones are less frequent than Bay of Bengal (1:4) and there is less exposure China/Japan Potentially high Potentially high Ongoing flood damage is reported in China Korea, Rep. of Low Low Region lacks large low-lying coastal areas, but this is changing owing to extensive land claim Thailand, Vietnam, Potentially high Medium to high Region is frequently impacted by typhoons, and population Philippines in deltas of low-lying areas is growing rapidly Pacific Islands Probably high High Limited historical information Australia and Low Low Region has limited habitation in low-lying coastal areas New Zealand Indian Ocean Islands Low Low Region has limited habitation in low-lying coastal areas Eastern Africa Low Low Habitation in low-lying coastal areas is not significant, but and Oman could increase Rio de la Plata Low Low Difficult to assess, owing to limited literature--may suggest (Argentina and limited impacts to date Uruguay) Caribbean Potentially high Medium to high Human activity is concentrated around the islands, and hence exposed to surge. However, the role of surge relative to other hurricane impacts is less clear Central America Potentially high Medium Human activity is often concentrated away from the coast, and Mexico in local areas to high which is atypical at the global level. Hence other hurricane impacts appear more important than in other regions (for exam- ple, Hurricane Mitch), although there are localized hotspots. United States-- Potentially high High Effective evacuation has reduced fatalities, but potential Gulf and East coasts hotspots remain Europe--Atlantic coast Potentially high Potentially high Hard defenses and improved flood predictions and warnings appear to have been effective in reducing this hazard Europe-- Locally high Medium to high Surges are not large, so deaths are unlikely, except in areas of Mediterranean coast land claim where flood depths could be substantial. However, significant damage and disruption can occur. Europe-- Potentially high Potentially high Hard defenses and improved flood predictions and warnings North Sea coast appear to have been effective in reducing this hazard Europe--Baltic Sea Locally high Medium to high Hard defenses and improved flood predictions and warnings coast appear to have been effective in reducing this hazard * In the column titled "Fatalities," "high" indicates the potential for more than 1,000 deaths in a surge event. Other damage estimates are based on the expert judgment of the author. Case Studies 103 6. The analysis was carried out in the context of civil and developing metropolitan areas is thus a component wars from 1983 to 2002. While natural disasters of development policy. accounted for 1,483 fatalities in this period, civil wars This report summarizes the findings of a preliminary accounted for more than 65,000. Wars and conflict study of the natural hazards faced by Caracas, Venezuela, complicate natural hazard and vulnerability analy- and proposes ways in which urban planning and design sis. However, the vulnerabilities created by the war can incorporate a qualitative natural hazards risk assess- make efforts to reduce disaster risks all the more ment. The report is designed to be illustrative rather important. than comprehensive, since the development of an urban plan is a complicated and organic task with many stake- Risks are calculated in a number of different ways, holders. However, the methodological approach and several alternative multihazard risk maps are pre- sented. Subject to data limitations, records of past dis- described in this work can serve as the basis for such a asters are used to weight for exposure and vulnerability plan and be applied to other cities and regions. to particular hazards. Located on the intersection of the South American One multihazard map is generated by weighting and Caribbean Plates, northern Venezuela faces extreme hazard frequency with historical disasters obtained from seismological hazards. Major earthquakes have destroyed the EM-DAT database (Figure 8.7). Multiple landslides Caracas three times in the last 400 years. The last large within a single year were treated as one event. earthquake (Mw = 6.5) came in 1967, killing an esti- Another multihazard risk map was calculated by mated 300 people and destroying four modern struc- weighting each hazard index by the disaster relief expen- tures built for earthquake resistance (Papageorgiou diture data from the Sri Lanka Department of Social Ser- and Kim 1991). In addition, the position of the north- vices for each hazard (Figure 8.8). This resulting map ern coast near 10°N ensures frequent heavy rainfall is heavily weighted toward droughts and cyclones, with events with strong erosion potential. In December 1999, landslides receiving a meager weight. a month of rain on the north central coast of Venezuela-- This hotspots map shows higher risk in the north including over 900 millimeters of rain in a 72-hour and north central regions and in Hambantota District period between 15 and 17 December--triggered land- in the southeast compared with the previous map. Deter- slides, mudflows, and debris flows on the north face of mining which methodology to employ will be based in the El Ávila range that killed an estimated 25,000 res- part on considerations regarding the application of the idents of the coastal state of Vargas. analysis for risk management. Since the last major earthquake in 1967, the popu- Weights based on relief expenditures obtained from lation of Caracas has doubled to five million people, the Sri Lanka Department of Social Services: Drought: with a population density of 12,000 persons per 126, Floods: 25, Landslides: 0.06, Cyclones: 60. The square kilometer and growth of 3.1 percent per year. weighted index has been rescaled to the range 0 to 100. Eighty-six percent of the Venezuelan population is urban, making it the seventh most urbanized country in the Multihazard Risks in Caracas, Venezuela world. The valley floor is well developed, with high- Cities are centers of economic opportunity and cul- rise buildings and densely packed apartment blocks ture, and are a natural focus for investment and devel- scattered unevenly throughout the city. These buildings opment. The role of cities is recognized globally in the are generally concentrated in the deepest part of the trend toward increasing urbanization in most countries. basin (where shaking is expected to be highest during However, the increased concentration of physical and an earthquake). cultural assets that accompanies increases in the spa- Barrios, or informal squatter settlements, dominate tial density of population also increases their exposure the landscape on the low-lying, rugged mountains to to geographically limited natural hazards. If these assets the east and west of the city center, where rainfall-induced are fragile (vulnerable), then the city is at risk. A region debris flows are expected to be greatest. To the south or country that depends on the sustainable growth of is a mixture of urbanizaciónes (similar to suburbs) and its cities shares the risk. Risk management for existing barrios. The individual building blocks of the barrios, 104 Natural Disaster Hotspots: A Global Risk Analysis Figure 8.7 Weights are based on relative frequency in EM-DAT: Drought: 9, Floods: 30, Landslides: 2, Cyclones: 3. The weighted index has been rescaled to the range 0 to 100. 80°0'0"E 81°0'0"E 82°0'0"E 9°0'0"N 9°0'0"N 8°0'0"N 8°0'0"N 7°0'0"N 7°0'0"N 6°0'0"N 6°0'0"N 10 20 30 40 50 60 70 80 90 80°0'0"E 81°0'0"E 82°0'0"E Case Studies 105 Figure 8.8 Weights are based on relief expenditures obtained from the Sri Lanka Department of Social Services: Drought: 126, Floods 25, Landslides: 0.06, Cyclones: 60. The weighted index has been rescaled to the range 0 to 100. 80°0'0"E 81°0'0"E 82°0'0"E 9°0'0"N 9°0'0"N 8°0'0"N 8°0'0"N 7°0'0"N 7°0'0"N 6°0'0"N 6°0'0"N 10 20 30 40 50 60 70 80 90 80°0'0"E 81°0'0"E 82°0'0"E 106 Natural Disaster Hotspots: A Global Risk Analysis known as ranchos, are constructed of unreinforced are assessed using a livelihood zonation data set that masonry, making them particularly vulnerable to earth- includes populated places. Flood inundation maps asso- quakes. ciated with the river depths for the 1961 and El Niño- Centuries ago, Caracas was purposefully built away related floods in 1997­98 were generated, and impacts from the coast and through steep terrain to deter seaborne assessed for moderate and severe flood event scenar- attacks on the city. However, this removal creates ios. The results are interpreted for use in contingency major transportation and utility infrastructure problems planning and preparedness. that are exacerbated by natural hazards. Caracas is linked The Tana River District in the coast province is divided to the world through its airport and seaport, both of into seven administrative divisions with a total area of which are located across El Ávila on the Vargas coast. 