97447 Better Risk Information for Smarter Investments 177°30'E 178°0'E River Network 17°30'S 18°0'S CATASTROPHE RISK ASSESSMENT METHODOLOGY 177°30'E 178°0'E © 2013 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 All rights reserved This publication 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 judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this 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. Designer: Miki Fernandez Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) Better Risk Information for Smarter Investments Risk Assessment - Summary Report CATASTROPHE RISK ASSESSMENT METHODOLOGY 3 Acknowledgements T he Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) is a joint initiative between the World Bank, the Secretariat of the Pacific Community through its Applied Geoscience & Technology Division (SPC/SOPAC), and the Asian Development Bank, with financial support from the Government of Japan, the Global Facility for Disaster Reduction and Recovery (GFDRR), the European Union (ACP-EU) and with technical inputs from GNS Science, Geoscience Australia, and AIR Worldwide. This report has been prepared by a World Bank team led by Iain Shuker and Olivier Mahul, comprising of Michael Bonte-Grapentin, Emilia Battaglini, Abigail Baca, Sandra Schuster, Cynthia Dharmajaya, and Sevara Atamuratova; an SPC/SOPAC team led by Mosese Sikivou, comprising of Litea Biukoto and Samantha Cook, and an ADB team led by Edy Brotoisworo and Jay Roop. The technical materials used in this report were produced by a team from AIR Worldwide led by Paolo Bazzurro, comprising Jaesung Park, Ivan Gomez, Bishwa Pandey, Daniel Duggan, Brent Poliquin and Yufang Rong. A team from GNS Science New Zealand led by Phil Glassey, provided ground truthing to the data used in the analysis. A team from Geoscience Australia, comprising of John Schneider and Alanna Simpson, provided technical support and advice throughout the project. The report greatly benefited from data, information and other invaluable contributions made by the Pacific Island Countries, development partners, donor partners and private sector partners. The team greatly appreciates the support and guidance received from Charles Feinstein, Ferid Belhaj, John Roome, Loic Chiquier, Francis Ghesquiere, and Abhas Jha. Table of Contents Acknowledgments .....................................................................................................................3 Executive Summary.....................................................................................................................7 Abbreviations and Acronyms......................................................................................................9 Outline, Objectives and Outputs...............................................................................................10 Catastrophe Risk Modeling.......................................................................................................12 1. Exposure Information...........................................................................................................12 1.1 Population......................................................................................................................12 1.2 Buildings........................................................................................................................14 a. Locations......................................................................................................................14 b. Field Surveys.................................................................................................................16 c. Occupancy Type and Construction Characteristics.........................................................18 d. Replacement Cost........................................................................................................20 1.3 Infrastructure..................................................................................................................22 1.4 Crops.............................................................................................................................24 1.5 Replacement Costs by Country.......................................................................................30 2. Hazard Assessment...............................................................................................................31 2.1 Tropical Cyclone Event Generation..................................................................................32 2.2 Tropical Cyclone Intensity Calculation.............................................................................35 a. Induced Winds.............................................................................................................35 b. Rainfall-Induced Inland Flood........................................................................................38 c. Coastal Flood................................................................................................................39 2.3 Earthquake Event Generation.........................................................................................39 2.4 Earthquake Intensity Calculation.....................................................................................47 a. Ground Shaking...........................................................................................................47 b. Tsunami Waves.............................................................................................................48 2.5 Ancillary GIS Data...........................................................................................................50 3. Damage Estimation..............................................................................................................53 3.1 Consequence Database..................................................................................................53 a. Data Sources................................................................................................................53 b. Explanation of Data Fields............................................................................................55 c. Economic Loss Trending................................................................................................56 d. Database Statistics........................................................................................................56 6 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) 3.2 Damage Functions..........................................................................................................59 a. Buildings......................................................................................................................60 b. Emergency Losses.........................................................................................................62 c. Infrastructure Assets.....................................................................................................63 d. Crops...........................................................................................................................63 e. Fatalities and Injuries....................................................................................................63 4. Country Catastrophe Risk Profiles.........................................................................................67 5. The Pacific Risk Information System (PacRIS).........................................................................74 6. Applications.........................................................................................................................75 6.1 Post Disaster Response Capacity and Disaster Risk Financing...........................................76 6.2 Disaster Risk Reduction and Urban/Infrastructure Spatial Planning...................................76 6.3 Post-Disaster Assistance and Assessment........................................................................76 6.4 Early Warning Systems and DRR Communication............................................................77 6.5 Reporting and Monitoring Agencies...............................................................................77 7. References............................................................................................................................78 Annex A: Field Survey Locations................................................................................................79 Annex B: Building Locations (Level 4 Methodology)..................................................................81 Annex C: Construction Type.....................................................................................................85 Annex D: Infrastructure Exposure Database...............................................................................87 Annex E: Example of Consequence Database............................................................................89 