37090 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 . 6 Natural Disaster Hotspots Case Studies 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 5 Natural Disaster Hotspots: A Global Risk Analysis Natural Disaster Hotspots Case Studies Disaster Risk Management Series Natural Disaster Hotspots Case Studies Edited by Margaret Arnold1 Robert S. Chen2 Uwe Deichmann3 Maxx Dilley4 Arthur L. Lerner-Lam5 Randolph E. Pullen6 Zoe Trohanis7 The World Bank Hazard Management Unit 2006 Washington, D.C. 1Hazard Risk Management Team, World Bank 2Center for International Earth Science Information Network (CIESIN), Columbia University 3Development Economics Research Group, World Bank 4United Nations Development Programme 5Center for Hazards and Risk Research (CHRR) and Lamont-Doherty Earth Observatory (LDEO), Columbia University 6Center for Inernational Earth Science Invormation Network (CIESIN), and Center for Hazards and Risk Research (CHRR), Columbia University 7Hazard Risk Management Team, World Bank ©2006 The International Bank for Reconstruction and Development / The World Bank 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 5 09 08 07 06 This volume is a product of the staff of the International Bank for Reconstruction and Development / The World Bank. The findings, interpretations, and conclusions expressed in this volume do not necessarily reflect the views of the Executive Directors of The World Bank or the governments they represent. 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 judgement 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 publication is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of applicable law. The International Bank for Reconstruction and Development / The World Bank encourages dissemination of its work and will normally grant permission to reproduce portions of the work 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; Internet: www.copyright.com. All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, The World Bank, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2422; e-mail: pubrights@worldbank.org. ISBN-10: 0-8213-6332-8 ISBN-13: 978-0-8213-6332-4 eISBN-10: 0-8213-6333-6 eISBN-13: 978-0-8213-6333-1 DOI: 10.1596/978-0-8213-6332-4 Cover photo: Margaret Arnold Library of Congress Cataloging-in-Publication Data Natural disaster risk hotspots case studies / edited by Margaret Arnold ... [et al.]. p. cm. -- (Disaster risk management series) Includes bibliographical references. ISBN 0-8213-6332-8 1. Natural disasters--Risk assessment. I. Arnold, Margaret, 1965- II. Series. BG5014.N35 2006 363.34--dc22 2005050448 Contents Preface xi Introduction xiii 1. Drought Disaster in Asia 1 Mathew Barlow, Heidi Cullen, Brad Lyon, and Olga Wilhelmi 2. Global Landslides Risk Case Study 21 Farrokh Nadim, Oddvar Kjekstad, Ulrik Domaas, Ramez Rafat, and Pascal Peduzzi 3. Storm Surges in Coastal Areas 79 Robert J. Nicholls 4. Natural Disaster Risks in Sri Lanka: Mapping Hazards and Risk Hotspots 109 Lareef Zubair and Vidhura Ralapanawe, Upamala Tennakoon, Zeenas Yahiya, and Ruvini Perera 5. Multihazard Risks in Caracas, República Bolivariana de Venezuela 137 Kristina R. Czuchlewski, Klaus H. Jacob, Arthur L. Lerner-Lam, Kevin Vranes, and Students of the Urban Planning Studio: "Disaster Resilient Caracas" 6. Reducing the Impacts of Floods through Early Warning and Preparedness: 165 A Pilot Study for Kenya Hussein Gadain, Nicolas Bidault, Linda Stephen, Ben Watkins, Maxx Dilley, and Nancy Mutunga Tables Table 2.1 Description of variables 25 Table 2.2. Classification of slope factor "Sr" for evaluation of susceptibility 26 Table 2.3. Classification of lithology factor "Sl" for evaluation of susceptibility 27 Table 2.4. Classification of soil moisture factor "Sh" for evaluation of susceptibility 28 Table 2.5. Classification of precipitation trigger indicator "Tp" 28 Table 2.6. Classification of seismicity trigger indicator "Ts" 31 Table 2.7. Classification of landslide hazard potential based on the computed hazard index originally suggested by Mora and Vahrson (1994) 31 Table 2.8. Classification of landslide hazard potential based on the computed hazard index used in this study 31 Table 2.9. Classification of slope factor "Sr" for snow avalanche susceptibility 31 Table 2.10. Classification of precipitation factor "Tp" for avalanche hazard evaluation 38 v vi Natural Disaster Hotspots Case Studies Table 2.11. Classification of temperature factor "Tt" for avalanche hazard analysis 38 Table 2.12. Classification of snow avalanche hazard potential 38 Table 2.13. Annual frequency of occurrence and typical return period (in years) for different classes of landslide and avalanche hazard 39 Table 2.A.1. Classes of frequencies 66 Table 2.A.2. Vulnerability indicators 72 Table 2.A.3. Exponent and p-value for landslide multiple regression 74 Table 2.A.4. Other exponents and p-values for landslide multiple regression 75 Table 3.1. Hurricane characteristics and indicative surge magnitudes based on the Saffir-Simpson scale 81 Table 3.2. Some major coastal cities and human-induced subsidence during the 20th century 82 Table 3.3. Generic approaches to hazard reduction based on purposeful adjustment. 85 Table 3.4. Regional contributions to coastal flooding in 1990 and the 2020s based on the analysis of Nicholls (2004). 87 Table 3.5. The range of scenarios used by Nicholls (2004) 88 Table 3.6. Estimates of the global exposure and incidence of flooding under the four SRES scenarios in the 2080s, plus 1990 estimates as a reference 88 Table 3.7. Global-mean sea-level rise scenarios (cm) used by Nicholls (2004) (referenced to 1990), including the IS92a GGa1 scenario as a reference 90 Table 3.8. The SRES Socioeconomic Scenarios for the 2080s: A Global Summary 90 Table 3.9. Deaths associated with major hurricanes, cyclones, and typhoons (MC) and extra- tropical storm (ETS) disasters (>1,000 deaths) since 1700. 92 Table 3.10. Deaths in storm surges around the North Sea from the 11th to the 18th centuries. All surges were due to extra-tropical storms 93 Table 3.11. An expert synthesis of storm surge hotspots around the world. 98 Table 3.12. Potential and actual hotspots vulnerable to flooding by the storm surge. 99 Table 5.1. Critical Facilities and Systems (Categories and Definitions) 142 Table 5.2. Studio estimates for the order of magnitude of losses for a generic city whose assets are valued at US $100 billion. 161 Table 6.1. Flood scenarios for a worst case and a moderate case 177 Table 6.2. Characteristics of the different livelihood zones analyzed 182 Table 6.3. Percentage of livestock contribution to cash income and food consumption of the livelihood zones analyzed 182 Figures Figure 1.1. Total Annual Precipitation, in millimeters. 3 Figure 1.2. Total number of drought disasters for all Asian countries with geo-referenced boundaries available 4 Figure 1.3. Number of drought disasters with month specified, for all countries listed in the Asia category in EM-DAT 5 Figure 1.4. Number of drought disasters for Asia and the maritime continent, summed by year and over all countries in the region 5 Figure 1.5. Number of drought disasters with months specified for Asia and the maritime continent 6 Contents vii Figure 1.6. Number of drought disasters for non-Asia countries in the EM-DAT database 6 Figure 1.7. Precipitation anomalies for the 1999­2001 period, divided by yearly standard deviation to facilitate comparison over diverse climate regimes 7 Figure 1.8. Reported drought disasters, 1999­2001 8 Figure 1.9. Match between drought disaster and climatic measure of drought (3 consecutive months with precipitation deficits meeting a set threshold). 10 Figure 1.10. Match between drought disaster and climatic measure of drought (4 out of 6 months with precipitation deficits meeting a set threshold). 10 Figure 1.11. Match between drought disaster and climatic measure of drought (12-month average of Weighted Anomaly of Standardized Precipitation (WASP). 11 Figure 1.12. Number of matches for 12-month WASP compared to the total number of drought disaster reports (with monthly data). 11 Figure 1.13. Correlation between the 12-month WASP calculated from two different precipitation data sets: the University of East Anglia (UEA) precipitation data and the CPC's Merged Analysis of Precipitation (CMAP). 13 Figure 1.14. Time series of drought disasters and climatic drought events (based on 12-month WASP) 13 Figure 1.15. Climate anomalies (12-month WASP) for two periods: 1982-1983 (red) and 1999-2000 (blue) 14 Figure 1.16. WASP estimate of climatic drought (shaded brown curve) and drought disaster declara- tions (red bars) for Central-Southwest Asia countries. 15 Figure 1.17. WASP estimate of climatic drought (shaded brown curve) and drought disaster declarations (red bars) for Laos and India. 16 Figure 1.A.1. Persistent deficit of precipitation 18 Figure 2.1. General approach for landslide hazard and risk evaluation 23 Figure 2.2. Global soil moisture index: 1961­1990 29 Figure 2.3. Expected monthly extreme values for a 100-years event. 30 Figure 2.4. Expected PGA with a return period of 475 years 32 Figure 2.5. Variation of slope factor, Sr, in Tajikistan and its neighboring regions 33 Figure 2.6. Variation of lithology factor, Sl, in Tajikistan and its neighboring regions 34 Figure 2.7. Variation of seismic trigger indicator, Ts, in Tajikistan and its neighboring regions 35 Figure 2.8. Variation of soil moisture factor, Sh, in Tajikistan and its neighboring regions 36 Figure 2.9. Landslide hazard zonation map obtained for Tajikistan and its neighboring regions 37 Figure 2.10. Example landslide hazard map for Central American and Caribbean countries 40 Figure 2.11. Example landslide risk map for parts of Central and South America 41 Figure 2.12. Historical rock avalanche events in Møre & Romsdal and Sogn & Fjordane Counties extracted from Norway's historical database (NGU/Astor Furseth) 42 Figure 2.13. Regional hazard zonation in Møre & Romsdal County in western Norway. 43 Figure 2.14. Landslide hazard map (landslide and rock fall hazards) for the western part of Norway based on the simplified model 44 Figure 2.15. Snow avalanche hazard zones for Norway based on the avalanche hazard model 45 Figure 2.16. Map of Armenia 47 Figure 2.17. Comparison of global landslide hazard mapping in Armenia using NGI model with the GEORISK landslide inventory 48 Figure 2.18. Major landslide events in Nepal during a 30-year time period (1971­2000) 49 Figure 2.19. Landslide hazard in Nepal predicted by the NGI model in this study 49 viii Natural Disaster Hotspots Case Studies Figure 2.20. Population density map of Nepal in 1995. 50 Figure 2.21. Landslide hazard in Georgia predicted by the model developed in this study 52 Figure 2.22. Snow avalanche hazard in Georgia predicted by the model developed in this study 53 Figure 2.23. Observed landslides in Sri Lanka between 1947 and 2003 and prediction of landslide hazard in Sri Lanka by the model developed in this study. 54 Figure 2.24. Historical landslide data in Jamaica 55 Figure 2.25. Prediction of landslide hazard in Jamaica with the model developed in this study 55 Figure 2.26. Global hotspot landslide hazard zonation for the world 56 Figure 2.27. Global hotspot landslide hazard zonation for Central Asia and the Middle East 58 Figure 2.28. Global hotspot landslide hazard zonation for Central American and Caribbean countries 59 Figure 2.29. Hotspot landslide risk zonation for Central America and Jamaica 61 Figure 2.30. Hotspot landslide risk zonation for Central Asia 62 Figure 2.31. Global hotspot snow avalanche hazard zonation for Central Asia 63 Figure 2.A.1. Distribution of risk utilizing a vulnerability proxy in Central America 69 Figure 2.A.2. Distribution of risk using a vulnerability proxy in South America 70 Figure 2.A.3. Distribution of risk utilizing a vulnerability proxy in Central Asia 71 Figure 2.A.4. Transformation for variables ranging between 0 and 1 73 Figure 2.A.5. Predicted killed versus observed for landslide 74 Figure 3.1. Areas in the southwest Netherlands flooded by the 1953 storm surge, February 1, 1953 (from Edwards 1953) 81 Figure 3.2. A simplified reconstruction of the November 1970 storm surge in Bangladesh. 82 Figure 3.3. Areas in Tokyo that are below normal high-water and surge levels with and without a 1-m rise in sea level. 83 Figure 3.4. Coasts affected by tropical cyclones 84 Figure 3.5. People at risk (that is, people potentially flooded) versus people in the flood hazard zone in 1990 for 20 global regions. 88 Figure 3.6. People at risk (that is, people potentially flooded) versus people in the flood hazard zone in the 2080s for 20 global regions. 91 Figure 3.7. Deaths by major hurricanes, cyclones, and typhoons (MC) and extra-tropical storms (ETS) from 1700 to 2000. 94 Figure 3.8. Number of "significant" events based on two thresholds of deaths: > 50,000 deaths, and > 20,000 deaths, as well as all events (>1,000 deaths) 94 Figure 3.9. Annual deaths due to surges, averaged over 50-year periods using the data in Table 3.8 95 Figure 3.10. Deaths per event for hurricanes making landfall in the United States 96 Figure 3.11. Flooding of the East Coast of England during the 1953 storm surge 100 Figure 3.12. Deaths per surge event in Bangladesh from 1800 to 2000 using the data in Table 3.9. 102 Figure 3.13. Surges on the U.S. Gulf and East Coast. Relative storm-surge potential (a), and surge graphs for six Atlantic coast locations (b), the hurricane of September 14­15, 1944. 103 Figure 3.14. Subsidence from the 1920s to the 1990s in Shanghai, China 104 Figure 4.1. The district boundaries of Sri Lanka are shown over the topography 110 Figure 4.2.a. The average annual rainfall climatology estimated based on data from 284 stations in the period between 1960 and 1990. Homogenous climatological regions as proposed by Puvaneswaran and Smithson (1993) are overlaid. 111 Contents ix Figure 4.2.b. The average monthly rainfall between 1869 and 1998 for Sri Lanka 112 Figure 4.3. The density of population in each of the 323 Divisional Secretarial Divisions based on data from the census of 2001 114 Figure 4.4. The food insecurity index of Divisional Secretariat Divisions (DSDs) as estimated by the World Food Programme 115 Figure 4.5.a. Sectoral breakdown of the GDP for 2001 117 Figure 4.5.b. Sectoral breakdown of the labor force for 2001 117 Figure 4.6. The gross domestic product (GDP) by province for 1995 118 Figure 4.7. The estimate of industrial output in the districts in 1995 119 Figure 4.8. Infrastructure density index estimated for each district, as described in the text 121 Figure 4.9. The drought hazard was estimated using a modified WASP index. 122 Figure 4.10. Drought disaster incidence frequency was constructed by aggregating the numbers of droughts that have been recorded in each district. 123 Figure 4.11. The flood hazard estimate based on the frequency of months of extreme rainfall using data between 1960 and 2000. 124 Figure 4.12. Frequency map of flood disaster incidence created by aggregating the numbers of floods recorded in each district between 1957 and 1995. 125 Figure 4.13. EM-DAT data on floods from 1975 to 2001 were used to estimate the monthly frequency of floods in the Western Slopes and Eastern Slopes regions. 127 Figure 4.14. A landslide hazard risk index was estimated based on frequency of incidence. 128 Figure 4.15. The storm and cyclone tracks for the last 100 years were used to create a cyclone hazard risk map. 129 Figure 4.16. The monthly count of storms and cyclones between 1887 and 2000 130 Figure 4.17. Multihazard index constructed by aggregating the hazard indices and scaling the result to range between 0 and 100 131 Figure 4.18. Multihazard risk estimated by weighting each hazard index by its frequency from 1948 to 1992 and rescaling the result to range from 0 to 100. 132 Figure 4.19. Multihazard risk estimated by weighting each hazard index by incidence frequency. 133 Figure 4.20. Multihazard risk estimated by weighting each hazard index by the associated relief expenditure between 1948­1992. 134 Figure 5.1. Regional elevation map of Caracas and Vargas State 139 Figure 5.2. Map of the Petare barrio of Caracas, illustrating the dual nature of the city. On the left is the open spacing of the planned, "formal" city. On the right are the densely packed squatter barrios of the "informal" city. 140 Figure 5.3. An example of the multihazard map produced by the Urban Planning studio. 145 Figure 5.4. Reserved open space in the Caracas Valley. 147 Figure 5.5. Hospitals, police stations, and fire stations in the Caracas Valley 147 Figure 5.6. Proposed interventions in a section of Petare to improve disaster preparedness with the allocation of reserved space 148 Figure 5.7. The Caracas water supply system, showing key infrastructure crossing fault lines 150 Figure 5.8. Seismic hazard affecting the city's transportation network. 152 Figure 5.9. Debris flows affecting the city's transportation network. 153 Figure 5.10. Flood hazard affecting the city's transportation network. The main east-west thoroughfare through the city is paralleled by the main, channelized river. 154 x Natural Disaster Hotspots Case Studies Figure 6.1. Location map of Tana River basin in Kenya with the river gauging stations 166 Figure 6.2. Location map of the Tana River and Garissa Districts with coverage of the Tana River basin in the Garissa District 168 Figure 6.3. Rainfall for selected stations during El Niño 1997­98 170 Figure 6.4. Data used in the creation of the DEM for flood-hazard mapping. 171 Figure 6.5. Variability of river stages at Garissa Town (1933­2001) with special focus on El Niño 1997­98 heights 172 Figure 6.6. Stream flow modeling at Garissa (1995­1999) 173 Figure 6.7. Flood hazard map for the 1961 flood (the case of a severe flood before construction of the dams) 176 Figure 6.8. Flood-hazard map for the El Niño 1997/98 flood (a worst-case scenario after the construction of the dams) 177 Figure 6.9. (a) Livelihood zones overlaid on El Niño floods case 178 Figure 6.9. (b) Livelihood zones overlaid on El Niño floods case 178 Figure 6.10. Livelihood zones overlaid on El Niño flood cases 180 Figure 6.11. Impacts of floods on market prices and livestock: (a) commodity prices (b) livestock losses 181 Preface This second volume of the Natural Disaster Hotspots Chapter One--Drought Disaster in Asia project presents a series of case studies undertaken to Written by Mathew Barlow of the University of Massa- support the global analysis, published in 2005 as Nat- chusetts, Lowell (formerly Atmospheric and Environ- uralDisasterHotspots:AGlobalRiskAnalysis. The Hotspots mental Research, Inc. and IRI), Heidi Cullen of the initiative aims to provide information to inform devel- Weather Channel (formerly National Center for Atmos- opment strategies and investments and to prioritize pheric Research (NCAR)), Brad Lyon of IRI, and Olga actions for reducing disaster risk. The initiative began Wilhelmi of NCAR. in 2001 under the umbrella of the ProVention Consor- tium as a collaborative effort of the World Bank, Columbia University's Earth Institute, and a number of Chapter Two--Global Landslides Risk Case Study international partners. Written by Farrokh Nadim of the International Center ThecorestudyteamfortheHotspotsinitiativeincluded for Geohazards at the Norwegian Geotechnical Insti- staff from the World Bank's Disaster Management Facil- tute (ICG/NGI); Oddvar Kjekstad, Ulrik Domaas, and ity (now the Hazard Risk Management Team) and the Ramez Rafat of NGI; and Pascal Peduzzi of the UNEP Development Economics Research Group (DECRG) Early Warning Unit DEWA/GRID-Europe. The Annex and from the Center for Hazards and Risk Research to Chapter Two was written by Christian Herold and (CHRR), the Center for International Earth Science Infor- Pascal Peduzzi of UNEP/DEWA/GRID-Europe. mation Network (CIESIN), the International Research Institute for Climate and Society (IRI), and the Lamont Doherty Earth Observatory (LDEO) at Columbia Uni- Chapter Three--Storm Surges in Coastal Areas versity. The project also benefited from close collabora- Written by Robert J. Nicholls of the School of Civil tion with the Norwegian Geotechnical Institute (NGI), Engineering and the Environment at the University of the United Nations Development Programme (UNDP), Southampton (formerly the Flood Hazard Research the United Nations Environment Programme (UNEP), Centre, Middlesex University, London). theUnitedNationsOfficefortheCoordinationofHuman- itarian Affairs (OCHA), the United Nations World Food Programme (WFP), the U.S. Geological Survey (USGS), Chapter Four--Natural Disaster Risks in Sri Lanka: the United Nations International Strategy for Disaster Mapping Hazards and Risk Hotspots Reduction (ISDR), and others. Written by Vidhura Ralapanawe of Ralapanawe Associ- Key contributors to the initiative were the authors of ates and Lareef Zubair of IRI with Upamala Tennakoon the case studies, each listed below with their affilia- of Natural Resources Management Services and Ruvini tions: Perera of IRI. Much of the analysis for the case study was carried out using the resources of IRI and its col- laborative project with the Mahaweli Authority of Sri Lanka, conducted by the Foundation for Environment, xi xii Natural Disaster Hotspots Case Studies Climate, and Technology (FECT). Samitha Jayamaha, of UNDP (formerly IRI); and Nancy Mutunga of the Brad Lyon, and Benno Blumenthal also contributed Famine Early Warning System Network Kenya. expertise and guidance to the study. Amara Samaras- inghe and Mahadevan Ramachandran provided back- In addition to the contributors listed above, the study ground of the work of the World Food Programme in team would like to thank Jeffrey Sachs, Director of the SriLanka.C.M.MaddumaBandaraoftheInterimNational Earth Institute; Katherine Sierra, World Bank Vice Pres- Water Resources Authority of Sri Lanka also provided ident for Infrastructure; Maryvonne Plessis-Fraissard, guidance to the project. Director of the World Bank's Transport and Urban Devel- opment Department; and Eleoterio Codato, Sector Man- ager for the Bank's Urban Unit, for their support of the Chapter Five--Multihazard Risks in Caracas, Hotspots initiative. The team is also grateful to Kathy República de Bolivariana Venezuela Boyer of the U.S. Federal Emergency Management Agency Written by Kristina R. Czuchlewski, Klaus H. Jacob, (formerly of the CHRR) for her extensive help with Arthur L. Lerner-Lam, and Kevin Vranes of the Lamont- project management and implementation, particularly Doherty Earth Observatory and the Center for Hazards relating to the case studies. Thanks are also due to and Risk Research of the Earth Institute at Columbia David Peppiatt, Manager of the ProVention Consortium University, and students and faculty of the Urban Plan- Secretariat, for his continued encouragement and sup- ning Studio: "Disaster Resilient Caracas," of the Gradu- port of the project. Funding for the initiative was pro- ate School of Architecture, Planning, and Preservation vided by the United Kingdom's Department for of Columbia University. International Development (DFID) and the Norwegian Ministry of Foreign Affairs. Their support is greatly appre- ciated. The team is also grateful to the CHRR, the Earth Chapter Six--Reducing the Impacts of Floods Institute, and the Lamont-Doherty Earth Observatory of through Early Warning and Preparedness: Columbia University for providing complementary sup- A Pilot Study for Kenya port to the project and support to the Caracas case study. Written by Hussein Gadain of the USGS Famine Early The support of the U.S. Agency for International Devel- WarningSystemNetwork;NicolasBidault,LindaStephen, opment (USAID) for the Kenya case study is gratefully and Ben Watkins of the World Food Programme, Vul- acknowledged. nerability Assessment and Mapping Unit; Maxx Dilley Introduction Natural disasters made 2005 an unforgettable year of The Hotspots initiative aims to contribute to efforts tragedy. It began in the aftermath of the December 26, to reduce disaster losses by identifying geographic areas 2004, tsunami in the Indian Ocean that devastated coun- that are most vulnerable to hazards and encouraging tries from Indonesia to Somalia, killing an estimated development agencies and policy makers to incorpo- 300,000 people and leaving 1.5 million people home- rate disaster risk management into investment plans less. In March 2005, another strong earthquake took and decisions. The project began in early 2001, when the lives of almost 2,000 people on the island of Nias the World Bank's Disaster Management Facility (DMF; in Indonesia. The year inaugurated a record hurricane now the Hazard Risk Management Team-HRMT), ini- season, with storms causing severe damages through- tiated discussions with the newly established Center for out the Caribbean, Mexico, and the United States' Gulf Hazards and Risk Research (CHRR) at Columbia Uni- Coast. On October 8, 2005, the world witnessed the versity to conduct a global-scale, multi-hazard risk analy- devastating impacts of another major earthquake in sis focused on identifying key "hotspots" where the risks the Kashmir region, which claimed more than 73,000 of natural disasters are particularly high. Discussions lives in Pakistan, and over 1,300 more in India. And culminated in the formulation of a project, implemented these were just a few major catastrophes that grabbed jointly by several departments within both the World the headlines for brief periods. All told, there were 360 Bank and Columbia University, with the participation natural disasters that killed more than 90,000 people of numerous other international partners. and affected more than 150 million lives in 2005. The project results consist of a global analysis of dis- Given the devastating losses of 2005, it would at least aster risks associated with six major natural hazards-- be a small comfort to consider the year an anomaly cyclones, droughts, earthquakes, floods, landslides, and unlikely to be repeated. However, the number of dis- volcanoes--accompanied by a series of case studies. aster events seemingly continues to rise, as do the social The global analysis, described in a separate volume, and economic costs. Disasters in 2005 caused some assessed the estimated spatial distribution of relative US$159 billion in damage (of which US$125 billion risks of mortality and economic losses associated with were losses caused by Hurricane Katrina in the United each hazard and all hazards combined. Risk levels are States), a 71 percent increase from the total losses of estimated by combining hazard exposure with histori- US$93 billion in 2004. And although the number of cal vulnerability for two indicators of elements at risk-- overall deaths caused by natural disasters is decreas- gridded population and gross domestic product (GDP) ing, the number of those affected in terms of disrup- per unit area. Calculating relative risks for each grid cell tions to daily life, loss of livelihoods, and deepening rather than for countries as a whole allows risk levels poverty continues to increase. The impacts of popula- to be compared at subnational scales. tion and economic growth, rapid urbanization, envi- The resolution of the global analysis is relatively ronmental degradation and climate change are a few of coarse, however, and global datasets inadequately cap- the factors that will continue to fuel this trend unless ture important factors affecting local risk levels. Spe- something is done to reduce disaster risks. cific limitations of the global analysis include the following: xiii xiv Natural Disaster Hotspots Case Studies · Global spatial datasets on vulnerability characteris- · ascertain what finer scale data may exist locally, for tics of the major sets of elements at risk to each hazard example, on vulnerability, response capacity, and do not exist, although vulnerability may be inferred poverty; from existing data on a limited basis in some cases. · identify cross-hazard dependencies and interac- · Existing global spatial datasets on major hazards and tions among hazards, exposure, vulnerability, and elements at risk are of coarse resolution, sufficient multihazard risk management at subnational scales; for resolving only relatively broad spatial patterns of · examine the policy context for risk management risk. and the degree to which multiple hazards are rec- · Existing global spatial datasets on major hazards cover ognized and addressed in an integrated manner; only limited time periods that may not fully charac- · engage national- to local-level stakeholders; and terize the probability of recurrence of hazardous · demonstrate that the theory and methods that guide events. the global analysis can be applied on a more regional · Global data on socioeconomic "outcome" variables-- or local geographic scale. for example, mortality, morbidity, economic losses, and impoverishment--are universally available only Key findings from the case studies are: at the country level in the form of national statistics. Yet such data are needed to verify the global risk Scale matters. Geographic areas that are identified as assessment (that is, assessed spatial patterns of dis- hotspots at the global scale may have a highly variable aster risk hotspots should correspond to historical spatial distribution of risk at finer scales. patterns of actual human and economic losses to some degree). Scale affects data availability and quality. Hazard, exposure, and vulnerability data are available at sub- To partially address these limitations, the case stud- national resolutions for individual countries and even ies in this volume were undertaken as the second com- cities, as the analyses for Sri Lanka and Caracas show. ponent of the Hotspots project to complement the More comprehensive, better quality data permit more global-scale analysis. Each case study was implemented complete, accurate, and reliable identification of multi- by a different set of researchers within a general frame- hazard hotspots at finer scales of resolution. work provided by the project. The case studies use the same theory of disaster causality as the global analy- Scale affects the utility of the results. Better data res- sis--namely, that the risks of a specified type of disas- olution and a richer set of variables contribute to results ter-related loss to a set of elements at risk over a given that are more relevant for national-to-local scale risk period are a function of the exposure of the specified management planning, as illustrated in the case study set of elements to natural hazards and their degree of from Caracas. This is very important, as decisions made vulnerability to the hazards to which they are exposed. at local and national scales have perhaps the greatest Three case studies address specific hazards: land- potential to directly affect risk levels, both positively slides, storm surges, and drought. Three other case stud- and negatively. ies address regional multihazard situations in Sri Lanka, the Tana River basin in Kenya, and the city of Caracas, The global- and local-scale analyses are comple- Repúblic Bolivariana de Venezuela. This small number mentary. In some instances, national-to-local level of geographically limited case studies was designed to: risk assessors and planners may be able to "downscale" · provide "ground truthing" for particular regions iden- global data for finer-scale risk assessment to compen- tified as potential hotspots; sate for a lack of local data. Ideally, however, global · explore specific cases where there are more detailed analyses would be scaled up--generalized from more loss probability data and models compared with what detailed larger-scale data. In practice, many barriers still is available globally; remain. The global infrastructure for systematically assembling and integrating relevant datasets for disas- Introduction xv ter risk assessment at multiple scales remains inade- nerabilities. Moreover, they need to understand the quate. Nonetheless, the fact that relevant datasets can potential interactions among these hazards, whether be obtained and integrated at various scales creates the direct--for example, storms that initiate both floods hope that data eventually can be collected and shared and landslides--or indirect--such as consecutive haz- routinely to improve disaster risk assessment globally ards that deplete resources and strain response capac- and locally. ities. We hope that in addition to providing interesting Taken together, the global analysis and case studies and useful results, the global analysis and case studies provide strong evidence of the importance of employ- will stimulate additional research, particularly at national ing multihazard approaches in disaster risk manage- and local levels, increasingly linked to disaster risk reduc- ment. Given resource constraints and the multiple roles tion policy making and practice. played by key infrastructure--such as roads, railroads, and ports in disaster preparedness, emergency response, reconstruction, and ongoing economic activity--it is vital that planners and decision makers at all levels have a sound appreciation of the hazards prevalent in their specific regions of concern, along with associated vul- Chapter 1 Drought Disaster in Asia Mathew Barlow, Heidi Cullen, Brad Lyon, and Olga Wilhelmi Drought ranks as the natural hazard with the greatest ing persistent, severe deficits in rain or snow, however, negative impact on human livelihood. Due to the com- can be reflected in all four categories and so may be more plexities of drought disasters, however, investigations amenable to generalization. The recent exceptionally of drought have been limited primarily to local case stud- severe drought in Central-Southwest Asia provides a ies. As global climate data are operationally monitored motivating example, where a clear signal is seen in both and forecast, a more regional-to-global scale perspective the climate and societal data across a wide variety of on the climatic signature of drought disasters could sectors--including agriculture, livestock, water resources, enhance ongoing efforts in drought monitoring, early and environmental management--as well as across warning, and mitigation efforts. This investigation under- several countries (Agrawala et al. 2001; Barlow et al. takes a preliminary examination of the climatic aspects 2002; Lautze et al. 2002). The first key question for the of drought disasters across a broad geographic range-- present investigation is whether there is a relationship the countries of Asia and the maritime continent (Indone- between severe and persistent precipitation deficits and sia and Malaysia)--with data and methodology that may the reported incidence of drought disaster at the coun- be easily extended to global consideration. try level. It may be that the disasters are too dependent A drought disaster is caused by the combination of on the local economic, political, and social circumstances both a climate hazard--the occurrence of deficits in rain- to show a clear and consistent link to the climate data, fall and snowfall--and a societal vulnerability--the eco- and it may be that the country-averaged climate data do nomic, social, and political characteristics that render not have sufficient resolution in space or time to cap- livelihoods susceptible in the region influenced by the ture the relevant climate fluctuations. The second key deficits. Global disaster databases such as the Emergency question is whether such a relationship could be gen- Events Database (EM-DAT), maintained by the Center eralized across the range of climatic and societal char- for Research on the Epidemiology of Disasters, include acteristics of multiple countries. global information on recorded drought disasters at the The EM-DAT database, which was used in this study, country level. The corresponding climate information contains drought disaster reports that meet at least one is not readily available, however, because, although of four criteria: 10 or more people reported killed, 100 several climate-based measures of drought exist, their or more people reported affected, a call for interna- general relevance to the incidence of drought disasters tional assistance, or a declaration of a state of disaster. is not well known. The basic spatial unit of the data is countries, although Both the climate dynamics and the societal impacts more specific geographic information is given when avail- of drought are highly complex, poorly understood, and able. Clearly, the identification of events based on these difficult to generalize. Drought is commonly divided into criteria depends strongly on the quality of reporting meteorological, agricultural, hydrologic, or socioeco- and may be critically influenced by political, social, and nomic categories, and there is great variation in both economic conditions. To avoid some of these complex- the climatic and the socio-political-economic character ities for this preliminary analysis, the disasters are con- between and within these categories. Droughts involv- sidered only in terms of whether one appears or does 1 2 Natural Disaster Hotspots Case Studies not appear in the disaster database for a given country, The results demonstrate a correspondence between unweighted by any other factor. The data are further severe, persistent precipitation deficits observed in the restricted to reports that include a month (some spec- available climate data and reported drought disasters for ify only a year), for unambiguous comparison to the 14 of the 27 countries during the 1979­2001 period. monthly climate data. The enhanced version of the EM- Global-scale climate fluctuations in 1982­1983 and DAT natural disaster database, which includes data from 1999­2001 strongly affected the occurrence of wide- 1975 to 2001, is used to provide incidence of drought spread precipitation deficits in the region and were also disaster for the 27 countries in the "Asia" category. The reflected in the disaster data, particularly in the later Asia region was chosen for this pilot effort because of period, which encompassed the largest values in both the large-scale, exceptionally severe drought that occurred climate and disaster data. There is some suggestion that from 1999 to 2001 as well as the availability of some the relationship is particularly strong in the semiarid geo-referenced data for the region. countries. The relationship is present in other climatic For the analysis of climate data, two estimates are used zones, however; Laos, for instance, shows a relation- based only on precipitation deficits and emphasizing ship over multiple drought events. multiple months of severe precipitation deficits. The first, Persistent Deficits of Precipitation (PDP), which was also used in the World Vulnerability Report, tracks the number Asian Drought Disasters in EM-DAT of consecutive months that the observed precipitation falls below a given threshold (for example, 75 percent The spatial distribution of the number of disaster reports of median). The second, the Weighted Anomaly of for each country over the 1975­2001 period is shown Standardized Precipitation (WASP), is an average of in figure 1.2 (Unfortunately, geo-referenced data are weighted (relative to the local annual cycle) and stan- not yet available for all the countries in the region. dardized (relative to the local standard deviation) pre- Although EM-DAT reporting began in 1975, no reports cipitation over a set number of months--12 for this specify a month until 1977.) Figure 1.3 shows the number investigation. This index is similar to the Standardized of reports for all countries in the region, with the fur- Precipitation Index (SPI) and, for averages in the 6- to ther restriction that only reports associated with a spe- 12-month range, correlates well with the Palmer Drought cific month are shown. These are the data that will be Severity Index (PDSI). (See Annex I.A for a discussion compared with the climate data. Note that the coun- of the calculation of the PDP and WASP.) Although many tries with the largest area and population (China and estimates of drought are available and would be high India) dominate the reports. priorities in future analyses, we begin here with two The year-to-year variation of the drought disaster that are used in climate monitoring by the International reports for the region is shown in figure 1.4 for all reports Research Institute for Climate Prediction (IRI). The cli- and in figure 1.5 for only the reports that specify a month. matic drought estimates are analyzed as country aver- Although there is modest variability throughout the ages for consistency with the disaster data, starting in period, the number of reports over the last three years 1979 (the beginning of consistently available satellite is notably larger. This increase is expected to some degree data to supplement the station data for precipitation). due to the severity of the recent drought in the region, These country averages are particularly problematic for which is one of the motivating factors for the study. How- large countries such as China and India, which range ever, the larger number of reports also raises questions over vast areas and encompass multiple climate zones, about changes in reporting over the period. As a first as evident in the range of annual average precipitation check on this, the time variation of drought disaster (see figure 1.1). The quality of the precipitation data in reports in all countries outside of Asia is shown in the region is also marginal in some areas, particularly in figure 1.6. Although there is still a maximum for the the north and west. The emphasis on severe, widespread recent period in the non-Asian reports, the largest values events may be expected to alleviate these problems some- for the period occur during 1983. Thus there does not what. appear to be a general bias in the EM-DAT reporting Dr ought Figure 1.1. Total Annual Precipitation, in millimeters. Due to the unavailability of geo-referenced data, this plot does not show Disaster several countries in the west and north of the study region: Israel, Jordan, Uzbekistan, Mongolia, Iraq, Tajikistan, Armenia, and Georgia. These countries all have annual averages of less than 500 mm, except for Georgia, which has an annual average of 720 mm. in Asia 3 4 Figure 1.2. Total number of drought disasters for all Asian countries with geo-referenced boundaries available Natural Disaster Hotspots Case Studies Drought Disaster in Asia 5 Figure 1.3. Number of drought disasters with month specified, for all countries listed in the Asia category in EM-DAT Figure 1.4. Number of drought disasters for Asia and the maritime continent, summed by year and over all countries in the region 6 Natural Disaster Hotspots Case Studies Figure 1.5. Number of drought disasters with months specified for Asia and the maritime continent Figure 1.6. Number of drought disasters for non-Asia countries in the EM-DAT database toward the end of the period. While the variations in case has precipitation deficits that average to near zero the non-Asian reports are outside the scope of this analy- for the country as a whole due to offsetting contribu- sis, we note that an exceptionally strong El Niño was tions from wet and dry regions.) Apart from Indochina, associated with global climate anomalies during the severe precipitation deficits are well represented by 1982­1983 and there were global-scale drought anom- the incidence of drought disaster for this recent period, alies during 1999­2001 (Hoerling and Kumar 2003). from 1999­2001 (see figure 1.8). The precipitation deficits for Asia in the 1999­2000 period are shown in figure 1.7. The precipitation deficits have been normalized by the yearly standard deviation Climate-Based Drought Estimates and to account for the dramatic changes in average precip- Drought Disaster Reports itation across the region (see figure 1.1). Severe deficits occurred during this period across the mid-latitudes and Persistent Precipitation Deficits subtropics of the entire region, except for parts of To look for severe and persistent events as they occur in southern China. (This highlights the problem of aver- the climate data, a tally may be kept of the number of ages for large countries such as China, which in this consecutive months that precipitation deficits exceed a Dr ought Disaster Figure 1.7. Precipitation anomalies for the 1999-2001 period, divided by yearly standard deviation to facilitate comparison over diverse climate regimes in Asia 7 8 Figure 1.8. Reported drought disasters, 1999­2001 Natural Disaster Hotspots Case Studies Drought Disaster in Asia 9 given threshold. As a starting point, we consider three increase in nonmatches. It appears that the comparison variations in the calculation of persistent precipitation works better in semiarid countries. deficits. For each, the threshold is set to be 75 percent of median (by calendar month), while the number of Weighted Anomaly of Standardized months considered is variously taken to be 3, 6, and 4 Precipitation out of 6. (See Annex 1.A for a description of this calcu- lation.) These criteria are applied to the gridded pre- Main Findings cipitation data, and then averaged to the country level, Another approach to estimating drought is to simply necessitating the choice of another threshold, here taken average the precipitation anomalies over a set number to be the requirement that a minimum of 50 percent of of months. In order to make such a calculation easier to the gridboxes within a country meet the criteria. Mul- interpret across different climate regimes and different tiple instances of meeting the climate criteria within a seasons, the anomalies can be normalized and weighted single 12-month period are taken to represent a single according to the average local magnitude and seasonal- climate event. For each observed climate event, the dis- ity. IRI uses the WASP for climate monitoring (see Annex aster reports are then searched to see if a disaster is 1.A for details of calculation). It is similar to another reported within three months. standard monitoring product, the SPI. Here, drought is The precipitation data are from the Climate Predic- identified in the climate data when the country average tion Center Merged Analysis of Precipitation (CMAP) of the WASP is less than ­1. As shown in figure 1.11, (Xie and Arkin 1996, 1997), which began in 1979. more than half the countries show a correspondence Figure 1.9 shows the number of matches (climate between 12-month WASP and drought disasters. The event and subsequent disaster report) and nonmatches number of matches is compared with the total number (climate event but no disaster report) between the three- of disaster reports (with a month specified) in figure month criteria and the disaster reports. Several coun- 1.12. The climate drought estimate, based on the 12- tries show matches, and both Afghanistan and Pakistan month WASP, identifies all reported drought disasters have matches with no nonmatches--that is, there is a for Israel, Afghanistan, Syria, Pakistan, Armenia, and one-to-one correspondence between climate events Malaysia (seven matches total) while also generating 10 and reported drought disasters. The countries have been nonmatches (identified climatic drought without a cor- ordered left to right based on their average annual pre- responding disaster report). cipitation, and there is some suggestion that the matches Statistical analysis, conducted by taking the observed are more frequent in the semiarid countries. WASP-identified drought episodes, randomly shuffling It is important to note that all the matches in the semi- them in time, and then recounting the matches with arid countries occur during the recent extended drought the disaster data, shows that in 1,000 random reshuf- period (1999­2001), so it is difficult to assess the flings the same number of matches (18) over all coun- reproducibility of this relationship. However, this drought tries did not ever occur by chance. This is consistent period encompasses 40 distinct drought disaster reports with the relatively few number of months where the from 20 different countries that represent a diverse array WASPthresholdisexceededandtherelativelyfewnumber of political, social, and economic conditions as well as of months with a disaster report (one to two months for a range of quality in climate data. So, the broad agree- most of the countries) over the 276 months and 27 coun- ment even for a single drought period is likely to be tries considered. Random matches are rare for each coun- meaningful. try (Afghanistan, for instance, would get two random The results for the six-month criteria are similar (with matchesonlyabout2percentofthetime),andthechances a smaller fraction of gridboxes required per country) and of randomly getting such matches in at least 14 of the are not shown. 27 countries are highly unlikely (less than 1 in 1,000). Figure1.10showstheresultswheredeficitsarerequired The WASP estimate picks up many of the same to meet the threshold in any four out of six consecutive droughts as the PDP, but there are some differences as months. Several more matches are present with little well; a more sophisticated analysis could refine the util- 10 Natural Disaster Hotspots Case Studies Figure 1.9. Match between drought disaster and climatic measure of drought (3 consecutive months with precipitation deficits meeting a set threshold). Number of matches (climatic drought events with subsequent disaster) are in red and non-matches (climatic drought events with no subsequent disaster) are in yellow. Countries have been ordered left to right based on annual average precipitation (green line, in mm). Figure 1.10. Match between drought disaster and climatic measure of drought (4 out of 6 months with precipitation deficits meeting a set threshold). Countries have been ordered left to right based on annual average precipitation (green line, in mm). Drought Disaster in Asia 11 Figure 1.11. Match between drought disaster and climatic measure of drought (12-month average of Weighted Anomaly of Standardized Precipitation (WASP). Matches are in red, non-matches in yellow. Countries have been ordered left to right based on annual average precipitation (green line, in mm). Figure 1.12. Number of matches for 12-month WASP compared to the total number of drought disaster reports (with monthly data). The scale is the same for both: Note that the 12-month WASP matches all 7 reported drought disasters for Israel, Afghanistan, Pakistan, Armenia, and Malaysia (with 10 non- matches for the same countries, cf. Figure 1.11). 12 Natural Disaster Hotspots Case Studies ity of multiple measures. Additionally, the behavior of umented for the La Niña event (Barlow et al. 2002; Hoer- the current estimates with different parameters (higher ling and Kumar 2003; Tippett et al. 2003). or lower thresholds, averaging periods) has not yet The 12-month WASP is shown in figure 1.15 for the been explored. two periods: in red for 1982­1983 and in blue for The limitations of the precipitation data are also a con- 1999­2000. As before, the countries are ranked based cern. Here we have primarily used the Climate Predic- on average precipitation. In this region, average precip- tion Center (CPC)'s Merged Analysis of Precipitation itation is closely related to latitude. A striking out-of- (CMAP) (Xie and Arkin 1996, 1997). To assess the effect phase relationship is present between the two periods: of uncertainty in the precipitation data on the corre- in 1982­1983, the countries in the south and east of spondence with the disaster, we have also calculated the region are dry, while the countries in the north and the 12-month WASP from the University of East Anglia's west are wet, and during 1999­2000 the reverse (UEA) 0.5x0.5 degree precipitation dataset (New et al. occurs. Although the climatic drought signal is large for 2000) for the 1979­1995 period (the overlapping period both years, the disaster signal is less in 1982­1983, with greatest underlying density of input stations in the consistent with the weaker relationship between climatic UEA data). The correlation between the two WASP esti- drought and disaster incidence for the tropical coun- mates is shown in figure 1.13. In the 1979­1995 period, tries. As a reminder, the current analysis is done with the overall correspondence with disaster data is some- respect to the simple measure of incidence of drought what better with the CMAP-based data. However, the disaster (whether a disaster is reported or not at the coun- two data sources identify different disasters--the UEA- try level). In terms of population affected, for instance, based data capture a disaster in both Nepal and the important events might be considerably different, Bangladesh, which the CMAP-based data do not. The depending to a large degree on what happens in only chance of a random match for a given country is small one or two countries, particularly in India. but not insignificant, so such differences must be viewed with caution. It is possible that a careful analysis of the Subregional Focus precipitation data, perhaps going back to the original We end with a brief consideration of specific countries. station data, might provide a better drought estimate in The Central-Southwest Asia countries are shown in figure data-scarce regions than any single gridded estimate. 