38,694 square kilometers. The topography, drainage The only road between Caracas and the airport and pattern, and soil determine the large extent of the intense seaport is a highway that travels through steep, land- flooding. The district is generally an undulating plain, slide-prone valleys crossing secondary faults that are which slopes southeast with an altitude ranging between part of the active San Sebastian fault system. 0.0 and 200 meters above sea level. The main physical Uncontrolled building and unenforced building geographical feature of this district is the Tana River. and zoning codes in this hazardous environment have The large floodbasin, whose width ranges from 2 to 40 led to human disasters and potential problems of great kilometers, provides fertile arable land and is the eco- magnitude. A lack of building codes and their enforce- nomic backbone of the district. The hinterland has ment allowed Vargas residents to build on active (but seasonal streams (lagas), which provide wet-season graz- quiescent for the previous 50 years) alluvial fans, which ing areas and are sources of inlets for earth pans. Soils reactivated during December 1999. Although various in Tana River district are divided into two groups: groups are working to repair and rebuild Vargas State well-drained sandy soils ranging in color from white to with new housing built in safe locations, poor plan- red, and poorly drained silty and clayey soils that are ning and code enforcement are allowing squatters to gray and black in color. return to the alluvial fans and streambeds where most Nomadic pastoralists who keep large herds of cattle, of the December 1999 destruction was concentrated. goats, and sheep mainly occupy the hinterland. In 1997, Seismic, landslide, and flood hazards affect a high during a three-month period, the district received over proportion of Caracas's urban infrastructure, including 1,200 millimeters of El Niño-related rainfall, triple its housing (Figure 8.9). The case study explores aspects annual average. The resulting floods destroyed many of risk management based on the risks identified. houses, damaged infrastructure, swept away crops, and killed livestock. Flood Risk Assessment for Contingency Planning Garissa district is one of three districts that make up in the Lower Tana River Basin, Kenya the North Eastern Province of Kenya. The total popu- Many Sub-Saharan African countries experience extreme lation of the district is 231,000, according to 1999 census weather conditions leading to severe flood events that population projections. About 40 percent of the pop- require humanitarian assistance. Emergency prepared- ulation resides within Garissa town. The district is pre- ness is a prerequisite for humanitarian response to be dominately inhabited by Somali people who traditionally effective, coordinated, dependable, and timely. A criti- practice livestock keeping. cal factor that has hampered responding agencies in The climate of Garissa is semiarid, and the long-term many countries is lack of information on who will be average rainfall is about 300 mm. Prior to the 1997­98 affected and what impacts are expected. El Niño rains, the greatest rainfall occurred in 1961 and This case study uses a streamflow model and flood 1968, when an average of 920 millimeters was meas- hazard mapping to generate flood scenarios for the lower ured at many stations. Unusually heavy rains in 1997 Tana River Basin, a flood-prone area in Kenya where totaled 1,027 millimeters; 925 millimeters occurred emergency assistance is frequently required (Figure between October and December 1997. This was a 8.10). Flood risks to the population and livelihoods huge amount of rainfall for an area receiving an annual Case Studies 107 Figure 8.9. Multihazard disaster risk, Caracas. Seismic shaking periods Hospitals (seconds) Schools 1.0­0.9 Population clusters 0.8­0.7 Fire stations 0.6­0.5 Police stations 0.4­0.3 Reserved open space 0.2­0.1 Urban outline Mud land slides risk: Major roads high medium low Flooding risk: high medium low Most affected buildings (height) Shaking period average of 300 millimeters. are more likely to have resources to cope and therefore This case study assessed risks of worst-case flood are less at risk of complete collapse of their livelihood. impacts on livelihoods using the El Niño flooding event The population in fisheries and subsistence cropping of 1997­98, with an estimated 35-year return period may find benefits in the floods, thanks to the likely as a scenario. The impact of floods on populations dif- increase in fish production, but they are also likely to fers depending on their livelihoods and wealth group. see their subsistence cropping resources affected. Among the different livelihood groups in both dis- Cash income for the pastoralist community is not tricts, the ones most exposed to flooding are pastoral- diversified at all, as 68 percent of total cash income comes ists, agro-pastoralists, and the dry riverine and Tana from livestock. This is also the case for agro-pastoral- Delta livelihood systems (Figure 8.11). ists, 40 percent of whose income comes from livestock. The livelihood zones directly on the river (dry river- To a lesser extent, people in the dry riverine zone also ine zone, Tana Delta zone, pastoralist, and agro- derive much of their income from livestock (22 percent). pastoralist--the latter mostly located in the hinter- During the El Niño floods of 1997­98, close to 90 per- land, except in the south part of the basin) are likely to cent of sheep and goats died in the Garissa and Tana be impacted by the direct destruction of their proper- River districts, resulting in complete collapse of that ties (such as houses, crop fields, and pumps). The source of income. For these three groups, sheep and population in the urban area (especially at Garissa town) goats represent close to 15 percent of their total income. is likely to be mostly affected through the indirect In addition, for larger animals, which were less directly effect of the floods, such as an interruption of access to affected by floods, mortality and morbidity increased markets and concomitant loss of income, though some dramatically as a result of diseases such as foot rot and may also lose property. However, people in urban areas pneumonia. In addition to the direct loss of animals, 108 Natural Disaster Hotspots: A Global Risk Analysis Figure 8.10. Location Map of Tana River and Garissa Districts with Coverage of Tana River Basin in Garissa District, Kenya GARISSA TANA RIVER Case Studies 109 Figure 8.11. Livelihood Zones Overlaid on El Niño 1997­98 Flood Case (Estimated Return Period 35 years) El Niño Floods Livelihood Zones Fishing & Subsistence Cropping Dry Riverine Zone Tana Delta Zone Pastoral Agro-Pastoral Wildlife Urban 0 25 50 km 110 Natural Disaster Hotspots: A Global Risk Analysis the decrease in livestock marketability also hurt income. (floods being the third)--are explored in case studies Because of the fear of Rift Valley fever, animals were not in this chapter. Work to improve hazard data must bought on the markets and income from animal sales continue. While storm surges are not incorporated was lost. The impact on livestock has hurt equally the directly into the global Hotspots analysis, the case study "very poor," "poor," and "middle" groups, who have seen here illustrates both their damaging potential and the their income from livestock reduced to zero. Most of spatial distribution of risk. Clearly an adequate base the very poor and poor have moved to the destitute cat- exists on which to build for the hazards involved in egory, while only the middle group with larger-sized most major natural disasters. cattle may have avoided destitution. Global data for estimating exposure and vulnerabil- The loss of livestock also had an important impact ity--of population, agriculture, urban areas, and infra- on food consumption. Food intake was reduced because structure--are also sufficiently developed for use in of the loss of animals that had provided meat and milk. assessing disaster risks. Data sets on historical disaster The loss of income also translated into a loss of pur- losses are being explored, improved, and integrated. chasing power, which together with higher commod- Over time, accumulated disaster and loss data provide ity prices put the commodities out of reach for these a dependent variable against which the contributions communities. As a result, their access to food was dras- of independent variable risk factors to spatial and tem- tically reduced. poral variations in losses can be calculated. Under the worst case scenario, pastoralists and dry One intended contribution of the multiscale nature riverine communities are expected to experience the of the hotspots analysis is to demonstrate the transfer- worst losses. Therefore, a response directed toward these ability of disaster causality theory and risk assessment groups, particularly the pastoralists, would be advis- methods between spatial scales. The same general able. Assistance should take the form of free food dis- approach--estimating exposure of elements at risk to tribution and income-generating activities, as the analysis an array of hazards