Annex F: Country Risk Profiles..................................................................................................91 CATASTROPHE RISK ASSESSMENT METHODOLOGY 7 Executive Summary T he Pacific Catastrophe Risk Financing and Insurance Initiative (PCRAFI), initiated upon the request of the Pacific Island Countries (PICs) in 2006, is an innovative program that builds on the principle of regional coordination and provides PICs with state-of-the-art disaster risk information and tools for enhanced disaster risk management and improved financial resil- ience against natural hazards and climate change. This initiative has been implemented in close collaborations between the World Bank, the Secretariat of the Pacific Community through its Applied Geoscience & Technology Division (SPC/SOPAC), and the Asian Development Bank, with financial support from the Government of Japan, the Global Facility for Disaster Reduc- tion and Recovery (GFDRR), the European Union (ACP-EU) and with technical inputs from GNS Science, Geoscience Australia, and AIR Worldwide. The following 15 PICs are involved in the program: Cook Islands (New Zealand), Federated States of Micronesia, Republic of Fiji, Republic of Kiribati, Republic of Nauru, Niue (New Zealand), Republic of Palau, The Independent State of Papua New Guinea, Republic of the Marshall Islands, Samoa, Solomon Islands, Democratic Republic of Timor-Leste, Kingdom of Tonga, Tuvalu, and Republic of Vanuatu. PCRAFI established the Pacific Risk Information System (PacRIS), one of the largest col- lections of geospatial information for the PICs. PacRIS contains detailed, country-specific in- formation on assets, population, hazards, and risks. The exposure database leverages remote sensing analyses, field visits, and country specific datasets to characterize buildings (residen- tial, commercial, and industrial), major infrastructure (such as roads, bridges, airports, ports, and utility assets), major crops, and population. More than 500,000 buildings were digitized from very-high-resolution satellite images, representing 15 percent (or 36 percent without Papua New Guinea) of the estimated total number of buildings in the PICs. About 80,000 buildings and major infrastructure were physically inspected to calibrate satellite based data. In addition, about 3 million buildings and other assets, mostly in rural areas, were inferred from satellite imagery. PacRIS includes the most comprehensive regional historical hazard cat- alogue (115,000 earthquake and 2,500 tropical cyclone events) and historical loss database for major disasters, as well as state-of-the art country-specific hazard maps for earthquakes (ground shaking) and tropical cyclones (wind). PacRIS contains risk maps showing the geo- graphic distribution of potential losses for each PIC as well as other visualization products of the risk assessments, which can be accessed, with appropriate authorization, through an open-source web-based platform. Country risk profiles were developed for each of the 15 PICs from the data contained in PacRIS. They can be used to draw attention to not only the risk that is faced by each country but also to give an indication of the frequency of these hazardous events and their associated economic and fiscal losses. Under this analysis, it was established that the average annual loss caused by natural hazards across all 15 PICs is estimated at USD 284 million, or 1.7% of the regional GDP. Vanuatu, Niue and Tonga experience the largest Average Annual Losses (AAL) from natural disasters in the region equivalent to 6.6%, 5.8% and 4.4% of their national GDP, respectively. This places them among those countries that experience the highest levels of AAL globally. There is a 2% chance that the Pacific region will experience disaster losses in excess of USD 1.3 billion from tropical cyclones and earthquakes in a given year. 8 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) PacRIS is also the platform for a series of applications to help the PICs and their partners better understand and assess countries exposure to natural disasters and provide unique and relevant information for their physical and financial management of natural disasters. (Figure 1). FIGURE 1. Pacific Risk Information System and Associated Applications The following applications are currently under development. They will strengthen PacRIS and demonstrate the use of the information: Rapid Post Disaster Estimation. The PCRAFI models can provide the basis for rapid post-disaster damage esti- mation and therefore have the potential to offer disaster managers and first responders with tools and information to quickly gain an overview following a disaster on areas and population affected. Urban Planning and Infrastructure Design. Applications for the mainstreaming of risk information into the urban and infrastructure planning aim to ensure that disaster risk and climate change information form an integral part of the urban and infrastructure planning process. Climate Change Adaptation. Under the climate change adaptation segment PacRIS is liaising with the Pacific Australian Climate Change Science and Adaptation Program to incorporate future tropical cyclone risk to critical assets into the PacRIS datasets. Disaster Risk Financing. The Disaster Risk Financing segment is designed to assist the PICs in increasing their financial resilience against natural disasters and improving their capacity to meet post-disaster funding needs with- out compromising their long-term fiscal balance. Rapid access to cash in the aftermath of a disaster is essential for the governments to ensure timely and effective post-disaster response. This application also tests the viability of market-based catastrophe risk insurance solutions for the governments. This report describes the development of the Pacific Risk Information System, from the collection and process- ing of the information to the variety of applications for disaster risk management and climate change adaptation. CATASTROPHE RISK ASSESSMENT METHODOLOGY 9 Abbreviations and Acronyms AAL Average Annual Loss CCA Climate Change Adaptation CK Cook Islands (New Zealand) DF Damage Function DR Damage Ratio DRM Disaster Risk Management FJ Republic of Fiji FM Federated States of Micronesia GDP Gross Domestic Product GFDRR Global Facility for Disaster Reduction and Recovery GIS Geographical Information System IBTrACS International Best Tracks Archive for Climate Stewardship KI Republic of Kiribati LULC Land Use / Land Cover MH Republic of the Marshall Islands NR Republic of Nauru NU Niue (New Zealand) OpenDRI Open Data for Resilience Initiative PacRIS Pacific Risk Information System PCRAFI Pacific Catastrophe Risk Assessment and Financing Initiative PG The Independent State of Papua New Guinea PGA Peak Ground Acceleration PICs Pacific Island Countries PW Republic of Palau SB Solomon Islands SOPAC Applied Geoscience and Technology Division, SPC SPC Secretariat of the Pacific Community TL Democratic Republic of Timor-Leste TO Kingdom of Tonga TV Tuvalu VU Republic of Vanuatu WS Samoa 10 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) Outline, Objectives and Outputs T This report he Pacific Region is one of the most natural disaster prone regions on earth. focuses on the The Pacific Island Countries (PICs) are highly exposed to the adverse effects of climate development of the change and natural hazards, which can result in disasters affecting their entire eco- country catastrophe nomic, human, and physical environment and impact their long-term development agenda. risk profiles, The average annual direct losses caused by natural disasters are estimated at US$284 million. the information Since 1950 natural disasters have affected approximately 9.2 million people in the Pacific Re- collected, how it gion, causing 9,811 reported deaths. This has cost the PICs around US$3.2 billion (in nominal was catalogued terms) in associated damage costs. and processed, The primary objective of PCRAFI was to develop risk profiles for earthquakes and now being (both ground shaking and tsunami) and tropical cyclones (wind and flood due to used for a variety precipitation and storm surge) for the following 15 PICs1: Cook Islands (CK), Federated of applications States of Micronesia (FM), Fiji (FJ), Kiribati (KI), Nauru (NR), Niue (NU), Palau (PW), Papua New in Climate and Guinea (PG), Republic of the Marshall Islands (MH), Samoa (WS), Solomon Islands (SB), Timor- Disaster Risk Leste (TL), Tonga (TO), Tuvalu (TV), and Vanuatu (VU). Management. This report focuses on the development of the country catastrophe risk profiles, the information collected, how it was catalogued and processed, and now being used for a variety of applications in Climate and Disaster Risk Management. The country risk profiles integrate data collected and produced through risk modeling and include maps showing the geographic distribution of assets