1.16 for both 12-month WASP (greens and browns) and the incidence of drought disasters (red bars). The Time Variations large-scale, severe, persistent drought at the end of the To examine temporal variability more closely, the time record is clearly seen and has a good association with series of the number of months the 12-month WASP the disaster reports. While general drought analysis across exceeds the ­1 threshold as well as the number of dis- multiple countries is important to identify appropriate aster reports is shown in figure 1.14 for the average monitoring and to forecast target variables, it should over all Asian countries. In the climate data, both also be connected with much finer analysis. Uzbekistan, 1982­1983 and 1999­2000 are notable drought peri- for instance, has a year of drought in 1986 that is simi- ods. There is a disaster maximum for both periods, but lar in magnitude to the recent drought, yet is not asso- the latter is much larger. A vigorous El Niño (warm ciated with a disaster report. Is this due to a change in episode) dominated the global climate during the first sociopolitical circumstances (independence from the period, while a vigorous La Niña (cold episode) strongly former Soviet Union)? Problems in the precipitation data? influenced the global climate during the later period, Subcountry variations in the drought? Only one year of suggesting the possibility of large-scale patterns across drought versus two? Lack of a drought this severe in the region for both periods. It should be noted that neighboring countries affecting the regional economy? both El Niño and La Niña episodes deviated somewhat Consideration of this level of analysis in conjunction with from their typical behavior, so extrapolation to future the large-scale analysis would considerably improve both. eventsisnotstraightforward.However,clearlinksbetween Figure 1.17 shows the same data for Laos and India. the Pacific climate and Asia precipitation have been doc- Both countries are particularly interesting as they show Drought Disaster in Asia 13 Figure 1.13. Correlation between the 12-month WASP calculated from two different precipitation data sets: the University of East Anglia (UEA) precipitation data and the CPC's Merged Analysis of Precipitation (CMAP). The correlation is computed on the monthly data from Dec. 1979-Dec. 1995. Figure 1.14. Time series of drought disasters and climatic drought events (based on 12-month WASP) 14 Natural Disaster Hotspots Case Studies Figure 1.15. Climate anomalies (12-month WASP) for two periods: 1982-1983 (red) and 1999-2000 (blue) agreement between climatic drought and disaster inci- months for every period that meets the climate criteria dence for several distinct events. India would benefit are listed, along with a notation as to whether the cli- tremendously from higher resolution analysis--and, of mate drought event was associated with a disaster report. course, there is a tremendous amount of information This database is intended only as a preliminary step, and analysis available for the country. Some province- needs careful validation and an estimate of data error, level data are present in the EM-DAT data; for instance, and should be examined only with extreme caution. Rajasthan is associated with several of the drought reports As another preliminary step, climate data were that show good agreement with the India-average WASP. added to the EM-DAT database for those drought dis- asters that could be identified as corresponding to the WASP-based drought estimate. A column was added Developing a Drought Hazard Database that gives, for those cases, the largest value of the WASP estimate recorded over the current or previous three As noted in the introduction, a drought disaster is caused months to the disaster report. These data should also by the combination of both a climate hazard and a soci- be considered only with extreme caution. etal vulnerability. Using a climate-based estimate of drought that is shown to have a relationship with the occurrence of drought disasters, a collection of drought Common Features of Drought Disasters events based on the climate definition will constitute a database of drought hazard. As a preliminary step, such Our preliminary results identify several recurring aspects a database has been constructed from the results of the of droughts associated with reported drought disasters: WASP analysis. For the countries that have a least one · Persistent and severe: At the country level, precipita- match between disaster report and WASP threshold, as tion deficits that are either less than 75 percent of described in the previous section, the starting and ending Dr Figure 1.16. WASP estimate of climatic drought (shaded brown curve) and drought disaster declarations (red bars) for ought Central-Southwest Asia countries. Green shading indicates wet periods. The -1 threshold is shown as a black line. Disaster in Asia 15 16 Figure 1.17. WASP estimate of climatic drought (shaded brown curve) and drought disaster declarations (red bars) for Laos and India. The -1 threshold is shown as a black line. Natural Disaster Hotspots Case Studies Drought Disaster in Asia 17 median for three consecutive months or have a WASP seven matches. Ten climatic droughts are also identi- value of less than ­1 in a 12-month average show a fied that do not have a corresponding disaster report statistically significant association with the reported (10 nonmatches). There is some suggestion that the rela- incidence of drought disaster. tionship is stronger in the semiarid countries. The relationship is present in other climatic zones, how- · Relatedtolarge-scaleclimatefluctuations:Thelargenumber ever; Laos, for instance, shows a relationship over mul- of drought disaster reports in 1999­2001 is closely tiple events. This link between climatic drought and related to widespread, severe climate anomalies. disaster reports is consistent through the two large · Present in a range of climate regimes: A normalized climate events that affected the region during the period definition of climatic drought shows agreement with of analysis. disaster reports across a wide geographic and climatic An understanding of the links between large-scale cli- range (figure 1.11), from the semiarid mid-latitudes mate data and the incidence of disasters could enhance to the tropics, over a 15-fold range in mean annual the utility of current climate monitoring and forecast- precipitation (figure 1.1). ing efforts. This pilot effort provides a preliminary identification of such links; further investigation is rec- · But apparently strongest in the semiarid countries: The ommended, as outlined in the next section. best association found in the current analysis is for countries with annual precipitation of less than 35 centimeters. Recommendations Although these features must be regarded as prelimi- nary, particularly given the limitations of the data and This preliminary analysis suggests a number of key exten- the prominence of only two climate events, these fea- sions: tures are not specific to Asia. The data from other regions · Subcountry analysis of Laos, Indonesia, Malaysia, and of the world would provide an effective testing ground. Bangladesh. Based on the matches between climate disaster data, these countries are ideal targets for forg- ing a link between large-scale analysis and local- Summary scale data and expertise. The 1982­1983 El Niño event and seasonality are key issues. Drought disaster reports are compared with precipita- · Analysis of regions outside Asia. These additional coun- tion-based estimates of drought at the country level for tries will allow further validation of the noted rela- the 27 countries listed in the Asia category in the EM- tionships, particularly in other semiarid countries and DAT database for the 1975­2001 period. The last three for the 1982­1983 versus 1999­2000 comparison. years of the record have, by far, the largest number of This is a natural extension of the current work, as reports. This pronounced maximum appears not to be the methodology, analysis tools, and datasets from the an artificial feature of EM-DAT reporting but rather phys- current analysis can be directly applied. ically linked to the exceptionally severe drought impact- ing Asia during that time. · Investigation of other parameters in drought disaster An objective comparison was undertaken between reports. Although this requires a large measure of cau- the monthly disaster reports and two climate-based esti- tion, the other information (people killed, people mates of drought. Even at the country level, and with affected, cost) should be assessed. limited data, a relationship can be discerned between · Assessment of the impacts of previous climatic droughts both climatic measures of drought and the incidence in Central-Southwest Asia. Why are there no drought of drought disasters in the region. In fact, the climate reports before the recent event in a region that com- drought estimate based on the 12-month WASP matches prises a wide range of sociopolitical situations? all reported drought disasters for Israel, Afghanistan, Syria, Pakistan, Armenia, and Malaysia, encompassing · Consideration of other estimates of climatic drought 18 Natural Disaster Hotspots Case Studies and precipitation station data. Other drought meas- References ures such as SPI, satellite-based vegetation indexes, and the PDSI should be included in the analysis. The Agrawala, S., et al. 2001. The Drought and Humanitarian Crisis parameter ranges of the estimates used here should in Central and Southwest Asia: A Climate Perspective. IRI Spe- also be explored more thoroughly. Comparing the cial Report 01-11:24. results based on different precipitation datasets, par- Barlow, M., H. Cullen, and B. Lyon. 2002. Drought in Central ticularly from station data only, rather than gridded and Southwest Asia: La Niña, the Warm Pool, and Indian data, would provide useful bounds on certainty in the Ocean Precipitation. Journal of Climate 15: 697­700. climate data. Hoerling, M., and A. Kumar. 2003. The Perfect Ocean for Drought. Science 299: 691­94. · Consideration of supplementary impact data. Crop fail- Lautze, S., et al. Qaht-e-Pool "A Cash Famine": Food Insecurity ure is a frequent result of drought, and crop estimates, in Afghanistan, 1999­2002. Washington, DC: The United to the extent they are available, could provide a com- States Agency for International Development. Medford, MA: plementary set of data. The Feinstein International Famine Center. · Regional vs. country-level analysis for China and India. New, M., M. Hulme, and P. D. Jones. 2000. Representing Twen- Given the often large discrepancy between the scale tieth Century Space-Time Climate Variability. Part 2: Devel- of a given country and that of climate variations (for opment of 1901­96 Monthly Grids of Terrestrial Surface example, drought) a closer examination of drought Climate. Journal of Climate 13: 2217­38. measures within specific, key regions of countries (for Tippett, M. K., M. Barlow, and B. Lyon. 2003. Statistical Cor- example, the North China Plain, Northwest India) rection of Central Southwest Asia Winter Precipitation Sim- would provide a necessary perspective on the coun- ulations. International Journal of Climatology 23: 1421­33. try-level analysis. Xie, P., and P. Arkin. 1996. Analyses of Global Monthly Precip- itation Using Gauge Observations, Satellite Estimates, and Figure 1.A.1. Persistent deficit of precipitation Source: Brad Lyon, IRI Drought Disaster in Asia 19 Numerical Model Predictions. Journal of Climate 9: 840­58. map (for example, a 25-mm monthly precipitation anom- Xie, P., and P. Arkin. 1997. Global Precipitation: A 17-Year Monthly aly may be relatively small for some regions but quite Analysis Based on Gauge Observations, Satellite Estimates and large for others). The standardized precipitation Numerical Model Outputs. Bulletin of the American Meteoro- anomalies are weighted according to the annual cycle logical Society 78: 2539­58. of precipitation at a given location based on average monthly precipitation values. This weighting reduces the tendency for standardized precipitation measures Annex 1.A to become artificially magnified at the start or end of the rainy season where there are distinct dry and wet Description of climate-based measures of drought Per- seasons. The standardized, weighted anomalies are sistent Deficit of Precipitation (PDP) summed over different periods of interest; at the IRI, This index measures the persistence of monthly WASP analyses are routinely produced for the most precipitation deficits for a given location. A deficit is recent 3-, 6-, 9-, and 12-month periods. defined to occur when the observed monthly-average The mathematical definition of the version of the index used precipitation falls below the long-term median value at in the study is described below. a given location. Variations of this index measure the number of consecutive months that the observed pre- 12 N PiiPi ­ Pi cipitation falls below different thresholds (percent of WASPN = N · · PA i=1 median). For example, thresholds used in the study include 75 percent of median for 3 and 6 consecutive Where, months and 75 percent of median for any 4 out of 6 consecutive months. A schematic of the method is shown WASPN = the N-month WASP, where N is the number below. of months (here, 12) over which the standardized, weighted anomalies have been integrated; The Weighted Anomaly Standardized Precipitation Index N = the standard deviation of the N-month (WASP) WASP over the historical record for the last month in the integration; This index was developed at the IRI as a simple, single- the observed precipitation for month i; variable index to measure the relative surplus or deficit Pi = the monthly climatological precipitation for of precipitation on different time scales. The index is Pi = month i; based solely on monthly precipitation but requires his- i = the monthly standard deviation in torical data (for at least a 25-year period) as well. The precipitation for month i; and basic idea in standardizing (by the appropriate monthly PA = the average annual precipitation. standard deviation) the data is to compare regions with different precipitation climatologies on a single Chapter 2 Global Landslides Risk Case Study Farrokh Nadim, Oddvar Kjekstad, Ulrik Domaas, Ramez Rafat, and Pascal Peduzzi Background on socioeconomic factors (population density, quality of infrastructure, collective organization) and the response The main objective of this study is to perform a data- capacity (prevention, capacity of aid intervention, and based, first-order identification of geographic areas that mitigation). The vulnerability evaluation was performed form the global landslide risk disaster hotspots on an in close cooperation with United Nations Environ- international scale, with the main emphasis on devel- ment Programme (UNEP)/GRID-Geneva. oping countries. This includes combining the identi- fied hazard and vulnerability, for people and infrastructure, to determine risk. The probability of General Approach and Terminology landslide occurrence is estimated from modeling of physical processes combined with statistics from past Definitions of hazard, vulnerability, and risk have evolved experience. The main input data for the assessment of during the last few years. In this study, we use the ter- landslide hazard are topography and slope angles, pre- minology adopted by United Nations International Strat- cipitation, seismic activity, soil type, hydrological con- egyforDisasterReduction(UN/ISDR) (http://www.unisdr. dition, and vegetation. Vulnerability mainly depends org/eng/library/lib-terminology-eng%20home.htm). Hazard A potentially damaging physical event, phenomenon, and/or human activity that may cause the loss of life or injury, property damage, social and economic disruption, or envi- ronmental degradation. Hazards can include latent conditions that may represent future threats and can have different origins: natural (geological, hydro-meteorological, and biological) and/or induced by human processes (environmental degradation and technological hazards). Hazards can be single, sequen- tial, or combined in their origin and effects. Each hazard is characterized by its location, inten- sity, frequency, and probability. Geological hazard Natural earth process or phenomenon that may cause the loss of life or injury, property (geohazard) damage, social and economic disruption, or environmental degradation. Geological hazards include internal earth processes of tectonic origin, such as earthquakes, geo- logical fault activity, tsunamis, volcanic activity and emissions, as well as external processes such as mass movements (landslides, rockslides, rockfalls or avalanches, surface collapses, and debris and mudflows). Hazard analysis Identification, studies, and monitoring of any hazard to determine its potential, origin, characteristics, and behavior. 21 22 Natural Disaster Hotspots Case Studies Hazard occurrence Probability of occurrence of a specified natural hazard at a specified severity level in a probability specified future time period. Risk The probability of harmful consequences, or expected losses (deaths, injuries, property, livelihoods, economic activity disrupted, or environment damaged) resulting from inter- actions between natural or human-induced hazards and vulnerable conditions. Con- ventionally, risk is expressed by the notation Risk = Hazard x Vulnerability. Elements at risk Inventory of people, houses, roads or other infrastructure that are exposed to the hazard. Risk assessment/ A process to determine the nature and extent of risk by analyzing potential hazards and analysis evaluating existing conditions of vulnerability that could pose a potential threat or harm to people, property, livelihoods, and the environment on which they depend. The process of conducting a risk assessment is based on a review of both of the following: the technical features of hazards such as their location, intensity, frequency, and probability; and an analysis of the physical, social, economic, and environmental dimensions of vulnerability. The risk assessment does so while taking into particular account the coping capabilities pertinent to the risk scenarios. Vulnerability A set of conditions and processes resulting from physical, social, economic, and envi- ronmental factors that increase the susceptibility of a community to the impact of haz- ards. Also the degree of loss to an element at risk should a hazard of a given severity occur. The present study focuses on rapid mass movements, General Approach for Snow Avalanche like rockslides, debris flows, snow avalanches, and rain- Hazard Evaluation fall- and earthquake-induced slides. The general approach adopted in the present study, The susceptibility to snow avalanche is derived from for the evaluation of global landslide hazard-prone areas the combination of all avalanche formation parameters, and risk hotspots, is depicted in figure 2.1. namely, terrain slope, precipitation, and temperature. These parameters are integrated to form grid maps with pixel values through Geographical Information System General Approach for Landslide Hazard (GIS) analyses. The corresponding probability of occur- Evaluation rence is found from statistical analyses of weather infor- mation for single grid cells, to obtain a return period Landslide hazard level depends on a combination of of the events based on precipitation. The probability trigger and susceptibility. In the first-pass estimate of may then be extrapolated globally. The product of prob- landslide hazard, five parameters are used: ability and susceptibility determines the hazard value for each grid cell. The initial prediction of the avalanche 1. slope factor within a selected grid (Sr); hazard uses three parameters: 2. lithological (or geological) conditions (Sl); 3. soil moisture condition (Sh); 1. slope within a selected grid (Sr); 4. precipitation factor (Tp); and, 2. precipitation values for four winter months (Tp); and, 5. seismic conditions (Ts). 3. temperature values (Tt). The landslide/avalanche models were validated and refined on the basis of historical data, through selected Global Landslides Risk Case Study 23 Figure 2.1. General approach for landslide hazard and risk evaluation national case studies in Norway, Armenia, Nepal, then performing a regression analysis using different Georgia, Sri Lanka, and Jamaica. sets of uncorrelated socioeconomic parameters in order to identify the best indicators of human vulnerability for a selected hazard in a given country. According to Approach for Vulnerability and the UNDRO definition, a formula for estimating the risk Risk Assessment can then be derived as follows: The landslide vulnerability and risk assessment has been R = H · Pop · Vul performed by UNEP/GRID in Geneva, following the Where: approach of the recent UNDP World Vulnerability Report. R = Risk, that is, the number of expected human Based on the UN definition (UN Disaster Relief Coor- impacts (killed); dinator [UNDRO] 1979), risk is determined by three H = Annual hazard occurrence probability; components: hazard occurrence probability, elements Pop = Population living in a given exposed area; at risk, and vulnerability (see definitions above). and For risk estimation, the computation is based on Vul = Vulnerability, depends on socio- human losses as recorded by various natural disaster politico-economic context. impact databases. The estimation of expected losses is achieved by first combining frequency and population Defining physical exposure (PhExp) as the annual exposed, in order to provide the physical exposure, and frequency of a hazard with specified severity multiplied 24 Natural Disaster Hotspots Case Studies by the number of persons exposed (PhExp = H · Pop), Model for Snow Avalanche Hazard the risk can be evaluated by logarithmic regression using Any model for snow avalanche should include param- the following formula: eters describing the terrain and the snow. Steep terrain is a necessary condition for avalanches to occur. Snow ln(R) = ln(PhExp) + ln(Vul) cover conditions during the winter, snow precipitation, wind conditions, and temperature development during In the case of landslides, once the average physical a storm can result in snow avalanches. The magnitude exposure is computed on the basis of past events, an of these different parameters controls the avalanche size, estimate of risk can be made using a proxy of vulnera- run-out distance, and return period (probability of occur- bility. rence). It is difficult to produce a global avalanche hazard map based on all these factors, so simplifications must be made. Description of Model for Landslide and The probability of a given amount of precipitation Avalanche Hazard Evaluation can be related to return period through analysis of long- term data from weather stations. However, due to a Landslide Model paucity of global data from weather stations, this is not The model developed for the study is based on a method feasible. Available information is restricted to average proposed by Mora and Vahrson (1994). The method was annual precipitation, monthly precipitation, and max- modified for use with the datasets available for global imum daily precipitation. These data are insufficient for application. The landslide hazard level H is defined by the estimation of return periods and probabilities. The a combination of susceptibility and triggering factors: calculation of probability can be avoided through cal- ibration of a susceptibility map in countries where the H = SUSC * TRIG avalanche history has been known for many centuries. Therefore, it is possible to combine some of the sus- where "SUSC" is the intrinsic susceptibility factor deter- ceptibility grid values to the known consequences and mined from a combination of slope factor "Sr," lithol- return periods. The estimated return periods for a number ogy (bedrock geology) factor "Sl" and relative soil moisture of locations in a country with long-standing records factor "Sh"; and "TRIG," which represents the trigger- may then be used to estimate the probability for ava- ing factor that initiates rapid movement and its proba- lanches in other areas. bility of occurrence, is determined from a combination A specific weight was assigned to each factor in the of seismic activity indicator "Ts" and precipitation (rain- avalanche model. Multiplying and summing these indexes fall) indicator "Tp." determines a relative avalanche hazard level Havalanche, For each factor, an index of influence is determined given by: by a reference value through a specific weighting (a weight of 1 for all factors was used in the first-pass analy- Havalanche = (Sr*0.4 + Tp*0.4 + Tt*0.2)*F sis). Multiplying and summing these indexes determines the relative landslide hazard level Hlandslide, given by: where "Sr" is the slope factor, "Tp" is a factor that depends on precipitation for four winter months, "Tt" Hlandslide = (Sr * Sl * Sh) * (Ts + Tp) is the temperature factor, and "F" is a factor that depends on the average temperature in winter months (F = 0 if The range for each parameter in the above equation average monthly temperature in winter months > 2.5°C; is discussed in the section on "Sources of Data and F = 1 otherwise). Data Processing Procedures" below. Global Landslides Risk Case Study 25 Description of Model for Vulnerability and to the content of the pixel, although it is not needed Risk Assessment for purposes of analysis. The vulnerability and risk model is based on a raster Identification of Vulnerability in Socioeconomic approach. Details of the model are provided in Annex Context 2.A. A raster file depicts the average frequency of land- slides (resolution/frequency/area). In general terms, the Once the raster grid of frequencies is established, the spatial resolution (pixel size) should be as detailed as resulting dataset is multiplied by the population dataset. possible, taking into consideration the quality of input The product is aggregated, at the national level, in order data. to obtain the average number of persons exposed per The content of the pixel should represent either: year. Historic records of casualties are then compared with this measure of physical exposure and with a series · The average frequency of occurrence and the aver- of national socioeconomic parameters that have been age area of landslide in the pixel (if different from previously transformed and standardized. A logarith- the pixel area). The unit of landslide area in this study mic regression is then performed to identify which is approximately 1 km2; or socioeconomic parameters are best linked with number · A range of frequencies can be provided to reflect of casualties. Coefficients (weights) are also associated uncertainty. In this case, the average frequency is used with the different components of the expression: in the Base Case calculation, and the difference between min. and max. frequencies is used for the computa- K = C (PhExplandslides)a· V1a1 a2 ap · ·V2 ...· Vp tion of error margins. Where: Note that the probability of occurrence (such as 50 K = Number of persons killed by a certain type percent probability of occurrence in the next 100 years) of hazard; is transformed into annual frequency by assuming that C = Constant; the hazard occurrence follows Poisson's law, and there- PhExp= Physical exposure (population living in fore: exposed areas multiplied by the frequency E(x) = = ­ln(1 ­ P(x 1)) of occurrence of the landslides); Vi = Socioeconomic parameters; and Where: i = Exponent of Vi, which can be negative E(x) = "Statistical expectance," that is the average (for ratio). number per year = ; and P(x) = Probability of occurrence. This approach enables one to test the quality of the link between socioeconomic parameters and physical Information on class of severity, type of landslide (for exposure as factors explaining casualties for a given example, avalanche, rockfall, mudflow, debris slide, and hazard. It also provides useful information on what con- so on), potential intensity/magnitude could also be added Table 2.1. Description of variables Information (values) X*: Longitude of pixel center. Y*: Latitude of pixel center. Area*: Measure of area affected by landslide within the pixel (either in km2 or in percentage of pixel surface). Frequency*: Average number of landslides per year. Min. frequency: Minimum value for the frequency range. Max. frequency: Maximum value for the frequency range. Landslide type: Numerical code for avalanche, mudflow, rockfall, and so on. Severity class: Numerical code for class of severity, otherwise anticipated magnitude or intensity. * Required information for vulnerability analysis 26 Natural Disaster Hotspots Case Studies ditions both increase the susceptibility and induce greater Considerations influencing the estimation of param- vulnerability in a society. eters are described below. Estimation of slope factor Sr Computation of Vulnerability Proxy and The slope factor represents the natural landscape rugged- Identification of Population at Risk ness within a grid unit. The approach described above is qualitative and should Source: GlobalelevationdatasetSRTM30fromISciences. not be used as a predictive tool. A more quantitative approach involves computing a proxy of vulnerability. Web site: http://www.isciences.com/ This proxy is based on the ratio between the number Description of people killed and the number of persons exposed, In February 2000, NASA collected elevation data for given by: much of the world using a radar instrument aboard the K Vulaproxy = space shuttle that orbited the earth. Raw data have been PhExp processed over the past three years. NASA has now released a global elevation dataset called SRTM30, refer- Where: ring to the name of the mission and the resolution of Vul = Vulnerability proxy; the data, which is 30 arc-seconds, or approximately 1 K = Past casualties as recorded in CRED; and km2 per data sample near the equator. The SRTM30 PhExp = Physical exposure. dataset is NASA's latest achievement in improving the Once the vulnerability proxy is computed, it is mul- quality of digital elevation data available for public tiplied by the physical exposure to produce a risk map use. The data cover a range from 60 degrees south lat- on a pixel-by-pixel basis. This method is, however, a itude to 60 degrees north latitude. generalized approach that cannot take into account Classification the significant vulnerability differences between a Slope data are reclassified on a geographical grid rural and urban population. This limitation can only (WGS84). Cells are distributed in five different cate- be overcome through use of subnational datasets on gories (0­4), as follows: socioeconomic features and on geo-referenced infor- mation on the number of casualties. Such an analysis Table 2.2. Classification of slope factor "Sr" for would also require more records than would the analy- evaluation of susceptibility sis of average vulnerability derived from the national- Range of slopes angle (unit: degrees) Classification Sr level values. 00­01 Very low 0 01­08 Low 1 08­16 Moderate 2 Sources of Data and Data Processing 16­32 Medium 3 Procedures 32­75 High­very high 4 Note: Sr is set equal to zero for slope angles less than 1° (that is, Landslides for flat or nearly flat areas), because the resulting landslide hazard is null even if the other factors are favorable. As mentioned earlier, a simplified model, similar to that proposed by Mora and Vahrson (1994), was adopted Estimation of lithology factor Sl for the study. In this model, the relative landslide hazard This is probably the most important factor and the most level Hlandslide is estimated through the following equa- difficult to assess. Ideally, detailed geotechnical infor- tion: mation should be used, but, at the global scale, only a general geological description is available. Rock strength Hlandslide = (Sr * Sl * Sh) * (Ts + Tp) and fracturing are the most important factors used to evaluate lithological characteristics. Since fracturing Global Landslides Risk Case Study 27 may occur in most rock types and is a local feature, the Estimation of soil moisture factor Sh rock strength will be the most important factor on a Sh is a soil moisture index, which indicates the mean global scale. humidity throughout the year and gives an indication of the state of the soil prior to heavy rainfall and pos- Source: Geological map of the World at 1/25,000,000 sible destabilization. scale published by the Commission for the Geological Map of the World and UNESCO (2000). The map is Source: Data are extracted from Willmott and Feddema's available on a CD-ROM. Moisture Index Archive. They are produced and doc- umented by Cort J. Willmott and Kenji Matsuura, at the Description Center for Climatic Research, Department of Geogra- This map is the first geological document compiled at phy, University of Delaware, Newark, USA. a global scale showing the geology of the whole planet, including continents and oceans. In the map, three main Web site: http://climate.geog.udel.edu/~climate/ types of formation are identified: sedimentary rocks, html_pages/README.im2.html extrusive volcanic rocks, and endogenous rocks (plu- Description tonic or strongly metamorphosed). Data cover the standard meteorological period Classification 1961­1990. Resolution of the grid is 0.5, 0.5 degrees. Five susceptibility classes have been identified. Usu- The gridded, mean monthly, total potential evapotran- ally old rocks are stronger than young rocks. Plutonic spiration (Eo) and unadjusted total precipitation (P) are rocks will usually be strong and represent low risk. taken from: Strength of metamorphic rocks is variable, but these · Terrestrial Water Balance Data Archive: regridded rocks often have planar structures such as foliation monthly climatologies, and and therefore may represent higher risk than plutonic · Terrestrial Air Temperature, monthly precipitation, rocks. Lava rocks will usually be strong, but may be and annual climatologies. associated with tuff (weak material). Therefore, areas with recent volcanism are classified as high risk. Sedi- These data can be downloaded from the Web site. mentary rocks are often very weak, especially young Estimates of the average-monthly moisture indexes for ones. The susceptibility classes are shown in table 2.3. Eo and P are determined only for land-surface grid points. There are 85,794 points. Average-monthly mois- Table 2.3. Classification of lithology factor "Sl" for evaluation of susceptibility Lithology and stratigraphy Susceptibility Sl · Extrusive volcanic rocks--Precambrian, Proterozoic, Paleozoic, Archean. Low 1 · Endogenous rocks (plutonic and/or metamorphic)--Precambrian, Proterozoic, Paleozoic and Archean. · Old sedimentary rocks--Precambrian, Archean, Proterozoic, Paleozoic. Moderate 2 · Extrusive volcanic rocks--Paleozoic, Mesozoic. · Endogenous rocks--Paleozoic, Mesozoic, Triassic, Jurassic, Cretaceous. · Sedimentary rocks--Paleozoic, Mesozoic, Triassic, Jurassic, Cretaceous. Medium 3 · Extrusive volcanic rocks--Mesozoic, Triassic, Jurassic, Cretaceous. · Endogenous rocks--Meso-Cenozoic, Cenozoic. · Sedimentary rocks--Cenozoic, Quaternary. High 4 · Extrusive volcanic rocks--Meso-Cenozoic. · Extrusive volcanic rocks--Cenozoic. Very high 5 28 Natural Disaster Hotspots Case Studies ture indexes are calculated according to Willmott and tribution approach. The results were divided into five Feddema (1992) using the gridded average-monthly classes and show that the two highest classes (4 and 5) total Eo and P values, at the same resolution as the water- cover 5 percent of the accumulated precipitation. The balance fields. susceptibility classes are shown in table 2.5. Classification Table 2.5. Classification of precipitation trigger indicator Five classes for soil moisture index are determined as "Tp" shown in table 2.4. 100-year extreme monthly rainfall (mm) Susceptibility Tp 0000­0330 Low 1 Table 2.4. Classification of soil moisture factor "Sh" for 0331­0625 Moderate 2 evaluation of susceptibility 0626­1000 Medium 3 Soil moisture index 1001­1500 High 4 (Willmott and Feddema 2002) Susceptibility Sh > 1500 Very high 5 -1.0 -0.6 Low 1 The map of the estimated 100-year extreme monthly -0.6 -0.2 Moderate 2 -0.2 +0.2 rainfall in the world is shown in figure 2.3. Medium 3 +0.2 +0.6 High 4 +0.6 +1.0 Very high 5 Estimation of seismic trigger factor Ts The map of the global soil moisture index is shown in Source: Peak Ground Acceleration (PGA) with a 475- figure 2.2. year return period (10 percent probability of exceedance in 50 years) from the Global Seismic Hazard Program Estimation of precipitation trigger factor Tp (GSHAP). Estimation of Tp is based on the 100-year extreme Web sites monthly rainfall. http://www.gfz-potsdam.de/pb5/pb53/projects/en/ Source: Monthly precipitation time series (1986­2003) gshap/menue_gshap_e.html from the Global Precipitation Climatology Centre http://www.dwd.de/en/FundE/Klima/KLIS/int/GPCC/ (GPCC), run by Germany's National Meteorological Ser- GPCC.htm vice (DWD). Description Web site: http://www.seismo.ethz.ch/GSHAP GSHAP was launched in 1992 by the International http://gpcc.dwd.de Lithosphere Program (ILP) with the support of the Inter- national Council of Scientific Unions (ICSU) and in Description the framework of the United Nations International DWD is a German contribution to the World Climate Decade for Natural Disaster Reduction (UN/IDNDR). Research Program and to the Global Climate Observ- The primary goal of GSHAP was to create a global seis- ing System. The data used are near real-time monitor- mic hazard map in a harmonized and regionally coor- ing products based on the internationally exchanged dinated fashion, based on advanced methods in meteorological data (GTS) with gauge observations from probabilistic seismic hazard assessments (PSHA). Modern 7,000 stations worldwide. The products contain pre- PSHA are made of four basic elements: earthquake cat- cipitation totals, anomalies, the number of gauges, and alogue, earthquake source characterization, strong seis- systematic error correction factors. The grid resolution mic ground motion, and computation of seismic hazard. is 1.0° 1.0° latitude/longitude. For the purposes of this study, the PGA with a 475-year Classification return period was used. Monthly values are available for 17 years, from 1986 Classification to 2002. The maximum registered values per annum The GSHAP PGA475 data are distributed within 10 classes, were used to calculate the expected 100-year monthly as shown in table 2.6. precipitation for every grid point using a Gumbel dis- Global Landslides Risk Case Figure 2.2. Global soil moisture index: 1961-1990 Study 29 30 Figure 2.3. Expected monthly extreme values for a 100-years event. Natural 0 Disaster Hotspots 0 Case Studies Global Landslides Risk Case Study 31 Table 2.6. Classification of seismicity trigger indicator "Ts" An example of how the different layers of input param- PGA475 (m/s2) Ts eters interact to produce a landslide hazard map is shown in figures 2.5 to 2.9. Tajikistan and its neighboring 0.00 ­ 0.50 1 0.51 ­ 1.00 2 regions are shown in the example. 1.01 ­ 1.50 3 1.51 ­ 2.00 4 2.01 ­ 2.50 5 Snow Avalanche 2.51 ­ 3.00 6 3.01 ­ 3.50 7 Susceptibility factors for snow avalanches are the slope 3.51 ­ 4.00 8 factor (Sr), temperature (Tt), and precipitations (Tp). 4.01 ­ 4.50 9 Relative avalanche hazard level is computed through > 4.50 10 the following equation: Figure 2.4 shows the global map, developed in GSHAP, of the PGA with a return period of 475 years. Havalanche = (Sr*0.4 + Tp*0.4 + Tt*0.2)*F Classification of landslide hazard The value of relative landslide hazard level Hlandslide obtained from the equation given in the "Landslide Estimation of slope factor Sr Model" section varies between 0 and 1500. In the orig- Source: The data were derived from NASA's SRTM30 inal Mora and Vahrson (1994) model, the landslide dataset (see description above for the reference source hazard classification shown in table 2.7 is suggested. In under the landslide hazard model). this study, the classification shown in table 2.8 was used. Classification Table 2.7. Classification of landslide hazard potential In total, nine categories were defined, as shown in based on the computed hazard index originally table 2.9. The most interesting categories for avalanches suggested by Mora and Vahrson (1994) are categories 7, 8, and 9. Classification of landslide Values for Hlandslide Class hazard potential Table 2.9. Classification of slope factor "Sr" for snow < 6 1 Negligible avalanche susceptibility 7­32 2 Low Range of slopes angle (unit: degrees) Slope factor "Sr" 33­162 3 Moderate 163­512 4 Medium 0­1 1 513­1250 5 High 1­2 2 > 1250 6 Very high 2­4 3 4­6 4 6­9 5 Table 2.8. Classification of landslide hazard potential 9­12 6 based on the computed hazard index used in this study 12­17 7 Values for Hlandslide Class Classification of landslide 17­25 8 hazard potential > 25 9 < 14 1 Negligible 15­50 2 Very low 51­100 3 Low Estimation of precipitation factor Tp 101­168 4 Low to moderate 169­256 5 Moderate Source: Mean monthly precipitation data from the IIASA 257­360 6 Medium Climate Database (International Institute for Applied 361­512 7 Medium to high System Analyses, Austria). 513­720 8 High > 720 9 Very high Web site http://www.grid.unep.ch/data/summary.php?dataid= The annual frequencies of landslide events correspon- GNV14&category=atmosphere&dataurl=http://www.grid. ding to these classes are given in table 2.13. unep.ch/data/download/gnv14.tar.Z&browsen= 32 Figure 2.4. Expected PGA with a return period of 475 years Natural Disaster Hotspots Case Studies Global Landslides Figure 2.5. Variation of slope factor, Sr, in Tajikistan and its neighboring regions Risk Case Study 33 34 Figure 2.6. Variation of lithology factor, S , in Tajikistan and its neighboring regions Natural Disaster Hotspots Case Studies Global Figure 2.7. Variation of seismic trigger indicator, Ts, in Tajikistan and its neighboring regions Landslides Risk Case Study 35 36 Figure 2.8. Variation of soil moisture factor, Sh, in Tajikistan and its neighboring regions Natural Disaster Hotspots Case Studies Global Figure 2.9. Landslide hazard zonation map obtained for Tajikistan and its neighboring regions Landslides Risk Case Study 37 38 Natural Disaster Hotspots Case Studies Description Table 2.11. Classification of temperature factor "Tt" for Grid resolution is 0.5° 0.5° latitude/longitude. The avalanche hazard analysis precipitation values for the four "winter months" in Mean monthly temperature in the northern and southern hemispheres have been added winter months (°C) Temperature factor "Tt" to the database. The amount of snow during the winter 2.5 30.5 0 months greatly affects the number and size of the ava- 1.5 2.5 1 0.5 1.5 2 lanches. Global differences in the expected winter snow 0.0 0.5 3 are given by a global precipitation map for the winter -0.5 0.0 4 months (December­March in the northern hemisphere -1.0 -0.5 5 and June­September in the southern hemisphere). This -1.5 -1.0 6 -2.0 -1.5 7 map does not show the best picture of the situation -3.0 -2.0 8 due to differences in the number of winter days. In < -3.0 9 reality, the more north the location of interest, the more it underestimates avalanche susceptibility. Classification of avalanche hazard The value of avalanche hazard level Havalanche obtained Table 2.10. Classification of precipitation factor "Tp" for from the equation given in the beginning of this sec- avalanche hazard evaluation tion varies between 1 and 9. Winter Precipitation (mm/year) Precipitation factor "Tp" 0­50 1 Table 2.12. Classification of snow avalanche hazard 50­100 2 potential 100­200 3 Classification of avalanche 200­300 4 Values for Havalanche Class hazard potential 300­500 5 500­750 6 4 1 - 750­1000 7 4.1­4.5 2 - 1000­1500 8 4.6­5.0 3 - > 1500 9 5.1­5.5 4 Negligible 5.6­6.0 5 Low 6.0­7.0 6 Moderate 7.0­7.5 7 Moderate to high Estimation of temperature factor Tt 7.5­8.2 8 High Source of the data Mean monthly temperature data from 8.3­9 9 Very high the IIASA Climate Database (International Institute for Similar to the landslide hazard, the avalanche hazard is Applied System Analyses, Austria). also divided into nine classes. The classes for ava- Web site lanche hazard based on the computed value of Havalanche http://www.grid.unep.ch/data/summary.php?dataid=GN are shown in table 2.12. The annual frequencies of V15&category=atmosphere&dataurl=http://www.grid.u (major) avalanche events corresponding to these classes nep.ch/data/download/gnv15.tar.Z&browsen= are given in table 2.13. Description A global temperature map, with a resolution of 0.5° Global Landslide and Avalanche Hazard 0.5° latitude/longitude, constrains the avalanche areas The slide/avalanche hazard has been classified into nine to colder regions. Areas with average temperature in at classes; that is, each pixel is assigned a value varying least one winter month (for example, January in the from 1 to 9. The nine classes roughly correspond to northern hemisphere) in the temperature range +5°C the annual frequency of occurrence for a 1-km2 pixel to 0°C or colder were studied. In mountain areas shown in table 2.13. above 1,000 m, precipitation occurs as snow when the temperature at sea level is less than +5°C. The model implies that the areas with longer cold periods have a greater potential to produce avalanches. Global Landslides Risk Case Study 39 Table 2.13. Annual frequency of occurrence and typical Norway return period (in years) for different classes of landslide and avalanche hazard Norway, a land with soft clay deposits, steep mountains, Class Annual frequency of Typical return period for and deep fjords, regularly experiences landslides, mainly occurrence (%) serious events (in years) due to quick clays, rockfalls (if the rock masses fall into 1 Virtually zero Not relevant fjords, they can lead to potentially devastating tsunamis), 2 Negligible 100,000­1,000,000 and snow avalanches. Although Norway is by no means 3 Very small 50,000­250,000 as much of a "hotspot" in terms of risk as many other 4 Small 20,000­10,000 Asian and Latin American countries, there are many 5 0.0025­0.01% 10,000­40,000 6 0.0063­0.025% 4,000­16,000 observations of Norwegian slides that can be used to 7 0.0125­0.05% 2,000­8,000 evaluate the reliability of the prediction model. 8 0.025­0.1% 1,000­4,000 9 0.05­0.2% 500­2,000 Landslides and Rockfalls Pixels in classes 1­4 have been ignored for the analy- The prehistorical and historical maps of rock-avalanche ses. A serious slide event would involve 10 percent to events in figures 2.12 and 2.13 show that the highest 100 percent of a pixel area. rock-slide frequency--occurring in the high-risk areas The combined annual frequency of landslide and in Western Norway--is concentrated in the inner fjords, avalanche events is approximately the sum of the fre- and mainly at the bottoms of the fjords (the areas sur- quencies for each event. The approximation is valid rounding the innermost parts of the fjords). In a very because the probability numbers are very small: few cases, slides were observed closer to the coast. The results of the regional zonation carried out by P[L or A] = P[L] + P[A] ­ P[L] · P[A] P[L] + P[A] the Geological Survey of Norway (NGU) and illustrated with red boundaries in figure 2.13 agree well with the where P[L] is the annual probability of a major land- observations close to the bottoms of the several fjords. slide event and P[A] is the annual probability of a A few of the large rockfalls/rock slides were, however, major avalanche event. not predicted with the regional mapping, especially in the northeastern part of the area shown and midway between Molde and Aalesund. Comparisons of Model Predictions with Figure 2.14 presents the mapping predictions for Actual Slide Events landslide hazard (both landslides and rockfalls) in West- ern Norway, where the more hazardous areas are given This section presents the results of some of the land- a relative hazard value of 4 to 5 on a scale of 1 to 9 and slide, rockfall, and avalanche mapping analyses done are located close to the bottoms of fjords or the arms in the study, with comparisons of observed hazards from of fjords. Otherwise, the model predicts that most of Norway, Armenia, Nepal, Georgia, Sri Lanka, and Jamaica. Western Norway has a hazard value of 3, which repre- These six countries were selected because relevant sents a low hazard. The model is probably too simpli- data for comparison of model predictions with actual fied to be able to predict the type of rockfalls and slide events were available. landslides that occur in Norway. Examples of global hazard and risk maps obtained by the models developed in this study are shown in Snow Avalanches figures 2.10 and 2.11. Figure 2.10 gives a hazard map Figure 2.15 illustrates a snow avalanche hazard map for Central American and Caribbean countries, and obtained from the snow avalanche hazard model figure 2.11 shows a risk map for Central and South described earlier. Snow avalanches are frequent in West- America. ern Norway, especially in the mountains close to the fjords. Looking at the map in figure 2.15, there is a very good correlation between the predicted Avalanche Hazard Classes 7 (moderate to high) to 9 (very high) 40 Figure 2.10. Example landslide hazard map for Central American and Caribbean countries Natural Disaster Hotspots Case Studies Global Figure 2.11. Example landslide risk map for parts of Central and South America Landslides Risk Case Study 41 42 Figure 2.12. Historical rock avalanche events in Møre & Romsdal and Sogn & Fjordane Counties extracted from Norway's historical database (NGU/Astor Furseth) Natural Disaster Hotspots Case Studies Figure 2.13. Regional hazard zonation in Møre & Romsdal County in western Norway. The hazard zones are characterized by the occurrence of a high number of both historical events and rock avalanche deposits. Global Landslides Risk Case Study 43 44 Figure 2.14. Landslide hazard map (landslide and rock fall hazards) for the western part of Norway based on the simplified model Natural Disaster Hotspots Case Studies Global Figure 2.15. Snow avalanche hazard zones for Norway based on the avalanche hazard model Landslides Risk Case Study 45 46 Natural Disaster Hotspots Case Studies and the areas where frequent occurrence of snow ava- Landscan 2001, Digital Chart of the World, and GLOBE lanches in Norway is observed. There may be other areas data. See Annex 2A for more details). GEORISK pro- with a high occurrence of snow avalanches, as not all vided NGI with the following information: of the territory is covered by observations; still, the agree- · Historical landslides; ment is very promising. · Landslide-prone zones: regions where landslide processes develop, regions of creep motion of the Armenia ground, regions of intense landslide processes, and regions of large seismic activity that involve the Landslides most hazardous landslides; Armenia (figure 2.16) is one of the most disaster-prone · Mudflows: Levels I, II, and III; countries in the world, as it features earthquakes, · Dams: high and low landslide hazard; landslides, hailstorms, droughts, strong winds, and · Population density in a 5 5 km grid; and floods. The average value of direct damage caused by · Population figures for cities, districts, and villages. landslides approaches US$10 million per year, affect- ing the social and economic infrastructure (Stephanyan Figure 2.17 presents the superimposition of the 2003). GEORISK landslide inventory (blue curves) onto the More than 3,000 large landslides have been reported global landslide hazard map obtained with the first-pass for Armenia, and one-third of the country is exposed model in this study. to landslide hazards. Nearly 470,000 people are exposed, Especially for the areas in the center of the region which represents about 15 percent of the total popula- mapped, the agreement between the NGI prediction tion. In five years, more than 2,000 families have been and the GEORISK inventory is very good. The NGI pre- left homeless as a result of landslide activity. The diction model assigns landslide values of between 4 and potential for future catastrophic landslide events is very 6 (a scale of 1 to 6, where 6 is the highest hazard, was significant. used in the previous study for Armenia) to all the land- Several landslide areas or groups of landslides in slide zones identified by GEORISK. The higher-hazard Armenia are considered to be the most dangerous for zones are well delimited by the areas characterized as the population: Vokhchaberd-Garni, Dilijan-Gosh, Aga- most susceptible to slides (values of 5 and 6). However iargan, Jermuk, Sunik (Sissian-Tolors), and Vanadzor. the NGI prediction model does not show the hazard Nearly 300 of the largest landslides are in an active stage area close to Yerevan, and can only indicate the south- of development. They include an area of about 700 km2, ern periphery of the hazard zone close to Azerbaijan involving 100 settlements, where nearly 400,000 people identified by GEORISK. live. About 1,500 km, of a total of 8,000 km of trans- port corridors in Armenia, are located in landslide-prone terrain. A typical huge landslide area covers a few km2. Nepal In some instances, a village with a population from a Landslides few hundred to a few thousand inhabitants is situated Data on observed landslides in Nepal were provided in an active landslide area. A typical landslide exhibits by Professor Narenda Raj Khanel of Tribhuwan Uni- a slow, creeping movement, with a thickness between versity in Katmandu (personal communication). 