and assessing their degrees of vul- has shown that for all of the groups, income, daily food nerability to the hazards they face--has been employed consumption, and nutrition are tied to livestock and in both the global analysis and the place-based case crop production, both of which may completely col- studies. It is hoped that the durability and rigor of the lapse in any flood scenario. Furthermore, assessments approach will lend itself to a continuing effort to improve during the 1997­98 El Niño floods showed that relief data quality and, more important, to a more systematic food enabled pastoralists to save their remaining live- approach to disaster risk management founded on sci- stock and to start rebuilding herds and livelihoods. entific risk assessment. For planning purposes, we know from our hazard maps The place-based case studies demonstrate how second- that food assistance in the short term and income regen- and third-order risk assessments can be conducted in erating activities in the long term would be required areas that include hotspots identified by the first-order for up to 70,000 persons. In a moderate-case scenario, global analysis. This multitiered approach creates oppor- the population in need would be 47,000. This finding tunities to achieve appropriate data density within provides core data for calculating the volume of food hotspots for a more precise analysis of risks and their commodities required and costs. causal factors. The case studies, both hazard-focused and place- based, also illustrate how single and multiple hazards Linkages to and Lessons for Global Analysis interact with exposed elements and their vulnerabili- ties to create complex patterns of risk. These interac- The global Hotspots analysis is intended as a first- tions among causal factors of disaster are an important order filter to reveal areas of highest disaster risk from topic for continuing research. The case studies suggest one or multiple hazards. To accomplish this, several that multihazard phenomena observed at coarse reso- hazard databases had to be created or substantially lution on global scales may lead to multihazard man- strengthened. Two of these--droughts and landslides agement problems at national and urban scales of analysis Case Studies 111 and decision making. Given resource constraints and dimension is particularly critical for hazards for which the multiple roles played by key infrastructure such as early warning is currently possible: droughts, floods, roads, railroads, and ports in disaster preparedness, cyclones, and volcanoes, as well as some landslides. emergency response, reconstruction, and ongoing eco- Cutting-edge work on hazard event prediction, partic- nomic activity, it is vital that planners and decision ularly if combined with more systematic exposure and makers at all levels understand the hazards prevalent vulnerability monitoring, could improve disaster early in their own regions. In particular, they need to under- warning on a temporal basis, with an emphasis on stand the potential interactions among these hazards, early warning systems in high-risk areas. whether direct (such as storms that trigger both floods As important as the incremental improvement in risk and landslides) or indirect (such as consecutive hazard assessment is fostering and promoting the ability to events that strain response capacities and exacerbate use information about risks effectively for risk reduc- vulnerabilities). tion and transfer. This application of the results of risk The combination of global and national or local-scale identification is explored in the Caracas and Tana analyses based on common theory, methods, and, in River case studies. The underlying rationale for risk some cases, data provides opportunities to integrate risk assessment is that it reveals where investments in risk management strategies at multiple scales toward a reduction are most needed and likely to have the biggest common objective of reducing disaster losses. Through payoff in terms of reduced losses. Much remains to be these analyses, international donors operating at the learned, however, about how to use this type of infor- global scale, for example, can focus attention and mation to best advantage, including the institutions, resources on high-risk regions. National and local author- policies, cost/benefit analyses, mitigation measures, and ities can use similar techniques to formulate proactive resource allocation decisions needed to convert risk and effective risk management plans that target verifi- information to disaster reduction. It is important that able risks transparently and objectively. Global/national this type of work be integrated into efforts to improve partnerships to reduce risks in highest risk areas may risk assessment and vice versa. be the only way for some disaster-prone countries to Quantitative data-based risk assessment, combined stem the tide of disaster losses that impede their social with successful efforts to reduce risks, creates greater and economic development. Focusing on risk man- potential for risk transfer through insurance and other agement rather than disaster relief would greatly ben- mechanisms. These offer the opportunity for popula- efit some countries, cut costs for donors, and free up tions at risk to transfer some of the risk to a wider base resources for promoting positive development. of risk-holders. While disaster risk will never be elim- Another factor that emerged in several of the case inated, an approach that combines risk identification, studies was the importance of temporal variations in reduction, and transfer offers the best possibility of min- risk. Now that the initial global risk assessment is com- imizing losses and repeated and expensive relief and plete and the case studies have demonstrated the use- reconstruction efforts. fulness of the theory to local risk assessment, a temporal dimension can be added. This dimension should be added to monitoring and forecasting risk levels, espe- cially in the highest risk, most disaster-prone areas. This 112 Natural Disaster Hotspots: A Global Risk Analysis Table 8.3. Potential and Actual Hotspots Vulnerable to Flooding by Storm Surge* Surge-Prone Regions Potential and Actual Hotspots Bay of Bengal (Bangladesh and Eastern India) Ganges-Brahmaputra mouth (Bangladesh and West Bengal); Mahandi delta (Orissa); and the Krishna and Godavari deltas (Andhra Pradesh) Western India/Pakistan Indus delta and Karachi (Pakistan); Mumbai (India) China/Japan Lower Liaohe River Plain (China); North China Plain (China); East China Plain and Shanghai (China); Hanjiang River deltiac plain (China); Pearl River deltaic plain; Guangzhou and Hong Kong (China); Guangxi coastal plain (China); North Hainan Plain (China); Taiwan (China) coastal plain; Taipei (Taiwan, China); Metropolitan Toyko (Japan); Metropolitan Osaka (Japan) Korea, Rep. of -- Thailand, Vietnam, Philippines Red River delta (Vietnam); Mekong delta (Vietnam); Metropolitan Manila (Philippines); Chaophraya delta and Bangkok (Thailand) Pacific Islands Most capital cities, all of which are on the coast; all atoll islands Australia and New Zealand -- Indian Ocean Islands -- Eastern Africa and Oman -- Rio de la Plata (Argentina and Uruguay) Buenos Aires (Argentina); Montevideo (Uruguay) Caribbean Most capital cities, all of which are on the coast Central America and Mexico -- United States--Gulf and East coasts New York City; Florida, particularly southern Florida and the Keys; New Orleans Europe--Atlantic coast -- Europe--Mediterranean coast Areas of land claim and high subsidence on the Northern Adriatic Coastal Plain in Italy (Nicholls and Hoozemans 1996) Europe--North Sea coast London and Kingston-upon-Hull (United Kingdom); the western Netherlands; Hamburg and Bremen (Germany) Europe--Baltic Sea coast St. Petersburg (Russia); potentially Helsinki (Finland) and Stockholm (Sweden) * This information is indicative rather than an exhaustive list of potential and actual hotspots. Note: -- none Chapter 9 Conclusions and the Way Forward This project has made a unique attempt to develop a The significance of high mortality and economic loss global, synoptic view of the major natural hazards, assess- risks for socioeconomic development indicated in this ing risks of multiple disaster-related outcomes and focus- analysis extends well beyond the initial direct losses to ing in particular on the degree of overlap between the population and economy during disasters. Covari- areas exposed to multiple hazards. This exploratory ate losses accompanying mortality, for example, include effort has used a range of existing and recently devel- partial or total loss of household assets, lost income, oped data sets to create an initial picture of the loca- and lost productivity. Widespread disaster-related mor- tion and characteristics of hotspots: areas at relatively tality can affect households and communities for high risk from one or more natural hazards. Although years, decades, and even generations. many researchers have justifiably critiqued the quality In