and people at risk (Section 1), hazards assessed (Section 2) and potential monetary losses and casualties (Section 3). The profiles also include an analysis of the possible direct losses (in absolute terms and normalized by GDP) caused by tropical cyclones and earthquakes, and their impact though severe winds, rainfall, coastal storm surge, ground shaking and tsunami waves. The expected return period indicates the likelihood of a certain specified loss amount to be exceeded in any one year. The country risk profiles developed can be used to improve the resilience of these 15 PICs to natural hazards and to help mitigate their tropical cyclone and earthquake risk (section 4). In addition, applications such as a risk information system and assessment tools were developed to better understand and assess the countries’ exposure to natural disasters. Disaster risk financing solutions and financial sector development (macroeconomic panning) are discussed. Further potential applications in disaster risk reduction and urban/ infrastructure spatial planning, post disaster assistance and assessment, early warning systems and communications are described (Section 5). For the purpose of this document the countries included in the initiative are referred to as the 15 PICs. 1 CATASTROPHE RISK ASSESSMENT METHODOLOGY 11 FIGURE 1. Location of the 15 PICs. u A list of selected references use is included (Section 7, page 78). Information in the Annexes (pages 79-91) contains details to the locations that were field surveyed by the project teams, further background to the development of the building location methodology, construction type and general condition of buildings, the infrastructure exposure database and examples of the consequence database. Most importantly, the 15 individual country risk profiles are included. 12 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) Catastrophe Risk Modeling T This study ropical cyclones and earthquakes are the most prominent natural hazards in considers the the Pacific Islands Region. This study considers the devastating effects of wind, flood, devastating effects and storm surge induced by tropical cyclones as well as earthquake ground shaking and of wind, flood, tsunamis. Other hazards, such as weaker but still potentially damaging local storms and vol- and storm surge canic eruptions, are not included in this study. The risk due to tropical cyclones and tsunamis induced by tropical is computed assuming current climate conditions and sea levels. The effects of climate change cyclones as well on risk, which can be addressed using a similar methodology, are left to future investigations. as earthquake The catastrophe models used to perform the risk analyses for the 15 PICs adopt ground shaking and state-of-the-art methodology summarized in Figure 2. Every step of the methodology is tsunamis. based on empirical data collected in the region, as described in the following sections. FIGURE 2. Risk modeling methodology 1. Exposure Information 2. Hazard Assessment 3. Damage 4. Catastrophe Estimation Risk Profiles INTENSITY INTENSITY CALCULATION CALCULATION EVENT INTENSITY GENERATION CALCULATION 1. Exposure Information Exposure forms part of the initial step in the risk analysis process. It includes information on the distribution of the population and characterization of the assets that are exposed to natu- ral hazards. 1.1 Population A population database, based on a Geographical Information System (GIS) was created in order to geographically identify the population and assets at risk in each PIC. The population for each of the 15 PICs is shown in Table 1. CATASTROPHE RISK ASSESSMENT METHODOLOGY 13 TABLE 1. Population projection for 2010 and administrative boundaries (resolution level) with census year and growth rates for each PIC 2010 Projected Census SPC Annual Country Administrative Boundary Levels Population Year Growth Rate CK 19,800 2006 0.32% Group Island Electoral Boundary Census District Enumeration Area FJ 846,800 2007 0.46% Province Tikina Enumeration Area - - FM 111,600 2000 0.42% State Municipality Electoral District - - KI 101,400 2005 1.85% Group Island Village - - MH 54,800 1999 0.69% Atoll Islet - - - NR 10,800 2006 2.08% Island District - - - NU 1,500 2006 -2.31% Village - - - - PG 6,405,600 2000 2.13% Province District Local Government Census Unit - Level PW 20,500 2005 0.59% State Hamlet - - - SB 547,500 1999 2.69% Province Ward Enumeration Area - - TL 1,066,600 2004 2.41% District Subdistrict Suco - - TO 103,400 2006 0.33% Division District Village Census Block - TV 10,000 2002 0.51% Island Village - - - VU 245,900 1999 2.54% Province Island Area Council Enumeration - Area WS 182,900 2006 0.30% Island Region District Village - Many sources were used to compile this data- FIGURE 3: Population database for TV base, including the bureaus of statistics of national (a) Village-resolution population distribution with governments, the TL GIS Web Portal, the Univer- associated Islands for TV sity of Papua New Guinea (UPNG) and SPC/SOPAC, VID Village IID Island Country Census 02 which provide population counts within each ad- ministrative boundary identified. In general, the 11 Hauma 1 Nanumea Tuvalu 181 population data collected is based on the population 12 Lolua 1 Nanumea Tuvalu 215 counts as of January 2011, detailed in the national cen- 13 Haumaefa 1 Nanumea Tuvalu 124 sus and consequently ten years old at most. To accu- 14 Vao 1 Nanumea Tuvalu 107 rately characterize the current (2010) population, each 15 Matagi 1 Nanumea Tuvalu 37 country’s population counts were trended to 2010 from the year of the last available official census. Country- 21 Tonga 2 Nanumea Tuvalu 281 specific Annual Growth Rates (AGR) were used from SPC 22 Tokelau 2 Nanumea Tuvalu 308 Statistics for Development Division (country-level AGRs 31 Teava 3 Niutao Tuvalu 439 for all nations except TL) and the TL 2010 preliminary 32 Kulia 3 Niutao Tuvalu 224 census (district-level AGRs for TL). The 2010 trended 41 Manutalake 4 Nui Tuvalu 140 population for the coarser resolutions were aggregated 42 Alamoni 4 Nui Tuvalu 176 from the finer resolutions. This aggregation technique ensures that each resolution within a particular country 43 Malaki 4 Nui Tuvalu 105 has consistent population counts, which was not always 44 Meang 4 Nui Tuvalu 127 the case with the original data. Figure 3 illustrates an Note: VID: Village ID, IID: Island ID, Census_02: total village population example of the population database for TV, illustrating from 2002 official census. the finer (village) and coarser (island) geographic levels. 14 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) (b) Screenshot of the population database for Niutao The design of the population database allows for Island quick and robust querying for statistical metric devel- opment and easy superimposition with the other GIS databases, e.g., the buildings, infrastructure, and crop databases. 1.2 Buildings The exposure database includes a comprehen- sive inventory of residential, commercial, public and industrial buildings. It consists of their location, structural characteristics that affect the vulnerability to the effects of natural disasters and replacement costs. a. Locations In developing the exposure database, the loca- tions of the estimated 3.5 million buildings were determined using four different levels of build- ing extraction methodologies. These four levels, ranked in order of resolution, are outlined below and The administrative boundaries of all the PICs are chosen to balance accuracy and economy. were acquired from different sources, which are country-specific and of varying granularity. For Level 1 some countries (CK, FM, KI, MH, WS, SB, TO, TV and Individual buildings were manually digitized from high- VU) the coastal boundaries that are widely used by resolution satellite imagery and surveyed in the field regional organizations (e.g., SPC/SOPAC) and local (about 80,000 buildings in PG, TO, VU, TV, SB, WS, government do not perfectly align with the coastal CK, FJ, KI, PW, and FM). Information on locations field boundaries on satellite imagery. The maximum dis- surveyed can be found in Annex A. tance of this misalignment was about two kilometers. This misalignment issue is non-negligible in the fol- Level 2 lowing ten countries: CK, FM, FJ, MH, KI, WS, SB, TO, Individual buildings were manually digitized from high- TV, and VU. In regions where there are offsets, the resolution satellite imagery but not field verified (about misalignment is generally in the order of 10 to 100 450,000 buildings in all 15 PICs). meters. Other countries may have the polygon repre- sentations (geometrical shapes) of the administration Almost all of the major urban areas in the 15 PICs boundaries in the population database not perfectly were digitized using level 1 and level 2 methodologies. aligned with true geography. This misalignment issue These total more than 