10 m and 100 m, and several, smaller, active creeping The results of the mapping of hazards in Nepal are zones inside the area. The ground movements are hor- given in figures 2.18 and 2.19. Figure 2.20 presents a izontal, vertical, and rotational, causing tension cracks demographic map of Nepal with population density in the ground, settlements, and rotational-slip surfaces. illustrated in different colors. A large proportion of the NGI previously produced a landslide hazard map for country has very low population density. Armenia with support from the Armenian Scientific Figure 2.18 plots all of the observed landslides in Research Company, GEORISK, and computations based Nepal between 1971 and 2000. Figure 2.19 presents on several datasets available on the Internet (Landsat7, the landslide hazard map predicted by the NGI model Global Figure 2.16. Map of Armenia Landslides Risk Case Study 47 48 Figure 2.17. Comparison of global landslide hazard mapping in Armenia using NGI model with the GEORISK landslide inventory Natural Disaster Hotspots Case Studies Global Landslides Risk Case Study 49 Figure 2.18. Major landslide events in Nepal during a 30-year time period (1971­2000) Figure 2.19. Landslide hazard in Nepal predicted by the NGI model in this study 50 Figure 2.20. Population density map of Nepal in 1995. Numbers refer to population count in a 2.5' x 2.5' grid cell. Natural Disaster Hotspots Case Studies Global Landslides Risk Case Study 51 developed in the present study. By superimposing the 80­100 mm means there is a high probability of debris two maps, one can conclude the following: flow activation. Figure 2.21 shows the landslide hazard predicted · The model defines an approximate band featuring by the model developed in the present study. medium and medium-to-high hazard, which is gen- The model in the present study can predict quite well erally consistent with observations of landslides. the areas with "strong" and "high" landslide hazard · There is good agreement between the prediction map susceptibility as suggested by the Georgian Geophysi- and the observations in defining a narrow band at cal Society, especially the general west-southeast trend the western end of Nepal with medium and medium- and the highly susceptible areas in southwest Georgia. to-high hazard. The model missed the "strong" hazard areas in the center · The area with the highest density of landslides (the of Georgia, and could not detect the "high" hazard area mid-section of the country) is only approximately in southeastern Georgia. identified by the prediction model. On the other hand, the prediction model in this study · The model predicts a high hazard area in the north- suggests a higher landslide hazard in the center of Geor- northeastern part of the country, where few landslide gia, assigning "medium to high" and even "high" hazard events have been registered. This could be explained labels, while the Georgian Geophysical Society charac- by the extremely low population density in that area, terizes the area as moderate to weak, with just a few which implies that landslides occurring in that region indentations with high landslide hazard. would go unnoticed and unreported. In summary, the prediction model yields a good first- Snow Avalanches pass approximation. Refinements would be needed to Most of the avalanches (70 percent of them) in Georgia enable the model to produce hazard maps that can are triggered from January to March. The probability capture additional features of the landscape as well as for avalanche occurrence is high if the snow cover geological and meteorological characteristics, and thus thickness is 1 m or more. During the last 30 years, the increase the model's ability to predict landslide sus- danger of avalanches has increased due to uncontrolled ceptibility in a reliable manner. forest harvesting activities in the Caucasian mountains. Increased avalanche activity was recorded in 1971, 1976­77, 1986­89, and 1996­97. The 1987 and 1989 Georgia winters were marked by extreme avalanche activity. Information on landslide and snow avalanche hazards In January 1987, Western Georgia experienced a in Georgia was obtained from the Georgian Geophysi- cyclone intrusion that covered the mountains in a cal Society Web site (http://www.ggs.org.ge/others- thick layer of snow (up to 3­5 m in the Svanety natural.htm). region). This led to the triggering of some very large avalanches, resulting in dozens of fatalities, destruc- Landslides tion of hundreds of buildings, and damage to infra- As expected for a mountainous country, Georgia is prone structure and lifelines. to massive landslides, debris flows, and mudflows. There Figures 2.22 shows the snow avalanche hazard pre- are some 10,000 potential landslide sites, 3,000 of which dicted by the model developed in the present study. are very active. Most of the active landslide sites are W hereas the Georgian Geophysical Society map- located in Western Georgia, where the climate is humid. ping on its web site characterizes the northern part of Most landslides and debris flows in Georgia are Georgia as having a "moderate" to "high" snow ava- triggered by heavy rainfall. The landslide activity in- lanche hazard, the prediction model developed in this creases when the accumulated annual precipitation study characterizes the same area as "moderate" only, exceeds the mean annual value by 200­400 mm. Sta- except the westernmost part of the country, which is tistical analyses show that 85 percent of debris flows categorized as having a "high" avalanche hazard. The originate after intense rain; a daily precipitation of trend showed by the two mappings is very similar. 52 Figure 2.21. Landslide hazard in Georgia predicted by the model developed in this study Natural Disaster Hotspots Case Studies Global Landslides Risk Case Study 53 Figure 2.22. Snow avalanche hazard in Georgia predicted by the model developed in this study A limited extent of the southwest part qualified as lation between the location of the observed landslides "low to moderate," which is also well predicted by the and the "relative" hazard higher class predicted by the new prediction model. The same remark also applies model at the same location. This example is a good to the zones with a minimal snow avalanche hazard. application of the global model to a region that, on a In summary, the model seems quite reliable for the global map, is not considered high risk. To make the mapping of both landslide and snow avalanche hazards results of mapping as meaningful as possible, the pre- in Georgia. diction in such regions would require local calibration and validation. Sri Lanka Jamaica Landslides Figure 2.23 presents a comparison between the obser- Landslides vations of landslides in Sri Lanka over a 57-year period Figures 2.24 and 2.25 compare historical landslide using NBRO data, with the landslide hazard map pre- observations in Jamaica with the landslide hazard dicted by the model developed in this study. The land- mapping predicted by the model developed in this study. slides that have occurred are shown as punctual There is, in general, a good correlation between the observations. The landslides group around the south- locations of historical slides and the hazard classes pre- ern, central part of Sri Lanka. dicted by the model. In particular, the model predicts Notwithstanding that the predicted hazard is low to well the large occurrence of landslides in eastern Jamaica negligible on a global basis, there is an excellent corre- and the extent of vertical hazard zones in the southern 54 Natural Disaster Hotspots Case Studies Figure 2.23. Observed landslides in Sri Lanka between 1947 and 2003 (a) and prediction of landslide hazard in Sri Lanka by the model developed in this study (b). center of Jamaica. The model could not, however, detect development banks and international agencies work- the frequent landslides on the western part of Jamaica, ing in developing countries. nor in the center of the country. Figures 2.26 to 2.28 illustrate the results obtained with the first-pass model for landslide hazard devel- oped in this study. Results of Global Analyses In figure 2.26, the hotspots are identified on a world map. The regions are characterized by landslide Hotspots for Landslide Hazards hazards between negligible and very high (white to red). The main areas with moderate to very high landslide Landslides contribute to major disasters every year on hazards include: a global scale, and the frequency of occurrence is on an upward trend. The increasing number of landslide dis- asters can be attributed in large part to the new reality · Central America of more extreme weather conditions combined with · Northwestern South America overexploitation of natural resources and deforestation, · Northwestern USA and Canada increased urbanization, and uncontrolled use of land. · Hawaii Recent examples are the mudflows of December 1999 · Antilles in Venezuela, involving over 20,000 deaths, and the El · The Caucasus region Salvador earthquakes of 2001, which caused 600 deaths · The Alborz and Zagros mountain ranges in Iran in just one landslide. Allocating resources for natural · Turkey hazard risk management is a high priority among the · Ukraine Global Landslides Risk Case Study 55 Figure 2.24. Historical landslide data in Jamaica (after Professor R. Ahmad) Figure 2.25. Prediction of landslide hazard in Jamaica with the model developed in this study 56 Figure 2.26. Global hotspot landslide hazar Natural Disaster Hotspots Case Studies Global Landslides Risk Case Study 57 · The Himalayan belt Results of Risk Analysis · Taiwan · Philippines and Celebes Hotspots for Landslide Risk · Indonesia The model used for evaluation of landslide risk was · New Guinea described earlier. Details of the model are provided in · New Zealand Annex 2.A. The equation below provided the most stable · Italy correlation between number of expected fatalities due · Iceland to landslides and socioeconomic parameters: · Japan · Kamtchatka 1n(K) = 0.661n (PhExp_all) + 0.701n (FCpc) + These areas are discussed further in the broader com- 0.361n(AR_Land) ­ 2.441n (HDI) ­ 14.98 panion report, Natural Disaster Hotspots: A Global Risk Analysis (Dilley et al. 2005). Where: Figure 2.27 provides a more detailed mapping for K = The expected number of fatalities due Central Asia and the Middle East. Countries with to landslides; "medium to high," "high," and "very high" landslide PhExp_all = Physical exposure including all classes; hazard scores include: FCpc = The transformed percentage of forest in the country; · Georgia HDI = Transformed Human Development · Armenia Index; and · Turkey Ar_Land = The percentage of arable land. · Lebanon · Iran Although around 73 percent of the variation is · A small part of southern Russia explained by the regression, one has to keep in mind · Tajikistan that this is not a predictive model, mostly because log- · Kyrgyz Republic arithmic regression prevents the use of "zero fatalities" · Afghanistan in the analysis. However, the model can still be used to · Nepal better understand the socioeconomic context of vul- · India nerability and risk, and allows a differentiation of the · Pakistan classes of countries at risk. · Southern China Results of the analysis confirm the relevance of Figure 2.30 presents the landslide hazard zonation identification of physical exposure (landslide hazard for Central American and Caribbean countries, where model). Nearly 98 percent of the recorded landslide vic- the following countries are mapped with "medium to tims lived in countries affected by landslide classes 5 high," "high," and "very high" landslide hazard scores. and higher. The process is validated by the good cor- relation observed between independent datasets such · Mexico as reported casualties in CRED and frequencies of · Guatemala landslides as computed by the model described in the · El Salvador report, together with national socioeconomic parame- · Honduras ters (such as HDI). · Nicaragua The risk evaluation study reveals that some countries · Costa Rica with recorded casualties do not have a high physical · Panama exposure. Issues related to frequency in different cli- · Colombia mate regimes, and vegetation cover, might explain such · Ecuador discrepancies and could represent interesting topics · Peru for future studies. 58 Figure 2.27. Global hotspot landslide hazard zonation for Central Asia and the Middle East Natural Disaster Hotspots Case Studies Global Figure 2.28. Global hotspot landslide hazard zonation for Central American and Caribbean countries Landslides Risk Case Study 59 60 Natural Disaster Hotspots Case Studies Conversely, countries with no recorded casualties est avalanche hazard value) include the border of over the 21-year period considered in this study Georgia and Russia, Tajikistan, Afghanistan, and the cannot be evaluated using the method of vulnerability Kyrgyz Republic. Each of these countries has areas proxy. Data for a longer time period should be obtained. with a "high" landslide hazard. The same countries, plus The strong correlation between high physical expo- Turkey, the Islamic Republic of Iran, Pakistan, India, sure and low HDI, with high risk, is relatively straight- Uzbekistan, and Kazakhstan, are characterized as "mod- forward to explain. However, the correlation between erate" snow avalanche hazard regions. high percentage of forest and high landslide risk is more difficult to explain. This high correlation might be due to the effects of deforestation on susceptibility to land- Recommendations for Further Studies slides, which manifests itself indirectly through the "per- centage of forest" parameter. Alternatively, countries The study presented in this report was a first-pass analy- with more forest coverage are likely to have more pre- sis intended to identify global landslide hazard and land- cipitation, and the effect of heavy precipitation is not slide risk hotspots, with an emphasis on developing adequately covered through the physical exposure param- countries. The maps developed represent first-order eter for these countries. The analyses demonstrate the identification of the geographic areas that constitute need for data on deforestation, which might improve global landslide disaster hotspots. The probability of the model and further explain vulnerability. landslide occurrence was estimated from modeling phys- Figures 2.29 and 2.30 illustrate some of the typical ical processes and combining this information with his- results obtained from the risk analyses. torical observations and geological characteristics. Figure 2.29 presents a map of the landslide risk in Rockslides, landslides, and snow avalanches were Central America and Jamaica. One can observe that, included in the study. The model was evaluated by com- with the model used, Colombia is the only country with paring observations of the intensity and frequency of greater than 10-2 risk of persons killed per year per square sliding events. The resulting landslide and avalanche kilometer. Mexico, Guatemala, El Salvador, Honduras, hazard maps constitute the input to the global hotspots Costa Rica, Panama, and Colombia show areas with risk multihazard analysis in the companion report by between 10-2 and to 10-3. Fairly large regions in nearly Dilley et al. (2005). every country show rather large areas with land- The model developed and the methodology used in slide risk of 10-3 to 10-4. The highest risk of persons the study can be improved. The basic input data for killed per year per square kilometer in Jamaica is 10-3 the models could also be augmented and made more to 10-4. reliable. The following factors contributed to uncer- Figure 2.30 presents a similar landslide risk map for tainty in the results of the study: Central Asia. In this case, Tajikistan, India, and Nepal show greater than 10-2 risk of persons killed per year · Scarcity of high-quality, high-resolution data on a per square kilometer. The same countries, plus global scale; Afghanistan and the Islamic Republic of Iran, show areas · Lack of a good-quality database and inventory of with risk between 10-2 and to 10-3. Only a few other landslides for statistical analysis; countries show areas with landslide risk of 10-3 to 10-4. · Meaningful measure of terrain topography for a 1 km 1 km grid cell; and, · A reliance on proxies when desired information is Hotspots for Snow Avalanche Hazard rarely available. How good are these proxies? In the same manner, it is possible to develop global snow Further studies are recommended on the following avalanche hazard maps. Figure 2.31 illustrates such global issues: results obtained for Central Asia with the simple snow avalanche prediction model developed in this study. · Application of more sophisticated theoretical models The more susceptible countries (those with the high- for evaluation of landslide hazards; Global Figure 2.29. Hotspot landslide risk zonation for Central America and Jamaica Landslides Risk Case Study 61 62 Figure 2.30. Hotspot landslide risk zonation for Central Asia Natural Disaster Hotspots Case Studies Global Figure 2.31. Global hotspot snow avalanche hazard zonation for Central Asia Landslides Risk Case Study 63 64 Natural Disaster Hotspots Case Studies · More focused regional and international studies to References calibrate and fine-tune the models for different regions of the world; Blikra, L.H, A. Braathen, and E. Skurtveit. 2001. NGU Report · Development of better databases of landslide inven- 2962.01. 108. Hazard Evaluation of Rock Avalanches, the Bar- tory, fatalities caused by landslides, and economic aldsnes Area. The Geological Survey of Norway, Trondheim. consequences of landslides, at both the national and (1NGU, 2NGI). international levels. Dilley, M., et al. 2005. Natural Disaster Hotspots: A Global Risk · Direct evaluation of the economic risk associated with Analysis. Washington, DC: International Bank for Recon- landslides and avalanches; and struction and Development. · More detailed analysis of the effects of deforestation Leemans, R., and Wolfgang P. Cramer, 1991. The IIASA Data- and vegetation cover on landslide hazard and vul- base for Mean Monthly Values of Temperature, Precipitation nerability. and Cloudiness of a Global Terrestrial Grid. Laxenburg, Aus- tria: IIASA, RR-91-18. Mora, S., and W. Vahrson. 1994. Macrozonation Methodology for Landslide Hazard Determination. Bulletin of the Associa- tion of Engineering Geologists 31(1): 49­58. Stephanyan, M., 2003. Regional Study. Managing Natural Dis- asters in Armenia. The World Bank Perspective. Country Risk Template.Yerevan, Armenia. Willmott, C. J., C. M. Rowe, and Y. Mintz. 1985. Climatology of the Terrestrial Seasonal Water Cycle. Journal of Climatology 5: 589­606. Willmott, C. J., and J. J. Feddema. 1992. A More Rational Cli- matic Moisture Index. Professional Geographer 44(1): 84­88. Global Landslides Risk Case Study 65 Annex 2.A-- Working Definitions and Formulae Risk and Vulnerability Hazards, Vulnerability, and Risk-- Identification for Landslides Definitions and Concepts The terminology used in this study is drawn from the Results and Conclusions from Statistical Analysis UN and other experts. The definitions of the concepts are provided in the following paragraphs: This appendix describes the method and results from the statistical analysis carried out to depict vulnerabil- · Risk: "The term risk refers to the expected losses from ity and approach the risk of casualties caused by land- a particular hazard to a specified element at risk in a slides. The research on landslide hazards was undertaken particular future time period. Loss may be estimated in by the Norwegian Geotechnical Institute (NGI). The terms of human lives, or buildings destroyed or in finan- team from UNEP/DEWA/GRID-Europe first computed cial terms" (UNDRO 1979; in Burton et al. 1993, the physical exposure and then attempted to identify p.34). the socioeconomic context that leads to higher vulner- Specificity in this research: The term "risk" is ability. The research provides interesting results and used to describe potential human losses (casualties) clear connections between socioeconomic context and resulting from an expected future hazard. vulnerability. The method used in this study is based · Hazard: "The hazard can be defined as a potential threat on the methodology developed during the project Global to humans and their welfare" (Smith 1996). The haz- Risk and Vulnerability Index Trend per Year (GRAVITY). ardous events vary in terms of magnitude as well as This project was described in the technical report in "frequency, duration, area extent, speed of onset, spa- (Peduzzi, Dao, Herold, Mouton (2002 and 2003)), which tial dispersion, and temporal spacing" (Burton et al. was made for the UNDP/BCPR and published in the 1993, p.34). report, Reducing Disaster Risk: A Challenge for Develop- Specificity in this research: Only frequencies and ment (UNDP 2004). area extent are considered in the model. Two different approaches were used for risk evalua- tion. The first method was based on observed casual- · Physical Exposure: "Elements at risk, an inventory of ties divided by physical exposure to map landslide risk those people or artifacts which are exposed to the hazard" distribution. Although this method allows for quan- (Coburn et al. 1991, p. 49). tification of vulnerability, it doesn't explain why one Specificity in this research: Computation of aver- population is more vulnerable than another. The second age population annually exposed to landslides. In method used selected socioeconomic parameters through this research the element at risk is the population. a statistical analysis in order to identify what particu- · Vulnerability: "Reflects the range of potentially dam- lar socioeconomic parameters lead to higher vulnera- aging events and their statistical variability at a partic- bility. The least developed and forested countries with ular location" (Smith 1996). "The degree of loss to each high physical exposure were identified as being the most element should a hazard of a given severity occur" (Coburn at risk. These first results are encouraging, but also et al. 1991, p. 49). demonstrate that further work is needed on the iden- Specificity in this research: The discrepancies of tification of frequencies for countries that include vic- casualties induced by different vulnerabilities are used tims but were not selected in the physical exposure. to identify socioeconomic indicators reflecting such This study also highlights the necessity to obtain accu- vulnerabilities. rate and relevant data on deforestation. 66 Natural Disaster Hotspots Case Studies By UN definition (UNDRO 1979), the risk is result- ing from three components: One way of estimating the risk is to look at impacts from previous hazardous events. The physical exposure "Hazard occurrence probability, defined as the proba- can be obtained by modeling the area extent affected bility of occurrence of a specified natural hazard at a spec- by one event. Using the area affected, the figure repre- ified severity level in a specified future time period, elements senting exposed population can be extracted using a at risk, an inventory of those people or artifacts which are Geographical Information System (GIS); the popula- exposed to the hazard and vulnerability, the degree of loss tion affected multiplied by the frequency provides the to each element should a hazard of a given severity occur" physical exposure. The identification of parameters lead- (Coburn et al. 1991, p. 49). ing to higher vulnerability can then be carried out by replacing the risk in the equation by casualties reported Formula and Method for Estimating Risk and in EM-DAT from CRED and running a statistical analy- Vulnerability sis for highlighting links between socioeconomic param- The formula used for modeling risk combines the eters, physical exposure, and observed casualties. three components of the UNDRO definition (UNDRO 1979): the risk is a function of hazard occurrence Computation of Physical Exposure probability, element at risk (population), and vulnera- bility. The following hypothesis was made for model- General Description ing the risk: the three factors explaining risk are In broad terms, the physical exposure was estimated multiplying each other. by multiplying the hazard frequency by the population living in the exposed area. The frequency of hazard R = H · Pop · Vul1 was derived for different strengths of events, and the physical exposure was computed according to the equa- Where: tion below: R = The risk, that is, the expected human impacts PhExpnat = F i·Popi (expected number of killed people); H = The hazard, which depends on the frequency Where: and strength of a given danger; PhExpnat = The physical exposure at the national level; Pop = The population living in a given exposed Fi = The annual frequency of a specific mag- area; nitude event in one spatial unit as pro- Vul = The vulnerability and depends on socio- vided by NGI; and politico-economic context of this population. Popi = The total population living in the spatial unit (divided by 10, following NGI's rec- From the previous discussion, the physical expo- ommendations). sure is defined as the combination of both frequency and population exposed to a given magnitude for a selected The frequencies used were the ones of classes 2 and type of hazard. The hazard multiplied by the popula- higher as described in table 2.A.1. tion can then be replaced by the physical exposure: Table 2.A.1. Classes of frequencies R = PhExp · Vul Class Annual frequency Typical return period for of occurrence (in %) serious events (year) Where: 9 0.05­0.2 500­2,000 PhExp = The physical exposure, that is, the fre- 8 0.025­0.1 1,000­4,000 7 0.0125­0.05 2,000­8,000 quency and severity multiplied by exposed 6 0.0063­0.025 4,000­16,000 population. 5 0.0025­0.01 10,000­40,000 4 0.001­0.005 20,000­100,000 1.The model uses a logorithmic regression; the equation is similar but with 3 0.0004­0.002 50,000­250,000 an exponent for each of the parameters. 2 0.0001­0.001 100,000­1,000,000 Global Landslides Risk Case Study 67 The total population annually exposed is slightly PhExpav = Kic PhExpic · higher than 4,350. This feature is very similar to the Ktot average number of people killed per year worldwide Where: (1,727). This is a good sign for the quality of physical PhExpav = Average physical exposure pondered exposure. Ninety-eight percent of the recorded victims by the casualties; are within the countries affected by landslides of class Kic = Killed from landslides for the year "i" 5 and over. The remaining 2 percent of casualties happen and the country "c"; in countries affected only by classes 2 to 4. PhExpic = Physical exposure for the year "i" and Extraction of population was based on the CIESIN, the country "c"; and IFPRI, and WRI Gridded Population of the World (GPW, Ktot = Total number killed from landslides Version 2) at a resolution of 2.5'2 (equivalent to 5 5 for the selected country. km at the equator). This layer was further completed by the Human Population and Administrative Bound- aries Database for Asia (UNEP) for Taiwan and the Identification of Risk to Landslide CIESIN Global Population of the World Version 2 (coun- Risk Distribution Using Vulnerability Proxy try-level data) for the former Yugoslavia. These datasets A quantitative approach can be used by computing a reflect the estimated population distribution for 1995. proxy of vulnerability. This proxy is based on the ratio Since population growth is sometimes very high in the between the number of people killed and the number 1980­2000 period, a correction factor using country of people exposed (see equation below). totals was applied in order to estimate current physical exposures for each year as follows: K Vulaproxy = PhExp PhExpi = Popi · PhExp1995 Where: Pop1995 Vul = Vulnerability proxy; Where: K = Past casualties as recorded in CRED; PhExpi = The physical exposure of the current and year; PhExp = Physical exposure: population living Popi = The population of the country at the cur- in exposed areas multiplied by the rent year; frequency of occurrence of the Pop1995 = The population of the country in 1995; landslides. and All three parameters can be at a certain power. PhExp1995 = The physical exposure computed with population in 1995. Once the vulnerability proxy is computed, it can be multiplied by the physical exposure to produce a risk To take into account the increase of population (hence, map at the pixel level. This method is, however, a gen- the increase of physical exposure), an average physical eralized approach that cannot take into account the exposure using the number of casualties is then com- significant vulnerability differences between rural and puted to better reflect the situation at the time the events urban populations. This limitation can only be over- occurred. The formula is similar to the one used to trans- come by the use of subnational datasets on socioeco- form socioeconomic values. nomic features and on geo-referenced information on the number of casualties. Such analysis would also 2.GPW2 was preferred to the ONRL Landscan population dataset, despite its five-times-lower spatial resolution (2.5' against 30"), because the orig- require more records than would the analysis of aver- inal information on administrative boundaries and population counts is age vulnerability derived from the national-level values. almost two times more precise (127,093 administrative units against 69,350 units). Furthermore the Landscan dataset is the result of a complex The data for the number of victims from the two most model that is not explained thoroughly and that is based, among other extreme events (Venezuela, 1999, and Colombia, 1985) variables, on environmental data (land-cover), making it difficult to use were removed because the events' magnitudes were for further comparison with environmental factors (circularity). deemed incompatible with the other events. 68 Natural Disaster Hotspots Case Studies Note: The limitation of such a method is that coun- · Relevance: vulnerability factors (outputs-orientated), tries without reported casualties are not represented in resulting from the observed status of the popula- the risk analysis. Physical exposure involving all the tion, not based on mitigation factors (inputs, action classes had to be taken into account in order to have taken). Example: school enrollment rather than edu- some measure of physical exposure in countries where cation budget. casualties are reported. Data quality and availability: data should cover the 1980­2000 period and most of the 249 countries and Examples of Distributions territories. Situations in territories were separated from those in countries; for example, the situation in Identification of Vulnerability in the Socioeconomic Martinique is valid only for this island and is not taken Context into account for computing socioeconomic average at Once the raster grid of frequencies is established, this the national level (France). dataset is multiplied by a dataset of population. The result is aggregated at the national level in order to obtain the average number of persons exposed per year. The Statistical Analysis: Methods and Results figures from past casualties are then compared with Defining a Multiplicative Model physical exposure and with a series of national socioe- The statistical analysis is based on two major hypothe- conomic parameters that have been transformed and ses. First, that the risk can be approached by the number standardized. A logarithmic regression is then performed of victims of past hazardous events. Second, that the to identify which socioeconomic parameters are best equation of risk follows a multiplicative model: linked with the casualties. Coefficients (weights) are also associated with the different components of the K = C (PhExp)a· V1 a1 a2 ap · ·V2 ...· Vp expression (see equation below): Where: K = C (PhExplandslides)a· V1 a1 a2 ap · ·V2 ...· Vp K = The number of persons killed by a certain type of hazard; Where: C = The multiplicitive constant; K = The number of persons killed by a certain PhExp = The physical exposure: population living type of hazard; in exposed areas multiplied by the fre- C = The multiplicative constant; quency of occurrence of the hazard; PhExp = The physical exposure: population living Vi = The socioeconomic parameters; and in exposed areas multiplied by the fre- i = The exponent of Vi , which can be nega- quency of occurrence of the landslide; tive (for ratio). Vi = The socioeconomic parameters; and i = The exponent of Vi, which can be nega- Using the logarithmic properties, the equation could tive (for ratio). be written as follows: This will enable a test of the quality of the link between the socioeconomic contextual parameters and physical 1n(K) = 1n(C) + 1n(PhExp) + 11n(V1) + exposure, both of which are features that help to explain 21n(V2) + ... p1n(Vp) casualties. It also provides useful information on what conditions increase societal susceptibility to land- This equation provides a linear relation between slides. Table 2.A.2., vulnerability indicators, provides logarithmic sets of values. Significant socioeconomic a list of socioeconomic parameters that were used in the parameters Vi (with transformations when appropriate) analysis. and exponents i could be determined using linear The factors considered for the analysis were selected regressions. according to the following criterion: Global Figure 2.A.1. Distribution of risk utilizing a vulnerability proxy in Central America Landslides Risk Case Study 69 70 Figure 2.A.2. Distribution of risk using a vulnerability proxy in South America Natural Disaster Hotspots Case Studies Global Figure 2.A.3. Distribution of risk utilizing a vulnerability proxy in Central Asia Landslides Risk Case Study 71 72 Natural Disaster Hotspots Case Studies Table 2.A.2. Vulnerability indicators Categories of vulnerability Indicators Source1 Economic Gross Domestic Product per inhabitant at purchasing power parity WB Human Poverty Index (HPI) Total debt service (% of the exports of goods and services), UNDP Inflation, food prices (annual %), WB Unemployment, total (% of total labor force) ILO Type of economical activities Percentage of arable land FAO Percentage of urban population UNPOP Percentage of agriculture's dependency for GDP WB Percentage of labor force in agricultural sector FAO Dependency and quality of the Forests and woodland (in percentage of land area), FAO environment. Percentage of irrigated land Human Induced Soil Degradation (GLASOD) FAO UNEP Demography Population growth, UNPOP Urban growth, GRID2 Population density, GRID3 Age dependency ratio, WB Health and sanitation Average calorie supply per capita, FAO Percentage of people with access to adequate sanitation, WHO/ UNICEF Percentage of people with access to safe water (total, urban, rural) WHO/ UNICEF Number of physicians (per 1,000 inh.), Number hospital beds WB Life expectancy at birth for both sexes WB Under five years old mortality rate UNPOP UNPOP Politic Index of Corruption WB Early warning capacity Number of Radios (per 1,000 inh.) WB Education Illiteracy Rate, WB School enrolment, UNESCO Secondary (% gross), UNESCO Labor force with primary, secondary or tertiary education WB Development Human Development Index (HDI) UNDP Risk Victims (killed by landslides) CRED 1.FAOSTAT (Food and Agriculture Organisation, FAO) / GRID: UNEP/Global Resource Information Database / WB: World Development Indicators (World Bank) / UNDP: Human Development Report (UNDP) / ILO: International Labour Office / UNPOP: UN Dep. Of Economic and Social Affairs/Population Division. Most of the data were reprocessed by the UNEP Global Environment Outlook team. Figures are available at the GEO Data Portal (UNEP), http://geodata.grid.unep.ch, CRED: Université Catholique de Louvain (as of 2002), EM-DAT: The OFDA/CRED International Disaster Database, http://www.cred.be/ 2.calculated from UNPOP data 3.calculated from UNEP/GRID spatial modelling based on CIESIN population data. Detailed Process · Countries with no physical exposure or no victims Data on Victims. The number of killed was derived reported (zero or null values). from the CRED database and computed as the average · Countries without all the selected socioeconomic number of killed per year over the 1980­2000 period. variables. · Eccentric values, when exceptional events or other Filtering the Data. The statistical models for each dis- factors would clearly show abnormal levels of vic- aster type were based on subsets of countries, which tims, for example, the landslides in Caracas (Venezuela, excluded: 1999) and in Armero (Colombia, 1985). Global Landslides Risk Case Study 73 Transformation of Variables. The average of socioe- Choice Between Variables. One important condition conomic parameters was computed for the 21-year when computing regressions is that the variables included period. The average of the socioeconomic parameters in a model should be independent; that is, the correla- was reached by using the number of victims to better tion between two sets of variables should be low. This reflect the situation at the time the events occurred. is clearly not the case with HDI and GDP per capita expressed in purchasing power parity (hereafter referred Kic · Vic to as GDPppp), which are highly correlated. In order to Vp_ av = Ktot keep the sample as complete as possible, a choice of available variables had to be made. This choice has been Where: performed by the use of both matrix-plots and corre- Vp_av = Socioeconomic value reached by number lation-matrixes (using low correlation and visualization of people killed for a selected country; of scatter plot as selection criteria). Kic = Killed from landslides for the year "i" and the country "c"; The Stepwise Approach. The validation of regression Vic = Socioeconomic value for the year "i" and was carried out using R2, variance analysis and detailed the country "c"; and residual analysis. Ktot = Total number killed in landslides for the Once the model was derived, the link between the selected country. estimated number of people killed and number of killed observed was provided by both graphical plots and com- For some of the indicators, the logarithm was com- putation of Pearson correlation coefficients in order to puted directly; for other parameters expressed in per- ease the visualization of the efficiency by the readers. centage form, a transformation was applied so that all This model allows the identification of parameters variables would range between - and +. This appeared leading to higher/lower risk, but should not be used as to be relevant as some of the transformed variables were a predictive model, because small differences in loga- proved to be significant in the final result. For others, rithm scale induce large ones in the number of killed. no logarithmic transformation was needed; for instance, the population growth already behaves in a cumulative Results way. The results following this method are relevant, espe- V'i = Vi cially considering the independence of the data sources (1­Vi) (no auto-correlation suspected). Where: During a multiple regression analysis, it is not pos- V'i = The transformed variable (ranging from - sible to test variables that are correlated together. Such to + ), and variables have to be separated into five different Vi = The socioeconomic variable (ranging from groups of analysis. 0 to 1). The equation below depicts the steadiest correlation Figure 2.A.3. Transformation for variables ranging found. The other variables showing relevant (although between 0 and 1 inferior) correlation are also provided. 1n(K) = 0.661n (PhExp_all) + 0.701n (FCpc) + 0.361n(AR_Land) ­ 2.441n (HDI) ­ 14.98 Where: K = The number killed in landslides; PhExp_all = Physical exposure, including all the classes; 74 Natural Disaster Hotspots Case Studies FCpc = The transformed percentage of forest Some countries were removed from the analysis, as in the country; they obviously didn't fit in with the rest: Egypt, Ethiopia, HDI = The transformed Human Development Guyana, and Sri Lanka. Aside from Ethiopia, all the Index; and others have low frequencies, ranging anywhere between Ar_Land = The percentage of arable land (AR). 1 and 4. Ethiopia and Egypt do not fit into the model, perhaps because of their low levels of forestation. Table 2.A.4. Exponent and p-value for landslide multiple regression Comments and Discussion R=0.852, R2=0.727 adjusted R2= 0.703 First, these results demonstrate that the work conducted Countries=53 B p-level3 on identifying physical exposure was relevant (p-value < 10-6). Although most of the countries affected by land- Intercept ­14.979 0.000011 slides of classes 5 and higher include 98 percent of their PHEX_all 0.6616 0.000000 recorded victims, some of the countries were missing, % forest 0.7026 0.000002 such as: Sri Lanka, Mozambique, Republic of Moldova, Liberia, Guyana, Egypt, and Angola. % AR_Land 0.3649 0.000168 To avoid exclusion of these countries in our statisti- HDI -2.4406 0.000104 cal analysis, we first used frequencies of classes 2 and over to compute physical exposure. Further analysis with physical exposure involving only classes 5 and over or 6 and over shows less correlation. The Figure 2.A.5. Predicted killed versus observed for landslide physical exposure of classes 2 and over produced the best results, although we had to exclude some of the countries with low hazard levels. This can be explained by the fact that some coun- tries (for example, Brazil, the Republic of Korea) with quite a large amount of casualties have a very large percentage of their hazard area in the lower classes (2­4). When including only high classes (5­9), these countries are not totally excluded like some others (for exam- ple, Sri Lanka, Mozambique). But, their physical-exposurelevelsaremuchsmaller than their recorded-casualty levels, a cir- cumstance that caused problems while doing the statistical analysis. In the end, four of them were rejected by the model (see list above), although Mozambique, Republic of Moldova, Liberia, and Angola were still included. Although at first sight, the results sug- gest that more forested countries are at greater risk, this is probably due to the fact that forested countries are subject to deforestation. 3.In broad terms, a p-value smaller than 0.05, shows the significance of the selected indicator, however this should not be used blindly. For obvious reasons, countries without forests cannot Global Landslides Risk Case Study 75 suffer from deforestation. In the absence of relevant data istan, and Vanuatu. For these areas, another model on deforestation, it is crucial that appropriate data on should be developed. deforestation be included in the model. Yet another Further research is needed to differentiate between explanation for landslide risk may stem from the link areas where earthquakes are the main trigger versus between hazard level and forestation level: large forests where precipitation is the main trigger. Areas that are are usually in wet areas, thus increasing the risk of land- more arid and humid might have a different level of vul- slides triggered by precipitation. This is still an open nerability. question that needs further evaluation. The variable "arable land" seems to indicate that rural Other Results populations are more vulnerable to landslides. The vari- Other interesting results were found but disregarded able also could be reflecting the types of activities in as they were either less significant (higher p-value, lower addition to the type of habitat. R2), included fewer countries, or were less meaningful More obviously, countries with a lower HDI are more in terms of interpretation. They are provided in the vulnerable to landslides. In developing countries-- tables below. featuring less-resilient infrastructure and lower levels of education--land planning is left to local authorities Table 2.A.4. Other exponents and p-values for landslide or even to individuals. In such cases, due to either igno- multiple regression rance or lack of choice, populations are settling in risk- 37 countries B p-value prone areas. However, just because less-developed Intercept -6.82 0.011 countries are more vulnerable, it does not necessarily Pop_loc 0.79 0.000 Percentage forest 0.48 0.006 mean that improving their development levels will dras- Corruption -0.75 0.040 tically decrease the number of casualties. Indeed, the R= 0.71, R2= 0.50, adjusted R2= 0.45 results in the UNDP report (UNDP/BCPR 2004) state that overly rapid development can lead to higher risk. 55 countries B p-value In terms of landslides, one can easily understand that Intercept -7.392 0.0626 development based on exportation of timber can lead Ln_nbEvent 1.081 0.000000 to a higher risk of landslides in forested areas. The tragedy Pop_loc 0.307 0.041 of Caracas (Venezuela) in 1999 is sadly explicit. Rapid Percentage forest 0.293 0.037 and inappropriate urban growth from new workers HDI -1.273 0.013 coming into the capital city led to devastating flood- R= 0.874, R2= 0.75, adjusted R2= 0.73 triggered landslides. As explained before, variables with auto-correlation 32 countries B p-value cannot be analyzed together. For this reason, groups of Intercept -7.26 0.000751 noncorrelated variables were made and statistical PhExp_6+ 0.727 0.000004 Percentage forest 0.545 0.001605 analyses were performed several times. In the tables pro- GDPppp 0.280 0.013075 vided hereafter, other socioeconomic contexts leading GLASOD_34 -0.827 0.006 to higher risk are shown: parameters such as corruption, R=0.86, R2=0.747, Adj R2=0.70 habitation on highly degraded soil (GLASOD_34), and GDP purchasing power parity. These parameters have 55 countries B p-value been selected using the relevant p-values. Habitation Intercept -14.225 0.00000 on highly degraded soil was always associated with PhExp_cred 0.635 0.00000 densely forested countries. This calls for a further Percentage forest 0.485 0.000137 analysis of deforestation. HDI -1.477 0.005955 Because of missing data, the model was not appli- R= 0.85, R2= 0.72, adjusted R2= 0.71 cable to the following countries: Afghanistan, Bosnia, Democratic Peoples Republic of Korea, Georgia, Kyrgyz 4.Highest correlation, but auto-correlation suspected between number of Republic, Lebanon, Liberia, Puerto Rico, Taiwan, Tajik- events and Pop_loc, intercept p-value > 0,05. 76 Natural Disaster Hotspots Case Studies Conclusions References The results from this analysis confirm that identifica- Anand, S., and A. Sen. 2000. The Income Component of the tion of physical exposure was relevant. The process is Human Development Index. Journal of Human Development validated by the good correlation between national 1(1). socioeconomic parameters (such as HDI) and inde- Blaikie, P., et al. 1996. At Risk: Natural Hazards, Peoples Vulner- pendent datasets such as reported casualties in CRED ability and Disasters. London and New York: Routledge. and frequency of landslides (as computed by the Bolt, B.A., et al. 1975. Geological Hazards. New York: Springer- Verlag Berlin-Heidelberg. model described in the main report using slopes, lithol- Burton, I., R. W. Kates, and G. F. White. 1993: The Environment ogy, level of precipitation, seismicity, and so on). as Hazard, Second Edition. New York/London: Guilford Press, Although around 73 percent of the variation is 31­47. explained by the regression, one has to keep in mind Carter, N. 1991. Disaster Management, a Disaster Manager's Hand- that this is not a predictive model, mostly because log- book. Manila: Asian Development Bank. arithmic regression prevents the use of "0" in the Coburn, A. W., R. J. S. Spence, and A. Pomonis. 1991. Vulner- ability and Risk Assessment. UNDP Disaster Management Train- analysis and minimizes the differences. However, classes ing Program, 57. of countries at risk can be established. The model can Dao, H., and P. Peduzzi. 2004. Global Evaluation of Human Risk be used to better understand socioeconomic context, and Vulnerability to Natural Hazards, Enviro-info 2004. Sh@ring, and, eventually, classes of countries at risk can be derived. Editions du Tricorne, Genève, Vol. I, 435­446. The study revealed that some countries with recorded Dao, H., and P. Peduzzi. 2003. Global Risk and Vulnerability Index casualties did not have appropriate physical exposure. Trends per Year (GRAVITY), Phase IV: Multiple Risk Integration. Scientific report. Geneva, Switzerland: UNDP/BCPR. The question of frequencies in different climates and at Peduzzi, P., H. Dao, and C. Herold. 2005. Mapping Disastrous different vegetation levels might be the source of such Natural Hazards Using Global Datasets, Natural Hazards, discrepancies and could constitute interesting future Vol. 35, Issue 2, 265­289. developments. Peduzzi, P., H. Dao, and C. Herold. 2002. Global Risk and Vul- Conversely, countries with no recorded casualties nerability Index Trends per Year (GRAVITY), Phase II: Develop- over the 21-year period cannot be considered using ment, Analysis and Results. Scientific report. Geneva, Switzerland: UNDP/BCPR. the method of vulnerability proxy. Data for a longer Peduzzi, P., et al. 2001. Feasibility Study Report on Global Risk period should be obtained. and Vulnerability Index Trends per Year (GRAVITY). Scientific Explanations for how high physical exposure and report. Geneva, Switzerland: UNDP/BCPR. low HDI lead to high risk are quite straightforward. The Peduzzi, P., et al. 2003. Global Risk and Vulnerability Index Trends selection of countries with high percentages of foresta- per Year (GRAVITY), Phase IIIa: Drought Analysis. Scientific tion is less easy to explain. Could this be because of report. Geneva, Switzerland: UNDP/BCPR. Smith, K. 1996. Environmental Hazards, Assessing Risk and Reduc- deforestation (which occurs more often in densely, as ing Disaster. London and New York: Routledge. opposed to sparsely forested countries)? Could this be Tobin, G. A., and B.E. Montz. 1997. Natural Hazards, Explana- an indirect way of measuring traditional activities in a tion and Integration. New York and London: The Guilford Press. country? In any case, the model failed to explain risk UNDRO (United Nations Disaster Relief Coordinator). 1979. for nine countries, demonstrating the need for data on Natural Disasters and Vulnerability Analysis in Report of Expert deforestation in order to improve the model and fur- Group Meeting (July 9­12, 1979). Geneva: UNDRO, 49. UNEP (United Nations Environment Programme). 2002. GEO: ther explain vulnerability. This was highlighted by the Global Environment Outlook 3: Past, Present and Future Per- selection of variables such as forested countries associ- spectives. 446 pp. ated with degraded soils. UNDP/BCPR (United Nations Development Programme/Bureau for Crisis Prevention and Recovery). 2004. Reducing Disaster Risk: A Challenge for Development. New York. 146 pp. Global Landslides Risk Case Study 77 Internet References Peduzzi, P. 2001. Project of Risk Evaluation, Vulnerability Indexing and Early Warning (PREVIEW). Geneva: Switzer- CIESIN, IFPRI, WRI. 2000. Gridded Population of the World land: UNEP/DEWA/GRID-Geneva. http://www.grid. (GPW), Version 2. http://sedac.ciesin.org/plue/gpw/ unep.ch/activities/earlywarning/preview/index.htm Deichmann, Uwe. 1996. GNV197--Human Population and UNDP. 2004. Reducing Disaster Risk: A Challenge for Administrative Boundaries Database for Asia. Geneva, Development. http://www.undp.org/bcpr/disred/rdr.htm Switzerland: UNEP/GRID. http://www.grid.unep.ch/data/ UNDP. 2002. Human Development Indicators. http://www. grid/gnv197.php undp.org/ International Decade for Natural Disaster Reduction. http:// UNEP, CGIAR, NCGIA. 1996. Human Population and www.unisdr.org/unisdr/indexidndr.html Administrative Boundaries Database for Asia.http://www. ISRIC, UNEP. 1990. Global Assessment of Human Induced Soil grid.unep.ch/data/grid/human.php Degradation (GLASOD). http://www.grid.unep.ch/data/grid/ UNEP/GRID (as of 2002). GEO-3 Data portal. http://geodata. gnv18.php grid.unep.ch/ OFDA/CRED (Office of U.S. Foreign Disaster Assistance/ Center for Research on the Epidemiology of Disasters). 