addition to mortality and its long-term conse- of these data for detailed quantitative analysis of the quences, both direct and indirect economic losses must risks posed by natural hazards, the data do permit dif- be considered (ECLAC and the World Bank 2003). Direct ferentiation of areas of relatively high hazard from losses are losses of assets, whereas indirect losses are areas of lower hazard. Combining these data across haz- the losses that accrue while productive assets remain ards using simple categorical methods thus enables damaged or destroyed. During disasters, both direct and objective identification of hotspots. indirect losses accumulate across the social, productive, The study also undertook a range of case studies and infrastructure sectors. The pattern of losses depends designed to provide important insights into the Hotspots on the type of hazard and the affected sectors' vulner- analysis, to test the applicability of the approach at abilities to the hazard. In large disasters, cumulative subglobal scales, and to explore the value of under- losses across sectors can have macroeconomic impacts. standing multihazard interactions at subnational scales. Disasters impose costs in addition to human and eco- nomic losses. Additional costs include expenditures for disaster relief and recovery and for rehabilitation The Costs of Disaster Risks and reconstruction of damaged and destroyed assets. In major disasters, meeting these additional costs can The combination of human and economic losses, require external financing or international humanitar- plus the additional costs of relief, rehabilitation, and ian assistance. reconstruction, makes disasters an economic as well Data on relief costs associated with natural disasters as a humanitarian issue. Until vulnerability, and con- are available from the Financial Tracking System (FTS) sequently risks, are reduced, countries with high of the United Nations Office for the Coordination of proportions of population or GDP in hotspots are Humanitarian Affairs (OCHA) for 1992 through 2003. especially likely to incur repeated disaster-related The FTS database contains information on all human- losses and costs. Disaster risks, therefore, deserve itarian aid contributions as reported to OCHA by serious consideration as an issue for sustainable international donors (http://www.reliefweb.int/fts/). development. Total relief costs for 1992 through 2003 are US$2.5 113 114 Natural Disaster Hotspots: A Global Risk Analysis Table 9.1. Countries Receiving High Levels of International Disaster Assistance, 1992 through 2003 Country Earthquakes Floods Storms Drought Volcanoes China X X India X X X Bangladesh X Egypt, Arab Rep. X Mozambique X Turkey X Afghanistan X X El Salvador X Kenya X X Iran, the Islamic Rep. of X Pakistan X X Indonesia X X X Peru X X Congo, Dem. Rep. of X Poland X Vietnam X X Colombia X Venezuela X Tajikistan X X Cambodia X Source: OCHA. billion. Of this, $2 billion went to just 20 countries dardized comprehensive methodology (ECLAC and the (Table 9.1). World Bank 2003). The assessment method allows losses The World Bank provided data for this study on emer- to be disaggregated by sector and into direct asset losses gency loans and reallocation of existing loans to meet as well as into indirect losses due to the loss of pro- disaster reconstruction needs for 1980 through 2003 ductive assets. A look at losses by sector and hazard type (http://www.worldbank.org/hazards). The total emer- for these six disasters clarifies the financial implica- gency lending and loan reallocation for 1980 through tions of future losses for the hotspots and suggests 2003 was US$14.4 billion. Of this, $12 billion went to what the actual losses might have been in thousands of the top 20 countries (Table 9.2). past disasters for which comprehensive assessments High proportions of the population, GDP per unit were not conducted.3 area, or land surface area in the countries listed in Tables Total direct and indirect losses for six major disasters 9.1 and 9.2 fall within areas identified above as high- were obtained from the Economic Commission for Latin risk hotspots. Presumably, as disasters continue to occur, America and the Caribbean (ECLAC) and the World these and other high-risk countries will continue to need Bank. These disasters were earthquakes in Turkey in high levels of humanitarian relief and recovery lending 1999 and in India and El Salvador in 2001; Hurricane unless their vulnerability is reduced. Keith in Belize in 2000; the Mozambique floods in 2000; Disaster relief costs drain development resources from and a drought in Central America in 2001 (Table 9.3). productive investments to support consumption over The total direct and indirect loss for these six disasters short periods. Emergency loans have questionable value alone was US$9.5 billion. Relief costs (OCHA) and recon- as vehicles for long-term investment and contribute to struction loans (World Bank) totaled $487.4 million and country indebtedness without necessarily improving economic growth or reducing poverty. The most significant implications of having large 3Due to the fact that data from comprehensive assessments of numbers of people, GDP, or land surface at risk can be direct and indirect economic losses have not been systematically seen in profiles of economic losses from six illustrative compiled and reported to date, economic loss estimates in EM- disasters in which losses were assessed using a stan- DAT, where they exist, are based on ad hoc reporting. Conclusions and the Way Forward 115 Table 9.2. Countries Receiving Emergency Loans and Reallocation of Existing Loans to Meet Disaster Reconstruction Needs, 1980 through 2003 Country Earthquakes Floods Storms Drought Volcanoes India X X X Turkey X X Bangladesh X X Mexico X X Argentina X Brazil X Poland X Colombia X X Iran, the Islamic Rep. of X Honduras X X China X X Chile X Zimbabwe X Dominican Republic X El Salvador X Algeria X X Ecuador X X Mozambique X X Philippines X Vietnam X Source: World Bank Hazard Management Unit (http://www.worldbank.org/hazards). $1.4 billion, respectively--5 percent and 14 percent, Implications for Decision Making respectively, of the total estimated loss. Neither the OCHA relief costs nor the World Bank The Hotspots analysis has implications for devel- reconstruction loan figures necessarily fully account for opment investment planning, disaster preparedness the total relief and reconstruction expenditures in and loss prevention. The highest risk areas are those these six disasters. Nevertheless, the above figures, where in which disasters are expected to occur most fre- data on all three variables are available, suggest that eco- quently and losses are expected to be highest. This nomic losses across all sectors in disasters may consid- provides a rational basis for prioritizing risk- erably exceed the costs of relief and reconstruction. reduction efforts and highlights areas where risk Thus, the greatest financial implications for the hotspot management is most needed. areas may be with respect to potential future economic For preparedness, identification of high-risk areas losses. provides a basis for contingency planning. The global Hazards are not the cause of disasters. By definition, analysis is appropriate for identifying which types of disasters involve large human or economic losses. Hazard hazards affect which parts of countries and groups of events that occur in unpopulated areas and are not asso- countries. This allows international relief organiza- ciated with losses do not constitute disasters. Losses tions to anticipate what types of problems might are created not only by hazards, therefore, but also by occur, and where, and plan accordingly. the intrinsic characteristics of the exposed infrastruc- For preventing losses, risk identification paves the ture, land uses, and economic activities that cause them way for risk reduction and risk transfer. Currently, to be damaged or destroyed when a hazard strikes. risks are so high in some areas that they are uninsur- This socioeconomic contribution to disaster causality able. Reducing them could create opportunities for at- is potentially a source of disaster reduction. Disaster risk populations or countries to sell part of their risk losses can be reduced by reducing exposure or vulner- instead of bearing it all themselves. ability to the hazards present in a given area. The resolution of the global data is most appropri- ate for only very general types of international-scale 116 Natural Disaster Hotspots: A Global Risk Analysis Table 9.3 Direct and Indirect Losses for Six Major Disasters Social Infrastructure Productive Environment Sectors Sectors Sectors and Other Total Hazard Year Country (106 US$) (106 US$) (106 US$) (106 US$) (106 US$) Earthquake 1999 Turkey (Marmara) 2,187 739 1,850 0 4,776 Earthquake 2001 India (Gujarat) 1,302 334 440 55 2,131 Earthquake 2001 El Salvador 472 398 275 68 1,212 Hurricane 2000 Belize 38 44 165 407 655 Flood 2000 Mozambique 69 133 281 5 488 Drought 2001 Central America 124 3 83 0 210 Total 