530,000 buildings, which rep- is limited only to the visual representation (and area resent approximately 15 percent of all the estimated of the land mass) of the population database. The buildings in the PICs. High-resolution satellite imagery exposure database and the computations in the tropi- was acquired from two main sources in order to manu- cal cyclone and earthquake risk assessment models ally digitized individual buildings. These were SPC/ are not affected by this issue. The aggregation of re- SOPAC’s high-resolution imagery repository, covering sults at fine level of granularity (e.g., loss estimates many urban centers of 14 countries (except TL), and aggregated at census district level) may sometimes be imagery purchased for this study from private ven- inaccurate due to the misalignment of administration dors. Geo-referenced high-resolution satellite images boundaries especially in urban areas where the census with pixel resolution of four meters or less were used districts are smaller. as backdrops to manually digitize building footprints CATASTROPHE RISK ASSESSMENT METHODOLOGY 15 by tracing polygons around the roof perimeter with Level 3: GIS software, which is a fairly straightforward but la- Clusters of buildings were extracted, outlined by poly- borious process. Select locations, typically high-density gons and manually counted from moderate to high- urban centers, were chosen for manual digitization. resolution satellite imagery. The coverage includes PG, Figure 4 shows Honiara, SB as an example. FJ, KI and, to a lesser extent, SB, CK, and MH. FIGURE 4. High-resolution satellite imagery and digitized Although the location of buildings in most of the building footprints (yellow outlines) in Honiara (SB) urban centers were digitized using high-resolution im- agery or captured during field surveys, buildings in a smaller number of second-tier urban areas with no coverage of high-resolution imagery (especially in PG) were interpreted using moderate-resolution imagery (e.g., more than four meter pixel resolution). For ex- ample, the colored polygons in Figure 6 are part of the urban areas surrounding the town of Lae in PG. The grey boxes indicate the boundaries of the high resolu- tion images where buildings were manually digitized. The red polygons fall outside those areas. They were processed by estimating the number of buildings with- in the urban polygon, where point locations of build- ings were randomly distributed over a 100 meter grid within the polygon boundary. Individual buildings were digitized with consistent criteria in order to minimize the digitization of non- building features such as farm equipment, cars, and FIGURE 6. The Lae urban area in PG (represented by the pavements. Figure 5 provides an example of how in- colored polygons) dividual building roofs were digitized. Generally, roof footprints were digitized as polygons for buildings approximately larger than 30 m2, whereas buildings smaller than 30 m2 were usually digitized as points. FIGURE 5. Detailed example of the building digitization process Note:The grey rectangles indicate the boundaries of high resolution sat- ellite imagery where buildings were manually digitized. The red poly- gons include buildings that were extracted from moderate satellite im- agery using the Level 3 methodology. Figure 7 shows how building counts and locations were determined by cluster polygons (i.e., counted manu- ally in groups without tracing individual roof footprints) in remote island areas (especially in western FJ) using mod- Note: Irrelevant features in the image are not digitized and adjacent erate to high-resolution imagery. Point locations of all the buildings are considered individually. buildings were aggregated to the centroid of the cluster. 16 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) FIGURE 7: Building clusters for Vakano, FJ The final building exposure database was supple- mented by ancillary data sets where available, includ- ing data that indicates the location of education and health facilities (for SB, VU, TL, PG, and PW) and re- sorts (for all countries except TV and TL). This data was collected from local government sources or generated and assembled during the development of the project. The number of buildings digitized in each country is listed in Table 2. The resulting maps of building locations for all 15 PICs can be found in Annex F (Country Risk Profiles). An example of the building locations in CK is shown in Figure 9. Note: Building clusters are outlined in yellow and are interpreted from moderate to high-resolution imagery using the Level 3 methodology. FIGURE 9. Map of the building locations in CK 166° W 164° W 162° W 160° W 158° W 0 150 300 600 Aitu- 012 4 Arutanga 8° S taki A land use analysis was used to develop the build- Kilometers Kilometers ing classification. Land use classes were devised ac- 10° S Atiu cording to inferred occupancy and construction type Mangaia 12° S prevalent in these countries. Figure 8 illustrates a land 012 4 use scheme for Wewak, PG. 0 1 2 3 14° S Buildings Residential Public Avarua Rarotonga 16° S Commercial Other FIGURE 8: Example of land use classes in buildings Industrial extracted from moderate resolution imagery – shown 18° S here for Wewak, PG. Atiu Aitutaki 20° S Rarotonga Mangaia 22° S Cook Islands 0 2 4 8 b. Field Surveys Field surveys were used to infer the characteris- tics of buildings whose location was either digi- n n Airport Commercial tized or statistically derived. To maximize the ben- n n Commercial agriculture Industrial efits of data collection within the constraints of budget Open land and time, most of the buildings in the field survey were n n Residential n Residential agriculture n Resort located in coastal urban areas which are more easily ac- Level 4: cessible, more prone to tropical cyclone and earthquake hazard, have a greater variety of building types and us- Buildings that are mostly located in rural areas were inferred age and have more costly structures. The field surveys using image processing techniques from low to moderate- conducted by teams of inspectors in PG, TO, VU, TV, resolution satellite imagery and/or census data. They were SB, WS, CK, FJ, KI, PW, and FM provided ground truth aggregated to uniform gridded polygons (“cells”) with verification. Even more importantly for the purpose associated building counts. The coverage includes PG, TL, of assessing risk, they provided a detailed inventory SB, VU, FJ, FM, MH, KI and, to a lesser extent, CK, TO, and of building characteristics, including occupancy type, TV. More details can be found in Annex B. CATASTROPHE RISK ASSESSMENT METHODOLOGY 17 TABLE 2. List of building counts per country Country Region Level 1 Level 2 Level 3 Level 4 Ancillary Total TL SE Asia – 96,539 – 300,791 1,355 388,685 FJ 18,622 79,545 8,214 158,436 1,323 266,140 PG 11,821 122,674 24,398 2,228,935 5,451 2,393,279 Melanesia SB 12,268 23,150 381 131,574 1,739 169,112 VU 10,661 21,883 – 66,782 1,420 100,746 FM 1,008 15,802 – 15,158 20 31,988 KI 746 12,137 2,139 12,562 5 27,589 MH Micronesia – 7,684 151 5,031 28 12,894 NR – 2,745 – – 10 2,755 PW 1,283 4,206 – 84 146 5,179 CK 5,044 4,889 100 357 212 10,602 NU – 1,105 – – 3 1,108 TO Polynesia 10,082 17,622 – 6,957 30 34,751 TV 956 1,258 – 804 – 3,018 WS 6,517 42,221 – – 93 48,831 All All 79,008 453,460 35,383 2,927,471 11,895 3,507,217 Note: The data bars indicate relative percentages of building counts, with blue bars = histogram showing distribution of level 1-4 across one country (to be read horizontally), yellow bars = histogram showing distribution of total building counts across countries (to be read vertically), and green bars = histogram showing total distribution of level 1-4 across all countries (to be read horizontally). construction type and structural characteristics, such FIGURE 10. Digitized roof footprints of building in Port as the buildings’ structural-frame, number of stories, Vila, VU, and a picture of a building taken by the field roof type, wall material, foundation type, presence of inspectors shutters, presence of defects, state of repair etc., which can only be extracted with certainty from field surveys. Each building visit was documented with photographs of the structure. In addition to residential and non- residential buildings, special structures and assets were also field surveyed, including infrastructure (airports, power plants, water facilities, etc.), bridges and farms/ gardens. Examples of the building footprint digitization exercise in Port Vila, VU, and a photograph of one of the inspected buildings is shown in Figure 10. No field surveys were conducted in NR, NU, MH, and TL. In NR and NU, which are very small countries (less than 10,000 inhabitants), building attribute data was collected using knowledge of local counterparts who processed the digitization. Very limited attributes were collected for MH and TL. Most of the building