2001. EM-DAT: The OFDA/CRED International Disaster Database. http://www.cred.be/emdat Peduzzi, P. 2000. Insight of Common Key Indicators for Global Vulnerability Mapping. Presentation for the Expert Meeting on Vulnerability and Risk Analysis and Indexing, September 11­12, 2000. Geneva, Switzerland: UNEP/DEWA/GRID-Geneva. http://www.grid.unep.ch/ activities/earlywarning/preview/appl/reports/ reports.htm Chapter 3 Storm Surges in Coastal Areas Robert J. Nicholls Flooding of low-lying coastal areas can occur for a host The paper examines the controls and occurrence of of reasons, such as tsunamis, intense local precipita- storm surges, including the following issues: tion, high river flows, and storm surges (Penning-Rowsell · The characteristics and magnitudes of surges around and Fordham 1994; Smith and Ward 1998; Parker 2000). the world; Some of these flood mechanisms may interact with each · Regional exposure and risk of flooding from storm other, as well as with other hazards such as human- surges, both now and further into the 21st century; induced subsidence. Coastal areas are characterized by · Damages, and especially fatalities, due to flooding by growing concentrations of human population and socioe- storm surges (this issue is based on historical expe- conomic activity (Sachs et al. 2001; Small and Nicholls rience, which has already identified some "hotspots"); 2003), which means such floods can have severe impacts, and including significant loss of life in certain situations. · Detailed case studies of selected surge-prone areas. Widespread efforts to mitigate coastal flood hazards are already apparent, and this need is likely to inten- It is important to recognize that the global datasets sify throughout the 21st century due to the above trends, on these issues are incomplete and that expert judg- as well as to a general increase in risks due to climate ment has been critical to developing the paper. There- change. fore, global analyses of flooding due to storm surges, This paper focuses on storm surges as a coastal hazard, such as the analysis developed by Nicholls (2004), have including identifying regions where the impacts of storm been important in the analysis. These regional analyses surges are potentially of particular significance and locat- cannot identify storm surge hotspots per se, but they ing potential "hotspots" within these regions (as much do indicate the regions where they are more likely to as the available data allow). Storm surges are generated occur now and in the future under a range of scenar- by tropical and extra-tropical storms. The low baro- ios. When this broad-scale analysis is combined with metric pressure and wind set-up combine to produce historical information on storm surge disasters, hotspots large temporary rises in sea level that have the capac- can be defined, and hence appropriate case studies can ity to cause extensive flooding of coastal lowlands. They be selected. are usually associated with strong winds and large onshore waves, which increase the damage potential relative to the potential damage caused by surge-induced high What Is a Storm Surge? water levels alone. The largest surges are produced by hurricane landfalls, but extra-tropical storms can also Surges are changes in sea level (either positive or neg- produce large surges in appropriate settings. Flooding ative) resulting from variations in atmospheric pressure by surges contributes to the damage and disruption and associated winds. They occur on top of normal tides, causes by coastal storms. It also threatens human and when positive surges are added to high tides they life--drowning by surges is generally the biggest killer can cause extreme water levels and flooding (flooding during coastal storms. is most severe when a surge coincides with spring tides). Surges are most commonly produced by the passage of 79 80 Natural Disaster Hotspots Case Studies atmospheric tropical or extra-tropical depressions.5 feet) above the predicted high tide, resulting in water Surges can occur in the open ocean, where the surge depths exceeding 4.9 m (16 feet) (figure 3.2). The occupies only part of the area (for example, a hurri- highest-ever recorded surge was in 1899 in Bathhurst cane landfall on ocean coasts), as well as in enclosed Bay, Australia, when a surge reached 13 m (http://www. basins such as the Baltic, where the surge event will ns.ec.gc.ca/weather/hurricane). influence most, or all, of the basin. The strong winds that contribute to surge events also The magnitude of the surge is controlled partly by the produce large storm waves. The offshore wave height storm track and intensity, and partly by the configura- is dependent upon the fetch, the wind strength, and the tion of the coastline and seabed. Onshore winds serve length of time the wind has been acting upon the sea to pile water against the coast and to generate surface surface. Waves increase sea levels and have significant currents and waves, which add to the maximum sea sur- potential to cause damage and exacerbate flooding. In face elevation. A depression also reduces the atmospheric particular, wave action can cause considerable erosion pressure, resulting in a rise in sea level (the inverted to protective backshore landforms (for example, bar- barometer effect). As a rule of thumb, an atmospheric rier islands, dune ridges), and damage artificial struc- change of 1 mb results in a sea level change of 1 cm. tural defenses. In the extreme, they can cause breaching Hence a deep depression with a central pressure of 960 of these defenses, enabling tidal waters to flood onto mb will cause the sea level to rise about 0.5 m above coastal lowlands in the lee of these defenses. Hence, what it would have been had the atmospheric pressure surges and the associated wave action need to be con- been at the average value of 1,013 mb. Coastlines fronted sidered together as part of the storm surge hazard. by a wide, shallow continental shelf experience larger The areal cover and depth of flooding due to storm surges than do coastal areas with steeper slopes and surge depends upon a range of parameters, including greater water depths. Coastal configuration is also impor- surge height and duration, defense standards, and land tant. The southern North Sea, for example, is open to elevation. In "natural" situations with little or no coastal the north and nearly closed to the south, thus amplify- defenses apart from natural dunes, such as those found ing the potential for surges. Given appropriate condi- on the U.S. east coast, a storm surge typically diminishes tions, surges due to extra-tropical storms can reach 2 to 0.2 to 0.4 m per km inland. Therefore, an extreme 6-m 3 m in the southern North Sea, as happened in the storm storm surge might reach 11 to 16 km inland if eleva- surge of January 31­February 1, 1953, and even more tions are low (only 1 to 2 meters), as is often the case in the German Bight, as happened in the 1962 surge. In (Pielke and Pielke 1997). However, steeper slopes will 1953, over 300 people lost their lives in the United King- curb inland penetration. In areas where land elevations dom (Kelman 2002), and nearly 2,000 people were killed are at or below sea level, surges could potentially create in the Netherlands (figure 3.1) (Smith and Ward 1998). bigger flood problems. In the Netherlands, over half In 1962, about 300 people were killed in Germany the country is threatened by flooding from surges and (Ascher 1991). Indicative surge magnitudes for hurri- rivers, but the flood defenses are built to a high stan- cane landfall are given in table 3.1, showing that surges dard (nominally up to a 1 in 10,000 year event) (Peer- of 6 m or more are possible. In the 1971 cyclone in bolte 1994). Many coastal areas that are threatened by Bangladesh, the maximum surge reached 3.8 m (12.5 surges have characteristics similar to those of the Nether- lands, with extensive low-lying areas of land claim,6 pro- tected by flood defenses. This has increased both the size 5.Tropical and extra-tropical storms are examples of weather systems that of the flood plain and the threatened population (for circulate in a counterclockwise direction in the Northern Hemisphere example, Germany and Bangladesh). Thus, the existing and a clockwise direction in the Southern Hemisphere. Extra-tropical situation has co-evolved into a potentially more vul- storms form over land or the ocean as the result of the temperature con- trast between the colder air at higher latitudes and the warmer air closer to the equator. Tropical cyclones form over the ocean waters of the trop- ics, and are termed "hurricanes" when sustained surface winds are 33 m/s 6.While the term reclamation is often used to describe land claim, this term or greater. In the eastern Pacific, hurricanes are termed typhoons, while is incorrect, as the process is usually the claiming of intertidal and wet- in the Indian Ocean they are termed cyclones. land areas, or even subtidal areas--that is, land claim (French 1997). Storm Surges in Coastal Areas 81 Figure 3.1. Areas in the southwest Netherlands flooded by the 1953 storm surge, February 1, 1953 (from Edwards 1953) Table 3.1. Hurricane characteristics and indicative surge magnitudes based on the Saffir-Simpson scale Scale Number Central Pressure (hPa) Wind Speed (km/hr) Surge magnitude (m) 1 >980 120­149 1.2­1.6 2 979­965 150­179 1.7­2.5 3 964­945 180­209 2.6­3.8 4 944­920 210­249 3.9­5.5 5 <920 >249 >5.5 Source: Smith and Ward 1998. nerable situation, compared to the natural situation. Figure 3.3 shows some of the threatened areas in Tokyo Human-induced subsidence has also increased the with and without sea-level rise. number of people potentially exposed to flooding by Flooding due to surges has a range of impacts, includ- storm surges. Table 3.2 lists some major coastal cities ing property damage and destruction, human distress that have experienced significant human-induced flood- and health effects, and, in the worst case, fatalities. Most ing due to groundwater withdrawal, and, hence, flood- of the world's coasts experience relatively small surges, ing due to surges has potentially been exacerbated. In and impacts might be quite localized with limited Japan, 2 million people live below the normal high water flood areas and shallow flood depths. However, even level due to subsidence and depend on flood defenses under relatively mild surge regimes (< 1 m), significant every day to stop floods, with a much larger population property damage can occur, as happens in the well- threatened by flooding due to surges (and other flood known flooding of the historic city of Venice (Penning- hazards, such as tsunamis) (Mimura et al. 1994). Rowsell 2000; Harleman et al. 2000). However, deep 82 Natural Disaster Hotspots Case Studies Figure 3.2. A simplified reconstruction of the November 1970 storm surge in Bangladesh. Circled data indicate the height by which astronomical high tide was exceeded; isolines show the depth of water above the ground surface. Table 3.2. Some major coastal cities and human-induced subsidence during the 20th century Date Human-Induced Megacity Maximum Subsidence (m) Subsidence Commenced Surge Potential Shanghai 2.80 1921 Tropical Storms Tokyo 5.00 1930s Tropical Storms and Extra-Tropical Storms Osaka 2.80 1935 Tropical Storms and Extra-Tropical Storms Bangkok 1.60 1950s Tropical Storms Tianjin 2.63 1959 Tropical Storms and Extra-Tropical Storms Jakarta 0.90 1978 Limited Metro Manila 0.40 1960 Tropical Storms Source: Adapted from Nicholls 1995. Storm Surges in Coastal Areas 83 Figure 3.3. Areas in Tokyo that are below normal high-water and surge levels with and without a 1-m rise in sea level. These low-lying areas have been largely created by human-induced subsidence. surges and fast-moving water can lead to death by drown- much of south, Southeast, and East Asia; and much of ing. It is noteworthy that millions of people have drowned the Pacific, including Papua New Guinea and Australia. due to storm surges around the world, with regular Extra-tropical storms affect mid- and high-latitude coastal recurrence in some notable hotspots around the North areas, with noteworthy surge potential in the North Sea, Sea, the Bay of Bengal, and East Asia (see the sections the Baltic Sea, and in the Rio de la Plata (between covering the definition of storm surge hotspots and the Argentina and Uruguay), to name just three locations. case studies). Figure 3.4 indicates those areas of the Parts of North America and East Asia are subject to both world's coasts that are affected by tropical cyclones tropical and extra-tropical storms, and, hence, surges and, thus, prone to significant surges. These areas include can result from more than one causal mechanism (for the Caribbean and North America; parts of East Africa; example, Zhang et al. 2000). 84 Natural Disaster Hotspots Case Studies Figure 3.4. Coasts affected by tropical cyclones It is important to note that extra-tropical and tropi- component of different disasters is simply not recorded, cal storms can produce a range of hazards in addition to except for those limited number of events where the surge surges and waves, particularly intense precipitation, wind impacts were dominant. When defining surge impacts, damage, and even tornadoes and water spouts. It is some- the most robust statistic is usually the number of fatali- times difficult to separate the impacts of these different ties, as other damages are integrated across all the haz- hazards, and the overall impacts of the storm event are ards produced by the storm event, as noted above. "integrated" into a composite set of impacts that does not distinguish between the contributions of the different hazards (Pielke and Pielke 1997). For instance, Hurri- Responding to Storm Surges cane Andrew produced a 4-meter surge in Biscayne Bay, southern Florida, but the major damage in Florida was While human processes such as land claim may have due to the hurricane-force winds. Similarly, Hurricane increased the areas threatened by storm surges, humans Mitch in Central America in 1998 produced a signifi- also have responded to this threat in various ways, and, cant surge that damaged and destroyed coastal homes hence, reduced the vulnerability of coastal populations and drowned many people on the coast. However, the to such flooding. There is a range of possible strate- main impact of the event was the intense precipitation gies for dealing with weather-induced hazards such as and run-off further inland, which caused most of the surges, as summarized in table 3.3. These strategies can damage and loss of life (UNEP 2002). Surges may also be described as follows. Choosing change means accept- interact with other types of flood mechanism, as ing the hazard and changing land use, or even relo- appears to happen in the Philippines (Perez et al. 1999). cating exposed populations. Reducing losses includes As a result of this, databases such as the EM-DAT Dis- trying to reduce the occurrence of the hazardous aster Database (http://www.cred.be/emdat/) have been event or, more commonly, reducing the impacts of a found to be of limited value to this study, as the surge hazardous event when it occurs. Both flood-protec- Storm Surges in Coastal Areas 85 Table 3.3. Generic approaches to hazard reduction flood protection such as dikes and flood walls--the based on purposeful adjustment flood forecast and warning system comes into public Purposeful adjustment Option action only if the flood protection is at risk of failure, Choose change Change location and its primary goal is to preserve life via evacuation. Change use Climate change and sea-level rise represent an addi- Reduce losses Prevent effects tional challenge around the world's coastal zones. Arti- Modify event cle 3.3 of the UNFCCC suggests that proactive adaptation Accept losses Share loss (as well as mitigation to reduce greenhouse gas emis- Bear loss sions) deserves particular attention from the interna- Source: Burton et al. 1993. tional climate change community given the threat of human-induced climate change: tion and flood-warning systems are approaches to reduce The Parties should take precautionary measures to antic- losses. Accepting losses includes bearing the loss, ipate, prevent or minimize the causes of climate change possibly by exploiting reserves, or sharing the loss and mitigate its adverse effects. Where there are threats through mechanisms such as insurance. Hence the abil- of serious or irreversible damage, lack of full scientific ity to recover from the disaster is of the utmost impor- certainty should not be used as a reason for postponing tance if losses are accepted. Note that these strategies such measures, taking into account that policies and are not mutually exclusive, and hazard reduction might measures to deal with climate change should be cost- include elements of all three approaches. These effective so as to ensure global benefits at the lowest approaches can also be applied at various levels, from possible cost.... the individual up to communities and beyond (for The threat of climate change is extending the scope example, large cities or even nations). of reduction strategies for weather-related hazards Over time, technology is increasing the options such as surges and focusing attention on the coming that are available for hazard risk reduction, particu- decades. The Intergovernmental Panel on Climate larly those strategies that reduce losses (Klein et al. Change (IPCC), Third Assessment Report (TAR) included 2000). In areas with large populations and strong a dedicated chapter on adaptation for the first time (Smit economies, there is usually a bias toward loss reduc- et al. 2001). Coastal zones constitute an area where there tion, and it can be argued that many of the populated has been particular interest in adaptation given the coastal areas threatened by storm surges would not have inevitability of global-mean sea-level rise (Klein et al. evolved in the way that we see today without the avail- 2000, 2001; Tol et al. forthcoming). Some coastal coun- ability of these hazard risk reduction strategies. Exam- tries such as the United Kingdom and Japan are at the ples of these approaches include warning systems, defense forefront of planning for climate change, with the major works, and resistant infrastructure. This approach is emphasis being on the implications for flooding of coastal most developed in urban areas around the North Sea, areas, as sea-level rise will increase the risk of flooding other parts of Europe, China, and Japan, where flood- due to surges and other flood mechanisms. ing by surges claimed many lives up to the middle of In general, proactive adaptation is aimed at reduc- the 20th century (see the section on defining storm surge ing a system's vulnerability by either minimizing risk hotspots). A particular problem is that while strategies or maximizing adaptive capacity. Five generic objectives to reduce losses (for example, flood defense) only reduce, of anticipatory adaptation can be identified (Klein and rather than remove, the risk, the measures are some- Tol 1997): times seen as invulnerable and, hence, encourage fur- ther development in what remain potentially hazardous · Increasing the robustness of infrastructural designs and areas (for example, Parker 2000). Therefore strategies long-term investments--for example, by extending the to respond to surges need to analyze the response to range of extreme water levels and wave loading that the full range of risk, including any residual risk. This a system can withstand without failure and chang- might mean combining a flood warning system with ing a system's tolerance of loss or failure (for exam- 86 Natural Disaster Hotspots Case Studies ple, by increasing economic reserves or insurance); seawalls and other coastal infrastructure. Efficient man- · Increasing the flexibility of vulnerable managed systems-- agement of beach and coastal sediments is also an impor- for example, by allowing mid-term adjustments tant strategy to maintain and enhance soft defenses, (including change of activities or location) and reduc- which can also sustain recreational and other functions. ing economic lifetimes (including increasing depre- A retreat strategy would serve to avoid placing vulner- ciation); able infrastructure and populations in the present and · Enhancing the adaptability of vulnerable natural sys- future flood plain. A strategy to accommodate could tems--for example, by reducing other (non-climatic) include increasing the flexibility or coping capacity of stresses and removing barriers to migration (such as managed systems. Examples include implementing flood- managed realignment); warning systems, raising buildings above flood levels to · Reversing trends that increase vulnerability ("maladap- minimize flood damage (as is already practiced in the tation")--for example, by introducing setbacks for United States as part of the National Flood Insurance new development or relocation of existing develop- Program), and sharing losses via insurance mechanisms. ment in vulnerable areas such as coastal flood plains; While protection has dominated the response to haz- · Improving societal awareness and preparedness--for ards in urban areas, proactive adaptation needs to con- example, by informing the public of the risks and sider the opportunities to retreat or accommodate in possible consequences of flooding by surge and by less-developed and developing coastal areas. setting up early-warning systems. A key point about the effective implementation of hazard reduction strategies is that they involve more As with the approaches listed in table 3.4, these than implementing a set of technical measures. They approaches are not mutually exclusive. need to be thought of as an ongoing process, including Each of these five objectives of adaptation is rele- planning, design, implementation, and monitoring (Klein vant for hazard reduction to surges. However, for coastal et al. 1999, 2001; Willows and Cornel 2002). The case zones, another classification of adaptation options is studies support this point. often used, one that distinguishes between the follow- Hence, there is a range of hazard reduction strate- ing three basic hazard management strategies (IPCC gies available for responding to the flood threat of surges. CZMS 1990): While continued technology development may further · Protect--to reduce the risk of coastal hazards by increase the detailed options that are available, new decreasing their probability of occurrence; problems and issues will continue to emerge: climate · Retreat--to reduce the risk of coastal hazards by change is only one example of how future conditions limiting their potential effects; can be expected to change. Reducing losses/protection · Accommodate--to increase society's ability to cope has been the main response in the past, and this seems with the effects of coastal hazards. likely to continue. However, the implementation of Klein et al. (2001) discuss these three strategies in proactive adaptation and the utilization of adaptive man- detail and provide examples of technologies for imple- agement principles raise opportunities to use other menting each of them. While the main hazard that is approaches in areas with lower levels of development considered is sea-level rise, aspects of the approaches and in areas that are developing or redeveloping, includ- that are discussed are relevant to all weather-related haz- ing recovery from disaster. This stresses that hazard ards in coastal areas, including surges. reduction is an ongoing process rather than a simple These strategies are applicable both for adaptation to set of technical measures, and it needs to be imple- climatic variations such as surges, and climate change mented on this basis. Lastly, hazard risk reduction strate- and sea-level rise. Protecting coastal zones would involve gies need to be implemented in the wider coastal context; increasing the robustness of infrastructural designs, thus, they comprise one issue within the broader goal and making long-term investments in construction of of integrated coastal zone management. Storm Surges in Coastal Areas 87 Table 3.4. Regional contributions to coastal flooding in 1990 and the 2020s based on the analysis of Nicholls (2004). Only population change is considered. PHZ--people in the hazard zone (that is, the potentially exposed population). PAR--People at risk, or the number of people potentially flooded per year. Analysis uses the A1 scenario for the 2020s, but all the SRES population scenarios are similar in the 2020s. 1990 2020s PHZ PAR PHZ PAR Region Millions % Thousands % Millions % Thouands % 1. North America 13.2 6.7 13 0.1 18.1 6.2 18 0.1 2. Central America 0.8 0.4 18 0.2 1.6 0.6 39 0.2 3. South America Atlantic Coast 4.6 2.3 33 0.3 6.7 2.3 36 0.2 4. South American Pacific Coast 1.4 0.7 13 0.1 2.3 0.8 21 0.1 5. Caribbean 1.2 0.6 10 0.1 1.5 0.5 13 0.1 6. Atlantic Small Islands 0.0 0.0 0 0.0 0.0 0.0 0 0.0 7. North and West Europe 19.0 9.6 19 0.2 21.6 7.4 22 0.1 8. Baltic 1.4 0.7 15 0.1 1.5 0.5 16 0.1 9. North Mediterranean 4.1 2.1 5 0.1 4.3 1.5 6 0.0 10. South Mediterranean 5.6 2.8 229 2.2 10.5 3.6 436 2.7 11. Africa Atlantic Coast 6.9 3.5 342 3.3 16.0 5.5 822 5.0 12. Africa Indian Ocean Coast 7.0 3.6 562 5.4 14.0 4.8 1,165 7.1 13. Gulf States 0.4 0.2 2 0.0 0.9 0.3 4 0.0 14. South Asia 52.3 26.5 4,292 41.6 90.8 31.0 7,461 45.5 15. Indian Ocean Small Islands 0.1 0.1 2 0.0 0.3 0.1 5 0.0 16. South-East Asia 26.5 13.5 1,874 18.2 39.3 13.4 2,742 16.7 17. East Asia 44.1 22.4 2,869 27.8 54.5 18.6 3,565 21.8 18. Pacific Large Islands 0.5 0.3 2 0.0 0.9 0.3 4 0.0 19. Pacific Small Islands 0.1 0.1 3 0.0 0.2 0.1 5 0.0 20. Former USSR 7.8 3.9 8 0.1 8.3 2.8 8 0.1 TOTAL (millions) 197 10.3 293 16.4 Source: Nakicenovic et al. 2000; Arnell et al. 2004. Regional Exposure to Storm Surges: of magnitude. While significant uncertainties remain 1990 to the 2080s (Small et al. 2000; Small and Nicholls 2003), these data and the analysis provide a broad "snap-shot" of Given the lack of consistent data on the flood impacts the regional exposure and frequency of flooding due to of storm surges, regional analyses are utilized to iden- storm surges in 1990, and how they might change under tify the regions where flooding due to surges is most a range of scenarios leading into the 2080s. The results common. These regions are likely to contain most of of this analysis are expressed in terms of number of the storm surge "hotspots." people exposed or impacted by flooding due to surges. The Global Vulnerability Analysis was developed to Two of these parameters are considered here: examine the impacts of sea-level rise at both regional7 · Peopleinthehazardzone: the number of people exposed and global scales (Hoozemans et al. 1993; Nicholls et to flooding by storm surges, ignoring sea defenses. al. 1999; Nicholls 2004). It includes estimates of the This is defined as the people living below the 1,000- coastal flood plain at risk from storm surges, flood plain year storm surge elevation and is a measure of expo- population, and return period of different events. Val- sure. idation suggests that the results are of the right order · Peopleatrisk(orAverageAnnualPeopleFlooded):anesti- mate of the average number of people who experi- ence flooding caused by storm surges each year, 7.The 20 regions listed in table 3.4 represent the smallest scale at which it is meaningful to report these results. including the benefits of protection from sea defenses. 88 Natural Disaster Hotspots Case Studies Figure 3.5. People at risk (that is, people potentially flooded) versus people in the flood hazard zone in 1990 for 20 global regions. Table 3.5. The range of scenarios used by Nicholls (2004) Environmental Changes Socioeconomic Developments Climate-Induced Global-mean sea level GDP/capita (which controls the upgrade of flood defenses to climate variability--no allowance for the effects of sea-level rise is made) Not Climate-Induced Vertical land movement Population Table 3.6. Estimates of the global exposure and incidence of flooding under the four SRES scenarios in the 2080s, plus 1990 estimates as a reference SRES Scenario People in the Flood Plain (millions) People at Risk (millions/year) 1990s 197 10 A1FI 314 7 A2 562 47 B1 304 3 B2 399 19 Source: Nicholls 2004. Storm Surges in Coastal Areas 89 Thisisestimatedasthenumberofpeopleinthehazard ative increase in the population exposed to flooding by zonemultipliedbytheriskofbeingfloodedand,hence, storm surges within the developing world, as these are measuresthelikelihoodoffloodingactuallyoccurring. the areas that are expected to experience the largest increases in population. Lastly, climate change and It should be noted that this analysis does not con- sea-level rise may exacerbate these flood risks as dis- sider the depth of flooding, due to the inadequacies of cussed below. the available data. Flood depth is an important param- The frequency, magnitude, and impacts of storm eter when considering the storm surge hotspots. surges will change through the 21st century due to a Considering the base year (1990), it is estimated that, combination of (1) sea-level rise and climate change, globally, a total of about 200 million people were living (2) increasing direct human modifications to coastal in areas vulnerable to flooding caused by storm surges. areas (for example, further land claim around estuar- Further, it is estimated that about 10 million people ies), and (3) socioeconomic changes (Warrick et al. potentially experience flooding from storm surge each 2000). Concern about increasing hurricane intensity year, which is about 5 percent of the exposed popula- was first raised by Emanuel (1988), who hypothesized tion. There are also important regional differences, which that in a globally warmed world, deeper depressions are summarized in table 3.5 and figure 3.5. Collectively, would be possible, thus potentially producing stronger the south, east, and south-east regions of Asia contained maximum winds, waves, and, hence, surges. Debate about 60 percent of the exposed population and nearly about the likely changes to hurricane intensity contin- 90 percent of the people who experience flooding. Other ues (for example, Henderson-Sellers et al. 1998; Knut- regions, such as North America and north and western son et al. 1998). An intensification of extra-tropical Europe, contain a large exposed population (13 mil- storms has also been suggested in some climate models lion and 19 million people, respectively), but due to under global warming (for example, Carnell and Senior higher defense standards vis-a-vis Asia, the incidence 1998). However, the IPCC TAR came to no firm con- of floods due to surges is small. However, despite the clusions on these changes (Houghton et al. 2001) and protection, the residual risk of flooding due to surges both increases and decreases remain possible, with still needs to be considered. Note that, in practice, there regional variability in the patterns of change very are important differences between Europe and North likely. Further, it is likely that long-term changes in surge America that the methods used do not explicitly address. frequency will be difficult to distinguish from the large In Europe, floods are mainly managed using hard defenses inter-annual and inter-decadal variations in storm fre- such as dikes and sea walls, with beach nourishment quency, intensity, and duration that the limited data increasingly being utilized in conjunction with the hard show (for example, WASA Group 1998; Alexanders- defenses. In contrast, the United States follows an son et al. 2000; Zhang et al. 2000; Araujo et al. 2002). approach based on accommodation of the surge hazard-- Hence, by far the most certain aspect of climate change all new buildings are raised above the 1-in-100-year that will influence surge characteristics is global-mean surge elevation. sea-level rise (Church 2001). The data in table 3.8 also show how both the expo- An analysis based on the SRES scenarios was per- sure and incidence of flooding are dynamic due to the formed using (1) increases in relative sea level, (2) rapid increase in global and, hence, coastal populations. changes in coastal population, and (3) improving defense By the 2020s, the number of people living in areas vul- standards (see table 3.5) (Nicholls 2004). The sea- nerable to flooding caused by storm surges could be level rise scenarios are derived from the Hadley Centre about 290 million people, or nearly a 50 percent increase (see Johns et al. 2003) (table 3.6). The global SRES over 1990 values. This increase assumes uniform changes socioeconomic scenarios are shown in table 3.7. The across countries and, hence, does not consider the poten- SRES regional scenarios were downscaled by the Center tial for coastward migration to increase exposure. for International Earth Science Information Network More rapid population growth in coastal areas is (CIESIN), and made available on the IPCC Data Dis- widely reported (Bijlsma et al. 1996; WCC'93 1994), tribution Centre (DDC) blue pages (http://ipcc-ddc.cru. but more quantification is required. It also shows a rel- 90 Natural Disaster Hotspots Case Studies uea.ac.uk) (see Arnell et al. 2004). The overall flood partly related to the magnitude of sea-level rise. In par- analysis simply assumed that rising relative sea level ticular, the A2 world appears to be inherently more vul- raises the surge uniformly, with no other physical changes. nerable to flooding caused by surges within the full The results show quite dramatic changes to the 2080s range of potential scenarios analyzable by this method as illustrated in figure 3.6 and table 3.8. The number (Nicholls 2004). However, it is important to note that of people who are potentially exposed to surges increases these results cover only a range of possible futures (Arnell by about 50 percent to 150 percent above 1990 values. et al. 2004; Nicholls 2004); thus, worlds with greater The incidence of flooding shows greater divergence, and lesser flood problems due to surges can be envis- with the A1FI and B1 worlds having a lower incidence aged.8 The overall conclusion is that the surge hazard of flooding due to surges compared to 1990 in terms will evolve significantly throughout the 21st century, of people affected, while the A2 and B2 worlds have a and new problems may emerge in areas where present greater incidence of flooding. In terms of regional effects, problems are relatively minor. These issues are further six regions are apparent in figure 3.6 to varying degrees: developed in the two following sections. (1) South Asia (which is the most consistently threat- ened region), (2) South-East Asia, (3) East Asia, (4) Africa Atlantic Coast, (5) Africa Indian Ocean Coast, Defining Storm Surge "Hotspots" and (6) the Southern Mediterranean. As will be dis- cussed in the following section, these Asian regions The previous section defined broad regions where storm already contain important surge "hotspots," but Africa surge flooding might be an important issue, both now does not. Hence, this analysis is suggesting that new and throughout the 21st century. However, flooding problems with surge hazards might emerge around the and its impacts are more localized than the regions con- continent of Africa through the 21st century. While sidered in the previous section. This reflects both the not apparent in the data presented here, small island occurrence of particular areas with significant surges regions (the Caribbean, Indian Ocean islands, and Pacific and high exposure to such events as flooding. Fur- small islands) also appear especially vulnerable to thermore, the implications of flooding need to be eval- increased flooding in relative terms (Nicholls 2004). uated in terms of flood depths, property damage, and These differences between the SRES worlds are only human health implications, including fatalities.9 In terms of the impacts of surge events, fatalities provide some of the most cer- Table 3.7. Global-mean sea-level rise scenarios (cm) used by Nicholls (2004) (referenced to 1990), including the IS92a GGa1 scenario as a tain information, which in some cases reference extends back almost 1,000 years (Lamb Year IS92a SRES 1995). Lists of major surge events that SRES caused substantial numbers of fatalities GGa1 A1FI A2 B1 B2 scenario range are widely published (for example, Ali 2020s 9 5 5 5 6 1 2050s 21 16 14 13 14 3 1999), but none appears to be compre- 2080s 37 34 28 22 25 13 hensive. Therefore, the author developed Source: Nicholls 2004. a synthesis of a number of sources that Table 3.8. The SRES Socioeconomic Scenarios for the 2080s: A Global Summary 8. Factors that control vulnerability include attitudes and implementation of environmental management, level Year and Population GDP GDP/capita of economic wealth, and population size. Greater global Scenario (billions) (trillion US 1990 $) (thousands US 1990 $) inequities could also be explored, such as a scenario 1990 5.3 20.1 3.8 where development in Africa follows the observations of the last 30 to 50 years, which would lead to a much 2080s A1 7.9 416 52.6 more vulnerable situation than exists in any of the SRES A2 14.2 185 13.0 scenarios (Nicholls 2003). B1 7.9 289 36.6 9.Flood damages are strongly linked to flood depth (for B2 10.2 204 20.0 example, Penning-Rowsell et al. 2003). Source: Nakicenovic et al. 2000. Storm Surges in Coastal Areas 91 Figure 3.6. People at risk (that is, people potentially flooded) versus people in the flood hazard zone in the 2080s for 20 global regions. These estimates consider four SRES futures: A1FI, A2, B1, and B2, plotted on the same scale for ease of comparison. The results assume that population change in the flood plain equals national change in population, and that protection standards are directly related to GDP/capita, with a 30-year delay. are all clearly indicated. Table 3.9 lists those surge events hensive, and undoubtedly surge events have been omit- over the last 300 years in which >1,000 people died ted due to the limited access to sources.11 Equally impor- and in which storm surge was a major or the major tant, it is likely that some events will have been poorly contribution to these deaths, mainly by drowning.10 documented and effectively "forgotten." A region where Known exceptions where other factors contributed to this seems particularly likely encompasses the Pacific most deaths--such as the hurricane flood around Lake islands. It is also noteworthy that the precise geographic Okeechobee, Florida, in 1928 (Pielke and Pielke 1997) location of many of the events is not well defined, and Hurricane Mitch (UNEP 2002)--are indicated. with the impacted area being reported only to the Note that table 3.9 is not considered fully compre- level of country. While this gives a better idea of the hotspot locations than did the indication in the previ- ous section, follow-up research would probably be able 10. In the United States, 90 percent of deaths in hurricanes were due to storm surge; while this cannot simply be applied to other areas, it gives an indication of the important role of surges in causing fatalities in low- 11. Data for North America and Bangladesh appear to be more compre- lying coastal areas during storms. hensive than data for China and Japan. 92 Natural Disaster Hotspots Case Studies Table 3.9. Deaths associated with major hurricanes, cyclones, and typhoons (MC) and extra-tropical storm (ETS) disasters (>1,000 deaths) since 1700. Events where surge is known to be only a minor cause of deaths are indicated. Note that there is a 1717 event in table 3.10 that is not included here as it is too imprecise in terms of number of deaths. Year Location Event Type Deaths Sources 1970 Bangladesh MC 300,000­500,000 1,3,4 1737 India MC 300,000 1,3 1881 China MC 300,000 1,3,4 1923 Japan MC 250,000 1,3,4 1584 Bangladesh MC 200,000 4 1897 Bangladesh MC 175,000 1,3,4 1991 Bangladesh MC 138,000­140,000 1,3,4 1694 Shanghai, China MC 100,000 10 1876 Bangladesh MC 100,000 1,3,4 1862 Zhujiang Delta, China MC 80,000 10 1847 India MC 75,000 1 1724 Jiangsu Province, China MC 70,000 10 1922 Santao, Guangdong, China MC 60,000­70,000 10, 12 1854 India MC 50,000 4 1912 China MC 50,000 12 1864 India MC 50,000 1,3,4 1833 India MC 33,000­50,000 1,3,4 1822 Bangladesh MC 40,000 1,3,4 1912 Bangladesh MC 40,000 4 1919 Bangladesh MC 40,000 4 1942 India MC 40,000 4 1780 Barbados, Martinique and MC 20,000­22,000 1,3,4,9, 11 St. Eustatius, Caribbean 1839 India MC 20,000 1,3,4 1789 (uncertain) India MC 20,000 1,3,4 1989 India MC 20,000 4 1965 (May 11) Bangladesh MC 19,279 1,3,4 1998* Honduras and Nicaragua (Hurricane Mitch) MC 10,000 ­17,000 5, 11 1965 (May 31) Bangladesh MC 12,000 4 1963 Bangladesh MC 11,500 1,3,4 1961 Bangladesh MC 11,468 4 1937 Hong Kong MC 11,000 4 1985 Bangladesh MC 11,000 3,4 1876 Bangladesh MC 10,000 1 1906 Hong Kong MC 10,000 4 1971 India MC 10,000 4 1999 Orissa, India MC 10,000 5 1974 Honduras MC 8,000­10,000 9, 11 1900 Galveston, Texas (U.S.) MC 8,000 (6,000­12,000) 6,9 1977 Krishna Delta, India MC 8,547­10,000 1,4,8 1930 Santo Domingo, Dominican Republic MC 8,000 9, 11 1941 Bangladesh MC 7500 4 1963 Cuba-Haiti MC 7,196­8,000 1,4,9 1991* Leyete, Philippines MC 6,000 12 1776 Guadeloupe MC 6,000 9 1988 Bangladesh MC 5,708 4 1960 (Oct 9) Bangladesh MC 5,149 1,4 1895 India MC 5,000 4 1959 Isle Bay, Japan MC 4,697 2 1775## Newfoundland Banks MC 4,000 9 1899 Puerto Rico & Carolinas (U.S.) MC 3,433 9 1928 Puerto Rico, Florida (U.S.) and Caribbean MC 3,411 9 1932 Cuba, Jamaica and Cayman Islands (U.K.) MC 3,107 9 Storm Surges in Coastal Areas 93 Table 3.9. continued Year Location Event Type Deaths Sources 1960 (Oct 30) Bangladesh MC 3,000 4 1934 El Salvador, Honduras MC 3,000 11 1934 Osaka Bay, Japan MC 2,702 2 1953 East Coast, UK and Delta Region, ETS 2,100 (1,800­2,300) 3, 7 the Netherlands 1945 Southern Kyushu, Japan MC 2,076 2 1893 Louisiana (U.S.) MC 2,000 6 1924 Leningrad (St. Petersburg), Russian Federation ETS 2,000 3 1893 South Carolina/Georgia (U.S.) MC 1,000­2,000 6 1928# Florida (U.S.), Puerto Rico, Guadelope MC 3,370 6, 11 1994 Fujian Province, China MC 1,216 10 1917 Tokyo Bay, Japan uncertain 1,127 2 1969 India MC 1,000 1 TOTAL (based on best/median estimate) 2.9 million fatalities Sources 1. Nicholls et al. (1995) 9. Elsner and Kara (1999) 2. Mimura et al. (1994) 10. Li et al. (2000) 3. Smith and Ward (1998) 11. Environment Canada Web Site www.ns.ec.gc.ca/weather/hurricane 4. Ali (1999) 12. http://www.noaa.news.noaa.gov/stories/s334b.htm 5. UNEP (2002) 13. * Deaths were mainly due to nonsurge effects, particularly high pre- 6. Pielke and Pielke (1997) cipitation and runoff 7. Kelman (2002) 14. # Deaths in Florida (1,836 people) were mainly due to flooding around 8. Winchester (2000) Lake Okeechobee, rather than ocean storm surge. 15. ## Assumed that deaths mainly due to shipwrecks. to better define the precise areas that were impacted Table 3.10. Deaths in storm surges around the in each case, and, hence, improve the capability to map North Sea from the 11th to the 18th centuries. them. All surges were due to extra-tropical storms. The data in table 3.9 show that at least 2.9 million Year Deaths people have died from storm surges since 158412, with 1200s >100,000 at least 2.6 million deaths since 1700. Surges due to 1200s >100,000 1200s >100,000 tropical storms are by far the major cause, with only 1200s 306,000 two extra-tropical storms being included in table 3.9 1446 >100,000 (and one event in table 3.10). The time series shows 1421 >100,000 that there has been a number of surge events with sig- 1570 400,000 1634 "some thousands" nificant fatalities throughout the last 300 years (figure 1671 "some thousands" 3.7). Smaller events are much better represented in the 1682 "some thousands" data after 1850, possibly indicating an increase in the 1686 "some thousands" number of surge events causing fatalities, or, more likely, 1717 "some thousands" TOTAL >1.2 million that more-comprehensive information is available for this more-recent period. This is shown in figure 3.8, Source: Lamb, 1995. where the number of events is presented for 50-year 12. Adding the data in table 3.10 would increase this number to nearly 4 million deaths since the 1200s. 94 Natural Disaster Hotspots Case Studies Figure 3.7. Deaths by major hurricanes, cyclones, and typhoons (MC) and extra-tropical storms (ETS) from 1700 to 2000. Data taken from Table 3.9, excluding those cases where storm surge was not the main cause of death. Figure 3.8. Number of "significant" events based on time periods based on different thresholds to define "sig- two thresholds of deaths: > 50,000 deaths, and > nificant" events. If all the surge events in table 3.9 are 20,000 deaths, as well as all events (>1,000 deaths) considered, the number of events increases substan- tially as we approach the present. This trend is not appar- ent if a higher threshold is considered; this circumstance would suggest that the number of "major" events peaked in the last 50 years of the 19th century and that the occurrence of "major" surges (in terms of fatalities) has declined subsequently. Figure 3.9 shows the average number of fatalities per year averaged over 50-year peri- ods. Since 1850, the long-term average is 10,000 to 15,000 deaths per year due to storm surges, although this method of presentation disguises the contribution of a few big events such as the 1970 cyclone in Bangladesh (figure 3.7). Only a few regions are represented in table 3.9, and these regions have some correspondence to those regions indicated in the previous section. Some impacts are apparent in the Caribbean and North America as well as in Europe. However, most fatalities have occurred in Asia, with the major hotspot for fatalities due to surges Storm Surges in Coastal Areas 95 Figure 3.9. Annual deaths due to surges, averaged over 50-year periods using the data in Table 3.8 is the Bay of Bengal.13 While the data quality is limited, small in contrast to the number at the Bay of Bengal, it is estimated that over the last 300 years about 1.7 mil- where a significant number of deaths has continued to lion people have been killed during cyclones, with the occur.15 Thus, table 3.9 identifies the Bay of Bengal, majority having drowned due to storm surges; this figure and especially Bangladesh, as the number-one surge translates into 65 percent of the global total of surge "hotspot" at the present time. deaths for that period. The most major event was the It is worth noting that the high incidence of death 1970 cyclone in which as many as 500,000 people due to surges around the Bay of Bengal is not unique may have been killed in a single event (Burton et al. in human history. There was similar loss of life due to 1993). The super-cyclone in Orissa, India, in 1999 storm surges around the North Sea in the late Middle that involved 10,000 fatalities serves as the most Ages (table 3.10). More recently, improved defenses recent major event (UNEP 2002).14 In East Asia (Japan have greatly reduced the death toll, but as demonstrated and China), there have been at least 800,000 deaths due in the 1953 storm surge (about 300 deaths in the United to surges over the last 300 years, comprising about 30 Kingdom and about 1,800 deaths in the Netherlands) percent of the global total. While the death toll has been and the 1962 storm surge (about 300 deaths in Ger- significant in some events, it is noteworthy that over the many), these areas remain threatened. Defenses have last 50 years, the number of deaths has been quite subsequently been improved to 1 in 1,000 standards or higher (making failure very unlikely), and opera- 13. All the fatalities in Bangladesh and most of those in India have occurred tional flood warning systems have been established, so on the coast of the Bay of Bengal, as there is a much lower frequency of cyclones (1:4) and less extensive coastal lowlands on the west coast of India. 15. However, subsidence has been a major problem, as illustrated by 14. Bangladesh is considered in more detail in the case studies section. Shanghai in the case studies section. 96 Natural Disaster Hotspots Case Studies Figure 3.10. Deaths per event for hurricanes making landfall in the United States the future death toll is likely to remain low. However, population lives in areas that would be seriously impacted the populations and investments in the flood-prone by storm surges during a category 3 or stronger hurri- areas are large and increasing (see the next section) cane; the estimate increases to 90 percent of the pop- (Nicholls and Branson 1998). It is probably no coinci- ulation in south-western Florida (Elsner and Kara 1999). dence that the United Kingdom, Germany, and the In the absence of hard defenses, evacuation is essen- Netherlands are preparing for sea-level rises with more tial, but could be problematic in some areas such as purpose than are most other coastal countries (Tol et the Florida Keys. New York City has been struck by a al. forthcoming). number of surges produced by extra-tropical storms The death toll from hurricanes has also declined on (so-called "northeasters"), including the December 1992 the U.S. coast over the 20th century (figure 3.10). In flood that came within 30 to 60 cm of causing wide- terms of response, this mainly reflects improved fore- spread flooding of the rail and tunnel infrastructure-- casts, warnings, and evacuation systems rather than this would have had severe impacts, including significant increased levels of protection. However, the decreased loss of life (Rosenzweig and Solecki 2001).16 death toll may also represent an element of luck, as there Several regions might be omitted by this analysis. In has been a lack of hurricane strikes on the most vul- particular, impacts on many of the small Pacific Islands nerable areas (Pielke and Pielke 1997). One concern is may not be adequately captured, although the death a potential direct hit on New Orleans, where much of the land lies below sea level and potential flood depths 16. Surges on the U.S. east coast are considered in more detail in the case are substantial. It is estimated that 25 percent of Florida's studies section. Storm Surges in Coastal Areas 97 toll appears uncertain. Major hurricanes can occur in due to surges is expected to fall relative to other related this region, as illustrated recently by Hurricane Zoe in hazards (although this is a heuristic rather than a December 2002, which impacted Tikopia and Anuta quantified concept at the present time). in the Solomon Islands with 148-mph (238 km/hr) Based on these observations, table 3.11 attempts to winds and an 18-ft (5.5 m) storm surge, plus waves. summarize the available information on surge hotspots Fortunately, this did not result in a large loss of life, but around the world. It distinguishes events that produce the damage to the islands' economies is immense, and fatalities from those events that produce other damages. it is unclear how rapidly they will recover. From 1965 For the same regions, table 3.12 identifies geographic to 1995, tropical cyclones in Fiji caused significant areas where the impacts from surge events could be par- economic damage, but most of this can be attributed ticularly significant. This judgment is based primarily to the non-surge impacts of the cyclone (Olsthoorn et on land elevation, population density, and historical al. 1999). The death toll was about 50 people, with experience, with some consideration of possible future surges being only a minor contributor. Equally note- conditions. worthy is that cyclone landfalls on the Indian Ocean In terms of information about the distribution of islands (for example, Mauritius) and on the East African storm surge hazard, tables 3.11 and 3.12 represent the coast do not appear to have produced significant num- most precise global view that has been possible to develop bers of fatalities to date. This probably reflects the lim- in the time available to prepare this paper. The detailed ited exposure to flooding at the present time. However, case studies in the following section are designed to this exposure might change, as might the frequency and illustrate the nature of storm surge hazard in a number magnitude of surge events (see the previous section). of the more vulnerable regions. We now consider other types of damage due to surges. Significantdamageduetohurricanesandtropicalcyclones has occurred in all of the areas defined in figure 3.1. Case Studies Damage due to surges produced by extra-tropical storms has occurred on the Atlantic coast of Europe, the North Four case study areas have been selected based on the Sea, the Baltic Sea, the Mediterranean, Northern China, preceding discussion. They serve to illustrate the nature Republic of Korea, Japan, and the eastern seaboard of of the surge hazard in four of the more vulnerable regions the United States. This damage can take many forms, identified in tables 3.11 and 3.12. Two of the areas are including direct damage to property, agriculture, and developed-world areas, reflecting the greater availabil- industry (especially coastally focused industries such as ity of information and the issues of high exposure and petrochemicals). Indirect damage due to disruption potential flood risk: (1) southern North Sea (United King- and dislocation also occurs. Lastly, there are the intan- dom, Netherlands, Germany), and (2) U.S. East Coast. gible impacts that are difficult, if not impossible, to meas- There are two developing-country areas: (3) Bangladesh, ure in economic terms, such as health effects and and (4) Shanghai (China). psychological well-being. The insurance industry has concerns about the financial damages, especially in devel- Southern North Sea oped countries where insurance coverage is high. However, as already noted, these impacts are related This region experiences significant surges due to extra- to a range of hazards, including storm surges that the tropical storms, and the locations susceptible to flood- tropical or extra-tropical storm generates. Therefore, it ing feature large populations and substantial investments. is more difficult to link these damages to surges. Of the Most of the flood-prone areas around the region are $30 billion or more in damages caused by Hurricane former wetlands that have been subject to land claim, Andrew, only about $100 million (or <3 percent) were which started in some areas 2,000 years ago under the directly linked to flooding due by surges (see table 7.3 Romans. Land claim created habitable areas that attracted in Pielke and Pielke 1997). In general, as the wind speeds larger populations to the coast. However, surges associated with a storm event increase, so the damage caused regular flooding and loss of life as illustrated in 98 Natural Disaster Hotspots Case Studies Table 3.11. An expert synthesis of storm surge hotspots around the world. Under 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. Surge-Prone Regions Hotspots Commentary Mentioned in Table 3.9 Fatalities* Other Damage Bay of Bengal (Bangladesh High High Improved flood warnings may reduce fatalities. uncertain and Eastern India) Western India/Pakistan ? Unclear ? Unclear Cyclones less frequent than Bay of Bengal (1:4) uncertain and less exposure. China/Japan Potentially high Potentially high Ongoing flood damage reported in China uncertain Rep. of Korea Low Low Lacks large low-lying coastal areas, but this is changing due to extensive land claim. yes Thailand, Vietnam, Potentially high Medium Frequently impacted by typhoons, and yes Philippines in deltas to high population of low-lying areas is growing rapidly, but not mentioned in Table 3.9, except for an event where surge was a minor cause of fatalities (hence, yes). Pacific Islands Probably high High Limited historical information. yes Australia and New Zealand Low Low Limited habitation in low-lying coastal areas yes Indian Ocean Islands Low Low Limited habitation in low-lying coastal areas yes Eastern Africa and Oman Low Low Habitation in low-lying coastal areas is not yes significant, but there is a potential to increase. Rio de la Plata (Argentina Low Low Difficult to assess due to limited literature-- yes and Uruguay) may suggest limited impacts to date. Caribbean Potentially Medium Human activity is concentrated around the uncertain high to high islands, and hence exposed to surge--however, the role of surge relative to other hurricane impacts is less clear. Central America and Potentially high Medium Human activity is often concentrated away uncertain Mexico in local areas to high from the coast, which is untypical globally. Hence, other hurricane impacts appear relatively more important than in other regions (for example, Hurricane Mitch), although there are localized hotspots. U.S. Gulf and East Coasts Potentially high High Effective evacuation has reduced fatalities, but uncertain potential hotspots remain. Europe--Atlantic coast Potentially high Potentially Hard defenses and improved flood predictions yes high and warnings appear to have been effective in reducing this hazard. Europe--Mediterranean Locally high Medium to Surges are not large, so deaths are unlikely, yes coast high except in areas of land claim where flood depths could be substantial. However, significant damage and disruption can occur. Europe--North Sea coast Potentially high Potentially Hard defenses and improved flood predictions uncertain high and warnings appear to have been effective in reducing this hazard. Europe--Baltic Sea coast Locally high Medium to Hard defenses and improved flood predictions and warnings appear to have been effective in uncertain high reducing this hazard. *Note: Under fatalities, high indicates the potential for more than 1,000 deaths in a surge event. Other damage estimates are based on the expert judg- ment of the author. Storm Surges in Coastal Areas 99 Table 3.12. Potential and actual hotspots vulnerable to flooding by the storm surge. This information is indicative; it is not an exhaustive list of all potential and actual hotspots. Surge-Prone Regions Potential and Actual Hotspots Bay of Bengal (Bangladesh Ganges-Brahmaputra mouth (Figure 3.2), (Bangladesh and West Bengal), Mahandi Delta (Orissa) and and Eastern India) 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 Deltaic Plain (China), Pearl River Deltaic Plain Guangzhou and Hong Kong (China), Guangxi Coastal Plain (China), North Hainan Plain (China), Taiwan Coastal Plain and Taipei (Taiwan), Metropolitan Toyko (Japan), Metropolitan Osaka (Japan) Republic of Korea inconclusive Thailand, Vietnam, Red River delta (Vietnam), Mekong delta (Vietnam), Metro Manila (Philippines), Chaophraya delta and Philippines Bangkok (Thailand) Pacific Islands Most capital cities which are all on the coast, and all atoll islands Australia and New Zealand inconclusive Indian Ocean Islands inconclusive Eastern Africa and Oman inconclusive Rio de la Plata (Argentina Buenos Aires (Argentina) and Montevideo (Uruguay) (assumed) and Uruguay) Caribbean Most capital cities which are on the coast Central America and Mexico inconclusive U.S. Gulf and East Coasts New York City, Florida, particularly southern Florida and the Keys, New Orleans Europe--Atlantic coast inconclusive 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 (U.K.), the western Netherlands, Hamburg, and Bremen Germany Europe--Baltic Sea coast Main hotspot is St Petersburg (Russian Fed.) with potential hotspots such as Helsinki, (Finland) and Copenhagen (Denmark). table 3.10. As technology improved, so did defenses. 