4,191 1,651 3,095 535 9,472 Sources: ECLAC and the World Bank. decision making, however, and the global map indi- disaster risk in terms of possible tradeoffs with long- cates the need for more localized work with better term socioeconomic goals. data. In particular, more localized work allows greater In high-risk regions and countries, it is particularly specificity in identifying vulnerability factors, which important to protect investments from damage or loss, identify the greatest opportunities for risk reduction. either by limiting hazard exposure or by reducing vul- As the previous chapters have shown, the methods used nerability. Risks of damage and loss should also be taken for assessing risks globally can be used for work at the into account when estimating economic returns during national and local levels. project preparation. Owing to intersectoral interactions, International development organizations are key large-scale covariate losses across multiple sectors can stakeholders with respect to the global analysis. The affect economic performance, even if those losses are analysis provides a scientific basis for understanding concentrated in sectors outside a particular investment where risks are highest and why, as well as a method- project. ological framework for regional- and local-scale analy- The theory of risk used in this report to identify dis- sis. The identified risks then can be evaluated further aster risks at the global scale can be applied to more using more detailed data in the context of a region's or localized areas, as demonstrated by the case studies in country's overall development strategy and priorities. the previous section. Similarly, the general methodol- This would serve development institutions and the coun- ogy of estimating hazard exposure and vulnerabilities tries in several ways to facilitate the development of can be applied to identify various risks more precisely better-informed investment strategies and activities. at national, subnational, and local scales. Such assess- Assistance Strategies. So for example, a develop- ments can then be used to set standards and imple- ment institution such as the World Bank may use the ment vulnerability reduction measures. analysis at the global and/or regional level to identify Sector Investment Operations. Investment proj- countries that are at higher risk for disasters and "flag" ect preparation, particularly in the high-risk areas iden- them as priorities to ensure that disaster risk manage- tified in the global analysis, would benefit from including ment is addressed in the development of a Country Assis- a risk assessment as a standard practice. This report's tance Strategy (CAS) or Poverty Reduction Strategy. theory and methods can be translated easily into terms While in some countries there can be a seemingly of reference for such assessments. Such assessments long list of urgent priorities to address in a CAS--e.g., should identify probable hazards, as well as their spa- reducing extreme poverty, fighting HIV/AIDS, promot- tial distribution and temporal characteristics (includ- ing education, achieving macroeconomic stability-- ing return periods), and should evaluate vulnerabilities managing disaster risk should be considered an integral to the identified hazards that should be addressed in part of the development planning to protect the invest- the project design. ments made, rather than as a stand-alone agenda. The Risk Reduction Operations. In high-risk countries CAS should consider the consequences of unmitigated and areas within countries, repeated, large-scale loss Conclusions and the Way Forward 117 events can harm economic performance (Benson and appraising" emergency loans, that is, designing a risk Clay 2004). It may be impossible to achieve develop- management strategy to guide the allocation of emer- ment goals such as poverty alleviation in these areas gency reconstruction resources should such resources without concerted efforts to reduce recurrent losses. become necessary, or to arrange for other types of con- Increasingly, risk and loss reduction are being seen as tingency financing with development banks. investments in themselves, and disaster-prone coun- The exercise of identifying risks and risk manage- tries are demonstrating a willingness to undertake ment opportunities would have benefits even if emer- projects in which disaster and loss reduction are the gency assistance is never needed, as it would create a principal aims. Such projects can include both hard and road map for reducing disaster risks. If a disaster did soft components: measures to reduce the vulnerability occur, the availability of an "off-the-shelf" recovery pack- and exposure of infrastructure, as well as emergency age would avoid starting the emergency loan appraisal funds, institutional, policy, and capacity-building meas- process from scratch and could identify previously ures designed to increase the abilities of countries to planned risk reduction measures. manage disaster risks. Countries with high disaster risks are candidates for these types of projects. Such countries may already expe- Information Development for Disaster Risk rience frequent disasters and significant losses. They Management may require financial assistance such as periodic and perhaps frequent restructuring of their development The Hotspots project provides a common framework portfolio to meet emergency needs, frequent emergency for improving risk identification and promoting risk borrowing, or both. In such cases, disaster risk reduc- management through a dialogue between organiza- tion projects offer a rational alternative to recurrent, tions and individuals operating at various geographic unplanned emergency spending. scales. The methods and results provide useful tools Contingency Financing. Emergency recovery and for integrating disaster risk management into devel- reconstruction needs after a major disaster may create opment efforts and should be developed further. a high demand for emergency financing. While such There is growing recognition of the need for better loans are usually appraised and approved relatively data and information on hazards and disasters at both quickly, at times there can be delays in disbursing the national and international levels. Within the United funds, which increases the social and economic impacts States, several recent reports by the U.S. National Research of the disaster. Council (NRC) and the U.S. government have high- In urgent disaster situations, there is little time to lighted the importance of both historical and current plan the allocation of resources for cost-effective risk data on hazard events and their associated impacts (NRC management over the longer term. Typically, a token 1999a, 1999b; Subcommittee on Disaster Reduction amount of emergency funding is earmarked for "disas- 2003). At the international level, there is strong inter- ter preparedness," but risk reduction is not necessarily est in improvement in disaster information systems and woven into the fabric of the reconstruction effort. At associated decision support tools (for example, ISDR worst, hastily planned reconstruction can simply 2003). result in rebuilding the same risks that led to the dis- A welcome shift in emphasis appears to be under aster in the first place. way from managing disasters by managing emergencies Advance planning for recovery and resource alloca- to managing disaster risks. This shift is evident in recent tion would allow for better targeting of resources toward publications such as the 2002 World Disasters Report: investments that would restore economic activity quickly Focus on Reducing Risk (IFRC 2002), Living with Risk (ISDR and relieve human suffering. This report's global disas- 2004), and Reducing Disaster Risk: A Challenge for Devel- ter risk analysis provides a basis for identifying situa- opment (UNDP 2004). Risk assessment, reduction, and tions in which future emergency recovery loans are likely transfer are the major elements of risk management to be needed. This creates an opportunity for "pre- (Kreimer and others 1999), offering a desirable alterna- 118 Natural Disaster Hotspots: A Global Risk Analysis tive to managing disasters through emergency response. hotspots approach to analysis and decision making at Risk reduction requires risk assessment in order to deter- regional, national, and local scales. The initial case stud- mine which areas are at highest risk of disaster and why, ies are promising, but they are certainly not on their so that appropriate and cost-effective mitigation meas- own sufficient to demonstrate the value of the overall ures can be identified, adapted, and implemented. approach or the specific data and methods under dif- We have designed the hotspots approach to be open- ferent conditions. More direct involvement of poten- ended to allow additional studies to be incorporated on tial stakeholders would be valuable in extending the an ongoing basis. As a global analysis conducted with approach to finer scales of analysis and decision