characteristics from these two countries were inferred from adjacent countries or census data. 18 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) The areas visited were selected according to the FIGURE 11. An example of the occupancy type of buildings following criteria: importance as a population and/or economic center, availability of high-resolution satellite imagery, exposure to significant natural hazards, inclu- sion of representative rural areas (proximity to supply markets), accessibility, security and cost efficiency. The spatial location of the digitized buildings was recorded Legend Footprints USE_GROUP and verified via portable GPS transmitters carried by n Other n Commercial/ the survey team during the field investigation. Build- agricultural n Critical facility n Education ings that no longer exist, features that were misin- n Hazardous facility n Health terpreted as buildings (pavements and equipment) n Public/community n Residential and newly constructed buildings were updated in the building database. Note: Residential buildings (single-family, multi-family, out-building), c. Occupancy Type and Construction commercial buildings (general, accommodation, gas station, out- Characteristics building), industrial buildings (general, food and drug, chemical pro- cessing), public buildings (education, emergency services, government, Of the 3.5 million buildings in the 15 PICs, 4.4 healthcare services, religion, general public, out-building) and other or unknown buildings are color coded according to the legend displayed. percent have verified basic attribute data (e.g., occupancy type and/or construction type) and 2.0 percent have verified detailed attribute data (e.g., building attributes collected from the field type statistics were determined from various sources. surveys). Characteristics of buildings were collected For single family homes (dwellings), construction type either from the field survey, local expertise, or inferred was determined from census data where available. from satellite images of building roofs. For an accurate This data included statistics of wall and roof materi- risk analysis, every modeled building should have de- als, along with other data, which was used to infer the tailed structural and occupancy use information. Build- construction type. Structural characteristics of single ing attributes were simulated using statistical tech- family houses, which represent more than 90 percent niques (a Monte Carlo method - essentially repeated of the enumerated buildings in the building exposure random sampling) based on empirical distributions of database, were statistically accurately represented over building characteristics extracted from data collected specific regions within each country. This detailed level from the building field surveys and/or country-specific of resolution allows for a comprehensive inventory of census reports. An example of the buildings occupancy building type characteristics and captures regional dif- type is shown in Figure 11. ferences within each country. Table 4 outlines the per- centages of construction type of single family houses The distribution of occupancy type is condition- for each state in FM, and shows that the western states al on the non-residential/residential and rural/ of Yap and Chuuk have a higher percentage of timber- urban distinction and uses statistics specific for frame houses, whereas the eastern states of Pohnpei each country. Table 3 shows the empirical distribu- and Kosrae have a higher percentage of masonry/con- tion of different types of occupancy in urban areas (for crete houses. These regional statistics provide valuable four PICs) and all rural areas (same values for all coun- insight into the building stock that could not be as- tries). About 90 percent of all buildings in the PICs are sessed strictly from the field-surveys, conducted mostly single family houses. in urban areas. For example, 74 percent of dwellings in rural areas of PG are traditional, as opposed to 10 The distribution of construction type is condi- percent in urban areas. According to the field survey, tional on the occupancy type and is distinct for only about 5 percent of the dwellings surveyed in PG different regions within each country (see Annex are constructed in the traditional style. C for more country specific details). Construction CATASTROPHE RISK ASSESSMENT METHODOLOGY 19 TABLE 3. Example of distribution of occupancy type per country and urban/rural location Country CK FJ FM PG ALL Location type Urban Urban Urban Urban Rural Commercial - Accommodation 36.5% 10.0% 4.1% 6.1% 2.6% Commercial - Gasoline station 0.6% 0.9% 1.2% 0.3% 0.8% Commercial - General commercial 29.1% 32.2% 40.8% 28.3% 34.6% Commercial - Out building 1.9% 1.4% 5.8% 3.5% 4.5% Industrial - Chemical processing 0.1% 0.6% 0.2% 0.1% 0.1% Industrial - Food and drug processing 0.2% 1.3% 0.2% 2.3% 0.1% Industrial - General industrial 2.2% 8.6% 1.5% 13.1% 0.8% Infrastructure 5.9% 5.8% 3.2% 4.5% 3.9% Other 3.0% 7.2% 1.8% 2.1% 6.8% Other - Out building 0.1% 1.1% 4.1% 1.8% 2.3% Public - Education 4.3% 14.7% 10.9% 20.1% 15.0% Public - Emergency services 0.1% 1.9% 1.5% 1.0% 0.8% Public - General public facility 5.9% 2.6% 2.6% 3.0% 2.3% Public - Government 4.1% 4.3% 12.4% 4.4% 9.4% Public - Health care services 1.8% 2.2% 2.2% 4.0% 4.1% Public - Religion 3.7% 4.7% 3.1% 4.1% 7.9% Public - Out buildings 0.7% 0.5% 3.4% 1.2% 4.1% Residential - Out building 13.3% 6.9% 8.4% 16.2% 0.0% Residential - Permanent dwelling multi family 1.5% 9.8% 7.2% 13.5% 0.3% Residential - Permanent dwelling single family 85.1% 83.3% 84.5% 70.2% 99.7% TABLE 4. Example of dwelling construction type statistics inferred from census data in FM Region name Yap Chuuk Pohnpei Kosrae Total HH 2,246 7,417 6,549 1,087 Multi-story masonry/concrete 4.5% 7.0% 8.2% 11.9% Multi-story timber frame with closed-under 1.4% 1.4% 1.0% 0.7% Multi-story timber frame with open-under 0.3% 0.3% 0.2% 0.2% Single story masonry/concrete 21.3% 32.7% 38.5% 55.6% Single story timber frame 51.5% 52.1% 36.7% 28.0% Traditional 21.0% 6.5% 15.4% 3.6% The distribution of secondary characteristics of are single story and less than 0.5 percent is taller than buildings (specific structural details, such as wall two stories. Most building foundations are concrete type, roof type, foundation type, and presence slabs or small posts that slightly elevate the structure of defects; as well as global characteristics such (less than one meter) above the ground, presumably to as number of stories and floor area) is condition- deal with flooding and pest issues. Gable roof shapes al on the construction type and non-residential/ are the most common, and over 90 percent of the non- residential distinction. Some characteristics are very traditional buildings have corrugated metal roofs, an common for the building stock throughout the PICs. For economical and reliable solution for roof cover in areas example, over 97 percent of the buildings in the PICs with high precipitation. 20 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) d. Replacement Cost pensive, while most buildings in rural areas lack mod- ern fixtures and use local materials and thus are much The economic losses from building damage are cheaper to construct. directly related to the replacement cost (or value) of the buildings. Replacement cost values for differ- Table 6 shows the building counts, total replace- ent types of buildings and occupancy types were col- ment cost, and average replacement cost per lected from a variety of sources, including a regional building for residential and non-residential build- construction cost management firm, government re- ings in urban and rural areas. The median replace- ports, interviews with local experts and historical disas- ment cost per building is much lower than the mean, ter reports. Table 5 shows examples of replacement indicating that a small percentage of buildings are very costs expressed in 2010 US dollars (US$) per square expensive, many of which have multiple stories and/or meter of the building floor. The total value of a build- very large floor areas. Figure 12 shows an expected ing is calculated as the product of the replacement correlation between the GDP per capita and the aver- cost, floor area and number of stories. It is very dif- age replacement cost per building in each country, in- ficult to accurately determine the replacement costs of dicating the validity of the building replacement costs. every possible building type and every location within a An example of the building replacement cost density country, since there is a large disparity in building value in SB is shown in Figure 13. Resulting maps of the within each PIC. For example, buildings constructed building replacement cost density for all 15 PICs can be with modern standards in urban areas tend to be ex- found in Annex F (Country Risk Profiles). TABLE 5. Example of