10,000-year standard for some defenses in the Nether- The catastrophic losses described in table 3.10 ceased lands.17 Equally important, an effective storm tide warn- in the 18th century, helped to some degree by a decline ing service has been developed that provides up to in the frequency of major coastal storms (Lamb 1995). 36-hours warning of a potential flood event. Collec- However, major floods continued up to the middle of tively, these new measures have been effective for the the 20th century: last 20 years18, and there has been no flooding around the southern North Sea, even though the extreme · The East Coast (United Kingdom) and the delta water levels of the 1953 event have been repeated and region of the Netherlands (figures 3.11 and 3.1) were even exceeded in some locations. However, the human last flooded in 1953. and infrastructural exposure remains substantial, poten- · Germany, including Hamburg, was last flooded in tially approaching 15 million people19 today, and there 1962. The response to these events was further massive 17.The only other place in the world where so many defenses are built to investment in flood-defense infrastructure, as illustrated a similar standard is Japan. by the mobile storm surge barriers on the Thames in 18.The new defenses were not completed until the early 1980s, suggesting Greenwich, London, and across the Western Scheldt that a 30-year time lag in response to flooding via major structural measures should always be assumed. in the Netherlands. Many of these defenses are built to 19.The population living beneath the 1-in-1,000-year storm surge a 1-in-1,000-year standard or higher, and up to a 1-in- elevation. 100 Natural Disaster Hotspots Case Studies Figure 3.11. Flooding of the East Coast of England during the 1953 storm surge Storm Surges in Coastal Areas 101 is rapid development occurring in these flood-prone Thames Gateway proposal reinforces the need for this areas. These areas also are subject to increasing risks upgrade, and it will be interesting to see if the design due to sea-level rise and climate change (Flather and standard (that is, the acceptable level of risk) is main- Smith 1998; Lowe and Gregory 1998). tained at the 1-in-1,000 level, or if it is increased fur- London, a city whose location was selected by the ther toward the levels seen in the Netherlands. The Romans 2,000 years ago, illustrates many of these issues. new flood strategy includes consideration for the first One million Londoners, or about 12 percent of the city's time of inland realignment of the flood defense line (that population, are potentially exposed to flooding caused is, a planned retreat policy [Klein et al. 2001]) as a by storm surges. They are defended by a complex system complimentary strategy to raising defenses. This reverses involving fixed flood defenses of varying standards along a long-term trend of encroachment and land claim into the entire length of the tidal Thames, the mobile Thames the tidal Thames (Shih 2002). Barrier at Greenwich (which is closed before a surge It is noteworthy that it is not widely appreciated that arrives), and a suite of warning systems that are used there is and always will be a residual flood risk for to decide when to close the Barrier (Gilbert and London. The operational flood management authority Horner 1984). While the possibility of a barrier was dis- is the (England and Wales) Environment Agency. It is cussed earlier in the 20th century, the decision to build trying to communicate this point to Londoners and to the present defenses was made in direct response to begin education about what might happen in the highly the 1953 storm surge. While there were a few deaths unlikely event of a flood. However, this is a difficult within London proper and while the city was largely issue that requires much more work-- emergency plan- spared on this occasion, the high vulnerability of London ning needs to be interwoven much more with the post- was apparent to all, and it galvanized collective action. 2030 plans for flood management. In terms of flood The new defenses became fully operational when the damages, all insured properties in London receive flood Barrier was completed in 1983--that is, 30 years after coverage as part of a standard home insurance policy. the decision to build was made. Parallel developments Therefore, losses would be shared in the unlikely event in storm surge warning were fundamental to the oper- of a flood event. However, it is worth noting that the ation of the Barrier. Since the Barrier was completed, U.K. insurance industry is concerned about its expo- London's derelict docklands have been regenerated with sure to flooding, as this is seen as one of the biggest new transport links, homes, and businesses, including potential losses that U.K. insurers face. Flood insur- the important new financial district around Canary ance is being selectively withdrawn in areas outside Wharf. Significant future development is planned along- London: this is a situation that would arguably reduce side the Thames, with 200,000 new homes proposed the resilience of England in the face of major flood in the next 15 years alone (termed the Thames Gate- events (Clark 1998). way proposal). This will extend London eastward toward In conclusion, the surge hazard co-evolved with the North Sea, and many of the proposed sites for build- human development of the coastal zone, and this con- ing are potentially flood prone. tinues today. Land claim created the conditions that The design life of the Barrier extends to 2030, when attracted high populations to the flood-prone areas and, rising flood levels due to a combination of global sea- hence, raised the exposure in a region where large (and, level rise and more local changes will reduce the resid- thus, potentially) killer surges can occur. The coastal ual flood risk to below a 1-in-1,000-year standard. Given populations adapted to these threats, but it is only in the long lead-time to upgrade the defenses, planning the last 50 years that the threat has really been reduced, of the upgrade of the flood defenses to the end of the primarily due to protection and warning measures. How- 21st century is already in its early stages. Substantial ever, as the long history of this region shows, compla- raising of the fixed flood defenses will be required, cency could be fatal, and flood management will need although it is hoped the Barrier can continue to oper- to keep developing to manage the changing risks of ate until 2100 with only marginal investment. The surge flooding. 102 Natural Disaster Hotspots Case Studies Bangladesh, Bay of Bengal from surge events, and the new population was highly vulnerable to coastal flooding (Burton et al. 1993). In terms of fatalities, Bangladesh is presently the dom- This is best illustrated in the 1970 cyclone event, in inant storm surge hotspot globally, as already discussed. which between 300,000 and 500,000 people drowned This concentration of fatalities reflects several interact- in coastal Bangladesh. ing factors that have a lot in common with the issues Figure 3.12 shows the death toll during cyclones from raised in section on the southern North Sea. These fac- the years 1800 to 2000. The 1970 event stands out as tors include (1) a high and rapidly growing coastal pop- the most significant event in terms of fatalities and would ulation, with little alternative land, (2) extensive coastal seem to be part of a rising trend in deaths due to surges.20 lowland areas that are close to sea level, (3) significant It can be argued that land claim has been one factor exac- and frequent landfall of tropical cyclones, and (4) shal- erbating vulnerability to surges in Bangladesh. The 1970 low coastal areas that exacerbate the surge potential of event caused a reassessment of preparations for surge the cyclones. Coastal Bangladesh is also an active delta flooding in coastal Bangladesh. There was the building (of the Ganges-Brahmaputra), and net accretion of of robust shelters for people and their animals, as well land is occurring (Nicholls et al. 1995). With the pres- as improvements in the forecast and warning systems, sure for land and the need for food, this has led to exten- with an emphasis on how to disseminate warnings sive land claim of these emerging areas and active promotion of accretion (for example, Koch 1986). How- ever, these new land areas were not well protected 20. The 1876 cyclone in Bangladesh may have had a death toll of up to 400,000, although the uncertainties are great. Figure 3.12. Deaths per surge event in Bangladesh from 1800 to 2000 using the data in Table 3.9. Storm Surges in Coastal Areas 103 from Dhaka (which is inland) to the threatened people in Bangladesh will continue to co-evolve with human along the coast. While the loss of life in the 1991 event use of the coastal zone throughout the 21st century. was again significant, it was substantially lower than in 1970, despite the event being comparable in terms of U.S. East Coast the surge characteristics (Kausher et al. 1996). Efforts to improve warnings have continued through the 1990s, The U.S. East Coast faces surges from hurricanes (in and it is suggested rather anecdotally that they are becom- summerandautumn)andextra-tropicalstorms(so-called ing more effective (White 2000). However, as the death northeasters) in the autumn to spring months. The analy- toll falls, there is concern that the lessons learned from sis of Zhang et al. (2000) shows that the relative impor- the avoided floods are not being considered. The tance of these two types of surges changes as you move underlying pressures remain, and without substantial northward: north of Hampton Road, Virginia, north- and continuing efforts, it is likely that significant deaths easters dominate, while to the south tropical storms and due to surges will recur, although probably not on the hurricanes are apparent. Figure 3.13 (a) shows the poten- scale of the 1970 and 1991 events. tial for surges on the Gulf and East Coasts, while figure While the death toll appears to be declining, other 3.13 (b) shows the surge that occurred during the hur- surge-related damages are likely to remain significant, ricane of September 14­15, 1944.21 A major surge due although, not surprisingly, most of the literature focuses to a northeaster occurred during the Ash Wednesday on the large number of deaths. Further efforts are likely to be necessary to mitigate flood damage and disruption. Hence, it would seem that the surge hazard 21. The smaller surges after the main surge are called the resurgences (Smith and Ward 1998). Figure 3.13. Surges on the U.S. Gulf and East Coast. Relative storm-surge potential (a), and surge graphs for six Atlantic coast locations (b), the hurricane of September 14­15, 1944. 104 Natural Disaster Hotspots Case Studies Figure 3.14. Subsidence from the 1920s to the 1990s in Shanghai, China storm of 1962. The surge was up to 2 meters high and lower than the population density around the North persisted for 5 high tides, resulting in major coastal ero- Sea and in Bangladesh. Hence, the death toll during sion along a large length of the U.S. east coast. More surge events has been relatively small and has dimin- recently, the December 1992 northeaster produced sub- ished over time, as shown for the East and Gulf Coasts stantial floods around New York City, including flood- in figure 3.10. On the East Coast, the largest death toll ing some important parts of the transportation in the 20th century occurred during the 1938 hurri- infrastructure, and coming close to flooding much more cane in which 600 people died (Pielke and Pielke 1997). critical parts of the underground transportation infra- This has led to a different approach to management of structure (Rosenzweig and Solecki 2001). the surge hazard. Instead of the large investment in dikes Apart from areas such as New York City, Miami, and and surge barriers seen around the North Sea, there New Orleans, the U.S. coastal population density is has been a focus on: Storm Surges in Coastal Areas 105 (1) effective warning systems and evacuation plans, and the area affected (Guo 1991).22 A range of new flood (2) flood-proofing and raising of new buildings above protection measures was implemented, including a lot the 1-in-100-year flood elevation. of small-scale measures such as flood barriers and sand bags for individual buildings. This culminated in large This has reduced the death toll in storm events, but new flood walls, built in the early 1990s, that protect storm damages continue to be significant (Pielke and the main city to a 1-in-1,000-year standard. However, Pielke 1997). An underlying driver of increased damage future subsidence problems remain possible. Anecdo- is a coastward relocation of the expanding national pop- tal reports suggest that illegal groundwater withdrawal ulation. This trend seems set to continue, and, hence, has increased in Shanghai over the last 10 years, and exposure to storms will inevitably continue to increase. the rate of subsidence has increased again. This illus- When combined with widespread erosional trends trates the ongoing nature of managing surge-induced and also rapid relative sea-level rises during the 20th flooding, as is apparent in all the case studies. century, this suggests that the impacts due to flooding This experience will have commonalities with many during surges will inevitably increase. While the United other large subsiding cities, all of which are in Asia (table States has the resources to respond to these challenges, 3.2). There are several other cities that might start to it will be interesting to see how rapidly the country experience subsidence as they develop and, hence, moves to increase protection along the U.S. East Coast. become more exposed to surges. Hanoi and Ho Chi The expansion of the use of beach nourishment might Minh City (Saigon) in Vietnam, and Yangon in Myan- be seen as a first step in this direction (Neumann et al. mar are potential examples, with Hanoi known to be 2001). More focused management is required where actively subsiding (Tom et al. 1996). critical infrastructure is threatened (as in New York City), as these areas are potential hotspots where major surge-related losses could occur (Rosenzweig and Solecki Conclusions 2001). This document represents a first attempt to draw together Shanghai, China the information related to storm surge "hotspots." The relevant information is widely scattered and often not Shanghai is a good example of a sinking coastal city in a form that can be readily synthesized and com- (table 3.2). It is built on geologically young deposits pared across regions. Therefore, it should be consid- of the Changjiang (Yangtze) delta, and it subsided as ered as a work in progress rather than as a definitive much as 2.8 m during the 20th century (figure 3.14) statement on storm surge hazard. It is also apparent that due to shallow, unregulated groundwater withdrawal only the broad regions that are vulnerable to surges (Han et al. 1995; Wang et al. 1998). The groundwater can be mapped with the present level of knowledge (see withdrawals were triggered by the growing city and tables 3.11 and 3.12), which points to the need to con- economy in the 1920s, and subsidence continued until tinue this type of analysis so that comparative studies the 1960s when groundwater withdrawal was regu- of hazards can be improved and developed. This pro- lated and subsidence rates were reduced to 3 to 4 mm vides an improved basis for sharing experiences and is per year--rates of subsidence one would expect in a fundamental to many international efforts that need deltaic setting. Therefore, while human action triggered objective methods to prioritize and target the limited the subsidence problem, this also made it possible to resources for hazard mitigation. greatly reduce the subsidence by managing the ground- Nonetheless, a number of important conclusions can water withdrawal. be drawn that are of relevance to the "hotspot" analysis: Shanghai was always flood prone due to both high river flows and typhoons. Table 3.9 includes 100,000 deaths due to flooding during a typhoon in 1694. How- ever, the subsidence promoted a substantial increase in 22. The subsidence has a range of other impacts, including damage to build- ings (Nicholls 1995). the incidence of flooding, the actual flood depths, and 106 Natural Disaster Hotspots Case Studies · Surges are a major issue in only a few global regions, ing the SRES Storylines. Global Environmental Change 14(1): with the Bay of Bengal being the most affected region, 3­20. and Bangladesh being the main "hotspot" for surge Ascher, G. 1991. The Case of Hamburg. In: Impact of Sea Level Rise on Cities and Regionsm, R. Frasetto (ed.). Proceedings of impacts. the First International Meeting "Cities on Water," Venice, · While surges are only one aspect of the impacts of a December 1989. Marsilio Editori, Venice, 75­80. storm, they are the main killer, and surges have led Bijlsma, L., et al. 1996. Coastal Zones and Small Islands. In: to several million deaths over the last two centuries, Impacts, Adaptations and Mitigation of Climate Change: Scien- mainly in Asia and particularly in Bangladesh. tific-Technical Analyses, R. T. Watson, M. C. Zinyowera, and · High death rates due to surges appear to be linked R. H. Moss (eds.), Contribution of Working Group II to the Second Assessment Report of the Intergovernmental Panel to land claim and substantial coastal modification, on Climate Change. Cambridge, UK: Cambridge University which have encouraged growth in vulnerable coastal Press, 289­324. populations without appropriate consideration of the Burton, I., R. W. Kates, and G. F. White. 1993. 2nd edition. potential for surges (for example, southern North Sea New York: The Guildford Press, 290. and Bangladesh). Carnell, M. G. R., and C. A. Senior. 1998. Changes in Mid-Lat- · The death toll in surge events appears to have fallen itude Variability due to Increasing Greenhouse Gases and Aerosols. Climate Dynamics 14: 369-383. substantially around the world as protection meas- Carter, R. W. G. 1988. Coastal Environments. Academic Press, ures and forecasts/warnings are improved, including 617. most recently in Bangladesh. Church, J.A. 2001. Changes in sea level. In: Climate Change 2001: · However, there is no room for complacency, and the The Scientific Basis, J. T. Houghton, et al. (eds.), Contribution surge hazard will continue to evolve throughout the of Working Group I to the Third Assessment Report of the 21st century due to changing socioeconomic condi- Intergovernmental Panel on Climate Change. Cambridge, UK: Cambridge University Press, 639­693. tions, coastal land use, and climatic risks. Clark, M. 1998. Flood Insurance as a Management Strategy for · Damages and disruptions due to surges are more UK Coastal Resilience. The Geographical Journal 164(3): difficult to define as they are one aspect of the storm 333­343. and as these impacts are often aggregated with other Edwards, K. C. 1953. The storm floods of 1st February 1953: damages, such as damages caused by tornados and VIII. The Netherlands Floods: Some Further Aspects and Con- other types of wind storms. sequences. Geography 38: 182­187. Elsner, J. B., and A. B. Kara. 1999. Hurricanes of the North Atlantic: Therefore, while it is useful to analyze surge by Climate and Society. New York and Oxford: Oxford University itself, it is also important to analyze the integrated impacts Press, 488. Emanuel, K. A. 1988. The Dependency of Hurricane Intensity of coastal storms, as these are what coastal communi- on Climate. Nature 326: 483­485. ties experience. In the future, a mixture of analyses is Flather, R., and J. Smith. 1998. First Estimates of Changes in required--one that considers each storm hazard, as well Extreme Storm Surge Elevations due to the Doubling of as the integrated impacts. CO2. The Global Atmosphere and Ocean System 6: 193­208. French, P. W. 1997. Coastal and Estuarine Management. London: Routledge, 251. References Gilbert, S., and R. Horner. 1984. The Thames Barrier. London: Thomas Telford, 182. Guo, H. 1991. Causes and the Performance of the Construction Alexandersson, H., H. Tuomenvirta, T. Schmith, and K. Iden. Work for Flood Tide Prevention in Shanghai. In: Impact of Sea- 2000. Trends of Storms in NW Europe Derived from an Updated Level Rise on Cities and Regions, R. Frassetto (ed.). Proceedings Pressure Dataset. Climate Research 14: 71­73. of the First International Meeting "Cities on Water" (Venice, Ali, A. 1999. Climate Change Impacts and Adaptation Assess- December 11­13, 1989), Marsilo Editori, Venice, 150­168. ment in Bangladesh. Climate Research 12: 109­116. Han, M., J. Hou, and L. Wu. 1995. Potential Impacts of Sea- Araujo, I., D. Pugh, and M. Collins. 2002. Trends in Compo- Level Rise on China's Coastal Environment and Cities: A nents of Sea Level around the English Channel. Proceedings of National Assessment. Journal of Coastal Research, Special Issue Littoral 2002 EUROCOAST 107­114. No. 14: 79­95. Arnell, N. W., et al. 2004. Climate and Socioeconomic Scenar- Harleman, D. R. F., et al. 2000. Blocking the Tide. Civil Engi- ios for Climate Change Impacts Assessments: Characteris- neering, October 2000: 52­57. Storm Surges in Coastal Areas 107 Henderson-Sellers, A., et al. 1998. Tropical Cyclones and Global Li, C. X., et al. 2000. Some Problems of Vulnerability Assess- Climate Change: A Post-IPCC Assessment. Bulletin of the Amer- ment in the Coastal Zone of China. In: Global Change and Asia- ican Meteorological Society 79: 19­38. Pacific Coasts, N. Mimura and H. Yokoki (eds.). Proceedings Hoozemans, F. M. J., M. Marchand, and H. A. Pennekamp. 1993. of APN/SURVAS/LOICZ Joint Conference on Coastal Impacts A Global Vulnerability Analysis: Vulnerability Assessment for Pop- of Climate Change and Adaptation in the Asia-Pacific ulation, Coastal Wetlands and Rice Production on a Global Scale, Region, Kobe, Japan. November 14­16, 2000. Asian-Pacific 2nd edition. The Netherlands: Delft Hydraulics. Network for Global Change Research, Kobe and the Center Houghton, J. Y., et al. (eds.). 2001. Contribution of Working for Water Environment Studies, Ibaraki University, 49­56. Group I to the Third Assessment Report of the Intergovern- (Downloadable at www.survas.mdx.ac.uk) mental Panel on Climate Change. Cambridge, UK: Cambridge Lowe, J., and J. Gregory. 1998. A Preliminary Report on Changes University Press. in the Occurrence of Storm Surges around the United King- IPCC CZMS. 1990. Strategies for Adaptation to Sea Level Rise. dom under a Future Climate Scenario. Hadley Centre for Cli- Report of the Coastal Zone Management Subgroup, Response mate Prediction and Research (unpublished). Strategies Working Group of the Intergovernmental Panel on Mimura, N., M. Isobe, and Y. Hosakawa. 1994. Impacts of Sea Climate Change, Ministry of Transport, Public Works and Level Rise on Japanese Coastal Zones and Response Strate- Water Management. The Hague, 122. gies. In: Global Climate Change and the Rising Challenge of the Johns, T. C., et al. 2003. Anthropogenic Climate Change from Sea, J. O'Callaghan (ed.). Proceedings of the Third IPCC CZMS 1860 to 2100 Simulated with the HadCM3 Model under Workshop, Margarita Island, Venezuela, March 9­13, 1992, Updated Emissions Scenarios. Climate Dynamics, submitted National Oceanic and Atmospheric Administration, Silver in revision. Spring, MD, USA, 329­349. Kausher, A., et al. 1996. Climate Change and Sea-Level Rise: Murty, T. S. 1984. Storm Surges--Meteorological Ocean Tides. The Case of the Coast. In: The Implications of Climate and Sea- Canadian Journal of Fisheries and Aquatic Sciences (Bulletin 12): Level Change for Bangladesh, R.A. Warrick and Q. K. Ahmad 897. (eds.). Kluwer, Dordrecht, 335­405. Nakicenovic, N., et al. 2000. Special Report on Emissions Sce- Kelman, I. 2002. CURBE Fact Sheet 3: U.K. Deaths from the narios: A Special Report of Working Group III of the Inter- 1953 Storm Surge. Version 2, January 11, 2003. Down- governmental Panel on Climate Change. Cambridge: Cambridge loaded from http://www.ilankelman.org/disasterdeaths/1953 University Press, 599. DeathsUK.doc Neumann, J. E., et al. 2001. Sea Level Rise and its Effects on Klein, R. J. T., and R. S. J. Tol. 1997. Adaptation to Climate Change: Global Resources. In: Climate Change: Science, Strategies and Options and Technologies--An Overview Paper. Technical Solutions, E. Claussen (ed.). Boston: Brill, 43­62. Paper FCCC/TP/1997/3, United Nations Framework Con- Nicholls, R. J. 1995. Coastal Megacities and Climate Change. vention on Climate Change Secretariat. Bonn, Germany, 33. Geojournal 37(3): 369­379. Klein, R. J. T., R. J. Nicholls, and N. Mimura. 1999. Coastal Adap- Nicholls, R. J. 2004. Coastal Flooding and Wetland Loss in the tation to Climate Change: Can the IPCC Technical Guide- 21st century: Changes under the SRES Climate and Socioe- lines be Applied? Mitigation and Adaptation Strategies for Global conomic Scenarios. Global Environmental Change 14, 69­86. Change 4: 51­64. Nicholls, R. J., and J. Branson. 1998. Coastal Resilience and Plan- Klein, R. J. T., et al. 2000. Coastal-Adaptation Technologies. ning for Uncertainty: An Introduction. Geographical Journal In: IPCC Special Report on Methodological and Technological Issues 164: 255­258. in Technology Transfer, B. Metz, et al. (eds.). Cambridge, UK: Nicholls, R. J., and F. M. J. Hoozemans. 1996. The Mediter- Cambridge University Press, 349­372. ranean: Vulnerability to Coastal Implications of Climate Change. Klein, R. J. T., et al. 2001. Technological Options for Adapta- Ocean and Coastal Management 31: 105­132. tion to Climate Change in Coastal Zones. Journal of Coastal Nicholls, R. J., F. M. J. Hoozemans, and M. Marchand. 1999. Research 17(3): 531­543. Increasing Flood Risk and Wetland Losses due to Global Sea- Koch, F. H. 1986. Land Development in the Ganges-Brahma- Level Rise: Regional and Global Analyses. Global Environmental putra Estuary in Bangladesh. Presented at Symposium on Low Change 9: S69­S87. Land Development, Indonesia, August 1986. In: Sea-Level Nicholls, R. J., N. Mimura, and J. C. Topping. 1995. Climate Rise: A Selective Retrospective, P. C. Schroder (ed.). Delft: Delft Change in South and South-East Asia: Some Implications for Hydraulics, 14. Coastal Areas. Journal of Global Environmental Engineering 1: Knutson, T. R., R. E. Tuleya, and Y. Kurihara. 1998. Simulated 137­154. Increase in Hurricane Intensities in a CO2-Warmed Climate. Olsthoorn, A. A., W. J. Maunder, and R. S. J. Tol. 1999. Tropi- Science 279: 1018­1020. cal Cyclones on the Southwest Pacific: Impacts on Pacific Lamb, H. H. 1995. Climate, History and the Modern World, 2nd Island Countries with Particular Reference to Fiji. In: Cli- edition. London: Routledge, 433. mate, Change and Risk, T. E. Downing, A. J. Olsthoorn, and R. S. J. Tol. (eds.). London: Routledge, 221­224. 108 Natural Disaster Hotspots Case Studies Parker, D. J. (ed.). 2000. Floods, 2 volumes. London: Rout- Smith, K., and R. Ward. 1998. Floods: Physical Processes and ledge, 431 and 317. Human Impacts. Chichester: Wiley, 382. Peerbolte, B. 1994. Hazard Appraisal: Modelling Sea-Level Rise Steers, J. A. 1953. The East Coast Floods, January 31­February and Safety Standards. In: Floods Across Europe, E. Penning- 1, 1953. Geographical Journal 119: 280­298. Rowsell, and M. Fordham (eds.). London: Middlesex Uni- Tol, R. S. J., R. J. T. Klein, and R. J. Nicholls. Forthcoming. versity Press, 107­134. Towards Successful Adaptation to Sea Level Rise along Europe's Penning-Rowsell, E. C. 2000. Has Venice Crossed the Rubicon? Coasts. Journal of Coastal Research. In: Floods, Volume 1, D. J. Parker (ed.). London: Routledge, Toms, G., et al. 1996. Coast Zone: Vulnerability Assessment. 277­287. First Steps Towards Integrated Coastal Zone Management. Penning-Rowsell, E. C., and M. Fordham (eds.). 1994. Floods Final Report. Government of the Socialist Republic of Viet- Across Europe. London: Middlesex University Press, 214. nam and Government of the Netherlands, UNEP, 2002. Global Penning-Rowsell, E., et al. 2003. The Benefits of Flood and Environment Outlook 3. London, UK: Earthscan, 446. Coastal Defence: Techniques and Data for 2003. Flood Hazard United Nations Environment Programme (UNEP). 2002. GEO: Research Centre, Middlesex University, 322. Global Environment Outlook 3: Past, Present and Future Per- Perez, R. T., L. A. Amadore, and R. B. Feir. 1999. Climate Change spectives. pp446. Impacts and Responses in the Philippines Coastal Sector. Cli- Wang, B., et al. 1998. Potential Impacts of Sea-Level Rise on the mate Research 12: 97­107. Shanghai Area. Journal of Coastal Research, Special Issue No. Pielke, R. A. and R. A. Pielke. 1997. Hurricanes: Their Nature 14: 151­166. and Impacts on Society. New York: Wiley, 279. Warrick, R. A., and Q. K. Ahmad (eds.). 1996. The Implications Rosenzweig, C., and W. D. Solecki (eds.). 2001. Climate Change of Climate and Sea-Level Change for Bangladesh. Kluwer, Dor- and a Global City: The Potential Consequences of Climate drecht, 415. Variability and Change--Metro East Coast. Report for the US Warrick, R. A., et al. 2000. Climate Change, Severe Storms and Global Change Research Program, National Assessment of the Sea Level: Implications for the Coast. In: Floods, Volume 2, Potential Consequences of Climate Variability and Change for D. J. Parker (ed.). London: Routledge, 130­145. the United States. New York: Columbia Earth Institute, 224. WASA Group. 1998. Changing Waves and Storms in the Sachs, J. D., A. D. Mellinger, and J. L. Gallup. 2001. The Geog- Northeast Atlantic. Bulletin of the American Meteorological raphy of Poverty and Wealth. Scientific America 284(3, March): Society 79: 741­760. 70­75. WCC'93. 1994. Preparing to Meet the Coastal Challenges of Shih, S. 2002. A Framework for Analysis of Managed Realign- the 21st century. Report of the World Coast Conference, Noord- ment for the Thames Estuary. MSc Thesis. Enfield, London, wijk, The Netherlands, November 1­5, 1993. Ministry of UK: Middlesex University, 88. Transport, Public Works and Water Management. The Hague, Small, C., V. Gornitz, and J. E. Cohen. 2000. Coastal Hazards 49 + apps. and the Global Distribution of Human Population. Environ- White, G. F. 2000. Lessons for Flood Hazard and Disaster Man- mental Geosciences 7: 3­12. agement from the International Decade­and Future Chal- Small, C., and R. J. Nicholls. 2003. A Global Analysis of lenges. In: Floods, Volume 1, D. J. Parker (ed.). London: Human Settlement in Coastal Zones. Journal of Coastal Research Routledge, 271­275. 19(3): 584­599. Willows, R., and R. Connell (eds.). 2002. Climate Adaptation: Smit, B., et al. 2001. Adaptation to Climate Change in the Con- Risk, Uncertainty and Decision-Making. UKCIP Technical text of Sustainable Development and Equity. In: Climate Change Report. (www.ukcip.org.uk). 2001: Impacts, Adaptation and Vulnerability, J. J. McCarthy, O. Winchester, P. 2000. The Political Economy of Riverine and F. Canziano, and N. Leary (eds.). Contribution of Working Coastal Floods in South India. In: Floods, Volume 1, D. J. Parker Group II to the Third Assessment Report of the Intergovern- (ed.). London: Routledge, 56­68. mental Panel on Climate Change. Cambridge, UK: Cambridge Zhang, K., B. C. Douglas, and S. P. Leatherman. 2000. Twenti- University Press, 877­912. eth-century Storm Activity along the U.S. East Coast. Journal of Climate 13: 1748­1761. Chapter 4 Natural Disaster Risks in Sri Lanka: Mapping Hazards and Risk Hotspots Lareef Zubair and Vidhura Ralapanawe, Upamala Tennakoon, Zeenas Yahiya, and Ruvini Perera Introduction SL Rs 89 million (approximately US$1 million), whereas floods accounted for only SL Rs 7.5 million. The goals for this case study of natural disasters in Sri The prevalence of drought may be surprising given Lanka were (1) to examine the methodologies needed for that Sri Lanka receives an average of 1,800 mm of rain- subnational assessments of hazard, vulnerability, and fall annually. However, it is distributed unevenly both hotspots; (2) to assess the interplay among hazards and spatially (figure 4.2.a) and temporally (figure 4.2.b). A vulnerability; and (3) to assess the consequence of com- large part of the island is drought prone from February binations of multiple hazards and vulnerability factors. to April and, if the subsidiary rainy season from May In the terminology used here, a "natural disaster" occurs to June is deficient, drought may continue into Sep- when the impact of a hazard is borne by "elements at tember. In our analysis, we use a regionalization of Sri risk" that may be vulnerable to the hazard. The elements Lankan climate into four climatologically homogeneous considered in this study are simplified into categories of regions (Puvaneswaran and Smithson 1993)--western people, infrastructure, and economic activities. and eastern slopes and northern and southern plains-- Sri Lanka has an area of 65,000 square kilometers as shown in figure 4.2.a. and a population of 18.7 million (Department of Census During the time frame of the study, disaster man- and Statistics 2001).The principal topographic feature agement has been carried out in Sri Lanka by the Depart- is an anchor-shaped mountain massif in the south- ment of Social Services under the Ministry of Social central part of the island (figure 4.1). The topography Services. Relief work for disasters is the responsibility and differences in regional climate (figures 4.2 a and b) of the parent body, the Ministry of Social Welfare. The are underlying causes of the contrasts in many facets of Government of Sri Lanka is currently revising its orga- the island. nizational structure for dealing with and planning for The most frequent natural hazards that affect Sri natural and manmade disasters. Lanka are droughts, floods, landslides, cyclones, vector- Our analysis is carried out in the context of civil wars borne epidemics (malaria and dengue), and coastal ero- that, together, extended from 1983 to 2002. During this sion (Tissera 1997). Tsunamis are infrequent but have period, natural disasters accounted for 1,483 fatalities, caused severe damage. Recent understanding of the tec- while civil wars accounted for more than 65,000. War tonics of the Indian Ocean region points to an increas- has devastated infrastructure and communities' ability ing risk of earthquakes. The risk of volcanoes is small. to deal with hazards, reduced incomes, weakened safety Here, we have addressed only those hazards related to nets, and undermined capacity to recover from hazard droughts, floods, landslides, and cyclones. We are map- events. For example, there has been a severe toll on hos- ping spatial risks of epidemics in a separate project to pital availability. Although there has been peace since develop an early warning system. 2002, longer-term consequences such as unexploded Drought is the most significant hazard in terms of landmines, war orphans, and the war-disabled continue. people affected and relief provided. The relief dis- The availability of data on hazards and vulnerability is bursements for drought between 1950 and 1985 were restricted in the war zones. The vulnerability analysis 109 110 Natural Disaster Hotspots Case Studies Figure 4.1. The district boundaries of Sri Lanka are shown over the topography Natural Disaster Risks in Sri Lanka: Mapping Hazards and Risk Hotspots 111 Figure 4.2.a. The average annual rainfall climatology estimated based on data from 284 stations in the period between 1960 and 1990. Homogenous climatological regions as proposed by Puvaneswaran and Smithson (1993) are overlaid. 112 Natural Disaster Hotspots Case Studies Figure 4.2.b. The average monthly rainfall between 1869 and 1998 for Sri Lanka Disaster-Related Data: The sources of data were the Sri Lanka Department of Social Services, Sri Lanka Depart- ment of Census and Statis- tics, and the Central Bank of Sri Lanka. These data were of varying resolutions, ranging in scope from the district level (droughts, floods, and cy- clones) to the DSD level (later instances of flood) to the GND level (landslides). Most dis- aster incidence data also con- tained relief expenditures. is complicated by the two-decade-long war. While the Climate Data: Data were obtained from the Sri Lanka precision of our analysis may be affected by the history Department of Meteorology and secondary sources. of war, vulnerabilities created by the war make efforts Although the country has around 400 functioning rain- to reduce disaster risks all the more important. fall stations, only a subset of these possesses uninter- The specific objectives for this study are as follows: rupted records. The records in the Northern Province were limited over the last two decades because of war. · To undertake a subnational analysis of droughts, We used data from 284 rainfall stations from 1960 to floods, cyclones, and landslides; 2000 to construct gridded data at a resolution of 10 km. · To assess vulnerability to these hazards at the sub- Using 1960 to 1990 as the base period, monthly cli- national level; matologies were calculated. Monthly anomalies were · To assess multihazard risks and hazard hotspots at calculated by deducting the climatology from observed the subnational level; and values (figure 4.2.a). · To assess methodologies for incorporating climatic information into hazard analysis. Hydrological Data: Data were obtained from the Sri Lanka We shall describe the data that were used, the method- Department of Irrigation and through secondary sources ologies used for hazard and vulnerability assessment, for monthly river flow measurements at 140 gauging and the analysis of multihazard risk in the following stations. These data had numerous gaps. sections. Landslide Hazard­Related Data: Data were obtained through the National Building Research Organization Data of Sri Lanka. In-Country Data Population, Social, Economic, and Infrastructure Data: The Department of Census and Statistics provided popula- The administrative divisions in Sri Lanka are provinces, tion data. Data at the DSD level were selected for com- districts, divisional secretariat divisions (DSDs), and parison and analysis. Gross domestic product (GDP) Grama Niladhari (Village Officer) divisions (GNDs). measures, including regional GDP, were obtained from There are 9 provinces, 25 districts, 323 DSDs, and 14,113 the Central Bank of Sri Lanka. GNDs, organized hierarchically. Food Security Data: An assessment of food security in Sri Lanka was conducted under the Vulnerability Assess- Natural Disaster Risks in Sri Lanka: Mapping Hazards and Risk Hotspots 113 ment and Mapping Program of the World Food Pro- four instances of epidemics. EM-DAT identifies 9 gramme (WFP), Sri Lanka office. The identification of droughts, 2 landslides, 3 cyclones and storms, and 33 DSDs with three levels of food insecurity was obtained flood events (including floods caused by the cyclones). from their maps. The dataset contains dates and affected areas and people. The United Nations Environment Programme/GRID Hazards and disaster records had good identification (UNEP/GRID) datasets include global cyclone tracks for of where these occurred, but often only the year when the period from 1980 to 2000. these occurred was available. The temporal resolution was improved by interrogation of multiple data sources and by consulting government officials. Exposure and Vulnerability Global Data Sources Exposure and vulnerability may be assessed for the three categories of elements at risk--people, economic Hazard Data­­Floods: Dartmouth Flood Observatory activity, and infrastructure. carries an archive of large flood events from 1985 onward. This database contains specific dates of the floods, sever- ity class, and affected area. However, the spatial reso- People lution is coarse, as the data have been derived from the Population: The population of Sri Lanka was 19.2 mil- district level. lion in 1998 (293 persons per km2) with an uneven dis- tribution (figure 4.3). Fifty-five percent of the population Climate Data: The data available at the International is concentrated in 20 percent of the land area (Depart- Research Institute for Climate and Society (IRI) Data ment of Census and Statistics 2001). Thirty percent of Library with long coverage for Sri Lanka is lower in the population resides in urban areas. The least-popu- resolution (250 km grid). lated districts (covering 40 percent of the island) host 10 percent of the population. In these districts, population Exposure Data­­Population, Social, Economic, and Infra- density ranges from 35 to 100 people per km2, which is structure Data: Center for International Earth Science still high by global standards (De Silva 1997). The high- Information Network's (CIESIN's) Gridded Population est population is in the Colombo, Gampaha, and Kalu- of the World (GPW2) dataset contains population data tara districts of the Western Province. There is a secondary on a 5 km grid. The gridding methodology of GPW2 population center in the Kandy District in the Central utilizes district-level population data. "Landscan Province and in the Galle District along the southern 2001" contains gridded population data on a 1 km coast. The high density of people in the wet parts of the grid calculated using population, roads, slope, land island increases the number of people who are vulnera- cover, and nighttime lights. ble to floods and landslides. Impoverishment and mortality are direct consequences Vulnerability Data: The United Nations Development of, as well as contributors to, natural disasters. In this Programme (UNDP) Human Development Reports pro- context, food security measures a community's resilience vide a number of key indicators at the national level. to the hazards and often its exposure. Food security cal- The UNDP Human Development Report for Sri Lanka culated by the WFP Sri Lanka office in 2002 was based provides most of these indicators at the district level. on the availability of food, access to food, and utiliza- tion of food (figure 4.4). Based on this study, 93 DSDs Disaster Data­­EM-DAT: The Office of U.S. Foreign Dis- out of 323 were categorized as "Most Vulnerable," 82 aster Assistance/Center for Research on the Epidemiol- as "Less Vulnerable," and 148 as "Least/Not Vulnerable" ogy of Disasters (OFDA/CRED) International Disaster (World Food Programme 2002). The spatial variability Database has recorded 48 natural disasters in Sri of the Least/Not Vulnerable category shows two con- Lanka during the period from 1975 to 2001, including tiguous regions and some scattered areas. One con- 114 Natural Disaster Hotspots Case Studies Figure 4.3. The density of population in each of the 323 Divisional Secretarial Divisions based on data from the census of 2001 Natural Disaster Risks in Sri Lanka: Mapping Hazards and Risk Hotspots 115 Figure 4.4. The food insecurity index of Divisional Secretariat Divisions (DSDs) as estimated by the World Food Programme 116 Natural Disaster Hotspots Case Studies tiguous region is the western coastal region, which has mining and minerals. Industries are heavily concen- higher rainfall, better infrastructure facilities, and indus- trated in Colombo, Gampaha, and Kalutara in the West- try. A second contiguous region with high food secu- ern Province. In the last two decades, industrial rity is the area around Kandy which also has higher production has shifted from heavy industries for domes- rainfall and better infrastructure facilities. A third con- tic consumption to export-oriented textile and other tiguous areas is the region around Anurhadhapura, processing. which has improved infrastructure, increased irrigation Industries are concentrated in a few regions in west- and lower population density. The higher food inse- ern Sri Lanka (figure 4.7) that are particularly prone to curity in the northern and eastern areas is due to a flooding. Drought in the Central Highlands can affect combination of war and dry climatic conditions punc- industry drastically through deficits in hydropower pro- tuated by cyclones and heavy rainfall. duction. A quarter of the manufactured products are from the processing of agricultural products (tea, rubber, and tobacco). Thus, these industries could be affected Economic activity by hazards that impact agricultural production. Industrial and infrastructure sectors account for the bulk of the national GDP (figure 4.5.a). Agriculture, Infrastructure animal husbandry, and fisheries provide livelihoods for one-third of the employed (figure 4.5.b), followed Infrastructure development, too, reflects a pattern of by employment in industries, infrastructure, and serv- heavy development in the Western Province with sub- ices. The disruption in agriculture, industry, and infra- sidiary development in the metropolitan districts of structure caused by natural disasters is addressed below, Kandy and Galle. along with descriptions of the salient features of these elements in relation hazards. Roads: Sri Lanka has an extensive road network with The Western Province had the largest provincial GDP better density and coverage compared with most devel- (figure 4.6) with SL Rs 180 billion (US$3.4 billion); oping countries. the Central Province came in second with SL Rs 46 bil- lion (US$0.88 billion) at constant 1990 prices (UNDP Electricity Generation and Distribution: As of 1995, 53 1998). percent of households had access to electricity. How- ever, the spatial distribution of electricity availability Agriculture: The primary food crop is paddy. The main ranges from more than 90 percent in Colombo and Gam- Maha cropping season commences with heavy rainfall paha to less than 40 percent for districts in the north starting in late September and ends in March. A sec- and east (Gunaratne 2002). Of the total nationally ondary season, Yala, extends from May to early Sep- generated electrical energy, approximately 60 percent tember, and during this season only half of the agricultural comes from hydropower, putting it at high risk during land is cultivated because of limited supply of water. drought periods. The droughts in 1995-96 and 2000- The major cash crops are tea, rubber, coconut, and 01 resulted in blackouts for the whole country. spices; and their cultivation is largely in the wet regions. The agrarian economy is thus susceptible to disrup- Telephones: The density of telephones is low with 41 tion through droughts and floods. Our previous work landlines and 23 cellular phones per 1,000 persons in has shown a link between rainfall variations and agri- 2000 (UNDP 1998). The spatial distribution of access cultural production (Zubair 2002). Note that there is indicates that Colombo has more than 50 percent of the an extensive irrigation network that modulates the landlines. spatial distribution of vulnerability. Separate indexes for roads, electricity, and telephone Industry: The major industries are textile and apparel, densities were analyzed to develop an infrastructure food and beverage processing, chemical and rubber, and density index. The road index was constructed by nor- Natural Disaster Risks in Sri Lanka: Mapping Hazards and Risk Hotspotss 117 Figure 4.5.a. Sectoral breakdown of the GDP for 2001 Figure 4.5.b. Sectoral breakdown of the labor force for 2001 118 Natural Disaster Hotspots Case Studies Figure 4.6. The gross domestic product (GDP) by province for 1995 Natural Disaster Risks in Sri Lanka: Mapping Hazards and Risk Hotspots 119 Figure 4.7. The estimate of industrial output in the districts in 1995 120 Natural Disaster Hotspots Case Studies malizing the length of different categories of roads (classes There is a stronger tendency toward drought in the A, B, and C) per district. The telephone and electricity southeastern district of Hambantota and the north- indexes were constructed as the number of house- western region, which includes the Mannar and Putta- holds that have access to these facilities in each dis- lam districts. The drought tendency is markedly less trict. These three indexes were evenly normalized and pronounced in the southwest corner of Sri Lanka where aggregated to create an infrastructure index (figure 4.8). there is heavy rainfall. There is a high concentration of infrastructure facil- A drought disaster risk map was constructed by ities in Colombo. This skewed distribution is largely weighting drought incidences for severity of the drought due to the heavy concentration of telecommunication in terms of relief expenditure (figure 4.10). The drought facilities. Electricity and telephone facilities have been hazard map constructed from rainfall data (figure 4.9) severely disrupted in the Northern Province because of is similar to the drought disaster incidence map (figure the war, and there are no estimates of recent conditions. 