making. very limited local-level participation and based on incom- To be effective, efforts to improve risk identification in plete data, the results presented here should not provide hotspot areas should be part of a complete package of the sole basis for designing risk management activities. technical and financial support for the full range of meas- The analysis does, however, provide a scientific basis for ures needed to manage risks, including risk reduction understanding where risks are highest and why, as well and transfer. as a methodological framework for regional- and local- Explore Long-term Trends. A third direction is to scale analysis. The identified risks then can be evaluated explore a key long-term issue: the potential effect of furtherusingmoredetaileddatainthecontextofaregion's underlying changes in hazard frequency (for example, or country's overall development strategy and priorities. due to human-induced climatic change) coupled with The hotspots analysis can be improved upon as a tool long-term trends in human development and settlement and developed in several directions. patterns. To what degree could changes in tropical storm Improve Underlying Databases. The first direction frequency, intensity, and tracks interact with contin- is to pursue the many opportunities in both the short ued coastal development (both urban and rural) to and long term to improve the underlying databases for increase risks of death and destruction in these regions? assessing disaster risks and losses. A range of new global- Are agricultural areas, already under pressure from scale data sets is currently under development, includ- urbanization and other land use changes, likely to become ing a new global urban-extent database being developed more or less susceptible to drought, severe weather, or by CIESIN in support of the Millennium Ecosystem floods? Could other hazards such as wildfires poten- Assessment. A joint project between the Earth Institute, tially interact with changing patterns of drought, land- the World Bank, and the Millennium Project will develop slides, deforestation, and land use to create new types a much more detailed and complete database on sub- of hotspots? Although some aspects of these questions national poverty and hunger. Much more comprehen- have been addressed in the general context of research sive regional data sets will become available in specific on climate change impacts, the interactions between areas of interest. On a regional scale, there are also climate change, the full range of hazards, and evolving much longer records of hazard events for specific haz- human hazard vulnerability have not been fully explored ards that could be harnessed to improve estimates of (for example, Brooks and Adger 2003; Chen 1994). hazard frequency and intensity in high-risk areas (for Pursuing work in these directions will necessarily example, O'Loughlin and Lander 2003). Significant involve a wide range of institutions--national, regional improvements could be made in characterizing flood, and international, public and private sector, academic, drought, and landslide hazards in particular. Existing and operational. We hope that the Hotspots project data on disaster-related losses are being compiled into has contributed a building block in the foundation of a multitiered system through which regularly updated a global effort to reduce disaster-related losses by man- historical data from multiple sources can be accessed. aging risks rather than by managing emergencies. We Additional work to link and cross-check existing data look forward to continuing collaboration with part- is needed, however, as is improvement in the assessment ners at all levels to put in place a global disaster risk and documentation of global economic losses. management support system in order to mobilize the Undertake Case Studies. A second direction is to knowledge and resources necessary to achieve this goal. explore more fully the applicability and utility of the Appendix A: Technical Appendix for Global Analysis A.1 Derivation of Tropical Cyclone, GDP Global GDP Surface Surfaces, and Agricultural Value Data Sources GDP: CIESIN: http://sedac.ciesin.columbia.edu/plue/gpw/ Tropical Cyclones index.html?main.html&2 Data Sources: CIA Factbook: http://www.cia.gov/cia/publications/factbook/ UNEP: http://www.grid.unep.ch/data/grid/gnv200.php World Development Indicators: http://www.worldbank.org/ Wind speed profile model: http://www.bbsr.edu/rpi/ data/wdi2000/ meetpart/land/holland2.html All available subnational GDP data (preferably purchasing Tracks of tropical cyclones are available from several differ- power parity [PPP] corrected, 2000) were collected for the ent centers. UNEP assembled a data set of all the wind storms world. Since these data come from different sources, the that occurred from 1981 through 2000. Wind speed profile subnational data were used only to calculate the share of the models are used to delineate buffer zones around the track national GDP of a subnational unit. The World Bank Indi- of a tropical cyclone. Hence the buffer zones represent the cators were used as a uniform data source for the national, areas affected by the tropical cyclone. After evaluating dif- PPP adjusted GDP in 2000. ferent wind speed profile models, the model of Greg Hol- The reallocation of the GDP to a subnational unit was land was chosen (Holland 1997). This model was also used based on the population distribution in that unit. The pop- by UNEP to generate asymmetric wind speed profiles. ulation in 2000 was projected based on CIESIN population The different wind speed buffers are translated into six database for 1995. These population numbers were adjusted categories using the Saffir Simpson Hurricane scale. The globe at national level using the U.N. population numbers for the was divided into small cells of one square kilometer. Over- year 2000. The result is a global dollar value surface with a laying the translated wind speed buffer zones of the storms five-kilometer cell resolution based on the GDP. that occurred in a given year with the one-kilometer resolu- tion grid results in a surface that represents how often a grid Agricultural Value cell is hit and the associated wind speed category. A combi- nation of the yearly surfaces gives a global tropical cyclone Data Sources: frequency grid for the available data during 1981 through GLCCD: http://edcdaac.usgs.gov/glcc/glcc.asp 2000. IFPRI: http://www.ifpri.org/pubs/books/page/agroeco_use.pdf FAOSTATS: http://apps.fao.org/page/collections?subset= Table A1.1. Available Tropical Cyclone Data by Region agriculture Region Data Availability Atlantic Ocean 1981­2000 The International Food Policy Research Institute (IFPRI) rein- Australia 1984­2000 terpreted the Global Land Cover Characteristics Dataset of Indian Ocean, North 1992­1997, 1999­2000 1998 focusing on the agricultural area in every cell of the global Indian Ocean, South 1981­2000 one-kilometer resolution grid. A cell value represents the Pacific Ocean, Northeast 1988­2000 Pacific Ocean, Northwest 1981­2000 percentage of area in that cell used for agricultural purposes. Pacific Ocean, South 1981­2000 The total agricultural area of a country was calculated. 119 120 Natural Disaster Hotspots: A Global Risk Analysis Table A1.2. Subnational GDP Data Argentina Chile United Kingdom Korea, Rep. of Romania Australia China Greece Lithuania Russian Fed. Austria Colombia Hungary Latvia Slovak Rep. Belgium Czech Rep. Indonesia Mexico Slovenia Bangladesh Germany India Mozambique Sweden Bulgaria Denmark Ireland Netherlands Thailand Bolivia Spain Iran, the Islamic Rep. of Peru Turkey Brazil Estonia Italy Philippines United States Canada Finland Japan Poland Vietnam France Kazakhstan Portugal South Africa Switzerland Production numbers for a specific crop in a country in 2000 the top three deciles, green the next three deciles, and blue the come from the FAO Statistical Database (FAOSTATS). Multi- bottom four deciles). plying these production numbers by IFPRI price per unit from For cyclones, there is a slight expansion of high-hazard 1989 through 1991 and summing them gives the total agricul- areas in the Caribbean and the southeastern United States, in tural value of a country. This value was redistributed spatially coastal areas of China, and in southern Africa. A few small areas over the agricultural area in a country to generate the agricul- in eastern India and Bangladesh also move into the highest tural value surface. three deciles under the influence of high population density. Flood areas also increase in most regions, at the expense of some relatively less densely populated areas along the U.S. Gulf A.2 Reclassification of Hazardous Areas Coast. Drought areas no longer considered as significant in terms Weighted by Exposure of population density include parts of the western United States, interior South America, Kazakhstan, and southern Australia. To characterize the effect of population density on the global dis- Only minor changes are evident for volcanoes. tribution of hazard, we divide our