ranges of replacement costs for different types of buildings in FJ and TO Fiji Tonga Characteristics Urban Rural Urban Rural Structure Category Height Quality Cost Lower Cost Higher Cost Lower Cost Higher Cost Lower Cost Higher Cost Lower Cost Higher Residential – High $800 $1,200 $275 $300 $800 $1,100 $225 $250 Residential – Medium $500 $800 $175 $275 $500 $800 $200 $225 Residential – Low $400 $500 $105 $175 $350 $500 $150 $200 Industrial – – $400 $650 $200 $325 $700 $900 $350 $450 Office Single story High $1,080 $1,215 $540 $608 $1,305 $1,350 $653 $675 Office Multy story High $1,080 $1,260 $540 $630 $1,350 $1,440 $675 $720 Office Single story Medium $810 $1,080 $405 $540 $1,170 $1,305 $585 $653 Office Multy story Medium $900 $1,080 $450 $540 $1,215 $1,350 $608 $675 Hotel Single story High $1,350 $1,500 $675 $750 $1,275 $1,425 $638 $713 Hotel Multy story High $1,500 $1,650 $750 $825 $1,350 $1,500 $675 $750 Hotel Single story Medium $1,200 $1,350 $600 $675 $1,050 $1,275 $525 $638 Hotel Multy story Medium $1,350 $1,500 $675 $750 $1,125 $1,350 $563 $675 Retail – High $600 $800 $300 $400 $800 $1,000 $400 $500 Retail – Medium $475 $600 $238 $300 $700 $800 $350 $400 Community – Medium $800 $1,100 $400 $550 $800 $900 $400 $450 Out/building – – $100 $160 $50 $80 $25 $50 $13 $25 Other – – $400 $500 $200 $250 $350 $500 $175 $250 Shack – – $30 $110 $30 $100 $20 $40 $20 $40 Traditional – – $67 $200 $50 $150 $133 $200 $100 $150 Infrastructure – – $1,100 $500 $500 $750 $1,000 $1,500 $500 $750 CATASTROPHE RISK ASSESSMENT METHODOLOGY 21 TABLE 6. Statistics of the PICs extracted from the exposure database (a) Building count (b) Replacement costs Building Count Building Replacement Cost Residential Non-Residential Residential Non-Residential Total Country Urban Rural Urban Rural Total Country Urban Rural Urban Rural (million USD) CK 53.2% 25.6% 18.5% 2.6% 10,602 CK 51.7% 6.0% 41.2% 1.1% $1,297 FJ 31.6% 58.9% 5.8% 3.6% 266,140 FJ 51.6% 16.7% 28.1% 3.5% $18,865 FM 17.7% 70.5% 3.9% 7.9% 31,988 FM 28.3% 37.0% 20.2% 14.5% $1,729 KI 30.0% 60.2% 5.4% 4.4% 27,589 KI 43.6% 21.4% 28.2% 6.8% $1,006 MH 48.9% 39.6% 9.2% 2.3% 12,894 MH 58.1% 10.9% 29.3% 1.8% $1,404 NU 82.9% 0.0% 17.1% 0.0% 1,108 NU 62.4% 0.0% 37.6% 0.0% $174 NR 83.1% 0.0% 16.9% 0.0% 2,755 NR 53.1% 0.0% 46.9% 0.0% $411 PG 6.7% 87.8% 1.2% 4.3% 2,393,279 PG 31.1% 29.3% 19.7% 19.9% $39,509 PW 52.3% 29.3% 17.5% 0.9% 5,179 PW 40.5% 4.3% 52.6% 2.5% $338 SB 18.2% 74.7% 3.2% 3.9% 169,112 SB 38.2% 25.7% 26.8% 9.4% $3,059 TL 31.9% 62.2% 3.0% 2.9% 398,685 TL 60.0% 13.8% 22.9% 3.3% $17,881 TO 67.8% 19.0% 12.2% 1.1% 34,751 TO 60.4% 4.5% 34.4% 0.8% $2,525 TV 38.0% 48.8% 9.4% 3.7% 3,018 TV 40.2% 18.7% 38.8% 2.3% $229 VU 26.6% 63.4% 6.4% 3.6% 100,746 VU 30.6% 25.8% 37.0% 6.6% $2,858 WS 22.2% 63.7% 4.9% 9.2% 48,831 WS 31.4% 18.7% 33.2% 16.6% $2,148 Total 14.2% 79.4% 2.3% 4.1% 3,507,217 Total 42.7% 21.6% 24.6% 11.0% $94,434 (c) Average replacement cost of buildings FIGURE 12. GDP per capita adjusted for purchasing power in the PICs parity (PPP) versus average building replacement cost per person Residential Non-Residential Country Urban Rural Urban Rural CK $118,882 $28,512 $272,076 $49,550 FJ $115,621 $20,151 $341,634 $69,150 FM $86,218 $28,384 $283,195 $99,136 KI $53,034 $12,936 $191,439 $56,164 MH $129,435 $29,878 $346,375 $82,611 NU $117,912 $- $346,065 $- NR $95,168 $- $414,425 $- PG $76,943 $5,510 $278,456 $75,689 PW $181,314 $34,707 $704,075 $628,987 SB $37,910 $6,235 $150,453 $43,090 Note: Data for NU is not included due to its low population count. GDP TL $84,213 $9,971 $345,196 $51,485 values are taken from the CIA Factbook. TO $64,701 $17,059 $205,396 $53,206 TV $80,395 $29,051 $311,896 $47,850 VU $32,619 $11,555 $164,976 $52,009 WS $62,096 $12,934 $298,657 $79,741 Mean $81,295 $7,340 $285,126 $71,682 Median $30,042 $3,149 $126,420 $44,794 22 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) FIGURE 13. Map of the building replacement cost density in SB Main Islands Honiara 0 1 2 4 0 50 100 200 Kilometers Building Replacement Kilometers Cost Density (million USD / km^2) 0 - 0.01 0.01 - 0.025 0.025 - 0.05 0.05 - 0.075 0.075 - 0.1 0.1 - 0.25 0.25 - 1 1 - 300 Solomon Islands Honiara 5° S 0 175 350 700 Main Islands Kilometers 10° S 155° E 160° E 165° E Note: Further maps for all 15 PICs are located in Annex F (Country Risk Profiles). 1.3 Infrastructure based data, Defence Imagery and Geospatial Organ- isation (DIGO) data, information issued by academia The infrastructure database was assembled using (reports, publications, maps, government agencies, similar techniques to those used for buildings, and firms), public databases, disaster reconnaissance comprises a detailed and extensive inventory reports and proprietary data. of major assets, such as airports, ports, power plants, dams, major roads, and bridges. For exam- While the infrastructure database is not ex- ple, Figure 14 shows the major infrastructure assets in haustive, it contains a comprehensive inventory PG. In addition to their locations, the infrastructure da- of major infrastructure facilities, with a higher tabase also includes estimates of the replacement costs level of detail in major urban centers. The main of such assets. types of infrastructure considered are airport, bridge, bus station, communications, dam, dock, generator, Information on the geo-locations and re- helipad, mine, oil and gas, port, power plant, water placement costs has been collected from a wide intake, storage tank, sub-station and water treat- ment. Total counts for each country and the gener- variety of data sources. These include field visits in al scope (and quality) of the data collection can be mostly major urban areas of 11 countries (PG, TO, VU, found in Annex D. TV, SB, WS, CK, FJ, KI, PW, and FM), manual inspection of publically available high-definition satellite imagery Different methods were used to calculate the (e.g., Google Earth), remote sensing techniques, GIS- replacement costs of infrastructure assets. For CATASTROPHE RISK ASSESSMENT METHODOLOGY 23 FIGURE 14. Location of major infrastructure assets in PG Major Infrastructure AIRPORT HELIPAD POWER PLANT PORT DOCK DAM MINE OIL & GAS COMMUNICATIONS STORAGE TANK GENERATOR SUB-STATION WATER TREATMENT WATER INTAKE BUS STATION ROAD BRIDGE example, airport costs are derived from the length and database. Significant infrastructure assets are typically condition (paved/unpaved) of the runway. Similarly, re- built to higher standards than residential structures, placement costs of bridges are derived from the length and their quality is similar throughout the entire PIC re- and material of the span. For ports, replacement costs gion. Therefore, our estimates of the unit replacement were obtained by multiplying the inferred unit cost by costs are independent of the location. An example of the area of the facility. For power plants, replacement the unit replacement costs of infrastructure in the PICs costs were estimated based on the energy output, as is shown in Table 7 while the total replacement cost of listed by the Carbon Monitoring for Action (CARMA) the infrastructure for each country is shown in Table 8. 24 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) TABLE 7. Unit replacement costs of infrastructure in PIC Type Cost (US$) Metric Large Airport 518 per linear foot of runway Medium Airport 366 per linear foot of runway Helipad 88,000 per unit (40 12.5’-by-20’ slabs) Airstrip 10,000 per unit Small Airport 100,000 per unit Dam 100,000,000 per unit Large Scale Mine 500,000,000 per unit Medium Scale Mine 100,000,000 per unit Small Scale Mine 10,000,000 per unit Steel/Concrete Bridge 10,000 per linear meter of span Non-Steel/Concrete Bridge 1,000 per linear meter of span Roads 500,000 per linear kilometer Railroads 100,000 per linear kilometer Dock 100,000 per unit Water Treatment 2,000,000 per unit Storage Tanks 10,000 per unit Water Intake 40,000 per unit Bus Station 30,000 per unit Communications 5,000 per unit Oil & Gas Facility 20,000,000 per unit Power Plant - Very Large 40,000,000 per unit Power Plant - Large 10,000,000 per unit Power Plant - Medium 5,000,000 per unit Power Plant - Small 1,000,000 per unit Power Plant - Very Small 500,000 per unit Generator 1,000 per unit Substation 500,000 per unit Port - Very Large 100,000,000 per unit Port - Large 50,000,000 per unit Port - Medium 10,000,000 per unit Port - Small 5,000,000 per unit Port - Very Small 1,000,000 per unit 1.4 Crops The crop exposure database is a subset of a more comprehensive Land Use / Land Cover The crop exposure database is a comprehensive (LULC) geo-database, in which other land use inventory of major cash crops – their location, categories were indexed (e.g., forests, lakes and types, and replacement costs. It does not directly rivers, sand, settlements, barren land, and grass consider subsistence crops or forestry. The spatial dis- land). The LULC maps were generated primarily using tribution of major