4.10), and this is evidence of the plausibility of hazard Thus, interpretation of the infrastructure index for these mapping. In the future, the drought mapping may be areas needs to be tempered with caution. improved by taking into account factors such as sur- Infrastructure elements that are at risk from natural face water availability. hazards include the road network (floods and land- The drought disaster risk map shows marked spa- slides), electrical distribution system (floods, landslides, tial variability. There is low drought disaster risk in and cyclones), electricity generation (droughts), and the western slopes and high drought disaster risk telephones (floods, landslides, and cyclones). in the southeastern, northern, and northwestern regions. The highest drought disaster risk is in the Anuradha- pura District followed by the Badulla and Batticaloa Analysis of Individual Hazard Risks Districts. Drought Hazard Flood Hazard Drought hazards can be estimated through the use of Rainfall, river flows, and topographical data can be used several methods. However the WASP indexes23 devel- to construct flood hazard maps. Such an effort needs oped by Lyon (2004) are the best option based on rain- hydrological modelling. An archive of satellite images, fall alone. Other indexes may be constructed by using too, may be used to identify flood-prone areas with stream flow, vegetation or soil moisture indexes, and higher resolutions. However, the stream-flow data needed so on, but these data are not available at adequate for hydrological modeling and satellite archives are not levels of resolution, reliability, and historical extent. Both available with required consistency, resolution, and his- 6- and 12-month WASP indexes were estimated for Sri tory to create high-resolution maps. Lanka (figure 4.9). Given the purposes of this study and the 10 km res- olution to which it is limited, flood hazards may be mapped by identifying instances in which extreme rain- 23. WASP is an acronym for the Weighted Anomaly Standardized Precipi- fall events were detected in the past. Flood hazard was tation index. This index gives an estimate of the relative deficit or sur- plus of precipitation for different time intervals ranging from 1 to 12 estimated by identifying instances of monthly precipi- months. In this case, analysis is based on 6-month and 12-month indexes. tation exceeding a threshold of 600 mm (figure 4.11). To compute the index, monthly precipitation departures from the long- A disaster incidence map of floods incurring losses term average are obtained and then standardized by dividing by the stan- was constructed by using the number of major floods dard deviation of monthly precipitation. The standardized monthly anomalies are then weighted by multiplying by the fraction of the aver- in the last 50 years at district level using data from the age annual precipitation for the given month. These weighted anom- Social Services Department and Dartmouth Flood Obser- alies are then summed over varying time periods--here, 6 and 12 months. vatory (figure 4.12). The frequency was normalized over On the plots, the value of the given WASP index has itself been stan- area and scaled from 1 to 100. There are similarities dardized. Regions with an annual average precipitation of less than 0.2 mm/day have been "masked" from the plot. between the essential features of the flood hazard esti- Natural Disaster Risks in Sri Lanka: Mapping Hazards and Risk Hotspots 121 Figure 4.8. Infrastructure density index estimated for each district, as described in the text 122 Natural Disaster Hotspots Case Studies Figure 4.9. The drought hazard was estimated using a modified WASP index. The details of the WASP index are provided in the text. The negative WASP values (dry) were averaged over a 12-month period to identify drought prone regions. The hazard values were normalized so that they ranged between 0 and 100. Natural Disaster Risks in Sri Lanka: Mapping Hazards and Risk Hotspots 123 Figure 4.10. Drought disaster incidence frequency was constructed by aggregating the numbers of droughts that have been recorded in each district. Major droughts as categorized by the Department of Social Services were weighted by 1.5, medium droughts by 1.0, and minor droughts by 0.5 124 Natural Disaster Hotspots Case Studies Figure 4.11. The flood hazard estimate based on the frequency of months of extreme rainfall using data between 1960 and 2000. The threshold chosen for extreme rainfall was 600 mm per month. Natural Disaster Risks in Sri Lanka: Mapping Hazards and Risk Hotspots 125 Figure 4.12. Frequency map of flood disaster incidence created by aggregating the numbers of floods recorded in each district between 1957 and 1995. Major floods, as categorized by the Ministry of Social Services, were weighted by 1.5 and minor floods were given a weight of 0.5. The index was normalized by area. 126 Natural Disaster Hotspots Case Studies mates based on rainfall data and the above disaster inci- Landslide Hazard dence estimate. Landslide hazards affect people, infrastructure, and eco- We estimated the seasonal distribution of floods in nomic activities. Most high-risk DSDs (except in the the western and eastern regions separately (figure 4.13) Kalutara District) are within regions of high food inse- based on 33 flood events in the EM-DAT database that curity. There is moderate economic activity in the high- had records of months of occurrence. risk regions. Transport by road and railway has frequently The flood hazard and disaster maps show high risk been affected, particularly in the hill country. in the western, southwestern, northern, northeastern, The National Building Research Organization (NBRO) and eastern parts of the country. The western slopes has undertaken a detailed study of landslide risk in Sri show the highest risk followed by the Batticaloa and Lanka. Landslide hazard mapping has been completed Badulla Districts. The most flood-prone districts are for five high-risk districts at a scale of 1:10,000. The Kegalle, Ratnapura, Kalutara, Kandy, Colombo, and NBRO methodology takes into consideration various Galle. These districts are located in the southwest part factors, including slope-gradient, geology, soil cover, of the island. Flood occurrences in the eastern slopes hydrology, and land use. Enhancement of this method- and the northern plains coincide with the period of ology is possible through the use of improved datasets heavy rainfall (September to January) during the for digital elevation modeling and hydroclimatic data Maha. In the western slopes, floods do occur during the and models. Maha, but are more common in the mid-Yala season, For this study, the potential risk zones were identi- which lasts from May to August. These trends are also fied at a resolution of 10 km in keeping with the reso- reflected in both hazard-risk and seasonal maps. Of lutions of the other hazard and vulnerability data. the 33 flood events in EM-DAT, 20 events occurred in Landslide incidence data from the NBRO was used to the November­January period (during the Maha rain- map the hazard risk. The event frequency data for each fall season), including 3 cyclone-induced events. grid cell between 1947 and 1996 was used as the risk These flood events affected both the western and east- factor for landslides (figure 4.14). ern parts of the island. Eleven events occurred during Eight districts in the central highlands are at risk. the May­July period (Yala rainfall season), which affected The highest risk is in the Kegalle District followed by only the western slopes (figure 4.13). Ratnapura and Nuwara Eliya Districts. Even within these Heavy rainfall in the eastern and southwestern slopes districts there is spatial variability at the DSD level. is a principal cause of the flood risk. The drainage and The Kalutara, Kandy, and Badulla Districts have mod- topography of certain districts and the land use patterns erate risk, and Matale and Kurunegala Districts have are also significant factors. For example, in the districts slight risk. of Kegalle and Ratnapura, people have settled in flood The frequency of landslides has increased in recent plains and steep hill slopes. The eastern slopes receive years. Changes in land use--including cultivation of most of the rainfall during the Maha season. This is tobacco on steep slopes, land clearing in the hills, block- also the cyclone and storm season that can bring heavy ing of drainage ways, and the impact of the large reser- rainfall in short time periods. The Vavuniya District voirconstruction--maybeduetotheincrease.Sometimes, shows a higher flood probability caused by cyclonic soil conservation programs, such as contour ditches, con- storms. Even though their annual rainfall is lower than tribute to increases in landslide hazard risk by increas- that of the western highlands, Vavuniya and Mullaitivu ing soil saturation (Madduma Bandara 2000a). in the north have recorded the highest rainfall intensi- ties on the island (Madduma Bandara 2000b). Floods affect people, economic activities, and infra- Cyclone Hazard structure. The high-risk regions in the western slopes have higher population densities, greater concentra- Cyclones affect people, infrastructure, and economic tions of industrial activity and infrastructure, and very activities. The high-risk areas in the north and the east- high GDPs. The north-eastern high-risk region has high ern seaboard have high food insecurity. Paddy fields are food insecurity. Natural Disaster Risks in Sri Lanka: Mapping Hazards and Risk Hotspots 127 Figure 4.13. EM-DAT data on floods from 1975 to 2001 were used to estimate the monthly frequency of floods in the Western Slopes and Eastern Slopes regions. in high concentration in the hazard-prone region. The Note that cyclone hazard mapping can be improved. storm surge at landfall can be devastating. The storm Wind-speed modeling techniques that estimate decel- surge of the 1978 cyclone extended up to 2 km inland eration after landfall can account for the diminishment in some areas. In addition to the storm surge, the intense of the intensity of storms over land. Wave and tidal gusting can be destructive. Intense rainfall that comes models can be used to identify the risks from storm along with cyclones creates floods and flash floods. surges. Elevation maps and hydrological analysis can Cyclones and storms have made landfall only in the be used to identify flood-risk areas. eastern coast of Sri Lanka, except for a single storm in 1967. The majority of cyclones and storms pass through the northern and north-central parts of the island. The Assessing Multihazard Hotspots cyclones that pass through Sri Lanka originate from the Bay of Bengal during the northeast monsoon. Inci- A multihazard map was constructed by aggregating dences of cyclones that pass through Sri Lanka in other the hazard indexes for droughts, floods, cyclones, and seasons are rare due to geography and the regional cli- landslides, with each hazard weighted equally (figure matology. There have been four severe cyclones during 4.17). This map shows the high risk of multiple haz- the last 100 years as well as a number of severe and ards in the north. The Anuradhapura, Polonnaruwa, moderate storms. Batticaloa, and Trincomalee Districts in the northeast The available cyclone tracks from 1900 to 2000 were also feature high risk, as do the southwestern districts used to construct a map of cyclone hazard (figure 4.15). of Kegalle, Ratnapura, Kalutara, and Colombo. Regions The frequency with which storms passed through a grid with sharp gradients along the mountain massifs (Nuwara point was estimated. The immediate adjoining grid Eliya, Badulla, Ampara, and Matale) also show high risk points were given an impact factor of half that given to of multiple hazards. grid cells that lay on the storm and cyclone track. Cyclones Disaster risk maps may be constructive by taking into were weighted three times as heavily as storms. The account exposure and vulnerabilities in addition to haz- northeastern seaboard has high hazard. ards. Exposure and vulnerability are more difficult to A cyclone seasonality graph was constructed by plot- quantify than hazards. A proxy for the combination of ting the number of cyclones and storms that occurred hazards and vulnerability may be constructed if it is in each month (figure 4.16). Cyclone incidence shows assumed that the history of hazards provides a repre- a strong seasonality, and 80 percent of all cyclones and sentation of future spatial variability. Such an approach storms occur in November and December. needs long records of disasters and is based on the 128 Natural Disaster Hotspots Case Studies Figure 4.14. A landslide hazard risk index was estimated based on frequency of incidence. Natural Disaster Risks in Sri Lanka: Mapping Hazards and Risk Hotspots 129 Figure 4.15. The storm and cyclone tracks for the last 100 years were used to create a cyclone hazard risk map. 130 Natural Disaster Hotspots Case Studies Figure 4.16. The monthly count of storms and cyclones between 1887 and 2000 assumption that the future occurrence of hazards, expo- Note that for the period from 1948 to 1992, the sure, and vulnerability is similar to past occurrences. EM-DAT data are weighted toward floods (Weights-- This assumption, while not precise, does enable us to Droughts: 9, Floods: 30, Landslides: 2, Cyclones: 3), provide an estimate of the variability of risk. More pre- whereas the data obtained from the Department of Social cise estimates must await more long-term data that have Services were weighted toward droughts and cyclones good spatial and temporal resolutions. (Weights--Droughts: 27, Floods: 24, Landslides: 17, Subject to the highlighted limitations, records of dis- Cyclones: 10). The difference may arise from differing asters may be used to weight for exposure and vulner- perceptions and criteria for identification as a disaster. ability to particular hazards. Figure 4.18 shows the The final multihazard risk map (figure 4.20) was cal- multihazard map with weights for each hazard based culated by weighting each hazard index by the disaster on the number of occurrences of each hazard from 1948 relief expenditure for each hazard. This hotspots map to 1992. Multiple landslides within a single year were is heavily weighted toward droughts and cyclones, with treated as one event. This map gives greater weight to landslides receiving a meager weight. This hotspots map droughts and less weight to floods. The result, however, shows higher risk in the north and north-central regions does not significantly differ from that produced with and in the Hambantota District (south-east) compared equal weight. There is high risk in some regions in the with previous maps. north and east in addition to the regions with the sharpest The various multihazard maps have differences but hill slopes in the south. The risks are also enhanced in also show commonalities. Three regions emerge as having the region around the Hambantota District in the south- high risk in all maps. One is the region with sharp east and around the Mannar District in the north-west. slopes in the south-west: the Kegalle District is the most The next figure (figure 4.19) is identical except that risk prone, with significant risk of landslides and floods the data for the frequency of hazards were obtained from and moderate risk for droughts. The Ratnapura and Kalu- the EM-DAT database. There is high weight toward floods tara Districts also have high risk of floods and land- in this dataset. This map shows very low risk in the south- slides. A second region is in the north-east: the Batticaloa, east and north-west and high risk in the north-eastern Trincomalee, Mannar, Killinochchi, and Jaffna Districts tip as well as the eastern and western slopes regions. along the north-eastern coast show high multihazard risk. Natural Disaster Risks in Sri Lanka: Mapping Hazards and Risk Hotspots 131 Figure 4.17. Multihazard index constructed by aggregating the hazard indices and scaling the result to range between 0 and 100 (Weights: droughts: 1, floods: 1, landslides: 1, cyclones: 1) 132 Natural Disaster Hotspots Case Studies Figure 4.18. Multihazard risk estimated by weighting each hazard index by its frequency from 1948 to 1992 and rescaling the result to range from 0 to 100. The hazard incidence data was obtained from the Department of Social Services. (Weights: droughts: 27, floods: 24, landslides: 17, cyclones: 10) Natural Disaster Risks in Sri Lanka: Mapping Hazards and Risk Hotspots 133 Figure 4.19. Multihazard risk estimated by weighting each hazard index by incidence frequency. (Weights: droughts: 9, floods: 30, landslides: 2, cyclones: 3) The result was rescaled to range between 0 and 100. 134 Natural Disaster Hotspots Case Studies Figure 4.20. Multihazard risk estimated by weighting each hazard index by the associated relief expenditure between 1948­1992. (Weights: droughts: 126, floods: 25, landslides: 0.06, cyclones: 60) Natural Disaster Risks in Sri Lanka: Mapping Hazards and Risk Hotspots 135 A third region is along the mountain massifs with the mate variability (such as the effects of El Niño). The sharpest hill slopes--this includes parts of the Nuwara ability to predict shifts in climate up to six months in Eliya, Badulla, Ampara, and Matale Districts. advance provides an opportunity to engage in predic- Some of the high-risk regions have concentrations of tive risk-mapping, as the climate of Sri Lanka is rela- economic output, agriculture, and industrial concen- tively predictable. trations. Some regions in the southwest with high multihazard risk also have high food insecurity. The Long-Term Climate, Environmental, and Social Change north shows high multihazard risk as well as high food Both the climate and environmental change, such as insecurity. Rice cultivation in these regions is particu- deforestation and urbanization, affect the hazard analy- larly vulnerable to drought and flood hazards. sis and, in ideal conditions, should have been included in the analyses. Such work is needed in the future. Here, we note that climate change is already making Discussion parts of Sri Lanka more vulnerable to drought. This is a development that shall have far-reaching ramifications. Spatial Variability Further investigation is required to build compre- hensive drought maps that take into account hydrolog- The use of data available in Sri Lanka enabled the con- ical and physical conditions that contribute to drought. struction of detailed hazard maps and the investigation Our vulnerability analysis, too, can be improved by taking of trends. The spatial hazard and disaster risk-map- account of long-term changes in demographics, urban- ping should be useful for local authorities as well as ization, migration, and the consequences of civil war. international relief organizations. Hazard Mapping Methodology Conclusions The hazard mapping methodology at the local scale needs to be fine-tuned to take advantage of the finer We have presented an example of the use of physical resolution of data and finer resolution of the results. and social data for fine-scale hazard and vulnerability An example of a good use of multiple datasets is the analyses. This case study has demonstrated that the landslide hazard-mapping project carried out by the Sri use of such fine-scale analyses recognizes crucial regional Lanka NBRO. variations and is more useful than relying on currently available global-scale data. We have presented methodologies for using fine-res- Vulnerability Analysis olution gridded climate data to estimate droughts and Hazard-specific vulnerabilities are needed at high res- floods and for using past-incidence data to estimate olution. Vulnerability analysis is more constrained by cyclones and landslides. Seasonal climate predictions data limitations than by hazard analysis. Notwithstanding can be factored into this methodology to yield hazard- these limitations, the vulnerability analyses provide a risk predictions that exploit the emerging technology broad initial assessment of the nature of hazard risks of seasonal climate predictions. and vulnerabilities at a national scale. Vulnerability analysis is much less precise than hazard analysis. The approach adopted here considers the specific elements of people, economic activity, and infra- Seasonality structure, and estimates these elements based on prox- Strong seasonality was evident in drought, flood, cyclone, ies, which has been shown to be viable with locally and landslide risks in Sri Lanka. Information about the available data. Crucial spatial variations in vulnerabil- seasonal risk levels of different disasters is useful for dis- ity emerged in the higher-resolution maps that were not aster risk management and should be provided. Our evident at coarser scales. Estimates of hazard-specific work has shown that the risk factors change with cli- vulnerabilities had to be based on the assumption that 136 Natural Disaster Hotspots Case Studies sufficiently long records of the past give an indication Program (ESMAP) Energy and Poverty Workshop, Addis Ababa, of future spatial variability. This is a reasonable assump- Ethiopia, October 23-25, 2002. tion when one considers that the topography, climate, Lyon, B. 2004. The Strength of El Niño and the Spatial Extent of Drought, Geophysical Research Letters. 31: L21204. and terrain, which are the basic causes of regional vari- Madduma Bandara, C.M. 2000a. Land Resources. Natural Resources ability, do not change substantively. However, long-term of Sri Lanka 2000. Colombo: National Science Foundation, p. climatic, environmental, and social change needs to be 53-73. investigated in the future and factored in the analyses. Madduma Bandara, C.M. 2000b. WaterResources,NaturalResources Multihazard mapping is subject to limitations in the of Sri Lanka 2000. Sri Lanka: National Science Foundation, p. types of data that are available, particularly for expo- 75-109. Puvaneswaran, K.M., and P.A. Smithson. 1993. An Objective sure and vulnerability. There were multiple ways to Classification of Homogeneous Rainfall Regimes in Sri weight the different hazards, each of which has argu- Lanka. Theoretical and Applied Climatology. 48: 133­145. ments in its favor. These different maps can suit differ- Tissera, C.H. 1997. Natural Hazards. In: Arjuna's Atlas of Sri ent purposes. Given the limitations in the methodology, Lanka, S. Somasekeram (ed.). Colombo: Arjuna Publishers, it is useful to focus on the commonalities from the maps. p. 76-78. United Nations Development Programme. 1998. National Human Development Report 1998: Regional Dimensions of Human Development, Sri Lanka: United Nations Development Pro- References gramme. World Food Programme. 2002. Identification of DS Divisions Department of Census and Statistics. 2001. Statistical Abstract Vulnerable to Food Insecurity. Vulnerability Analysis and Map- of the Democratic Socialist Republic of Sri Lanka, Department of ping Unit. Sri Lanka: World Food Programme. Census and Statistics, Colombo. Zubair, L. 2002. El-Niño-Southern Oscillation Influences on De Silva, Soma. 1997. Population. In: Arjuna's Atlas of Sri Rice Production in Sri Lanka. International Journal of Clima- Lanka, S. Somasekeram (ed.). Columbo: Arjuna Publishers. tology 22(2): 249­260. Gunaratne, Lalith. 2002. Rural Electrification of Sri Lanka, pres- entation at World Bank Energy Sector Management Assistance Chapter 5 Multihazard Risks in Caracas, República Bolivariana de Venezuela Kristina R. Czuchlewski, Klaus H. Jacob, Arthur L. Lerner-Lam, Kevin Vranes, and Students of the Urban Planning Studio: "Disaster Resilient Caracas" Catastrophic risk management associated with natural For instance, flood height may depend on the amount hazards has developed over the last few decades into a of rain that has fallen over time and space. Other con- mature, quantitative discipline. This preliminary case tributing factors are the topography of the landscape study employs quantitative methods where feasible, but and river network: the degree to which the ground time and resource constraints have prevented a more surface can absorb and retain the precipitation (types systematic application. Nevertheless, we present a of soils and ground cover, such as vegetation versus brief overview of these methods as they are used in this urbanized parking lots); and the shape, gradient, and report. smoothness of drainage and river channels. Mud- or debris flows require, in addition to the variables men- tioned for floods, the availability and size distributions General Definitions Used in This Report of solid materials that can be mobilized. The nature and severity of the debris flow hazard is different when tree Perils. Perils refer to the different natural processes that trunks and house- or room-sized boulders are available, can potentially provide hazards and cause losses. Such as compared to when only clays, silts, sands, or perhaps perils can be earthquakes, floods, and land- or mud- gravel make up the suspended solid bulk of the flu- slides, for instance. In other areas, cyclones, coastal idized debris flow. Large boulders and tree trunks con- storms, volcanoes, hail, tornados, or snow avalanches tributed to the severe damage during the massive debris may be important perils. For Caracas, the most impor- flows observed during the Vargas disaster that killed tant perils are earthquakes, mudflows and landslides, many tens of thousands of Venezuelans in December and stream flooding. 1999. In the case of seismic shaking, the magnitude of Hazards. Each peril poses hazards that are more or causative earthquakes, their depth and distance, and less quantifiable. For example, earthquake shaking can the firmness of ground (rock versus soil, among other be measured on the Modified Mercalli Intensity Scale factors) will determine the level of ground shaking. (MMI, based on damage to structures), by peak ground The frequency of earthquakes as a function of their mag- acceleration (PGA), or by some other measurable ground nitude determines the probability with which set motion variable. In the case of river flooding, the levels of ground shaking will be exceeded. "flood stage" (that is, the height to which a river may rise) or the flow rate (volume of water passing per unit Risk. The risk to a region or community is the expected of time) could become the measure for the severity of future loss defined in terms of probability of losses vs. the hazard. Similar units could be used for mud- or their magnitude. Since risk implies future losses, it is debris flows. For winds, the hazard may be character- inherently uncertain because hazard and other risk-con- ized by wind speed. tributing elements are uncertain. Generally, risk is defined Once a hazard parameter is chosen it can be traced in terms of the value of assets, their fragility (in terms of back to the interaction of many contributing factors. direct loss or loss of function) in the face of particular 137 138 Natural Disaster Hotspots Case Studies hazards, and the exposure of those assets to multiple haz- tude between the national government and nongovern- ards. In general, while quantitative risk assessment pro- mental organizations.) The heavy rains also caused fatal- vides a motivation for implementation of specific ities in the Caracas basin. interventions, the identification of critical assets and a High slope angles along El Ávila (some up to 80 per- qualitative, experience-based assessment of their fragility cent) allow for immediate acceleration of surface fluids. and exposure constitute the bulk of this report. We use The December 1999 event saw evidence of hyper-con- empirical methods to estimate the occurrence of perils, centrated flows, alluvial fan reactivation, and some evi- and geographic analysis to identify the exposure of assets. dence of prior, larger flows (Salcedo 2000, 2001). The This report makes qualitative assessments of the poten- Vargas coast is lined with extensive alluvial fans, nine of tial fragility of structures and infrastructure. This analy- whichreactivatedduringtheDecember1999event(Larsen sissupportspreliminaryrecommendationsontheinclusion et al. 2001). The Caracas basin contains at least three allu- of risk management interventions in a Caracas urban vial fans with clear signs of earth movement in the past plan. In concluding remarks, we develop a set of rec- 100years.Larsenetal.(2001)presentanexcellentoverview ommendations on quantifying the risk analysis. of El Ávila geology and mud flow processes during the December 1999 disaster. Although rainfall data are sparse, historical records The Caracas Urban and Environmental compiled by Salcedo (2000) show that either the Cara- Context cas or Vargas area has been severely affected by debris flows roughly every 25 years since the start of recorded Natural Hazards history. Previous empirical work by local geologists illus- trates that any rainfall exceeding 100 mm in 24 hours Located at the intersection of the South American and will cause damaging mud- and debris flows; any rain- Caribbean Plates (figure 5.1), the northern area of fall exceeding 300 mm in 72 hours is considered cata- República Bolivariana de Venezuela faces extreme seis- strophic. An analysis of 25 years of regional rainfall data mological hazards. Perez et al. (2001) report a 2-cm produced during the Columbia University Urban Plan- per year rate of plate motion at the offshore boundary. ning Studio extrapolates a frequency of recurrence of Half of this boundary is accommodated by the San Sebas- 100 mm per 24 hours to once every 5 to 10 years; the tian fault, which likely comes ashore under the Simon probability of 300 mm per 24 hours is once in 25 years. Bolivar International Airport in Vargas State (Audemard et al. 2000). The fault zone is diffuse, containing the Tacagua-El Ávila and La Victoria fault systems that sur- Built Environment round Caracas to the north and south. Major earth- Since the last major earthquake in 1967, the popula- quakes have destroyed the city three times in the last tion of Caracas has doubled to 5 million people, with 400 years. The last large earthquake (Mw=6.5) came in a population density of 12,000 persons per km2 and 1967, killing an estimated 300 people and destroying growth of 3.1 percent per year. Eighty-six percent of the four modern structures built for earthquake resistance Venezuelan population is urban, making it the seventh (Papageorgiou and Kim 1991). most urbanized country in the world. The valley floor The natural hazards faced by this area are not limited is well developed, with high-rise buildings and densely to earthquakes. The position of the northern coast near packed apartment blocks scattered unevenly through- 10°N ensures frequent heavy rainfall events with strong out the city. These buildings are generally concen- erosion potential. In December 1999, a month of rain trated in the deepest part of the basin (where shaking on the north-central coast of the country--including over is expected to be highest during an earthquake). 900 mm of rain in a 72-hour period between December Barrios, or informal squatter settlements, dominate 15 and 17--triggered landslides, mudflows, and debris the landscape on the low-lying, rugged mountains to flows on the north face of the El Ávila range that killed the east and west of the city center, where rainfall-induced anestimated25,000residentsofthecoastalstateofVargas. debris flows are expected to be greatest. To the south (Estimates of this number vary by one order of magni- exists a mixture of urbanazacions (similar to suburbs) Multihazard Risks in Caracas, República Bolivariana de Venezuela 139 Figure 5.1. Regional elevation map of Caracas and Vargas State and barrios. The individual building blocks of the bar- which are located across El Ávila on the Vargas coast rios, known as ranchos, are constructed of unreinforced (figure 5.1). The only road between Caracas and the air- masonry, making them particularly vulnerable to earth- port and seaport is a single highway that travels through quakes. The ranchos are highly visible from every point steep, landslide-prone valleys crossing secondary faults in the city as they carpet the hills, creating a starkly con- of the active San Sebastian fault. trasting landscape of a dual urban fabric (figure 5.2). Uncontrolled building and lack of enforcement of While the formal city averages 6,000 persons per km2, building and zoning codes in this hazardous environ- similar to the average urban density worldwide, the bar- ment have led to human disasters and potential prob- rios approach 25,000 persons per km2. Vargas State is iso- lems of great magnitude. A lack of building code and lated physically from the Caracas basin by El Ávila (figure enforcement allowed Vargas residents to build on active 5.1). However, Vargas is inextricably linked to Caracas. (but quiescent for the previous 50 years) alluvial fans, It serves as Caracas' economic connection to the rest of which reactivated during December 1999. Although var- the Caribbean, and acts as the social "relief valve" for the ious groups are working to repair and rebuild Vargas State citybyofferingweekendrecreationforresidents.Although with new housing built in safe locations, a general lack separated geographically, Vargas and Caracas are eco- of planning and enforcement is allowing squatters to nomically and culturally intertwined. return to the alluvial fans and stream beds where most of the December 1999 destruction was concentrated. Collision of the Built Environment and Natural Hazards Addressing the Risk: Results of the Urban Planning Studio Centuries ago, Caracas was purposefully built away from the coast and through steep terrain to deter sea-borne It is clear that natural hazards affect the Caracas urban attacks on the city. However, this distance creates environment. To address this complex problem, a strate- major transportation and utility infrastructure problems gic planning process, designed to identify the city's abil- that are exacerbated by natural hazards. Caracas is linked ity to cope with the forces of nature, was initiated. In order to the world through its airport and seaport, both of to accomplish this, certain critical facilities (for example, 140 Figure 5.2. Map of the Petare barrio of Caracas, illustrating the dual nature of the city. On the left is the open spacing of the planned "formal" city the right are the densely packed squatter barrios of the "informal" city Natural Disaster Hotspots Case Studies Multihazard Risks in Caracas, República Bolivariana de Venezuela 141 the Caracas-Vargas highway) were identified based on a severe water shortage. After a disaster, water is consid- their importance both during and after a potential natu- ered essential to human survival, medical response, fire raldisaster.Thesefacilitiescanbeoverlaidontopofhazard fighting, and infrastructure/services recovery. In addition, maps to identify specific locations in need of special atten- a contaminated water supply system could lead to wide- tion. By analyzing such maps, it is possible to determine spread disease. a city's strengths and weaknesses pertaining to natural disasters. This allows decisions to be made about both Transportation: The surface vehicular transportation large-scale systems and individual entities. network is considered to be extremely fragile. The system Table 5.1 and accompanying explanations act as the lacks redundancy, and current critical links may be ren- foundation for the recommendations made in the Urban dered useless in a natural disaster. Survival of a cata- Planning Studio action plan. They were developed strophic disaster within Caracas may require immediate through a series of discussions that analyzed the city's and then continued evacuation out of the Caracas utility and public service infrastructure--facilities crit- basin and Vargas. There is a strong possibility that a ical for survival in the event of a natural disaster. large natural disaster affecting Caracas will also cut its link with the airport and seaport and possibly cut the secondary supply routes to the west (Valencia). In addi- Ranking Explanations tion, the airport and the seaport are fragile facilities Tier I due to their location, potentially rendering them use- Tier I facilities are critical for human survival immedi- less in the event of a large earthquake (for example, on ately after a natural disaster. The extent and severity of the San Sebastian Fault). Clear transportation is essen- human casualties and injuries as well as property damage tial for evacuation, fuel importation (assuming cut will depend primarily on the quality of these facilities. pipelines), food distribution, fire and hazardous mate- rials (HAZMAT) response, as well as the movement of Medical: The ability of existing medical facilities to temporary sanitary and medical facilities. absorb a massive influx of triage and the ability to create temporary overflow space in the immediate aftermath Shelter/housing: Much of the housing in Caracas is of a disaster will be crucial for the survival of the city. considered extremely fragile due to the quality of con- Early indications are that up to 60 percent of the exist- struction in residential areas. Therefore, there will be a ing medical facilities in the greater Caracas region are need for the absorption of large numbers of displaced in some state of disrepair. Unknown factors include people after a natural disaster. The facilities for tempo- the availability of temporary medical facilities (that is, rary shelter must exist, be ready for deployment, and field hospitals), the stockpiling of medicine and med- be fully accessible by emergency response teams. Cur- ical supplies, and the training of medical personnel in rent open spaces (parks, stadiums, and reinforced audi- responding to a disaster (which will likely displace them toriums) must be able to absorb large numbers of people. from their normal place of work). However, these Designated shelters and controlled security will dis- functions are critically needed in any large city, such as courage people from returning to precarious struc- Caracas, in the event of a disaster. tures immediately after a disaster. Water: Should the city become cut off from its water Communication: Open communication is essential for supply, it will have less than one full day of stored water. coordinated response and recovery following a disas- The water supply system is extremely fragile, consider- ter. This includes radio/CB/cellular lines as well as TV ing that the three incoming supply routes, Tuy 1, Tuy 2, and commercial radio. The fragility of these facilities is and Tuy 3, originate in one general location to the south largely unknown, but it is assumed that the robustness of the city (Tuy Valley) and that the current maximum and location of communication transmission sites (towers supply rate barely exceeds current consumption levels. and antennae) will determine the quality of the system. Blockage or destruction of even one aqueduct will cause People need to be informed on a continuous basis. 142 Natural Disaster Hotspots Case Studies Table 5.1. Critical Facilities and Systems (Categories and Definitions) TIER I Medical Hospitals, ambulances, clinics, Red Cross installations, depots of supplies (medicine and equipment), mobile equipment (field hospitals) Water Water supply infrastructure, aqueducts, storage tanks, water trucks, water treatment plants, impoundments Transportation All roads, rail lines, airports, bridges, heliports, seaports, evacuation routes, trucks Shelter/housing Existing structures, including barrios and inner-city apartment blocks (for fragility), that can serve as secure shelters. Tents, cots, blankets, gas lamps, camping equipment (for response) Communication Emergency broadcast system, cell phone/ radio/CB/TV transmission towers and infrastructure TIER II Fuel Fuel storage tanks, delivery mechanisms, pipelines Fire/HAZMAT Fire stations, hydrants, fire equipment/apparatus, HAZMAT equipment/apparatus, response system Electricity Power generation stations, transmission stations/nodes/infrastructure, backup generators, batteries Food Storage warehouses, distribution points TIER III Reserved space Stadiums/parks/fields for construction of temporary tent cities, field hospitals, and central gathering points Sanitary facilities Treatment facilities, temporary facilities SYSTEMS Management system Competent, cohesive command and control center with high-level authority (mayor or president) in central location Search and Rescue Trained personnel, dogs, heavy lifting and clearing equipment, truck evacuation system, boats, and building inspection teams Law enforcement/ Military, police, National Guard Security Tier II Fire/HAZMAT: Most formal homes in Caracas use Tier II facilities may not directly impact civilian lives in natural gas for heating/cooking, so it is assumed that the immediate aftermath of a natural disaster. However, extensive fires could occur after an earthquake. The these facilities are key components of a coordinated and HAZMAT situation is unknown. The fragility of fire efficient emergency response system, aimed at mini- stations, fire fighting equipment, and HAZMAT appa- mizing the severity of the disaster in terms of human ratus is also unknown. The fire authority is assumed to loss and property damage. play a leading role in disaster response, but it is unknown whether they currently accept and practice for this Fuel: Fuel will be critical for emergency electricity role. generators, general transportation, and recovery vehi- cles. The fragility of the supply is considered high, due Food: The supply of food is important but not as crit- to the presence of fuel storage facilities in hazardous ical as other basic needs such as water and shelter. It is areas; however, not much is known about this factor. assumed that food is not stored within the city, but rather The availability of mobile supply equipment is also follows some dynamic market path where food in the unknown. It is assumed that the current supply for the city is replaced as it is consumed (much like the water Caracas area is via fuel pipeline, which should be con- situation). Food supply is also not centralized, but rather sidered fragile. spread among many thousands of grocers and possibly Multihazard Risks in Caracas, República Bolivariana de Venezuela 143 some warehouses. The fragility of the food supply is Management system: Survival after a disaster will therefore considered low. The reintroduction of food require highly coordinated management of emergency into Caracas will require a functioning transportation response and recovery. This should be staged from a system, including airport access. hardened, robust central location, staffed by trained per- sonnel and run by an official in the upper government Electricity: Electricity will be critical for some missions level (that is, a mayor or higher). Staffing must consist (medical, police, and rescue coordination) and unnec- of well-trained government employees whose job require- essary for others (basic personal survival). It is assumed ments will immediately be shifted to emergency response. that temporary and mobile generators will fill inevitable All groups of personnel--police, fire, military, emer- gaps in the power supply, but the availability of these gency medical service (EMS), search and rescue (SAR)-- facilities is unknown. The fragility of the existing power must be under the control, and be totally responsive supply in terms of generation and distribution is also to, the central command authority. This system does largely unknown. Since most earthquake disasters involve not exist, but its importance cannot be overstated. extensive loss of electricity, this must be assumed for Caracas as well. However, much of the electricity for Search and rescue (SAR): There must be in place a Caracas originates in southern Venezuela, and the supply deployable, trained volunteer SAR force pulled from might not be affected in a natural disaster, although dis- all sectors of the community. SAR cannot rely solely on tribution may be. one branch of government service (that is, fire or mil- itary), as each will have other duties to fulfill. SAR will Tier III save lives in the days after the event, but will not be the Tier III facilities are not critical to human survival in most crucial component in life-saving (in terms of quan- the event of a natural disaster; at least, they are not tity) immediately following the event. The fragility of immediately necessary in an emergency response situ- SAR should be low to allow heavy moving equipment ation. However, these facilities are important parts of to be dispersed and undamaged by the event. a coordinated response system. Law enforcement/security: Control of the security Reserved space: The evacuation and subsequent shel- situation will be challenging, requiring coordination tering of displaced persons will require available open by all departments. The fragility of stations and bar- space. This will likely include parks, stadiums, and racks is unknown. This will require an intact backup any other open areas. These areas must absorb tempo- electricity supply and functioning communication lines, rary shelters (tent cities) and field hospital facilities. as well as a working central management system. These areas will become central locations for informa- However, the military, National Guard, and police are tion dispersion, missing persons/split family reunions, functioning entities that may be mobilized in the event and social services. of a natural disaster. Sanitary facilities: Some provision for temporary san- itation must be made before a disaster occurs. Public Strategic Planning Process--Development sanitary facilities are few or non-existent, so they are of the Plan not considered fragile facilities. Without such provi- sion, conditions can become untenable after a few days, The hazard mitigation plan calls for a targeted and delib- and may cause outbreaks of disease. erately focused planning methodology. This was pro- vided by the strategic planning process, which assumes a critical situation and seeks an effective and direct Systems path toward solutions. It prioritizes efforts to attain the Systems refer to operational organizations that require best possible results with the means available. effective coordination between decision-making per- The analysis involved with the strategic planning sonnel and a corps of trained workers. method is focused on identifying strengths, weaknesses, 144 Natural Disaster Hotspots Case Studies opportunities, and threats (SWOT). The threats exam- promotion of open space. ined were possible earthquakes, floods, and land- · El Ávila National Park acts as a natural growth bound- slides/debris flows. Also examined were the strengths, ary to the north. weaknesses, and opportunities in the urban structure · Investment in rail transportation infrastructure opens as well as the current and future socioeconomic and regions of the Tuy Valley and points west for devel- political conditions that could influence the ability to opment. cope with the threats. The following short lists are exam- ples of the items that were generated. Threats · Branches of the San Sebastian and Tacagua-Ávila Strengths faults traverse the Caracas region, posing a threat of · Moderately diversified economic profile with strengths both minor and major earthquakes in the region in in the energy sector. the next few decades. · Modern and efficient public transit system with growth · Climatic variability and steep terrain presents an potential. ongoing threat of small and large landslides and debris · Developed intellectual capital based in universities flows to populated areas of the region. and private institutes. · Extensive rainfalls could result in flooding condi- · Establishment of successful barrio intervention models. tions. · Political receptivity to issues of economic and land- use development. Based on the SWOT analysis, a hazard mitigation · Presence of urban airport and military base in plan was developed that addressed critical weaknesses Caracas. and leveraged strengths and opportunities present in · Presence of major seaport to the west of Caracas: dis- the Caracas region. Using ARCView, a geographic infor- aggregation of wealth and resources. mation system (GIS), base maps were prepared of regional land-uses, service infrastructure, housing typologies, and critical emergency facilities (such as police stations, Weaknesses fire stations, and hospitals). These base maps were · Water system infrastructure is underdeveloped, and then overlaid with maps identifying hazards risks--spe- supply channels traverse fault lines. cific areas where steep topography suggested that · Highway systems lacks sufficient redundancy to over- landslides might occur, and low-lying areas that were come traffic congestion and are vulnerable to clo- prone to flooding. In addition, maps of soil depth, indica- sure. Parts of the Cota Mil, a major highway along tive of earthquake shaking periods, were created and the northern rim of the city, are incomplete, and the overlaid on maps of the city. The composite map bridge that connects Caracas with the Vargas coast resulting from the layering process highlights, at the is experiencing structural duress. metropolitan level, neighborhoods with specific vul- · Communication technology lacks a public emergency nerabilities that might require targeted intervention broadcast system. (figure 5.3). · The sanitation and water run-off system is underde- The studio did not develop micro-level plans for each veloped, lacking sewage treatment and possibly posing of the vulnerable areas of the facilities, but rather focused a health risk. on creating a comprehensive schematic plan for the · Pervasive fear of crime and corruption. Caracas metropolitan area that addressed cited critical weaknesses. In addition, the studio selected a few exam- ples of micro-interventions as demonstration models of Opportunities the broader concepts. The development of small open- · Newly consolidated metropolitan government. spaces in select urban neighborhoods is one such exam- · Mixed-density development allowing for in-fill and ple. The bi-level approach of the plan and Multihazar Figure 5.3. An example of the multihazard map produced by the Urban Planning studio. The map is a compilation of urban-facilities research and natural-hazards research. This map relied upon existing estimates of ground shaking period from FUNVISIS and estimated flooding extent, based on local topography. d Risks in Caracas, República Bolivariana de Venezuela 145 146 Natural Disaster Hotspots Case Studies recommendations, also referred to as "mixed scanning," resistance (by reviewing their previous performance was best suited for the large region under considera- during earthquakes). This information can be used for tion. The broad scope was intentional--at the onset of determining the disaster response capabilities of key the project the studio participants understood the lim- systems and to assist in the decision-making process for itations they faced as "foreign specialists." It was felt that retrofitting critical facilities. the broad plan was a platform from which local plan- ning professionals could develop specific project plans Reserved Space using a more participatory planning approach that included the direct stakeholders from each area of inter- Reserved spaces are open spaces such as parks and recre- vention. The principal aim was to develop the method- ational fields as well as buildings like community cen- ology within which detailed intervention could be ters and school gymnasiums. In the event of a natural introduced and the obtained outputs perhaps modi- disaster, these places will serve a dual function as evac- fied interactively with stakeholders. uation centers, providing temporary emergency serv- The plan focused primarily on physical facilities ices to surrounding residents. Therefore, reserved spaces that required additional assessment, but also touched must be created and/or enhanced in areas that can be upon social and administrative programs that would easily accessed by people and vehicles. Additionally, build resiliency in the community. These included aspects they should be equitably dispersed throughout the city of a land tenure program that would stimulate improved so that every neighborhood has a predetermined evac- building construction and insurance coverage, as well uation site. Larger spaces, such as Parque del Este, Parque as a public hazards awareness campaign, which would Central, and Universidad Central, could serve as evac- occur in the popular media and through education (pri- uation areas for large populations for long stays (figure marily at the school level). 