grid cells into deciles based A more noticeable difference is evident for the earthquake on population density and use the resulting index (1­10) to data, which demonstrate a clear shift towards more densely pop- weight each hazard individually. A grid cell with a drought decile ulated areas of the northeast United States, Europe, the coast of value of 8 might therefore have a drought-population index rang- West Africa, India, China, and the Koreas. Less populated ing from 8 to 80. Figures A2.1a­f illustrate the results for each areas, such as western South America, Indonesia, and New hazard, using the same grouping of deciles for the population- Zealand, rank lower on this scale. weighted indexes as for the hazard-only indexes (red indicates Appendix A Figure A2.1. Single-Hazard Exposure Index Based on Top Three Population-Weighted Deciles a) Cyclones Cyclone Hazard Exposure Population Weighted Deciles 1st ­ 4th 5th ­ 7th 8th ­ 10th 121 122 Figure A2.1. Single-Hazard Exposure Index Based on Top Three Population-Weighted Deciles b) Drought Natural Drought Hazard Exposure Disaster Population Weighted Deciles 1st ­ 4th Hotspots: 5th ­ 7th 8th ­ 10th A Global Risk Analysis Appendix A Figure A2.1. Single-Hazard Exposure Index Based on Top Three Population-Weighted Deciles. c) Floods Flood Hazard Exposure Population Weighted Deciles 1st ­ 4th 5th ­ 7th 8th ­ 10th 123 124 Figure A2.1. Single-Hazard Exposure Index Based on Top Three Population-Weighted Deciles d) Earthquakes (pga) Natural Earthquake (PGA) Hazard Exposure Disaster Population Weighted Deciles 1st ­ 4th Hotspots: 5th ­ 7th 8th ­ 10th A Global Risk Analysis Appendix A Figure A2.1. Single-Hazard Exposure Index Based on Top Three Population-Weighted Deciles e) Volcanoes Volcano Hazard Exposure Population Weighted Deciles 1st ­ 4th 5th ­ 7th 8th ­ 10th 125 126 Figure A2.1. Single-Hazard Exposure Index Based on Top Three Population-Weighted Deciles f) Landslides Natural Landslide Hazard Exposure Disaster Population Weighted Deciles 1st ­ 4th Hotspots: 5th ­ 7th 8th ­ 10th A Global Risk Analysis Appendix A 127 A.3 World Bank Country Income Classifications Table A3.1. World Bank Country Income Classifications: High Income European Monetary Union (12) OECD (24) Others, non-OECD (28) Country 2000 GDP (Millions of Country 2000 GDP (Millions Country 2000 GDP (Millions Current US$) of Current US$) of Current US$) Austria 202,954 Australia 410,590 Andorra .. Belgium 247,634 Austria 202,954 Aruba .. Finland 130,797 Belgium 247,634 Bahamas, The .. France 1,409,604 Canada 715,692 Bahrain Germany 1,976,240 Denmark 174,798 Bermuda .. Greece 132,834 Finland 130,797 Brunei .. Ireland 119,916 France 1,409,604 Cayman Islands .. Italy 1,180,921 Germany 1,976,240 Channel Islands Luxembourg 20,062 Greece 132,834 Cyprus .. Netherlands 413,741 Iceland 8,608 Faeroe Islands Portugal 121,291 Ireland 119,916 French Polynesia .. Spain 649,792 Italy 1,180,921 Greenland .. Japan 3,978,782 Guam Korea, Rep. of 476,690 Hong Kong, China 161,532 Luxembourg 20,062 Israel .. Netherlands 413,741 Kuwait .. New Zealand 58,178 Liechtenstein Norway 189,436 Macao, China Portugal 121,291 Monaco .. Spain 649,792 Netherlands Antilles .. Sweden 229,772 New Caledonia .. Switzerland 268,041 Northern Mariana Islands United Kingdom 1,552,437 Qatar .. United States 10,416,820 San Marino Singapore 86,969 Slovenia 21,108 United Arab Emirates .. Virgin Islands (U.S.) .. 128 Natural Disaster Hotspots: A Global Risk Analysis Table A3.2. World Bank Country Income Classifications: Low and Middle Income Low Income (65) Lower Middle Income (52) Upper Middle Income (38) Country 2000 GDP (Millions of Country 2000 GDP (Millions Country 2000 GDP (Millions Current US$) of Current US$) of Current US$) Afghanistan .. Albania 4,695 American Samoa - Angola 11,380 Algeria 55,666 Antigua and Barbuda 710 Armenia 2,367 Belarus 14,304 Argentina 102,191 Azerbaijan 6,090 Belize 843 Barbados .. Bangladesh 47,328 Bolivia 7,678 Botswana 5,188 Benin 2,690 Bosnia and Herzegovina 5,249 Brazil 452,387 Bhutan 594 Bulgaria 15,608 Chile 64,154 Burkina Faso 2,839 Cape Verde 631 Costa Rica 16,887 Burundi 719 China 1,237,145 Croatia 22,421 Cambodia 3,677 Colombia 82,194 Czech Rep. 69,590 Cameroon 9,060 Cuba .. Dominica 254 Central African Rep. 1,075 Djibouti 597 Estonia 6,413 Chad 1,935 Dominican Rep. 21,285 Gabon 4,971 Comoros 256 Ecuador 24,347 Grenada 414 Congo, Dem. Rep. of 5,704 Egypt, Arab Rep. 89,845 Hungary 65,843 Congo, Rep. 3,014 El Salvador 14,287 Isle of Man - Côte d'Ivoire 11,717 Fiji 1,878 Latvia 8,406 Equatorial Guinea 2,173 Guatemala 23,252 Lebanon 17,294 Eritrea 582 Guyana 710 Libya .. Ethiopia 5,989 Honduras 6,594 Lithuania 13,796 Gambia, The 388 Iran, the Islamic Rep. of 107,522 Malaysia 95,157 Georgia 3,324 Iraq .. Malta .. Ghana 6,021 Jamaica 8,001 Mauritius 4,532 Guinea 3,174 Jordan 9,296 Mayote - Guinea-Bissau 216 Kazakhstan 24,205 Mexico 637,205 Haiti 3,590 Kiribati 44 Oman 20,073 India 515,012 Macedonia, FYR 3,712 Palau 130 Indonesia 172,911 Maldives 618 Panama 12,296 Kenya 12,140 Marshall Islands 108 Poland 187,680 Korea, Dem. People's Rep. of .. Micronesia, Fed. Sts. 232 Puerto Rico .. Kyrgyz Rep. 1,632 Morocco 37,263 Saudi Arabia .. Lao People's Dem. Rep. of 1,680 Namibia 2,793 Seychelles 630 Lesotho 730 Paraguay 5,389 Slovak Rep. 23,700 Liberia 564 Peru 56,901 St. Kitts and Nevis 340 Madagascar 4,514 Philippines 77,076 St. Lucia 660 Malawi 1,880 Romania 44,428 Trinidad and Tobago 9,372 Mali 3,163 Russian Federation 346,520 Uruguay 12,325 Mauritania 983 Samoa 261 Venezuela 94,340 Moldova 1,621 South Africa 104,235 Mongolia 1,262 Sri Lanka 16,373 Mozambique 3,920 St. Vincent and Grenadines 361 Nepal 5,493 Suriname 895 Nicaragua .. Swaziland 1,177 Niger 2,170 Syrian Arab Rep. 21,872 Nigeria 43,540 Thailand 126,407 Pakistan 60,521 Tonga 136 Papua New Guinea 2,793 Tunisia 21,169 Rwanda 1,736 Turkey 182,848 São Tomé and Principe 50 Turkmenistan 7,672 Senegal 4,940 Vanuatu 234 Sierra Leone 789 West Bank and Gaza 3,015 Solomon Islands 240 Yugoslavia, Fed. Rep. 15,555 Somalia .. (Serbia/Montenegro) continued Appendix A 129 Table A3.2. continued Low Income (65) Lower Middle Income (52) Upper Middle Income (38) Country 2000 GDP (Millions of Country 2000 GDP (Millions Country 2000 GDP (Millions Current US$) of Current US$) of Current US$) Sudan 13,490 Tajikistan 1,208 Tanzania 9,383 Timor-Leste 388 Togo 1,384 Uganda 5,866 Ukraine 41,380 Uzbekistan 9,713 Vietnam 35,110 Yemen, Rep. of 10,395 Zambia 3,683 Zimbabwe 8,304 References Advanced National Seismic System. 1997. Composite Earthquake Catalog. Available at http://quake.geo. berkeley.edu/anss/ (accessed December 2003). Benson, C., and E. J. Clay. 2004. Understanding the Economic and Financial Impacts of Natural Disasters. Dis- aster Risk Management Series No. 4. Washington, DC: The World Bank. 134 pp. Available at http://www. wds.worldbank.org/servlet/WDS_IBank_Servlet?pcont=details&eid=000012009_20040420135752. Brooks, N., and W. N. Adger. 2003. Country level risk measures of climate-related natural disasters and implications for adaptation to climate change. Working Paper 26, Tyndall Center for Climate Change Research. Norwich, UK: University of East Anglia. 25 pp. Burton, I., R. W. Kates, and G. F. White. 1993. The Environment as Hazard. 2d Ed. New York: Guilford Press. Center for International Earth Science Information Network (CIESIN), Columbia University; International Food Policy Research Institute (IFPRI); and World Resources Institute (WRI). 2000. Gridded Population of the World (GPW), Version 2. Palisades, NY: CIESIN, Columbia University. Available at http://sedac.ciesin. columbia.edu/plue/gpw. Center for International Earth Science Information Network (CIESIN), Columbia University; International Center for Research on Tropical Agriculture (CIAT). 2004. Gridded Population of the World (GPW), Ver- sion 3 (beta). Palisades, NY: CIESIN, Columbia University. Available at http://beta.sedac.ciesin.columbia. edu/gpw. Central Intelligence Agency Factbook. 2004. Washington, D.C. Available at http://www.cia.gov/cia/ publications/factbook Chen, R. S. 1994. The human dimension of vulnerability. In Industrial Ecology and Global Change, R. Socolow, C. Andrews, F. Berkhout, and V. Thomas, eds. Cambridge, UK: Cambridge University Press. pp. 85­105. Coburn, A. W., R. J. S. Spence, and A. Pomonis. 1994. Vulnerability and Risk Assessment. 2d Ed. Disaster Man- agement Training Programme. New York: United Nations Development Programme. 69 pp. Economic Commission for Latin America and the Caribbean (ECLAC) and the World Bank. 2003. Handbook for Estimating the Socio-Economic and Environmental Effects of Disasters. Mexico City and Washington, DC: ECLAC and the World Bank. Gaffin, S. R., C. Rosenzweig, X. Xing, and G. Yetman. 2004. Downscaling and geo-spatial gridding of socio- economic projections from the IPCC Special Report on Emissions Scenarios (SRES). Global Environmental Change 14(2): 105­123. Holland, G. 1997. Horizontal wind structure. Paper presented at the Workshop on Windfield Dynamics of Landfalling Tropical Cyclones, 28­30 May 1997. Available at http://www.bbsr.edu/rpi/meetpart/land/ holland2.html (accessed 30 August 2004). Hubert-Ferrari, A., A. Barka, E. Jacques, S. S. Nalbant, B. Meyer, R. Armijo, P. Tapponnier, and J. P. King. 2000. Seismic hazard in the Marmara Sea region following the 17 August 1999 Izmit earthquake. Nature 404: 269. International Federation of Red Cross and Red Crescent Societies (IFRC). 2002. World Disaster Report, Focus on Reducing Risk. Geneva: Switzerland: International Federation of Red Cross and Red Crescent Societies, 239 pp. 130 References 131 International Strategy for Disaster Reduction (ISDR). 2001. Natural disasters and sustainable development: Understanding the links between development, environment, and natural disasters. Background Paper No. 5, Commission on Sustainable Development, Second Preparatory Session, 28 January­8 February 2002 (DESA/DSD/PC2/BP5), 10 pp. ------. 2002. Living with Risk: A Global Review of Disaster Reduction Initiatives. Preliminary Version. Geneva, Switzerland: Inter-Agency Secretariat of the International Strategy for Disaster Reduction. 382 pp. Avail- able at http://www.unisdr.org/eng/about_isdr/bd-lwr-eng.htm. ------. 2003. Information systems and disaster risk reduction. Background paper contributed to the World Summit on the Information Society. Available at http://www.unisdr.org/news/WSIS/WSIS.pdf. ------. 2004. Living with Risk: A Global Review of Disaster Reduction Initiatives. 2004 Version. Volume 1. Geneva, Switzerland: Inter-Agency Secretariat of the International Strategy for Disaster Reduction. 454 pp. Kreimer, A., M. Arnold, C. Barham, P. Freeman, R. Gilbert, F. Krimgold, R. Lester, J. D. Pollner, and T. Vogt. 1999. Managing Disaster Risk in Mexico, Market Incentives for Mitigation Investment. Washington, DC: The World Bank. Mileti, D. S. 1999. Disasters by Design: A Reassessment of Natural Hazards in the United States. Washington, DC: Joseph Henry Press. 351 pp. Mimura, N. 2000. Distribution of vulnerability and adaptation in the Asia and Pacific region. Asia-Pacific workshop. Available at www.survas.mdx.ac.uk. Murty, T. S. 1984. Storm surges--meteorological ocean tides. Canadian Journal of Fisheries and Aquatic Sci- ences (Bulletin 12). 897pp. Nadim, F., and others. 2004. Global Landslide and Avalanche Risk Hotspots. Washington, DC: The World Bank. Nadim, F., and O. Kjekstad. 2005. Assessment of global landslide hazard hotspots. Proceedings of the 16th International Conference on Soil Mechanics and Geotechnical Engineering, Osaka, Japan. http://www.icsmge2005. org/02/02lect.shtml National Geophysical Data Center database. http://www.ngdc.noaa.gov/seg/hazard/hazards.shtml National Research Council (NRC). 1999a. Reducing Disaster Losses through Better Information. Washington, DC: National Academy Press. 61 pp. ------. 1999b. The Impacts of Natural Disasters: A Framework for Loss Estimation. Washington, DC: National Academy Press. 68 pp. Nghiem, S., D. Balk, C. Small, U. Deichmann, A. Wannebo, R. Blom, P. Sutton, G. Yetman, R. Chen, E. Rodriguez, B. Houshmand, and G. Neumann. 2002. White Paper, Global Infrastructure: The Potential of SRTM Data to Break New Ground. Available at http://www.ciesin.columbia.edu/whitepaperfinal.pdf. Nicholls, R.J. and Hoozemans, F.M.J., 1996: The Mediterranean: Vulnerability to coastal implications of cli- mate change. Ocean and Coastal Manage, 31:105-132. Norwegian Geotechnical Institute (NGI). 2004. First-Order Identification of Global Slide and Avalanche Hotspots. Report No. 20021613-1 (31 March). Oslo: NGI. O'Loughlin, K. F., and J. F. Lander. 2003. Caribbean Tsunamis: A 500-Year History from 1498-1998. Doredrecht, The Netherlands: Kluwer Academic Publishers. 263 pp. Papageorgiou, A. S., and J. Kim. 1991. Study of the propagation and amplification of seismic waves in Cara- cas Valley with reference to the 29 July 1967 earthquake: SH waves. Bulletin of the Seismological Society of America. 81(6): 2214­2233. Parsons, T., S. Toda, R. S. Stein, A. A. Barka, and J. H. Dieterich. 2000. Heightened odds of large earthquakes near Istanbul: An interaction-based probability calculation. Science 288: 661­665. Risk Management Solutions. 2004. RMS Catastrophe Maps (online). Available at http://www.rms.com/ Publications/Maps.asp (accessed 17 August 2004). Sachs, J. D., A. D. Mellinger, and J. L. Gallup. 2001. The geography of poverty and wealth. Scientific Ameri- can 284(3): 70­75. Sapir, D. G., and C. Misson. 1992. The development of a database on disasters. Disasters 16(1): 74­80. 132 Natural Disaster Hotspots: A Global Risk Analysis Schneider, S. H., and R. S. Chen. 1980. Carbon dioxide warming and coastline flooding: Physical factors and climatic impact. Annual Review of Energy 5: 107­149 Simkin, T., and Siebert, L. 1994. Volcanoes of the World, 2nd edition. Tucson, AZ: Geoscience Press in associ- ation with the Smithsonian Institution Global Volcanism Program, 368 p Subcommittee on Disaster Reduction (SDR). 2003. Reducing Disaster Vulnerability through Science and Tech- nology. Washington, DC: National Science and Technology Council, Committee on the Environment and Natural Resources. 42 pp. Available at http://sdr.gov/SDR_Report_ReducingDisasterVulnerability2003.pdf. Tobler, W., U. Deichmann, J. Gottsegen, and K. Maloy. 1995. The Global Demography Project. Technical Report 95-6. National Center for Geographic Information and Analysis. 75 pp. + diskette. Available at http://www.ncgia.ucsb.edu/Publications/Tech_Reports/95/95-6.pdf. United Nations Development Programme (UNDP). 2004. Reducing Disaster Risk: A Challenge for Development. New York: United Nations Development Programme, Bureau for Crisis Prevention and Recovery. 146 pp. United Nations Disaster Relief Organization (UNDRO). 1979. Natural Disasters and Vulnerability Analysis. Report of Expert Group Meeting. Geneva, Switzerland: Office of the United Nations Disaster Relief Coordi- nator. United Nations Environment Programme: http://www.grid.unep.ch/data/grid/gnv200.php United Nations Population Division. 2004. World Urbanization Prospects: The 2003 Revision. Data Tables and Highlights. ESA/P/WP.190. New York: United Nations. 195 pp. Available at http://www.un.org/esa/ population/publications/wup2003/2003WUPHighlights.pdf. Warrick, R. A. and Ahmad, Q. K., eds. 1996. The Implications of Climate and Sea-Level Change for Bangladesh. Dordrecht, The Netherlands: Kluwer Academic Publishers. 415 pp. White, Gilbert F. and J. Eugene Haas. 1975. Assessment of Research on Natural Disasters, Cambridge, Mass.: MIT Press. World Bank. 2000. World Development Indicators. Available from http://www.worldbank.org/data/wdi2000/. The Natural Disaster Hotspots report is a path-breaking effort and a wonderful scientific accomplishment. I m certain that it will prove to be a crucial tool and will stimulate further research in the area. Applying risk analysis to disasters such as earthquakes, drought, and other natural hazards using rigorous science will have huge benefits for policymakers and for the world. Jeffrey Sachs Director, The Earth Institute at Columbia University The tragic impacts of the earthquake and tsunami that occurred on December 26, 2004, threw many around the world into a state of disbelief. As shocking as the tsunami disaster was, however, it s important to remember that the events of this magnitude have happened in the past and they will happen again. In 1984, persistent droughts in Ethiopia and Sudan killed 450,000 people. In Bangladesh in 1991, nearly 150,000 lives were taken by a cyclone. Hundreds of natural disasters, both large and small, occur each year. While the largest capture the attention of the global media, there are many more events that we don t hear about. The cumulative effect of these smaller and medium-sized disasters have equally devastating impacts on developing countries: loss of development gains, torn communities, and increased impoverishment. The poor in these countries are consistently the most severely affected. Natural Disaster Hotspots presents a global view of major natural disaster risk hotspots areas at relatively high risk of loss from one or more natural hazards. It analyzes the location and characteristics of hotspots for six hazards earthquakes, volcanoes, landslides, floods, drought, and cyclones. Data on these hazards are combined with state-of-the-art data on the subnational distribution of population and economic output and past disaster losses to identify areas at relatively high risk from one or more hazards. Areas at risk from different hazards are depicted in nearly 60 full-color global maps. The book also summarizes a range of case studies designed to complement the information provided by the global analysis. They highlight practical examples at the local level to help the reader understand multihazard interactions in the context of policy decisions, investment planning, and prioritization of areas where risk reduction is necessary. THE WORLD BANK TMxHSKIMBy359303zv,:&:*:.:! 1818 H Street, N.W. Washington, D.C. 20433 U.S.A. Telephone: 202-473-1000 Internet: www.worldbank.org E-mail: feedback@worldbank.org ISBN 0-8213-5930-4