5.4). The plan was outlined in a flexible 20-year time frame Smaller spaces, likely located in barrios, are intended to allow for a phased implementation. The timing of the to accommodate a designated neighborhood for a few plan considered the present condition of any specific days. Because of the smaller scale of reserved spaces in factor under consideration, the resources required for barrios and the high population density of barrio improvement, and the extent of development time needed neighborhoods, a larger number of reserved spaces must for implementation. Qualitative factors such as eco- be strategically created there. nomic and political conditions also influenced the phas- Open spaces may be the facilities that could be located ing sequence. in high earthquake risk areas. However, the same cannot be said for their location within areas subject to hydro- logical hazards. Facilities Buildings used for evacuation centers must be rein- forced to withstand the strongest of earthquakes and Emergency facilities are crucial to minimizing loss of may be a better option in barrios, considering the scarcity life immediately following a natural disaster. Their impor- of land. Regardless of their location, reserved spaces tance is heightened in situations where the entire city must be large enough to allow for temporary shelter, is affected, such as after a large earthquake. New facil- hospitals, and information centers. A key concept in ities must be built in areas where populations are under- both hazard mitigation and disaster response for served, and all facilities, whether located in a high-hazard Caracas is the proposal to create Plazas de Seguridad. area or not, must be structurally sound to withstand Similar to a traditional Latin American parroquia in powerful earthquakes. (See table 5.1 for a list of design--an open space surrounded by civic buildings needed facilities during a disaster and for the locations and community facilities--the Plaza de Seguridad is a of the existing facilities [figures 5.4 and 5.5].) In order practical option, primarily in barrios, where public serv- to determine the structural integrity of existing facili- ices and neighborhood centers are lacking (figures 5.2 ties, all buildings in Caracas should be inventoried based and 5.6). on their age and assessed based on their earthquake- The Plazas de Seguridad should include police and Multihazard Risks in Caracas, República Bolivariana de Venezuela 147 Figure 5.4. Reserved open space in the Caracas Valley. The box to the lower right indicates the location of the Petare District. Figure 5.5. Hospitals, police stations, and fire stations in the Caracas Valley fire stations, a medical facility, a community center, nity and enhancing the quality of life, their clustered and technical units when possible. Therefore, while design would help facilitate coordinated emergency the Plazas de Seguridad would function regularly in the response efforts when an earthquake or flood strikes. absence of a natural disaster, supporting the commu- Plazas de Seguridad should be located in low hazard 148 Figure 5.6. Proposed interventions in a section of Petare to improve disaster preparedness with the allocation of reserved space Natural Disaster Hotspots Case Studies Multihazard Risks in Caracas, República Bolivariana de Venezuela 149 zones if possible. Nevertheless, the civic buildings that areas such as military bases. Access to major trans- accompany the plazas should be strong enough to with- portation links should be a major factor in determin- stand the strongest of earthquakes. At varying capaci- ing their locations. ties, both open spaces and buildings used as evacuation centers will function as: Command and Control Center · temporary shelters, equipped with a supply of tents A command center should be created in a central loca- if necessary; tion. It must be a self-sufficient structure, capable of · field hospitals, possibly in existing buildings with absorbing and disseminating massive amounts of infor- stored supplies and basic medical equipment; mation in a short period of time. The center will be used · information centers, with uninterrupted linkages to to coordinate immediate emergency relief efforts and the central communications system; long-term disaster relief programs; it should also serve · supply distribution points for basic survival supplies, as a permanent installation able to manage routine emer- such as water, food, and blankets; and gency situations. · sanitary facilities, including toilets, showers, and waste disposal units. Cultural Buildings Cultural buildings such as museums, libraries, gov- Medical, Police, and Fire Buildings ernment buildings, and universities hold the country's These critical facilities have to be self-sufficient struc- history and heritage within their walls. They are tures, able to withstand strong earthquakes and remain important symbols of national identity and pride, and functional. Therefore, they should have structural many government buildings and libraries store official integrity, backup electrical generators, and a separate records, legal documents, and personal identification and sufficient water supply and storage space. As with information. These buildings should receive high pri- the Plazas de Seguridad, their effectiveness in terms of ority for retrofit efforts. disaster mitigation depends, to a large extent, on their proximity to populations. Thus, proportionate distri- Water Distribution System bution of hospitals, police stations, and fire stations based on population should be encouraged. Where As mentioned previously, the water distribution this is not feasible, smaller medical facilities can fill in system in Caracas is inadequate, considering that cur- gaps as long as they have access to locally stored med- rent supply lines provide insufficient amounts of water ical equipment and supplies. The hardening or relo- to the valley and are susceptible to earthquakes. As a cating of critical facilities must be considered on a result, more distribution pipes and aqueducts built to case-by-case basis, as the decision would depend upon withstand seismic events must be constructed, and the factors such as accessibility and need in the case of a system must be made more redundant in case of mal- disaster. function of one of the lines. Additionally, all existing lines must be hardened, particularly at the points where the aqueducts intersect seismic faults (figure 5.7). While Storage Depots these measures will help to ensure that Caracas has a It is necessary to construct major assembly points for sufficient and consistent water supply, the critical the maintenance and storage of equipment and supplies function of water, both during and after a natural dis- that will be utilized during the post-disaster period. This aster, necessitates extra precautionary measures. equipment includes heavy machinery such as cranes, Because water must be pumped up into Caracas, bulldozers, and trucks; hazardous materials cleanup which is located ~900 meters above sea level, water dis- apparatus; a fleet of helicopters and buses; and tempo- tribution capability is directly tied to the robustness of rary shelters and hospital equipment. The storage depots the electrical power system. As a result, backup power should be placed in strategically located and secured is critical to ensuring continued water delivery in the 150 Figure 5.7. The Caracas water supply system, showing key infrastructure crossing fault lines Natural Disaster Hotspots Case Studies Multihazard Risks in Caracas, República Bolivariana de Venezuela 151 case of a general power failure. All water pumping sta- Bridges should be given special attention, as they tions should have individual generators and be pro- are the most fragile elements of the surface trans- vided with additional alternate electrical lines, as portation system and are under heavy seismic threat. appropriate. The bridge connecting the Caracas valley to Maique- Additionally, since one or more of the water supply tia/La Guaira should receive immediate attention, con- lines may fail during a strong seismic event (even if they sidering it is the only connector to the coast and to are well built), ample water storage must be available the major international airport and port. Its vulnera- within the city. Currently, there is less than a one-day ble status has been documented for some time, and supply of water on reserve, and the major storage facil- structural failure could occur, even without a major ities are located outside of the valley. Consequently, sites disaster event. should be found, within the city, where large holding tanks or reservoirs can be built. The total water reserve Communication capacity should be no less than the equivalent of three days worth of potable and fire-fighting water. In the minutes and hours directly following a major emergency, communication can be a matter of life or death. The general public must be able to receive Surface Transportation information about what has happened, as well as instruc- Surface transportation is a critical infrastructure tions for further action. An emergency broadcast system system that must be functioning well at all times, but should be created so that information is disseminated is especially crucial in case of emergency. Many lives as quickly and efficiently as possible. will be saved directly following a major disaster if emer- Television and radio broadcasts as well as announce- gency vehicles carrying personnel and supplies are ments over public address systems and megaphones are able to move efficiently throughout the city. all effective means of disseminating information. The Therefore, the key network should be planned to system should be set up with a strict hierarchy of deci- withstand natural hazards and to handle emergency sion making focused on what information is to be dis- responses (figures 5.8­5.10). Sections of the road net- tributed and what personnel should be responsible to work under risk must be well engineered to handle avoid confusion. both earthquake shaking and mudslides/debris flows The general public should be made aware of where caused by heavy rains. Elevated sections are under high- to receive information and instructions immediately fol- est threat during an earthquake, and should be hard- lowing a disaster. Public education and awareness ened to prevent major structural failure. Some sections campaigns, along with regular drills, can help to acquaint of the highway built at grade are under threat of flood- residents with the sources and processes, and particu- ing. The river must be channeled and controlled in larly with the emergency broadcast system. these areas to prevent water from blocking all vital Authorities responding to an emergency need to be movements. in constant communication, as well. A unified and A road network with a high degree of redundancy permanent emergency communication center should should be in place both within the Caracas valley and be created that will handle all communication between along the major highways that connect Caracas to the the police, civil defense, fire fighters, medics, and rest of the region so that alternate paths may be open other authorities in case of an emergency. A clear hier- under the most serious disaster conditions. Many of archy of instruction, procedure, and personnel should these major connectors are under threat, and a robust be established to avoid mass confusion and wasteful network must be created and emergency routes estab- duplication of effort following the disaster. lished to prevent gridlock should one of these main Finally, the city should adopt software systems to pre- arteries fail. Without redundancy, the entire city could vent communication gridlock, maintain a protocol, and be paralyzed following a major event, preventing assis- give priority to the appropriate personnel. Gridlock hap- tance from reaching victims. pens during post-disaster periods when the use of tele- 152 Figure 5.8. Seismic hazard affecting the city's transportation network. Five points of vulnerability to shaking were identified in the major transportation infrastructure. Natural Disaster Hotspots Case Studies Multihazar Figure 5.9. Debris flows affecting the city's transportation network. The Cota Mil highway is especially vulnerable to debris flow as it traces d the southern border of the steep El Ávila mountain range. The Caracas-Vargas highway is also threatened by landslides. Risks in Caracas, República Bolivariana de Venezuela 153 154 Natural Disaster Hotspots Case Studies Figure 5.10. Flood hazard affecting the city's transportation network. The main east-west thoroughfare through the city is paralleled by the main, channelized river. phone lines and other communication links increases design. Many light and heavy rail lines are currently drastically. Some lines must be reserved for official emer- being built or proposed for the valley of Caracas. Among gencyuseonlyinordertopreventacompletebreakdown. them are the Railroad Cua-Tuy Medio, the new METRO lines, and the Los Teques line. To prepare these sys- tems for a natural disaster, it is necessary to assure that METRO System structures are built robust enough to withstand an earth- The structural strength of the METRO system is most quake. important in terms of earthquake risk. Operational It is also necessary to assure that the operational prob- elements must be engineered to handle these forces. lems the systems may encounter during a natural dis- Operational safety of the system is also crucial in case aster, including issues of citizen safety, availability of of a natural disaster. The METRO system should be pre- evacuation routes, and accessibility for repairs, are antic- pared to move and protect its users by providing clear ipated. All these issues should be taken into account in and unobstructed emergency escapes and evacuation the design and construction phases so that time and routes. These routes should be equipped with com- money can be saved later. munication devices connecting it to the emergency broadcast system. Air and Seaports Since the METRO was built recently, it can be assumed that proper safety factors have been incorporated in its Air and seaports become critical in the event of a dis- Multihazard Risks in Caracas, República Bolivariana de Venezuela 155 aster. They are the major ports of entry and exit for prevent leakage in case of earthquake or flooding. Con- personnel and equipment. They must be fully opera- crete and steel are generally used for sewage pipes. Con- tional in order to absorb national and international relief crete has poor tensile strength and can be highly assistance, and to evacuate people out of the area if vulnerable to ground shaking. Steel has much better necessary. tensile force, but the joints can be vulnerable and may have to be retrofitted to handle earthquake stress. Cast iron pipes are brittle and perform poorly during Power System earthquakes. The power system must be designed to withstand a disaster, and to be operational should one occur. Par- ticularly during large earthquakes, it is common to have Housing massive power failures in multiple locations, making restoration efforts difficult. To prepare the system for a The overarching goal of housing policy is to equip all natural disaster, it is necessary to make an assessment existing residential structures for hazard resiliency and of the existing conditions and to reinforce the available to guide future disaster-resistant housing development. infrastructure in relation to its importance in the system. Housing is at once both inherently physical and social, This includes insuring the structural stability of power and the programs presented here reflect this duality. generation stations, transmission towers, transformer stations, switchyards, and distribution lines. The con- Physical Programs trol and related communication systems must be hard- ened and redundant. Appropriate building codes: these should be devel- olped based upon a complete hazard assessment of the given area as well as information about the existing struc- Natural Gas Distribution tures. Requirements should include building and site- Caracas has a natural gas distribution system that serves grading ordinances along with design and construction most of the valley, yet only limited information has been regulations. All categories of local housing and the array obtained so far on the specific location and condition of local materials used in its construction have to be of these lines in the city. They carry a highly flamma- considered. Mortgage lending, permits, training, and ble substance, thus representing a major fire threat in quality control inspections can be tied into the admin- the case of disaster. The system should be analyzed to istration of building codes as well. A failure to include assess its structural soundness, identifying possible leak- disaster-resistant standards for low-cost housing increases age and break-up scenarios. In some cases, retrofitting the vulnerability of citizens, and further excludes them may be necessary. In order to reduce fire risk, emer- from the formal housing market. Strict codes not only gency shut off valves--which would allow authorities ensure that existing structures will have a better to stop the flow of gas from the source--should be chance in the event of a major disaster, but they also considered. These valves should exist at least at central guide future hazard-resistant building. Enforcement is distribution nodes, but may be considered at major con- critical, and a well-trained, well-staffed agency is needed sumption points as well. to accommodate demand. Structural reinforcement: Hazard maps and building Sewage history can be used to establish priority structures for Although sewage is not an essential element in disas- hazard abatement programs. In the most hazardous ter preparedness and mitigation, it becomes a danger areas, local governments can adopt mandatory retrofit to human health if raw sewage enters the water supply programs. Retrofitting measures may include the or the environment. It is necessary to strengthen the insertion of walls on the outside or in the interior of sewage system, especially the major outflow pipes, to buildings, buttresses, specially anchored frames or the 156 Natural Disaster Hotspots Case Studies construction of a new frame system, covering of columns taining high performance standards in workmanship and beams, portico fill-in walls, and tie-rods and safety is crucial (possibly more effective than building codes) "wraps." Government subsidies may be necessary to in assuring that structural standards are met. Programs promote this program in the form of low-interest should be monitored to ensure that the instruction is loans, to ensure that people are not displaced by rising effective. Testing and licensing procedures should be rents. instituted, and appropriate quality control standards should be in place. Cooperative building groups, Land use and zoning: Hazard risk mitigation should material subsidies, or community work programs will be a deciding factor when choosing new building sites. benefit low-income families who cannot afford exten- These issues should be considered at the early stages of sive home improvements or construction loans. site selection to ensure that hazards will be weighed against the strategic advantages of a given location. Thus, Social Programs in addition to abiding by building codes, structural design considerations should include the location and Land title: The Housing Policy Law of 1989 recog- height of the building, the structural system, building nizes ranchos as a legitimate form of housing, and materials, functional relations between various sections mandates the granting of property rights to established and building composition, vulnerability to specific barrios as a part of the process of barrio upgrading. Barrio disasters, possible impacts of disasters on occupants, upgrading has become the focal point of national and special needs of residents. housing policy, and hazard risk mitigation should be fused into these programs. Several government agen- Training: A great number of homes in Caracas are cies, states, and municipalities contribute to barrio built outside of the formal construction process with- upgrading, including the Comision Nacional de Equipa- out the involvement of lending institutions, trained mentodeBarrios (National Commission on Barrio Upgrad- architects, or experienced builders. Therefore, building ing), which was created in 1995 with a mandate to codes must be supplemented by a training program that coordinate policies and investments in barrios. Grant- teaches disaster-resistant building techniques to all ing land tenure to barrio residents explicitly acknowl- builders within the private and informal sectors, as well edges that city growth and improvement will be linked as to self-help labor (those who build their own homes). to barrio development. As such, the immense informal The most important component of a training program housing stock can be transformed from a problem to is local participation. People must be aware of the hazard be resolved into a resource to be utilized. Additionally, risks, believe that the implementation of certain con- the provision of property rights can facilitate the com- struction techniques will add to the safety of their homes, munity participation necessary to pursue effective hazard- and be offered the opportunity to either build or improve resistant planning. Land title not only gives legitimate their homes to meet building codes at affordable costs. value to investment, but also can be used to leverage A higher level of training is necessary for architects, home improvements, loans, and disaster insurance. builders, masons, and other construction personnel. In Caracas, 58 percent of barrios are located on public Trainers from this pool may be selected to instruct other lands; 27 percent are on lands with a mixture of builders or to provide technical assistance to local public and private ownership; and 15 percent are located communities. They also can be certified as building on private lands. The process of transferring public inspectors. A second level of training should focus on lands to agencies involved with barrio improvements self-help builders. This may include specific training must be streamlined and made more efficient. Private in basic building techniques, as well as raising aware- land owners should be reasonably compensated for ness of the importance of hazard-resistant building. The relinquishing their land, and a government agency in success of any housing improvement program is depend- cooperation with local communities should oversee the ent upon the extent to which communities seek to transfer of land titles to residents. The extensive involve- increase the safety and stability of their homes. Main- ment of each community is essential for legitimizing Multihazard Risks in Caracas, República Bolivariana de Venezuela 157 the ownership structure, thereby reducing subse- who undertake loss prevention measures. Structures quent property-related disputes. It also will encour- that meet certain criteria would qualify for financial age community policing of further invasions. The incentives from banks, contractors, and insurers. Group Housing Policy Law has begun this process, but it should insurance schemes can be incorporated into the train- be modified and given additional legal and institutional ing and education programs aimed at low-income com- support if it is to successfully incorporate disaster-resist- munities. Incentives to exceed minimum safety standards ant measures. can be built into the rate structure. The government Hazard risk mitigation is an underlying factor in the could provide tax incentives to companies who insure transfer of land rights. Landowners and real estate bro- very low-income communities. kers should be legally responsible for full disclosure of hazard assessments of the property. Planning and provision: New barrio development is constantly underway and must be targeted for imme- Housing finance: Volatile conditions within the finan- diate planning for service provision to prevent the prob- cial sector have direct and adverse effects on the hous- lems with lack of access, services, and extreme hazard ing sector. The high annual inflation rate (averaging 52 risk that affect many existing settlements. Service and percent between 1990 and 1997) hinders the develop- infrastructure provision in newly emerging barrios will ment of the mortgage market; overdependence of the be far more cost effective than late interventions. Munic- economy on oil leads to the fluctuation of housing ipalities can work together with these new communi- subsidy levels. This causes stagnation in the private ties to develop good design ideas and to establish effective financial sector and the deterioration of real income zoning codes. levels. All these factors serve to cripple the efficient development of a strong housing sector. Unfortu- Land Use Regulations and Relocation Principles nately, housing loans will continue to be in limited supply as long as interest rates continue to be subsidized, and Regulating land use, both in built and open areas, is a many of the proposals suggested here require a supply crucial component of hazard risk mitigation. In open of low-interest loans. To reconcile these conflicting areas, good urban design guidelines, enforceable trends, attention must be given to finding ways to intro- building codes, and limits on population settlement in duce greater competition into the housing finance market. high hazard areas can ensure that the city is a less risky One possible solution is to eliminate direct housing place, even as population grows. In built areas, atten- subsidies and, instead, to subsidize loans for safe tion to the adequate provision of public services and housing construction and disaster-resistant upgrading. space, as well as the identification of particular facili- These loans should first be made available to those ties that need structural reinforcement or relocation, with the greatest need, such as low-income barrio or can help reduce the vulnerability of large sectors of the "vertical" communities (in low-income, multi-story res- city's population. Finally, attention to land use regula- idences). Low-interest loans can also be tied to train- tions can make Caracas a more livable city, both before ing programs as an added incentive for builders to and after a natural disaster strikes. undergo training in the construction of hazard-resist- ant buildings. Enforcement Insurance: Disaster insurance should be introduced At the outset, it needs to be emphasized that no regula- and tied to housing finance. Real estate agents could tion should be enacted unless there is an immediately be required to generate risk disclosure statements, workable enforcement procedure (including adequate offer special insurance coverage and policy riders, main- funding and involvement of the local community in both tain affordable premiums, and introduce mandatory drawing up the regulations and implementing them.) purchase requirements in lending agencies. Insurers Regulations enacted without the consent of the com- could provide premium reductions for policy-holders munity are likely to be challenged or ignored, while unen- 158 Natural Disaster Hotspots Case Studies forced regulations can actually increase overall risk, as in areas of moderate risk, except where it is related to they lead to complacency on the part of government the introduction of new infrastructure. officials and the affected population; such a circumstance When buildings are removed to introduce roads gives the sense that the problem had been solved when and other access routes into the barrios, moderate-risk it was actually worsening during periods of perceived areas should receive priority--ensuring that the safety.Land-useregulationsthatinvolvecommunitypolic- people are relocated out of risky areas, but that new ing and enforcement efforts should receive priority. infrastructure is not located within the areas of highest risk. Such roads and access routes should lead to newly established Plazas de Seguridad, which should Open Areas be located within areas of low risk, if possible. This For the open areas of the Caracas metropolitan area-- will not be feasible in many instances because these particularly at the edges of barrios and in locations plazas must be easily accessible to the population that subject to formal private real estate development--high- they are expected to serve. risk areas, as determined by thorough hazard assess- ments, should be off limits to residential construction and to nearly all commercial activity. Such areas can still Relocation Principles be used; potential uses include agricultural production, recreation, and other open spaces with activities that In some areas of Caracas, only extremely costly engi- do not place people and significant investments in danger. neering techniques can save the lives and property of Areas of medium risk should be reserved for roads and many people living in risky areas. Where it is a matter other access routes, manufacturing and industrial activ- of life and death, relocation of people and buildings ity, and low-density commercial uses. Low-risk areas should be considered. Relocation may also be warranted should be open to residential construction and serve as where the entire community would benefit--that is, for the sites of Plazas de Seguridad, higher-density com- the introduction of access roads and services into mercial activity, schools, hospitals, and other critical barrio areas. facilities. However, they, too, should be built recogniz- In these cases, it is recommended that certain prin- ing potential dangers. Additionally, it is possible to con- ciples be followed. First, the members of the popula- sider the construction of some non-vital structures with tion at risk should be active participants in the relocation an expected short life span and limited investment, process. This population should be involved not only recognizing that they may be destroyed. in determining who needs to be relocated, but also deter- mining what constitutes a risky area. Different popula- tions have different perceptions of risk, and this needs Built Areas to be taken into account. Often low-income popula- For the built areas of the Caracas metropolitan area, tions, in particular, knowingly bear risks in order to sat- there is likely to be tension between the hardening of isfy immediate living and work objectives. existing structures in high-hazard areas and the relo- Second, priority should be given to maintaining the cation of people and buildings out of those areas. No familial, social, and economic support networks of the blanket rules can be established to resolve this difficult people who are moved. Experience with natural disas- choice. ters and relocation both in Repúplica Bolivariana de Nevertheless, certain guidelines can be recommended. Venezuela and around the world has shown that if these Areas that are high risk both for seismic and hydrolog- networks are disrupted, people will move back into ical reasons ought to have a bias toward relocation. Areas risky areas. Residents of Caracas should be allowed to of high risk for either, but not both, should have a bias remain in Caracas, preferably within their existing toward hardening, with an emphasis not only on tra- neighborhoods. ditional engineering techniques, but also on the use of This can be achieved by increasing density through alternative materials. No relocation should take place appropriate housing design within the lower-risk parts Multihazard Risks in Caracas, República Bolivariana de Venezuela 159 of neighborhoods affected by relocation. Finally, relo- ing a family hazard drill and plan--would be both an cation for natural-hazard mitigation should not be used entertaining and effective way of making all those who as an excuse for mass relocation or to justify "urban live in Caracas aware of the dangers that surround them. renewal" programs. If the above principles are followed, Penetrating people's consciousness in such a creative only a small percentage of the population at risk will way can take many different forms. Since so many homes need to be moved. in Caracas are built by the residents themselves, it is crucial to link some form of education in proper build- ing techniques to the sale of home-building materials. Education and Outreach Such an effort should be just a small part of an overall education and outreach scheme targeting the profes- Education is one of the most fundamental and impor- sional community (who both prepare for and respond tant hazard mitigation strategies. It is also an area where to natural disasters), including builders, architects, health innovation, creative thinking, and experimentation can care professionals, emergency response technicians, and be employed at a low cost to raise the collective aware- so on. ness about the dangers posed by natural hazards. A pop- Much of this specialized education could be admin- ulation that is cognizant about the risks associated istered by technical units, which should be established with their environmental surroundings is more likely throughout the entire metropolitan area. These units to be willing to accept and participate in the imple- would not only provide education and training at the mentation of an overall hazard risk mitigation plan. grassroots level, but also could serve as an operations Hazard education in Caracas should generally fall base for a trained volunteer corps such as the Red into two categories: emergency response, or what to Cross and other nongovernmental support groups. do when disaster strikes; and the teaching of basic mit- If all these tasks can be accomplished, not only will igation techniques, or what to do to lessen the impact Caracas be a safer place in which to live, but also the of a natural disaster. city could become a regional center for hazard educa- Crucial for both of these areas is the introduction of tion and research, attracting international attention and a mandatory natural-hazard curriculum into the schools funding for further innovation and experimentation in of Caracas. Children can be taught both what to do in hazard education and outreach. a disaster and how to prepare for it ahead of time. Such instruction can eventually be mainstreamed into the overall school curriculum so that subjects such as sci- Management Structure ence and history can take on disaster-related themes. Raising awareness and preparing the population in An effective management structure is crucial to plan- general to respond to natural disasters is a crucial task ning and implementing hazard mitigation plans as for the mass media. Public service announcements and well as for preparing for disaster response. While some short instructional messages on television and radio and of the recommendations here are important independ- in the print media can go a long way toward educating ent proposals, if implemented in their entirety, the sum the public about their city and its dangers. These mes- will be greater than the parts. Thus, coordination and sages should be incorporated into the preparation of oversight are essential. Nevertheless, it is not the intent hazard drills and the formulation of family, school, com- to propose additional layers of bureaucracy, both for munity, and business hazard response and mitigation fiscal and programmatic reasons. Many of the policies plans. called for in this plan can be implemented by existing Yet, even deeper than such traditional methods, government departments--they simply need to be prop- natural hazards and disasters can become a theme in erly oriented to undertake current and new programs popular culture. Incorporating disaster subject matter with disaster mitigation and prevention in mind. Fur- into soap operas, movies, and advertisements--most thermore, a new bureaucracy could inhibit many of any domestic story line could accommodate formulat- the community-based initiatives proposed in this plan. 160 Natural Disaster Hotspots Case Studies When governments are forced to respond to natural ensuring that NGOs and community groups are fully disasters, agencies and organizations must quickly coop- integrated into the planning and implementation of erate to enact coordinated actions. This type of coop- hazard-mitigation policies. eration could work for the planning for natural-hazard The section of the Commission responsible for dis- mitigation as well. Therefore, a Presidential Commis- aster response should have at its disposal the resources sion on Disaster Preparedness and Response should be of both Defensa Civil and Guardia Nacional. It would created that would signal and provide leadership from also, through an NGO/Community Group liaison office, the top, but assign both planning and response largely be able to integrate the expertise and abilities of these to existing organizations. types of organizations as well as call upon, in the event At an organizational level, the Commission would of a disaster, experts from other government depart- be run by an executive director, who would have two ments through an additional liaison office. Within this assistant directors--a civilian official responsible for sector it is crucial that clear lines of authority, respon- hazard mitigation (that is, continuous long-range sibility, and procedure be established, practiced, and planning), and a military officer responsible for disas- monitored. ter response (that is, action under emergency condi- tions). While in many cases mitigation and response measures can overlap, one aiding the other, in terms of Cost-Benefit Considerations planning and implementation they are largely separate activities. In the Venezuelan context, the programs Preparing a city for a disaster is a major task, requiring needed for mitigation are generally implemented by a serious investment in preparedness, mitigation, and civilian government agencies and local community emergency response measures. Before undertaking such groups, while the manpower and logistics needed to a project, a full cost-benefit analysis would allow pri- respond adequately to natural disasters are largely located orities to be defined and effective programs to be struc- within military institutions and domestic and interna- tured. Such an effort was not possible to accomplish in tional nongovernmental organizations (NGOs). this preliminary study, but should be considered in a Aiding the executive director would be two inde- full-scale disaster preparedness effort. In the meantime, pendent offices--one focusing on research and infor- based upon past experiences in other cities, some esti- mation (and linked with those doing hazard awareness mates can be made as to the losses incurred by major and disaster preparedness education, as well as with earthquake events, both in terms of property damage those working on barrio integration), and the other on and loss of life. This point can be illustrated by explor- finance and fundraising, with an emphasis on securing ing the impact of magnitude 5, 6, and 7 earthquakes in monies from the international community and sug- a city whose physical assets are valued at US$100 bil- gesting ways to raise funds internally for mitigation and lion (table 5.2). preparedness. It is in these two broad areas, informa- These figures are limited to physical losses as well tion and funding, that mitigation and preparedness often as losses associated with suspended use and service overlap. after a seismic event. They do not address the loss of The section of the Commission responsible for hazard life, which can be much more difficult to estimate. mitigation should work directly with the various gov- Though the benefits of preparedness, mitigation, and ernment ministries and independent government agen- emergency response are nearly impossible to quantify, cies responsible for the built environment and social the costs associated with these activities can be esti- policies and welfare. Each one of these ministries and mated. In order to make the process more cost effec- agencies would create a department of hazard mitiga- tive, it is necessary to first inventory all existing buildings tion, which, upon direction from the Commission, would and infrastructure. ensure that all activities of the ministry are consistent These structures and systems, summarized in our with proper hazard-mitigation procedures. In addition, action plan, can then be prioritized for retrofitting and this sector would have an office expressly dedicated to other mitigation measures. A system of "rapid screen- Multihazard Risks in Caracas, República Bolivariana de Venezuela 161 ing," which is now underway Table 5.2. Studio estimates for the order of magnitude of losses for a generic in the United States, allows city whose assets are valued at US $100 billion. Approximate losses are given for qualified professionals to three earthquake magnitudes. Actual losses can vary widely depending on geol- ogy, fragility of structures, and the location of the earthquake relative to the observe any structure from the center of the city. Estimates are based on observed losses from recent earthquakes outside and to enter informa- in several countries. tion into a standardized form via portable computer. This Assumed Asset Value: US $100 billion allows an engineer to cover a earthquake magnitude: recurrence period: percent loss: dollar loss: lot of territory in a short period 5 10 years < 0.1% < $0.1 billion of time, and information is 6 10­100 years 1-5% $1­5 billion directly transmitted to those 7 100+ years 10-20% $10­20 billion making systems decisions. This is a relatively low-cost approach that has enormous long- social, economic, and physical risks that need to be term benefits. managed to transform Caracas into a disaster-resilient Constructing new buildings to withstand natural dis- metropolis for the 21st century. But before realistic dis- asters is essential to lowering the future cost of prepar- aster risk management policies for this region can be ing a city for natural disaster or of rebuilding it. On implemented, it will be necessary to repeat this effort in average, the added cost associated with building seis- a fully quantitative mode. Such an effort will require fully mic resistant structures from the start is estimated at 4 probabilistic methodologies of hazard and risk assess- percent. This figure varies depending upon location, ment; cost-benefit analyses; and the use of improved, hazard, and type of structure. This can be compared to more comprehensive datasets on the physical, demo- the cost of retrofitting, which can be as high as 25­30 graphic,social,andeconomiccharacteristicsoftheregion. percent of the replacement value. We briefly list some of the key methodologies that will Instituting a system of disaster insurance means that need to be applied for such an effort. the financial burden of a disaster is shifted away from the victims themselves alone. The insurance industry Quantitative Risk Assessment could be tied in to mortgage lending, permits, and financing in order to insure that each new structure Risk is generally defined as the product of three locally built in Venezuela is properly insured in the event of varying factors, integrated or summed over the region a disaster. or subjects of interest. The three elements are assets, Studio estimates for the order of magnitude of hazard, and fragility of the assets to the hazard. As an losses for a generic city whose assets are valued at US$100 equation, one can write this definition of risk: billion. Approximate losses are given for three earth- Risk = Regional Sum of the Local Products of quake magnitudes. Actual losses can vary widely depend- (Assets x Hazard x Fragility) ing on geology, fragility of structures, and relative location The assets are taken as the (dollar) value of any of of the earthquake to the center of the city. Estimates are the objects or subjects at risk. They can be lives lost, based on observed losses from recent earthquakes in or they can be the replacement value of built structures several countries. such as buildings, their contents, or infrastructure. It is easier to estimate the value of physical structures, and, hence, compute the losses from physical damage Future Disaster Risk Management Tasks to the built environment. It is much harder to esti- for Caracas mate the indirect economic losses that follow from physical damage, not to speak of the even more diffi- The Urban Planning Studio has made largely qualitative cult estimation of the intangible losses such as loss of assessments of the most important hazards to which Cara- lives or the impact on the culture and social fabric from cas is exposed, and it has provided some sense of the natural catastrophes. 162 Natural Disaster Hotspots Case Studies The hazard, as described above, is the level of the ture of disaster preparedness and mitigation; to trying hazard parameter at each location of an exposed asset, to largely eliminate the risks through radical rebuild- likely to be exceeded at a given probability level. If there ing and restructuring the communities in a truly dis- are multiple perils (floods, earthquakes, landslides), aster-resilient way at obviously high up-front social and then the above risk equation has to be computed for financial costs. In reality, some optimization will take each hazard and chosen exceedance probability sepa- place that attempts to balance affordable costs to the rately, and the losses from the different types of haz- community with the achievable benefits of risk and loss ards must be added to obtain a combined multihazard reduction. risk at this exceedance probability. When the losses are To achieve such optimization in disaster mitigation calculated at different exceedance probabilities, one can on sound scientific/technical grounds, it will be neces- construct a probabilistic loss or risk curve (Loss vs. Annual sary to: Probability of Exceedance). The averaged total of · Build the institutional and personal knowledge infra- annualized losses24 is related to the area under the risk structure for risk assessment and management so it curve. can become effective in those populated regions of The third variable is the fragility of a given asset to a country that are most threatened by natural haz- the given hazard. Fragility is defined as the fraction of ards and disasters. the asset's replacement value that was damaged and, · Establish sound data acquisition and data manage- hence, lost. A fragility of 1 represents a total loss, and ment programs on the natural environment and a fragility of 0 means no damage, and, hence, no loss. processes that are needed to quantify at least the most The fragility varies for each hazard and hazard level, important hazards (that is, meteorological, hydro- and as a function of the type of asset. For the same shak- logical, seismic, and geological data). ing level, an adobe building may collapse, a concrete · Assemble databases for the exposed assets (or future building may show cracks, and a well-built steel build- planned assets) and their current (or planned) fragili- ing may have no damage. But for the most severe shak- ties to the various hazards. ing, all three types of structures may collapse. · Based on the quantitative risk assessments, develop Modern computer-based techniques have been devel- indigenous, cost-effective policy options to advance oped to quantify the risks according to the above rela- optimized, region-specific, natural-hazard, and risk- tion for risk. The losses can be quantified either for given mitigation procedures that are worth financing and scenario events, or as annualized average losses to the that contribute to disaster-resilient sustainable devel- region. Physical losses, economic losses, and demands opment. in terms of emergency resources needed (for example, available versus needed hospital beds; amount of debris to be removed; functionality of infrastructure systems Specific Recommendations for Caracas as a function of reconstruction time; and so on) are useful outputs of these computer-aided risk assess- As can be seen from the previous section, those who ment tools. are planning, building, and administering the city of Caracas and the country of Venezuela have many tasks Developing Optimal Risk Management Plans ahead of them if they are to realize the goal of creating a city that is prepared to withstand natural disasters. It Once the risks from natural hazards to the region are is difficult to argue that certain tasks are more essential quantified, then one can develop informed plans to than others. It is hard to imagine a resilient and strate- manage these risks and to mitigate them. Many options gic road network being built without the issues of access are available, ranging from just waiting and then deal- and land use in the barrios being adequately addressed, ing with the consequences of unmitigated events; to a mix- for example. Yet we are certain that one thing needs to be done first. This planning exercise needs to be dupli- 24.Summing up the annualized loss contributions from all probability levels. cated, in Caracas, on a larger scale, and by those who Multihazard Risks in Caracas, República Bolivariana de Venezuela 163 have a better understanding of the issues surrounding Infrastructure implementation than we do. It cannot be stressed enough 1­5 years: Inventory existing infrastructure; harden and that the plan offered here is more an example of the retrofit the most critical infrastructure such as the issues that need to be addressed than a prescription that road between Caracas and the Vargas coast; develop should be followed. Our data were limited--terribly appropriate building codes considering hazard condi- so in some cases--and the numerous assumptions that tions, building history, and building location; deter- we have had to make to conclude this plan may have minethelocationandprogrammingofnewopenspaces, caused it to have some fatal flaws. critical facilities, and other needed infrastructure. Because of these data inadequacies, such a full- 5­10 years: Harden and retrofit second-tier-risk infra- scale, natural-hazard, mitigation-planning task would structure; begin relocation/redundancy schemes for have to begin with a better set of data. In the science risky areas; strengthen water infrastructure, includ- section of this publication, we have already outlined ing the construction of water storage facilities some of the issues that need to be addressed in this area, within the valley; construct new open spaces. such as obtaining more accurate measurements of cur- 10-plus years: Achieve a redundant and resilient road rent plate motion and rainfall rates, a better historical network, water system, power grid and communi- record of major earthquakes and debris flow events, and cations infrastructure; relocate all critical facilities a more complete methodology for determining the rela- currently located in high-risk areas; create open space tionship between soil type and the magnitude of earth in all areas where needed; establish critical facilities shaking and debris flows. in all areas where needed. Similarly, the city's infrastructure and housing need to be completely inventoried, with particular attention paid to building conditions and how they intersect with Housing the hazard risks for each individual location. This data 1­5 years: Inventory existing structures; develop appro- collection is just the beginning of what we see as an inte- priate building codes; streamline transfer of land grated implementation timeline that simultaneously tenure; organize technical design units and the dis- works on many levels in a diverse set of areas. Each of tribution of safe materials; begin hardening and retro- these areas--infrastructure, housing and land use, sci- fitting of residential units; plan and organize relocation entific inquiry, education, administration--should be strategies. addressed immediately; furthermore, each area has some 5­10 years: Continue barrio and infrastructure upgrad- short-term goals that both address particular problems ing; fully integrate the work of technical units; devel- and further the completion of some medium- and long- opment of hazard-based zoning code; complete term goals as well. relocation; continue transfer of land title; begin to The recommendations and goals we have arrived at establish a real estate market in tandem with title require many people to begin to make hazard planning transfer. a part of their daily lives at both work and home. They 10-plus years: Legitimate land title; low-interest-loan also call for important steps to be made in areas that options; working real estate market. are not in the traditional domain of hazard planning. But in this way, all those concerned with the health and welfare of the entire population of Caracas can be Scientific Inquiry part of achieving the overall goal of building a safe and 1­5 years: Data mining; exchanges between profes- livable city. sionals and technicians; purchase of appropriate tech- nology. Goals and Time Frame 5­10 years: Complete set of accurate hazard maps; established system for ongoing data gathering, hazard For the different sectors below we propose the follow- mapping and data analysis; ongoing information ing short-, medium-, and long-term goals. exchange. 164 Natural Disaster Hotspots Case Studies 10-plus years: Hazard information clearinghouse for ministry; streamline funding through centralized scientific community and others interested in the system. field. 10-plus years: Clearly organized management system in place with active participation of elected officials, In the area of scientific inquiry, we propose the fol- military, community groups, local and international lowing short-, medium-, and long-term goals. nongovernmental organizations, the scientific com- munity, and the general public. Public Education and Outreach 1­5 years: Establish hazard curriculum in schools; begin References public outreach/education programs; mainstream the hazard message into pop culture; begin public serv- Audemard, F. A., et al. 2000. Map and Database of Quaternary ice announcements via mass media; hazard drills, Faults in Venezuela and its Offshore Regions. United States family, school, community, business hazard plans; Geological Survey Open File Report 00-018. Reston, VA: USGS. Begin training for professionals (builders, archi- Larsen, M. C., et al. 2001. Venezuelan Debris Flow and Flash tects, health care, and legal) on hazard-specific issues. Flood Disaster of 1999 Studied. Eos Trans., American Geo- 5­10 years: Hazard message mainstreamed into schools; physical Union: 82: 572­573. develop training programs for the technical design Papageorgiou, A. S., and J. Kim. 1991. Study of the Propaga- units, the public and other professionals; establish tion and Amplification of Seismic Waves in Caracas Valley trained volunteer corps. with Reference to the 29 July 1967 Earthquake: SH Waves. 10-plus years: Make Caracas a regional center for hazard Bulletin of the Seismological Society of America 81(6): 2214­2233. education and research. Salcedo, D., and J. Ortas. 1992. Investigation of the Slide at the Southern Abutment Hill of Viaduct No. 1, Caracas-La Guaira Administration Highway, Venezuela. Proceedings of the Sixth International Symposium, Landslides, 189­198. 1­5 years: Establish constitutional and legal legitimacy Salcedo, D. 2001. Aspectos Socio-económicos y Socio-ambien- for disaster management; convene meetings of experts tales de los Flujos Catastróficos de Diciembre 1999 en el Estado in all related fields; establish a funding authority for Vargas y el Area Metropolitana de Caracas. Proceedings of III disaster management; mobilize grassroots commu- Panamerican Symposium on Landslides, Cartagena, Colombia, nity groups; train military and civil defense forces in 291­317. disaster response. Salcedo, D. 2000. The December 1999 Catastrophic Debris Flows 5­10 years: Establish a hazard-coordination authority at Vargas State and Caracas, Venezuela, Characteristics and with clear legal and organizational authority; set up Lessons Learned. Proceedings of the XVI Seminario Sociedad, hazard mitigation groups within each government Venezolana de Geotecnia, 28­175. Chapter 6 Reducing the Impacts of Floods through Early Warning and Preparedness: A Pilot Study for Kenya Hussein Gadain, Nicolas Bidault, Linda Stephen, Ben Watkins, Maxx Dilley, and Nancy Mutunga Kenya is a drought-prone country and is reasonably about the potential impacts of floods on people and prepared for drought emergencies. However, Kenya livelihoods. is also prone to very serious flood risks, especially in The Tana River basin (figure 6.1) is one of the biggest the lowlands in northeastern Kenya--particularly, river basins in Kenya, with an estimated river length of the Garissa, Ijara, and Tana River districts; the areas 1,000 kilometers (km) and a drainage area of approx- surrounding Lake Victoria (Nyanza Province); and imately 126,000 km2. It runs from the Aberdare and Nairobi. Despite warnings of strong El Niño-related Mount Kenya ranges of central Kenya through the arid weather anomalies in 1997, Kenya was unprepared for and semiarid lands in the eastern part of the country the floods that occurred in 1997 and 1998. The mag- and into the Indian Ocean through a fan-shaped Delta, nitude of flooding necessitated a massive relief oper- which covers approximately 1,300 km2. The Delta has ation. Ironically, floods are often more destructive in unique, fragile, and vulnerable ecological characteris- a low-rainfall environment where drainage infra- tics. The Tana basin supports the livelihoods of more structure, control, and coping strategies tend to be than four million people; most of them are pastoralists, under-developed. In the Tana River basin, El Niño farmers, and fisher folk. It is the only permanent river 1997­98 floods displaced thousands. Properties in this extremely dry region and constitutes a vital water were destroyed and the livelihoods of the riverine pop- resource for all sectors of the human population. At ulation were severely impacted. the middle and lower parts of the Tana basin lies a flat An impact analysis of the El Niño flood event clearly flood plain, 20 km wide. Irrigated agriculture is prac- demonstrates that Kenyans were inadequately prepared ticed along the river with pastoralists occupying the rest to cushion these adverse impacts. Up to that point, the of the basin. The lower reaches of the river pass through Government of Kenya (GoK) had neither a flood dis- semiarid land populated by pastoralists and riverine aster management policy nor an institutional frame- people. The delta also has a high tourism potential. work to monitor and manage flood disasters. This has Over the last 50 years, the Tana basin has under- had serious implications as floods have recurred, espe- gone major changes in land use and cover. The loss of cially in western Kenya, causing displacement and death forest in the headwaters to smallholder farming and on an annual basis. timber harvests has increased surface runoff and flood- The Government of Kenya is trying to strengthen its ing during the rainy seasons and sediment deposition disaster management capability, with an emphasis on in the storage reservoirs, drastically decreasing dry- preparedness and risk management. A flood contin- season flows. The construction of dams for hydropower gency planning activity for Kenya is already underway. generation in the late 1960s, 1970s, and 1980s resulted Data resources are reasonable, but data management in decreased outflows downstream during the dry period and analysis remain weak. While much has been achieved with high outflows during the high-flows period. This in modeling floods and predicting climate, less is known resulted in spill-water levels that severely compromised 165 166 Natural Disaster Hotspots Case Studies Figure 6.1. Location map of Tana River basin in Kenya with the river gauging stations dam safety. In 1961, the most dangerous flood recorded in the region occurred. The meteorological conditions associated with this extremely wet period were experi- enced over a wide area. In the middle and lower sec- tions of the Tana basin, however, the impacts of the flood were probably more severe, with the Tana River district considered as the most seriously affected Food Some of these activities were not fully implemented, and Agriculture Organization ([FAO] 1967). The impacts however. Only hydropower potential was fully investi- of the 1997­98 El Niño events were less than those of gated. To date, five major reservoirs have been built on the 1961 flood. This is mainly attributed to the dam the upper reaches of the Tana: Kindaruma (1968), Kam- construction process that took place in the 1970s and buru (1975), Gitaru (1978), Masinga (1981), and 1980s. Kiambere (1988). Dam construction has had a major Watershed planning and management in the Tana influence on the river's downstream flow and physical basin has traditionally been initiated and implemented characteristics, most notably through regulating water at the national level with little subnational-level input. flow and decreasing the frequency and magnitude of In 1967, the Government of Kenya initiated develop- flooding. The Masinga dam is the biggest storage reser- ment of Tana River water resources for hydropower voir for hydropower generation in Kenya, and also is development, flood control, and irrigation (FAO 1967). the main cause of downstream flooding during the high- Reducing the Impacts of Floods through Early Warning and Preparedness: A Pilot Study for Kenya 167 flow periods. When the dam spills, a huge amount of rain and satellite data; water is released. The spill waters from the dam cause · Linkage of the flood hazard maps with livelihoods flooding downstream in Garissa town and parts of the along the flood plains in the Tana River and other Ijara and Tana River districts. The question of "who infrastructure data to assess differential flood impacts; will be flooded downstream?" still remains, despite · Scientifically tested strategies to guide policy, con- efforts in recent years by the Kenya Power Generation tingency, and response planners in the flood early Company (KenGen) to issue warnings to the down- warning process in the Tana River basin; and stream settlements through radio, TV, and newspa- · Increased capacity among local and national insti- pers; through daily dissemination of information on tutions for sustainable planning and management dam levels; and through provision of data on spill levels of the Tana River waters. This objective is running to the disaster management unit in Kenya. in parallel to this study and being handled by the This study focuses on the middle and lower parts of U.S. Geological Survey and the Drought Monitoring the Tana River basin that are prone to recurrent floods: Center in Nairobi (DMCN). mainly the Tana River and Garissa districts. Using a semidistributed hydrologic model, this study aims to quantify the economic losses arising from flood events Study Area using flood mapping as an input to livelihoods and eco- nomic analysis. The outcomes of this study are flood The study area encompasses the Tana River and hazard maps linked with different livelihoods for use Garissa districts in Kenya (figure 6.2). in future flood-contingency planning and preparedness in the Tana River and Garissa districts. The case study Tana River District uses a semidistributed stream flow model and flood hazard maps to generate flood-level scenarios for the The Tana River District in the Coast Province is divided lower Tana River Basin, where emergency assistance is into seven administrative divisions with a total area of frequently required due to flood events. Flood impact 38,694 km2. The topography, drainage pattern, and soil risks to the population and livelihoods are assessed using contribute to a potentially large areal extent of flood- a livelihood zoning dataset that includes populated ing. The district is generally an undulating plain, places. The results are interpreted for use in contin- which slopes southeast with an altitude ranging between gency planning and preparedness. 0.0 and 200 meters above sea level. The main geo- graphical feature of this district is the Tana River. The large flood basin, which ranges from 2 to 40 km in Objectives width, provides fertile arable land and is the economic backbone of the district. The hinterland has seasonal The overall objective of this study is to contribute towards streams (lagas), which provide wet-season grazing areas sustainable flood preparedness for improved livelihoods and serve as sources of inlets for earth pans. Soils in protection against any future flood events. The target the Tana River district are divided into two groups: well- beneficiaries are local communities that live along the drained, sandy soils ranging in color from white to red, riverine and flood plains in the Tana River and Garissa and silty, clayey, poorly drained soils that are gray and districts. This includes farmers, pastoralists, and other black in color. The nomadic pastoralists--who keep stakeholders in the basin. The study aims to provide large herds of cattle, goats, and sheep--are the main better tools for contingency and response planning; inhabitants of the hinterland. In 1997, during a three- management of water, agriculture, and livestock; and month period, the district received over 1,200 mm of awareness creation. El Niño-related rainfall, which was triple its annual aver- Other specific objectives include the following: age. The resultant floods destroyed many houses, damaged infrastructure, swept away crops, and killed · Generation of flood hazard maps using a semidis- livestock. tributed hydrologic model and high-resolution ter- 168 Natural Disaster Hotspots Case Studies Figure 6.2. Location map of the Tana River and Garissa Districts with coverage of the Tana River basin in the Garissa District Garissa District percent of the population is below 20 years of age, and about 40 percent of the population resides within the The Garissa district is one of three districts that make environs of Garissa town. The district is predominately up the North Eastern Province of Kenya. The total area inhabited by Somali people who traditionally practice of the district is 43,931 km2, and it is divided into 14 livestock-keeping. administrativedivisions.MostofGarissaconsistsofnearly The climate of Garissa is semiarid, and the long-term level and featureless plains. The plains generally slope average rainfall is about 300 mm. Rainfall normally toward the southeast and south from about 300 m occurs in high concentrations and intensities, allow- above mean sea level to gradients of 0.5 to 0.7 meters ing temporal accumulation of excess moisture for per kilometer. Most of the district is drained by broad, drought-adapted vegetation and surface-water storage. shallow, and poorly defined streams that flow for only Since 1961, average annual rainfall has been above a few hours at a time, once or twice per year. The Tana the mean rainfall of the previous 30 years (UNICEF River crosses the district to its outlet in the Indian Ocean. 1998). Prior to the 1997/98 El Niño rains, the great- Soils in Garissa are divided into two groups: well-drained, est rainfall events occurred in 1961 and 1968, when sandy soils ranging in color from white to red, and an average of 920 mm was measured. Unusually silty, clayey, poorly drained soils that are gray and heavy rains occurred in 1997, totaling 1,027 mm; 925 black in color. mm occurred between October and December 1997. The total population of the district is 231,000, accord- This was a huge amount of rainfall for an area receiv- ing to 1999 census population projections. Almost 60 ing an annual average of 300 mm. Reducing the Impacts of Floods through Early Warning and Preparedness: A Pilot Study for Kenya 169 Impacts of the 1997 El Niño Floods flooding in Eastern Africa. The Tana River has been monitored in this process since 2001. During El Niño-related floods in the month of October The FEWS NET hydrological model was built to pro- 1997, the catchments of the Tana River received heavy vide a continuous simulation of stream flow, on a daily rainfall. As a result, the river levels rose within the first time step, for approximately 5,600 basins on the African 10 days of October. By the end of the first week of continent. The model is a physically based, catch- November 1997, the river had risen by more than 4 ment-scale hydrologic model (semidistributed hydro- meters to three times its average level (UNICEF 1998). logic model). It consists of a GIS-based module used The peak of the flood wave was in December when the for model input and data preparation, and the rainfall- river reached 6.65 meters and eventually covered the runoff simulation model. The rainfall-runoff model is measuring staff. The effect of this forced the riverine comprised of a module for soil water accounting that people to flee for their lives. The river maintained produces surface and subsurface runoff for each sub- flood levels for more than three months. In southern basin, an upland headwater basins routing module, and Garissa, most of the flooding was caused by direct a major river routing module. The model also gener- rainfall (figure 6.3). Virtually all the farms along the ates flood inundation maps for specified river depths river, pump- and flood-receding irrigation systems, and at different cross sections. To allow for evacuation and water supply systems were swept away during the relief services, the maps can easily be linked with peak floods. Traveling was extremely difficult; people livelihoods and infrastructure to provide hazard maps were marooned in camps for more than three months; that indicate, three days in advance, which areas and and health conditions were very poor, with high out- population centers would likely be inundated. breaks of rashes and cholera, which severely effected on children. Data and Methodology Hydrologic Model Data The U.S. Geological Survey Earth Resources Observa- The data used in this study to create the flood hazard tion Systems Data Center (USGS/EROS/EDC), through maps consist of approximately 150 topographic maps its support to the U. S. Agency for International Devel- at a 1:50,000 scale. The maps were scanned, digitized, opment (USAID), Famine Early Warning System Net- and geo-referenced in a GIS to form a mosaic of con- work (FEWS NET), is implementing activities related tours, spot heights, and rivers at 20-meter intervals to hydrological modeling and flood forecasting in the (figure 6.4). Using procedures developed by the Envi- Greater Horn of Africa (GHA). The activity is being ronmental Systems Research Institute (ESRI), finer ter- implemented by a number of international organiza- rain data were generated from the mosaic to form a tions and other partners in the region. Digital Elevation Model (DEM) containing a 100- FEWS NET, in cooperation with the USGS/EROS, meter spatial resolution (figure 6.4). The DEM was has undertaken efforts to enhance flood preparedness. further processed to remove topographically incorrect With hydrologic modeling techniques, it is possible to sinks to obtain a hydrologically corrected DEM for use better predict and react to such events. The FEWS in the creation of the flood hazard maps. NET Geo-Spatial Stream Flow Model (GeoSFM) (Artan Although the DEM was created from 1:50,000-scale et al. 2001) is a geo-spatial model based on the use of maps, the contour interval is coarse (10-m interval near satellite remote sensing, numerical weather forecast the coast). This caused outliers in the SFM analysis in fields, and geographic datasets describing the land sur- low-lying areas due to the flat terrain toward the ocean face. The model is currently operational in Eastern and in relation to the contour interval. Southern Africa. Since the severe flooding in Kenya Data on observed stream flow were available as gauge- that resulted from the El Niño rains in 1997/98, USGS height observations at Garissa through the Depart- and FEWS NET have begun to regularly monitor ment of Water in the Ministry of Water Resources covering 170 Natural Disaster Hotspots Case Studies Figure 6.3. Rainfall for selected stations during El Niño 1997­98 the period 1933­2001. These data (figure 6.5) show population centers and schools were digitized by the historical variability of river levels before and after WFP/VAM, and data on administrative boundaries the construction of the dams upstream of Garissa. From and population came from the latest census in 1999. the figure, it is clear that the river recorded high levels (These were updated during the 2002 elections by the during the 1961 and 1997/98 periods, when severe Central Bureau of Statistics [CBS].) flooding occurred in the middle and lower reaches of the river. Satellite Rainfall Estimates (RFE) for the same Stream Flow Modeling period were available through USGS. The U.S. National Oceanic and Atmospheric Administration Climate Stream flow modeling was done using the USGS GeoSFM Prediction Center (NOAA/CPC) developed these data to generate forecast stream flows at Garissa. The model for the USAID-funded FEWS NET project starting in was calibrated for the period from 1995 to 1998. The 1995 (Xie et al. 1998). Other spatial data on terrain, modeled stream flows mimicked the observed situation soils, land use, and land cover were available through to a reasonable level of accuracy, suggesting predictive USGS as well. skill by the model for use in the future for flood fore- The data used for impacts analysis were obtained casting. Model flow estimates were converted into equiv- from the World Food Programme Vulnerability Assess- alent river stages for mapping of inundated areas, since ment and Mapping unit (WFP/VAM). These data were inundation is more directly related to river stage than comprised of a GIS layer on livelihood zones developed to flow. The model efficiency criterion used to judge the by WFP/VAM, FEWS NET Kenya and the Arid Lands model performance was the coefficient of determina- Resources Management Project (ALRMP) in the Office tion, R2, proposed by Nash and Sutcliffe (1970). An of the President (OP) for the whole of Kenya. Data on R2 of 0.72 was obtained for the calibration period. Since Reducing the Impacts of Floods through Early Warning and Preparedness: A Pilot Study for Kenya 171 Figure 6.4. Data used in the creation of the DEM for flood-hazard mapping 172 Natural Disaster Hotspots Case Studies Figure 6.5. Variability of river stages at Garissa Town (1933­2001) with special focus on El Niño 1997­98 heights the river is controlled at a series of dams upstream in Flood Hazard Mapping Garissa, the modeled observed flows are more influ- Rapid progress has been made in the use of remote sens- enced by the dam releases than by the actual response ing techniques for flood inundation mapping. Appli- of the catchment to rainfall events. According to a cations of satellite imagery from optical sensors, including Ministry of Water report (UNICEF 1998), in 1997­98 SPOT (Blasco et al. 1992), AVHRR (Zhou et al. 2000), the gauge at Garissa washed away and a gauge height and LANDSAT (Mayer and Pearthree 2002), for flood of 6.65 meters was observed before the hydraulic struc- inundation mapping are well documented. The key lim- ture measuring the water levels failed. The model was itation of these remote-sensing techniques is their inabil- able to predict this high flood amplitude with reason- ity to estimate inundation for areas for which there is able accuracy (figure 6.6). Since there were no observed no existing satellite imagery of the river in a flooded gauge heights available during the flood for this state. Flood inundation maps based on topography period, it is difficult to predict what level the river are, therefore, useful for flood-warning purposes. One reached. For this purpose, we restricted our flood example is the USGS experience in flood forecasting mapping to an arbitrary height as is presented below. and mapping for Mozambique. In Mozambique, where Reducing the Impacts of Floods through Early Warning and Preparedness: A Pilot Study for Kenya 173 Figure 6.6. Stream flow modeling at Garissa (1995­1999) a high resolution DEM was available, a detailed flood- DEM, the DEM is used in the USGS GeoSFM interface forecasting system was implemented by the local water to generate hazard maps for various depth increments agency (ARA-Sul) and USGS. Fine-resolution DEM data in the channel. (90-meter resolution) were created from topographic The method maps flood inundation from a one- sheets by the local remote-sensing lab (CENACARTA) dimensional flow routing model onto a digital elevation and USGS. With the high-resolution DEM data, inun- model assuming hydraulic connectivity, while respect- dation maps were created and linked with model fore- ing the effects of impediments to flow within the area cast river levels for flood early-warning purposes. of inundation. The method makes use of the three- Mild, moderate, and severe flood scenarios were gen- dimensional topographic information contained in the erated and linked to data for settlements, schools, and DEM. The user identifies the level of water that must other livelihood zones. The result was comprehensive be attained in the river channel before increments of information for flood contingency planning, prepared- the adjacent land surface become inundated. ness, and response in Mozambique. The DEM is the main data input for terrain analy- The methodology adopted and implemented in this sis. The analysis begins with the establishment of the study was developed and implemented by USGS in raster equivalent of the river line, which is the imagi- Mozambique after the two devastating floods events in nary line connecting the lowest points along the river 2000 and 2001 (USGS 2003). Most of the analysis and channel. The direction of flow, characterized by the methodology presented in this study is taken from USGS eight-direction pour-point scheme (Jenson and (2003). After correcting topographic sinks in the Domingue 1988), is computed. Each DEM grid cell is 174 Natural Disaster Hotspots Case Studies assigned one of the eight compass directions based on stage grid. All cells with flow values of flow depth greater the assumption that water flows in the direction of than zero are identified as inundated, while those with steepest descent. River cells are identified by comput- values of zero are classified as dry cells. The cells along ing flow accumulation-- defined as a grid cell count the flow path of least resistance are also identified as of the upstream area--followed by an application of a minimally inundated, since the newly established minimum threshold of the contributing area to differ- hydraulic conductivity requires flow through these cells. entiate between river cells and non-river cells. The grid The procedure is done iteratively for multiple depths. cells along the flow path represent the minimum extent of inundation required for creating a raster represen- Livelihood Data Collection and Mapping tation of the river. Assigning an increment depth to all non-river cells within the analysis area creates an incre- The livelihood zoning of Kenya was undertaken through mental flow depth grid. The incremental flow depth interagency collaboration with the objective of col- grid is then added to the original DEM to create a flow lecting, in a cost-effective way, basic livelihood data at stage grid. the smallest, workable administrative unit: the sub-loca- An increment value that is more than the vertical res- tion (administrative unit six). The sub-location unit olution of the original DEM, that is, a 1-meter eleva- covers approximately three kilometers by three kilo- tion, results in a disruption of the hydraulic connectivity meters, in which only three to four settlements, on aver- over the DEM surface through the formation of flow age, can be found. sinks adjacent to the raised river cells. A method is Three teams were sent to cover a total of 30 dis- then applied that allows water to be distributed over tricts. In each visit to the districts, the district officials the land surface in a manner that approximates a nat- were met, including the District Agricultural Officer ural maintenance of hydraulic connectivity, taking into (DAO), the District Livestock Officer (DLO), and the consideration the newly incremented river stage rela- District Crops Officer. The officials were given a ques- tive to the ground surface elevation of all surrounding tionnaire and an A3 map of their district showing the cells. boundaries of the divisions, locations, and sublocations. After each river depth increment is applied, disrup- The questionnaire was designed to gather basic liveli- tions to hydraulic connectivity are identified by com- hood data, including the following: puting flow direction based on the new flow stage grid. · Main sources of income and food This follows the same basic process that was initially · Crop production per season used in computing the flow direction grid and the flow · Livestock and poultry ownership path of least topographic resistance from the original · Labor patterns DEM. The algorithm assigns a flow direction from a · Expenditure patterns given cell toward whichever one of the neighboring grid · Market(s) serving the livelihood zone cells has the lowest potential energy, whether or not it · Settlement and migration patterns of the is a river cell. A disruption to flow (flow sink) is iden- inhabitants of the livelihood zone tified whenever a grid cell cannot be assigned a flow · Society and ethnicity direction because all neighboring cells have a higher · Historical patterns of hunger flow stage or ground surface elevation. The disruption · Hazards and constraints to main economic activities is resolved by simulating the accumulation of water in the flow sink until its stage rises sufficiently for it to The officials were asked to identify the different liveli- discharge to one of the neighboring cells (Hutchinson hood zones in their respective districts and to answer 1989). A new flow path of least resistance is traced and the questionnaire for each livelihood. This was done in a raster representation is created. consultation with experienced field staff at sublocation The depth of flow on each grid cell is subsequently levels. Answers relied on available data, especially for determined by subtracting the original DEM elevations questions concerning crop production. When the data from the hydraulically connected, sink-filled flow were not available, officials were asked to use their "best Reducing the Impacts of Floods through Early Warning and Preparedness: A Pilot Study for Kenya 175 guess." This particularly applied to questions on pat- Using the methodology described above for the Tana terns of hunger, hazards, and constraints to main eco- River basin application, flow depths of 1 to 15 meters nomic activities. For each sublocation, officials were were applied in 1-meter increments, resulting in 15 also asked to identify on the map that was provided to inundation polygons associated with 15 different depths them the main livelihood system representative of that of flow at the channel centerline. These flood maps are sublocation. not associated with a specific flood; the user can iden- Two weeks were given for the completion of the exer- tify the bank levels and generate depth-specific maps. cise, and the data were sent back to the WFP Kenya In the Tana River case, flood inundation maps associ- office, where they were entered into a database. Map- ated with the river depths for the 1961 and El Niño ping outputs were produced with ArcView GIS soft- floods were generated. Considering the 1997­98 ware. levels as the highest recorded and comparing this level A total of 78 different and highly specific liveli- to the GeoSFM results, hazard maps from the GeoSFM hoods were identified in the 30 districts visited. Data results were generated and considered to be the worst- collection for mapping livelihoods in the remaining 40 case scenario for the past 35 years after the construc- districts of Kenya is currently underway. The liveli- tion of the dams upstream in Garissa. A more moderate hood zoning exercise serves national and district plan- flood event, taking into consideration the presence of ning in a number of ways. It helps to identify a new the dams, was also generated. unit of analysis--the livelihood zone--at the sub-loca- tion level; it improves the ability to analyze and assess Historical Floods the impact of floods, droughts, and other hazards on the population living in the livelihood zone; and it allows To model the impacts on the Tana River and Garissa dis- more meaningful analysis of price-related, agricul- tricts, we intersected the basin boundaries with the tural, and socioeconomic data. two district boundaries. This resulted in a smaller por- tion of Garissa district contained within the Tana River drainage boundaries (figure 6.2). As explained in the Results background section, the floods in the Tana River basin are the results of the rainfall runoff from the Mount In this analysis, two severe floods were investigated: the Kenya area. Before 1968, when the first dam was con- 1961 flood and the El Niño of 1997­98. River depths structed at Kindaruma, the absence of dams downstream associated with the 1961 flood were the highest river from Mount Kenya resulted in severe flooding in the depths recorded over a 70-year period, with a maxi- Tana River basin. Using the stream-flow model, the 1961 mum depth of 7.0 meters. The deepest El Niño 1997­98 flood depths were mapped, as shown in figure 6.7. In recorded river depth was 6.65 meters. This occurred order to give an idea of the impact of these severe floods, on the night of December 3, 1997. According to the locations of settlements from the Central Bureau of Water Department, the river level rose rapidly to three Statistics (CBS) 1999 census, schools, and roads--which times its long-term average (figure 6.5), forcing people have been recently digitized from the 1:50,000 topo- to flee for their lives. A second flood came in January graphic maps--were overlaid onto the flood map as 1998. The river maintained flood levels for more than shown in figures 6.7, 6.8, and 6.9. It is clear that the three months. Although rainfall levels associated with flooded areas are more contained in the Tana River dis- the 1997­98 El Niño were much higher than was the trict, and that the floods due to Tana River flows had case in 1961, lower river depths were recorded. This is less impact in the Garissa district. The floods of the mainly attributed to the Masinga dam, which controls Ewaso Nyiro River, which passes through most of Garissa river flow upstream of Garissa town. Massive rainfall and drains into the Indian Ocean, had more of an impact associated with the 1997­98 El Niño resulted in severe in Garissa than did the Tana River (UNICEF report floods in many parts of the country, including the Tana 1998). The low elevation in the middle and lower River basin. parts of the basin allows the floods to spread widely, 176 Natural Disaster Hotspots Case Studies Figure 6.7. Flood hazard map for the 1961 flood (the case of a severe flood before construction of the dams) reaching a maximum extent of more than 10 km reduced compared to that of the 1961 historical across at its maximum width in the Tana delta. The wide floods. The presence of the dams managed to mini- extent of these floods resulted in a large number of set- mize the extent of El Niño floods in the basin in 1997­98, tlements being entirely underwater, as well as main roads avoiding an even more serious disaster. However, flood- being completely submerged, rendering access by road ing in the Tana River has become recurrent, with a return to these areas extremely difficult for emergency-assis- period of 1 to 2 years, constituting shocks that hamper tance personnel. sustainable development along the river. Better control El Niño-related flood levels for 1997­98 were also of the water spilling over at the dam to minimize the mapped and overlaid with the data for settlements, floods would help plan more sustainable use of the land, schools, and main roads, as was done for the 1961 floods thus decreasing vulnerability for the groups living in (figure 6.8). These floods were quite severe, and a large the area. number of settlements, schools, and portions of main roads were once again underwater. According to the Flood Scenarios Kenya Meteorological Department, more rainfall was received during the 1997­98 El Niño (figure 6.3) than Areas and Population Affected in 1961. Comparing the two maps, we can observe As previously mentioned, flooding is a regular occur- that the overall flooded area of El Niño was significantly rence in the Tana River basin, as almost every year inhab- Reducing the Impacts of Floods through Early Warning and Preparedness: A Pilot Study for Kenya 177 Figure 6.8. Flood-hazard map for the El Niño 1997/98 flood (a worst-case scenario after the construction of the dams) itants of the basin have to prepare for losses to prop- assumed the total population in a sub-location is con- erty, livestock, and crops. However, the amplitude of tained in the villages located in that sub-location. One the floods is not the same from year to year. Floods of drawback in this methodology is that the data on vil- the magnitude experienced during the 1997­98 El Niño lages and schools were digitized from old maps. The are, fortunately, a rare occurrence and unlikely to happen school locations, therefore, are different from the vil- more than once every 35 years, given the cascade of lage locations. In reality, these schools are associated dams upstream. On the contrary, moderate floods, char- with certain villages. We subdivided the population by acterized by a maximum water level of approximately the total number of villages in the sub-location and com- 5 meters on the river gauge at Garissa, could strike at puted the average population per settlement (table 6.1). a frequency of once every 2 to 5 years. Figures 6.8 and These figures are only for riverine areas; in contrast, 6.9 show the effects of moderate and severe floods. Considering the 1997­98 El Niño as the worst-case scenario, the map in figure 6.8 shows the flooded area Table 6.1: Flood scenarios for a worst case and a as modeled by the SFM and the impact on the popula- moderate case tions. For each scenario, we computed the total flooded Flood type Villages People Area Km2 area. Villages that are underwater and the total number Severe 73 70,000 5,377 of people affected by the floods were identified. We Moderate 51 47,000 4,612 178 Natural Disaster Hotspots Case Studies Figure 6.9. (a + b) Livelihood zones overlaid on El Niño floods case the 1997­98 El Niño floods occurred due to direct and access to food other than through food aid relief brought high rainfall on non-riverine areas. by helicopters. Pockets of population in Dujis and In a moderate flood, the total riverine area that would Modogashe (Garissa district) were marooned for more be flooded was found to be 4,600 km2. Out of a total than two months as result of roads being cut off. number of 148 villages located in the flood plain, 51 For most communities, access to markets was villages would be under water, resulting in approxi- highly reduced, resulting in decreased food access. This mately 47,000 people affected by these kinds of circumstance was aggravated by lower purchasing power floods. In a severe flood, the impact would obviously due to loss of income and conjugated high commodity be larger, with an approximate total flooded area of prices resulting from increased logistical costs. 5,400 km2 (17 percent higher than a moderate flood); Most schools were damaged, resulting in high dropout 73 villages would be underwater and a total of 71,063 rates (65 percent of the boys and 50 percent of the girls people (50 percent higher compared to a moderate flood) in primary school in the Garissa district). A large number would be affected. of water sources were washed away, submerged, or rendered unsafe for drinking through contamination, General Impacts of the 1997­98 El Niño Floods mostly from the destruction of latrines. This was espe- The 1997­98 El Niño floods drastically affected liveli- cially the case in the permanent settlements along the hoods in the area, especially along the river in both the river. Garissa and Tana River districts. The floods caused major During the 1997­98 floods, there was a change in infrastructure damage, and all roads were severely dam- the overall physical environment, resulting in an increase aged. Most roads were gullied and silted, becoming in humidity to above 80 percent of saturation in both impassable. Some roads became part of the riverbed districts, with a peak of 97 percent humidity in Decem- and some bridges were washed away. This resulted in ber 1997. This resulted in a drastic increase in human some communities being completely isolated, with no and livestock diseases, an increase in vectors (mosqui- Reducing the Impacts of Floods through Early Warning and Preparedness: A Pilot Study for Kenya 179 toes, tsetse flies), and a severe outbreak of Rift Valley ment, destitution, and increased insecurity; reports of Fever. There were massive outbreaks of malaria, diar- these maladies have been especially prevalent in the rheal disease, skin disease, ARI, worms, and cholera. Tana River district. Malnutrition rates of children under five years of age reached incredibly high levels, with reported overall Impacts of Floods on Livelihoods rates of 56 percent and 48 percent, respectively, in the The impacts of floods on populations differ depending Garissa and Tana River districts (using mid-upper-arm on the members' livelihoods and wealth-group affilia- circumference [MUAC] measured at <13.5cm). tions. In order to better identify the impacts of the floods Disease had a considerable economic impact on live- on the different livelihoods in the Tana River and Garissa stock exports. Saudi Arabia banned the import of live- districts, we mapped the main livelihoods using the data stock from the east Africa region. According to a flood described above at the sublocation levels in both dis- emergency assessment report in Kenya prepared by tricts. The result is shown in figure 6.10. These two UNICEF, GoK, and UNDP (1998), immediately after districts are characterized by several livelihoods. On the the 1997­98 El Niño floods, shoats died in masses in Garissa district side, three livelihood groupings can be the Garissa district, which accounted for an unbeliev- distinguished: fishing and subsistence cropping, able 70 to 95 percent of dead shoats reported. This urban (around Garissa town), and pastoralist. On the predominantly affected the pastoralists' communities Tana River district side, three livelihoods zones can in the basin. Figure 6.11.a shows the number of live- also be found: dry riverine zone (in the north and cov- stock before and after the floods. The loss of livestock ering a great length of the river), agro-pastoralists, and in the Tana River district was reported to equal 50 to Tana Delta zone (in the south). The floods affected all 90 percent of all animals (depending on particular areas), of these zones. with an average figure of over 70 percent for the whole Among the different livelihood groups in both dis- district. Larger animals did not die in as large numbers tricts, the ones most exposed to flooding are pastoral- as small animals. Some 10­20 percent of camels and ists, agro-pastoralists, dry riverine, and the Tana Delta 0­5 percent of cattle were reported to have died as a livelihood system. For that reason, we have focused our result of the floods in the Garissa district. However, in analysis on these groups. The analysis begins with a some areas of the Tana River district, the reported description of the baseline livelihood. The key charac- cattle loss was as high as 70 percent. The ground became teristics considered are cash income, expenditure sources, saturated with water, which meant that large animals income diversification, food consumption sources, live- had to stand or lie in water, resulting in increased live- stock ownership and its contribution to cash income stock mortality from foot rot or pneumonia. Moreover, and food consumption, and market dependency. A the stress on livestock resulted in mass abortion rates, wealth breakdown was applied for agro-pastoralists, reportedly affecting 80 to 100 percent of pregnant ani- pastoralists, and dry riverine livelihoods to highlight mals of all species in both districts. This, together with disparities within these groups. The analysis aims at the high morbidity, resulted in low milk production. building a picture of the impacts of floods on the dif- Crops were destroyed en masse. Up to 1,200 hectares ferent elements that constitute the livelihoods of the of bananas, tomatoes, and vegetables were reportedly exposed populations taking into account the immedi- washed away in the Garissa district, while 100 percent ate, short-term and long-term impacts. of the bananas, mangoes, rice, maize, and pulses were The livelihood zones directly on the river (dry river- destroyed in the Tana River district. The destruction of ine zone, Tana Delta zone, pastoralists, and agro-pas- crops resulted in a drastic increase in commodity prices, toralists, as they are mostly located in the hinterland as indicated in figure 6.11.b. The prices of most com- except in the south part of the basin) are likely to be modities doubled, making these essential commodi- affected through the direct destruction of their proper- ties out of reach for most of the population. ties (houses, crop fields, pumps, and so on). The pop- The overall impact on the livelihoods has been impov- ulation in the urban area (especially at Garissa town) is erishment of the population, hunger stress, displace- likely to be mostly affected through the indirect impacts 180 Natural Disaster Hotspots Case Studies Figure 6.10. Livelihood zones overlaid on El Niño flood cases Reducing the Impacts of Floods through Early Warning and Preparedness: A Pilot Study for Kenya 181 Figure 6.11. Impacts of floods on market prices and livestock: (a) commodity prices (b) livestock losses Source: UNICEF 1998 182 Natural Disaster Hotspots Case Studies of the floods, such as through an interruption of access As described earlier, during the El Niño floods close to markets and concomitant loss of income, though to 90 percent of shoats died in the Garissa and Tana some may also encounter loss of properties. However, River districts, resulting in a complete collapse of the people in urban areas are more likely to have resources main source of income. Shoats represent close to 15 to cope and, therefore, are less at risk of a complete percent of the total income for the agro-pastoralist, dry collapse of their livelihoods. The population in fisheries riverine, and pastoralist groups. In addition, in both the and subsistence cropping may find benefits in the floods short and long term, mortality and morbidity among thanks to the likely increase in fish production, though larger animals increased drastically as a result of dis- they are also likely to see their subsistence cropping eases such as foot rot and pneumonia. On top of direct resources being affected. loss of animals, the decrease in livestock marketability The different characteristics of the livelihoods are has also hurt income. Due to the fear of Rift Valley Fever, summarized in table 6.2. Unless specified, the data rep- animals were not bought on the markets, and income resent an average. from animal sales was lost. The impact on livestock As we can see in table 6.2, the main income source has hurt equally the "very poor," "poor," and "middle of the pastoralists, agro-pastoralists, and the people from income" groups, who have seen their income from the dry riverine communities comes from livestock. This livestock reduced to zero. Most of the "very poor" and is also a major source of food consumption. Table 6.3 "poor" have moved to the destitute category, while summarizes the percentage of livestock contribution only the "middle income" group, featuring larger cattle to cash income and food consumption. sizes, may have avoided destitution. Table 6.2. Characteristics of the different livelihood zones analyzed Characteristics Agro-pastoralists Dry riverine zone Pastoralists Tana delta zone Cash income Livestock prod: 40%, Livestock prod: 22%, Livestock prod: 68%, Food crop prod: 40% sources Food crop prod: 10% Firewood collection: 12%, remittances and gifts: 10%, (Mangoes: 37%), (Maize: 30%), Food crop prod: 10% Firewood collection: 5% Formal wage labor: 15%, Poultry prod: 10% (Maize: 30%) Livestock prod: 10% Hunting and gathering: 10% Expenditure of Maize: 50% Maize: 50% Milk prod: 40% Maize: 47% poor HH Rice: 10% Rice: 10% Meat: 20% Pulses: 10% Milk: 10% Milk: 10% Vegetables: 10% Food consumption Maize: 40% Maize: 50% NA Maize: 41% Maize sources for Own farm: 40% Own farm: 30% Market purchase: 60%, Own farm: 60% consumption Market purchase: 40%, Market purchase: 10%, Gifts and food aid: 40% Market purchase: 20%, Gifts and food aid: 40% Gifts and food aid: 60% Gifts and food aid: 20% Livestock ownership: Poor Middle Poor Middle Poor Middle Cattle 20­50 70­100 0­2 2­5 5­20 30­50 2 Shoats 25­40 75­125 5­10 10­20 15­60 70­120 5 Source: Livelihoods data base 2004 (FEWS NET Kenya, WFP/VAM and GoK) HH = Household Table 6.3. Percentage of livestock contribution to cash income and food consumption of the livelihood zones analyzed Percent Agro-pastoralists Dry riverine zone Pastoralists Tana delta zone Cash income Cattle 15 33 70 20 Shoats 40 65 20 70 Camels 15 - - - Food Cattle 15 15 40 40 consumption Shoats 60 80 40 53 Camels 10 - - - Reducing the Impacts of Floods through Early Warning and Preparedness: A Pilot Study for Kenya 183 The loss of livestock also had an important impact Conclusions on food consumption. Food intake was reduced due to the direct loss of animals, from which people were get- The above discussion has demonstrated that some sec- ting meat and milk. Secondly, the loss of income trans- tors of the population will be more affected than others lated into a loss of purchasing power that, when combined during flood events in the Tana River basin. This find- with higher commodity prices, put basic commodities ing is critical for contingency planning, in which the out of reach for these communities. As a result, their objective is to anticipate and therefore better manage a access to food was drastically reduced. potential crisis. The central components of a contin- Moreover, these populations depend highly on maize gency plan are to understand who (who is affected?), as their major source of food, either from their own farm where (where are they located?), what (what types and production (40 percent agro-pastoralists, 30 percent modalities of emergency assistance are needed?), how dry riverine, 60 percent delta Tana) or from market pur- (how, if roads are impassable, will food be delivered and chase (60 percent pastoralists). The direct destruction people evacuated?), and how much (how much are the of crops reduced availability of maize in the area, with associated costs for implementing an emergency oper- the consequence of very high prices for available maize ation?). on the market. In all cases, maize has become unavail- Drawing from our analysis here and past experi- able for these populations. As a result most people in ence, under the worst-case scenario, pastoralists and these livelihood zones require food aid to survive. It is dry riverine communities are expected to experience interesting to note that all the livelihoods in the Tana the worst losses. Therefore, a response directed toward River basin seem to be heavily dependent on food aid. these groups of people, and the pastoralists in particu- This is the result of consecutive years of drought in the lar, would be advisable. Assistance should take the form area. Up to 60 percent of the maize consumption for of free food distribution and income generating activ- the dry riverine people already comes from food aid dis- ities, as the analysis has shown that for all of the tribution. However, due to the difficulties in shipping groups, income, daily food consumption, and nutrition food into the areas and the increased competition for are tied to livestock and crop production, both of which food aid, the threat of starvation for these populations may completely collapse in any flood scenario. Fur- is drastically increased. thermore, assessments during the 1997­98 El Niño The population less likely to see a complete col- floods showed that relief food enabled pastoralists to lapse of livelihoods is that of the Tana Delta zone, where save their remaining livestock and to start rebuilding 40 percent of the income comes from crop production, herds and livelihoods. For planning purposes, we know especially from mangoes (37 percent), tomatoes (20 from our hazard maps that food assistance in the short percent) and bananas (10 percent). Any impact on these term and income regenerating activities in the long term products will negatively impact this group. Through would be required for up to 70,000 persons. In the mod- more diversified income sources, however, with only erate case scenario, the population in need would be 10 percent linked to livestock production, these 47,000 people. This provides core data for calculating people are more likely to be able to cope. For example, costs and the volume of food commodities required. 15 percent of their income comes from formal wage On the logistical side, a workable solution to pro- labor. Yet even this income may not be guaranteed, as, viding timely assistance in both worst-case and mod- after a flood shock, there is likely to be increased com- erate-case scenarios would be to deliver commodities petition for this income source. Similarly, this popula- and evacuate by air, although accessibility from the tion is also dependent on maize as a source of food airstrip may not be guaranteed. During December and consumption, with 20 percent coming from food aid; January of 1997­98, WFP was able to use helicopters so they are likely to require immediate and short-term to airlift food to over 60 sites within the Garissa dis- food assistance to survive. trict. Due to the limited capacity of the helicopter (payload of 2.5 tons) and the large number of needy sites, the amount delivered per site was very limited. 184 Natural Disaster Hotspots Case Studies Even so, the relief food probably averted total famine. FAO (Food and Agriculture Organization). 1967. Survey of the One way of alleviating the logistical burden would Irrigation Potential of the Lower Tana River Basin. Prepared be to gather the affected people into displacement camps by ACRES International Limited, Canada; International Land located in easily accessible areas. This requires moving Development Consultants N. V., the Netherlands; and the Gov- large numbers of people at once, which was not possi- ernment of Kenya. ble during the 1997­98 El Niño floods due to the status Hutchinson, M. F. 1989. A New Procedure for Girding Eleva- of the roads. Affected people in the Garissa area tion and Stream Line Data with Automatic Removal of Spu- poured into town, increasing the stress on the local pop- rious Pits. Journal of Hydrology 106: 211­232. ulation. Jenson, S. K., and J. O. Domingue. 1988. Extracting Topographic Given the analysis presented here, it is also possible Structure from Digital Elevation Data for Geographic Infor- to estimate logistical costs should air transport be the mation System Analysis. Photogrammetric Engineering and only option. As an example, during the 1997­98 Remote Sensing 54(11): 1593-1600. emergency (which serves as the worst-case scenario Mayer, L., and P. A. Pearthree. 2002. A Method for Mapping here), the operational cost to the World Food Programme Flood Inundation in Southwestern Arizona Using Landsat TM was US$1,100 per ton for each airdrop. A total of Data. In: Ancient Floods, Modern Hazards: Principles and Appli- 7,414.23 MT was distributed at a cost of US$4,117,734. cations of Paleoflood Hydrology, vol. 5, P. K. House, et al. (eds.), This food reached up to 641,451 beneficiaries in Garissa, Washington, D.C.: AGU, 61­75. Wajir, Mandera, Isiolo, Marsabit, Moyale, and the Tana Nash, J. E., and J. Sutcliffe. 1970. River Flow Forecasting through River. Conceptual Models. Part 1: A Discussion of Principles. Jour- For policy makers, planners, and the humanitarian nal of Hydrology 10: 282­290. aid sector as a whole, the value of this analysis, which UNICEF (United Nations Children's Fund), UNDP (United has linked flood forecasting and livelihood zone data, Nations Development Programme), and Government of Kenya. is a forecast of the scale of a hazard, estimated costs, 1998. A Flood Assessment Report. Prepared after the El Niño and, most important, an identification of vulnerable 1997/98 Floods. groups. This information will improve emergency pre- USGS (U.S. Geological Survey). 2003. U.S. Geological Survey paredness, response, and assistance. Technical Support to the Mozambique Integrated Informa- tion Network for Decision-Making (MIND). Project Comple- tion Report, submitted to the U.S. Agency for International References Development. Prepared by K. Asante and J. Verdin. Xie, P., and P. A. Arkin. 1998. Global Precipitation Estimates Artan, G., J. Verdin, and K. Asante. 2001. A Wide-Area Flood from Satellite Observed Outgoing Long Wave Radiation. Jour- Risk Monitoring Model. Proceedings of the Fifth International nal of Climate 11: 137­164. Workshop on Application of Remote Sensing in Hydrology, Zhou, C. H., et al. 2000. Flood Monitoring Using Multi-tem- Montpellier, France. poral AVHRR and RADARSAT Imagery. Photogrammetric Engi- Blasco, F., M. F. Bellen, and M. U. Chaudhury. 1992. Estimating neering and Remote Sensing 66(5): 633­638. the Extent of Floods in Bangladesh Using SPOT Data. Remote Sensing of Environment 39: 167­178. THE WORLD BANK 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 0-8213-6332-8