crops was derived from moderate- remote sensing and were supplemented with various resolution satellite imagery and image detection tech- sources. To our knowledge, this is the first time that niques. To the extent possible, results were validated LULC maps were derived and made available to the using high-resolution imagery, agricultural census data, public for the PIC region. Similar to the identification of ancillary data collected during the course of the project building locations in rural areas, the LULC maps were (FJ, TO and VU) and feedback from local experts. developed using satellite imagery (mostly moderate- TABLE 8. Total replacement cost of infrastructure per country Type CK FJ FM KI MH NI NR PG PW SB TL TO TV VU WS Total Airport $20,891,472 $54,013,080 $40,641,392 $40,259,053 $43,875,113 $13,019,960 $11,988,400 $204,069,571 $12,439,320 $21,587,600 $24,818,248 $29,389,536 $8,497,824 $36,707,648 $23,063,306 $585,261,521 Bridge $2,910,000 $180,399,974 $22,968,139 $12,150,590 $525,300 $– $– $139,044,869 $23,149,939 $18,990,926 $151,749,790 $1,459,700 $– $7,155,132 $13,310,230 $573,814,589 Bus station $– $210,000 $– $– $– $– $– $30,000 $– $– $– $– $– $– $30,000 $270,000 Communications $60,000 $125,000 $– $5,000 $25,000 $– $5,000 $190,000 $15,000 $230,000 $– $140,000 $– $160,000 $45,000 $1,000,000 Dam $– $300,000,000 $– $– $– $– $– $400,000,000 $– $– $– $– $– $– $– $700,000,000 Dock $2,700,000 $3,100,000 $10,100,000 $900,000 $2,600,000 $100,000 $600,000 $6,400,000 $3,900,000 $3,900,000 $100,000 $6,200,000 $400,000 $1,600,000 $600,000 $43,200,000 Generator $1,000 $17,000 $– $– $– $– $– $56,000 $– $14,000 $– $5,000 $– $5,000 $4,000 $102,000 Helipad $– $– $– $– $616,000 $– $– $– $– $– $– $– $– $– $– $616,000 Mine $– $– $– $– $– $– $– $2,210,000,000 $– $100,000,000 $10,000,000 $– $– $– $– $2,320,000,000 Oil & gas $– $40,000,000 $– $– $– $– $– $200,000,000 $– $100,000,000 $20,000,000 $– $– $20,000,000 $20,000,000 $400,000,000 Port $10,000,000 $482,000,000 $85,000,000 $24,000,000 $182,000,000 $1,000,000 $5,000,000 $641,000,000 $27,000,000 $101,000,000 $15,000,000 $25,000,000 $5,000,000 $65,000,000 $30,000,000 $1,698,000,000 Power plant $13,500,000 $339,000,000 $41,000,000 $20,500,000 $28,500,000 $5,000,000 $10,000,000 $685,000,000 $15,000,000 $61,500,000 $18,500,000 $17,000,000 $9,500,000 $35,000,000 $76,000,000 $1,375,000,000 Water intake $720,000 $880,000 $– $– $– $840,000 $– $320,000 $– $360,000 $– $7,040,000 $– $240,000 $160,000 $10,560,000 Storage tank $2,290,000 $4,560,000 $600,000 $850,000 $1,730,000 $30,000 $1,880,000 $3,490,000 $390,000 $1,200,000 $410,000 $1,690,000 $310,000 $1,150,000 $640,000 $21,220,000 Sub-station $3,500,000 $2,500,000 $– $– $– $– $– $3,000,000 $– $500,000 $5,000,000 $– $8,000,000 $5,000,000 $– $27,500,000 Water treatment $– $20,000,000 $– $– $2,000,000 $– $2,000,000 $10,000,000 $2,000,000 $6,000,000 $4,000,000 $– $– $4,000,000 $4,000,000 $54,000,000 Rail $– $32,100,000 $– $– $– $– $– $– $– $– $– $– $– $– $– $32,100,000 Roads $61,000,000 $1,635,000,000 $112,500,000 $65,500,000 $24,000,000 $54,000,000 $10,500,000 $2,136,500,000 $76,000,000 $5,000,000 $1,911,000,000 $171,500,000 $8,000,000 $244,000,000 $299,500,000 $6,814,000,000 Total $117,572,472 $3,093,905,054 $312,809,531 $164,164,643 $285,871,412 $73,989,960 $41,973,400 $6,639,100,440 $159,894,259 $420,282,526 $2,160,578,038 $259,424,236 $39,707,824 $420,017,780 $467,352,536 $14,656,644,110 CATASTROPHE RISK ASSESSMENT METHODOLOGY 25 26 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) FIGURE 15. LULC map for WS with locations of major cash crops Note: Maps for all 15 PICs can be found in Annex F (Country Risk Profiles) resolution and, when unavailable, low-resolution im- for all 15 PICs and their estimated areas are shown in ages). Other ancillary datasets were used to validate Table 9. the accuracy of LULC maps, including the Global Land The unit replacement cost of different cash Cover 2001 classification, topo-sheets, classified maps crops was derived from crop production budgets and reports. Data reflecting land surface characteris- issued by local governments. This data is current as tics, which include land suitability maps, agriculture of July 2010 and available for FJ, PG and TO, providing and vegetation maps, digital elevation maps, and slope representative crop production and cost information and aspect maps were also used as ancillary data in the for the key agricultural producing countries in the re- classification. This allowed for identifying crops com- gion. Crop production budgets are good proxies to de- monly grown in certain terrain conditions. An example rive replacement costs after a disaster strikes since they of a LULC map, one output of the analysis, is displayed indicate the total cost per hectare incurred by a farmer in Figure 15. It shows the major crops for WS. LULC if they were to completely rebuild their production sys- maps for all 15 PICs can be found in Annex F (Country tem. Also, the total cost per hectare is useful as a proxy Risk Profiles). for assessing business interruption losses (which are, The LULC maps developed herein are suitable however, outside of the scope of this study), especially for characterizing a crop exposure database and for fruit trees and permanent plantations affected by to be used as an input for a disaster risk analysis. cyclones. These costs also provide a measure of the loss The primary crop types included in the final LULC maps incurred for renewing the operation while the crops TABLE 9. Major crop types and estimated area per crop type in each country as per the LULC database Estimated country specific crop type area in hectares (HA) # Crop type CK FJ FM KI MH NR NU PG PW SB TL TV TO VU WS 1 Banana 141688 56 7 7 2 Cassava 5 298 12 820 12 3 Cocoa 8222 4 Coconut crops 691 2025 1533 1778 1806 152 3560 17383 1008 14455 8337 569 11363 11716 6861 5 Coconut forest 2323 6221 2577 28951 9568 82533 861 51782 5456 16 7380 43692 25914 6 Coconut plantation 1102 21453 4350 29557 12801 335 25336 1754 40867 1450 1192 16383 1211 7 Cultivated land 14658 272 320772 54852 12494 48 8 Grazing land 13 9 Legumes 1 10 Low intensity mixed crops 142824 11 Mixed crop 825 30679 12 Nori tree 87 13 Navel nut tree 2 14 Nut tree 356 22 6 335 38 144 15 Orchard 2 187 23 16 Palm oil 997 28 98 579 175 234976 6 43775 1495 28 25 14 17 Plantation 1 28463 1476 951 361 7105 2606 18 Rice 98 1059 61 9 1277 133 3530 39670 338 19 Rubber 4035 20 Sago 40546 21 Scattered coconut plantation 734 22 Squash 11093 23 Sugarcane 20 137084 14786 24 12 3 24 Sweet potato 199473 25 Taro/Chinese taro 135 726 186 136116 446 5 80 1 26 Tea/Coffee 70524 27 Unknown crops 733 30 236 686 224 428 277 28 Vanilla 4068 29 Yams CATASTROPHE RISK ASSESSMENT METHODOLOGY 27 28 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) reach their full productivity. To quantify the importance inputs intensively and obtain higher prices when sell- of each crop for the agricultural sector of the PICs, in- ing their products in the export markets. In order to formation on production value information for all the account for the different types of production systems crops grown in each country was also collected. in the PICs, the following were considered during the development of the average replacement costs shown Table 10 shows the replacement costs per in Table 10 : hectare computed for the key crops under pro- duction in the PICs. The average replacement cost ■■ Subsistence farmers are assumed to invest only a estimates are representative of production systems fraction of the costs incurred by an average pro- with average production and management practices. ducer in the region. Therefore, the replacement These average costs are not representative of subsis- costs for subsistence producers have been reduced tence farmers that use fewer inputs and therefore have to one fourth of the traditional cost. less production costs, or commercial farmers that use TABLE 10. Replacement costs for key crops under different production systems in the PICs Average Replacement cost Replacement cost replacement cost subsistence commercial farmer Crop type (US$ per hecatre) (US$ per hecatre) (US$ per hecatre) Banana 4,065 1,016 6,098 Breadfruit 386 97 579 Cassava 2,468 617 3,702 Cocoa 1,766 442 2,649 Coconut (Copra) 294 74 441 Coconut (Fresh Nut) 504 126 756 Coconut (Mature Nut) 504 126 756 Coffee 1,512 378 2,268 Ginger 7,697 1,924 11,546 Gourd/Squash 1,213 303 1,820 Kava/Yaqona 3,532 883 5,298 Lemon 966 242 1,449 Mango 375 94 563 Nut Tree 1,750 438 2,625 Oil Palm 5,300 1,325 7,950 Papaya 3,039 760 4,559 Pineapple 2,009 502 3,014 Pumpkin 2,999 750 4,499 Rubber Tree 504 126 756 Sago Palm 1,488 372 2,232 Sugarcane 1,234 309 1,851 Sweet Corn/Maize 1,822 456 2,733 Sweet Potato 1,474 369 2,211 Giant Taro/Ta’amu 1,365 341 2,048 Taro 2,993 748 4,490 Tobacco 9,080 2,270 13,620 Vanilla 1,243 311 1,865 Yam 9,843 2,461 14,765 CATASTROPHE RISK ASSESSMENT METHODOLOGY 29 ■■ Commercial producers that invest heavily in tech- indicated in the LULC maps, which sometimes included nology and whose production is oriented towards multiple crops in one area, were mapped appropriately export markets are assumed to have higher re- to a similar crop classification in which the replacement placement costs than the average crop production costs (see later in this section) and damage functions systems. For commercial farmers, the replacement could be easily assigned. costs have been increased by half of the traditional The agricultural sector of the PIC’s is prone to cost. crop losses from recurring natural disasters, es- For the purpose of the risk assessment, re- pecially in FJ, SB, CK, TO, WS and VU, which are placement cost estimates used for cash crops are located in cyclone-prone regions. In general, the dif- consistent with those for commercial farmers, ferent crops in the PIC’s react distinctly when affected by which is appropriate when estimating the losses cyclones, tsunamis or flooding. For example, it has been caused to cash crops by tropical cyclones. In addi- documented that recently introduced crops to serve ex- tion, the values in Table 10 were modified slightly to port markets are very susceptible to damage from cy- correspond with each country’s GDP for the agriculture clonic winds or salt spray compared to more resilient, sector as given in the CIA World Factbook. This final native crops or those that have been cultivated over step ensures the validity and consistency of the total centuries by the PIC’s farmers. Table 11 shows a classifi- crop asset value for each country. cation of crop groups according to their vulnerability to adverse weather events. Due to the scarcity of data for The LULC databases, which contain informa- crop damage and associated damage functions, three tion on all vegetation, were used to create the main groups of crops are considered in the risk analysis cash crop exposure database. Cash crops were in- – tree crops, root crops and annual crops. These crop dexed by sampling the LULC data on an 80-by-80 me- groups were chosen since they represent three general ter grid for most countries. For the larger countries (PG, vulnerability classes of crops. For example, root crops are WS and FJ), the sampling grid was taken at 270 by 270 more susceptible to flood damage, while tree crops are meters. These different sampling resolutions balanced more susceptible to wind damage. Another category is accuracy and economy, allowing for the detection of categorized as “inter crops,” which is some combina- cash crops in small atolls. In addition, the crop types tion of these three main crop types. TABLE 11. Crop groups according to vulnerability to natural disasters in the PICs Crop groups Crop types Vulnerability to natural disasters Root crops Yam, Taro/Dalo, Xanthosoma, Highly vulnerable depending on stage of development, losses will be Cassava, Kumala, Sweet Potato, reduced if near harvest. Damage often requires total replanting and 10-12 Kava/Yaqona, Pukala/Giant Taro/Kape months to reach full production potential. Tree crops Coconut, Palm Oil, Breadfruit, Resillient, will resist moderate wind speed without uprooting but leaves Cocoa, Coffee, Mango, Papaya, and fruits can be completely destroyed. When uprooting occurs economic Pandanus damage is large since it takes 3-4 years for a mature tree to reach peak of production. Plantation Sugarcane, Coffee, Palm Oil, Less vulnerable given distance to coast and good mechanical protection crops Coconut, Cocoa, Papaya, Citrus, structures (drainage, wind barriers, supports). If event is severe, losses will Banana, Vanilla be large due to higher exposure aggregation at one single spot. Substinence Yam, Taro/Dalo, Xanthosoma, Cas- Local varieties developed by the polynesian and Melanesian civilizations crops sava, Kumala, Sweet Potato, Kava/ over the centuries are quite resilient to natural disasters. When crop foods Yaqona, Pukala/Giant Taro/Kape, are destroyed, people often rely on these varieties for sustenance for Banana months until help is delivered to remote locations. Annual crops Pepper, Gourd, Squash, Tomato, Highly vulnerable depending on stage of development, losses will be Capsicum, Cabbage, Corn, Peanut, reduced if near harvest. Damage often requires total replanting and 10-12 Rice, Okra, Eggplant, Ginger, months to reach full production potential. Watermelon, Pumpkin. 30 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) 1.5 Replacement Costs by Country comprises US$94 billion in buildings, US$15 billion in infrastructure assets, and US$4 billion in major crops. Millions of building and infrastructure assets A breakdown of the replacement costs by country and hundreds of thousands of hectares of cash is shown in Table 12. The exposure database, as crops with their characteristics and location well as the hundreds of satellite imageries acquired, are included in a geo-referenced database. organized, and processed for this project, are hosted To date, this database is the most comprehensive and maintained by SPC-SOPAC. This wealth of data exposure dataset for the pacific. The estimated can support multiple applications, such as in urban total replacement cost of all the assets in the and development planning that benefit both public 15 PICs is about US$113 billion, an amount that and private stakeholders (see Section 5). TABLE 12. Replacement costs by country Total replacement cost (million US$) Country Buildings Infrastructure Cash crops Total CK 1,296.8 117.6 7.8 1,422.2 FJ 18,865.2 3,093.9 216.1 22,175.3 FM 1,729.0 312.8 5.8 2,047.3 KI 1,006.1 164.2 11.3 1,181.5 MH 1.404.1 285.9 5.7 1,695.7 NR 410.6 42.0 0.1 452.7 NU 173.8 74.0 1.2 248.9 PG 39,509.0 6,639.1 3,060.7 49,208.8 PW 1,338.5 159.9 2.5 1,500.8 SB 3,058.7 420.3 11.7 3,490.7 TL 17.881.3 2,160.6 102.9 20,144.8 TO 2,525.2 259.4 31.9 2,816.5 TV 229.3 39.7 1.2 270.2 VU 2,858.4 420.0 56.0 3,334.4 WS 2,147.9 467.4 24.7 2,639.9 Total 94,434.0 14,656.6 3,539.5 112,630.1 CATASTROPHE RISK ASSESSMENT METHODOLOGY 31 2. Hazard Assessment The hazard estimation is the second building block in the risk assessment methodol- ogy shown in Figure 2. The hazard assessment module comprises two main components: the simulation of future events that may cause damage to the PICs and the prediction of the intensity of such simulated events in the region affected. They form what is often referred to as the stochastic event set. The Pacific Region is prone to a variety of natural hazards with tropical cyclones, earthquakes and tsunamis assessed in this initiative. The effects of tropical cyclones are wind speed, precipitation and coastal surge. For earthquakes they are ground shaking and in certain cases tsunami waves, which in this study are gauged by wave height and velocity. The models that characterize these effects are based on empirical data and on the underlying physics of the phenomena. The tropical cyclone and earthquake hazard models have been peer reviewed by scientists at Geoscience Australia, which found them of “high standard, thorough and representative of best practice.” The resulting tropical cyclone hazard map, for example for VU, is shown in Figure 16 and the earthquake hazard map for PG in Figure 17. Tropical cyclone and earthquake hazard maps for all 15 PICs can be found in Annex F (Country Risk Profiles). This section provides the detail on how those maps were created. FIGURE 16. Tropical cyclone hazard map for VU 0 75 150 Kilometers 300 Shefa Malampa 0 75 150 300 0 15 30 60 14° S Sanma Kilometers Kilometers 16° S Malampa Shefa 18° S Port Tafea Vila Vanuatu 20° S 168° E 170° E 0 1.5 3 6 0 15 30 60 Kilometers Kilometers Tafea 0 15 30 60 Sanma Port Vila Kilometers 0 25 50 75 100 125 150 175 200 Maximum Wind Speed Note: Maximum 1-minute sustained wind speed (in miles per hour) with a 40 percent chance to be exceeded at least once in the next 50 years (100 year mean return period). Tropical cyclone hazard maps for all 15 PICs can be found in Annex F (Country Risk Profiles). 32 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) FIGURE 17. Earthquake hazard map for PG 145° E 150° E 155° E 0 100 200 400 Kilometers 5° S 5° S Lae Port Moresby 10° S 10° S Papua New Guinea 0 3 6 12 Kilometers Port Moresby 145° E 150° E 155° E Very Perceived shaking Not felt Weak Light Moderate Strong strong Severe Violent Extreme Moderate/ Potential damage none none none Very light Light Moderate heavy Heavy Very heavy Peak ACC (%g) <0.17 0.17-1.4 1.4-4.0 4.0-9 9-17 17-32 32-61 61-114 >114 Peak Vel. (cm’s) <0.12 0.12-1.1 1.1-3.4 3.4-8 8-16 16-31 31-59 59-115 >115 Instrumental intensity I II-III IV V VI VII VIII IX X+ Note: The peak horizontal acceleration of the ground (Note: 1g is equal to the acceleration of gravity) with about a 40 percent chance to be exceeded at least once in the next 50 years (100-year mean return period). The scale is based upon Wald et al. 1999. Earthquake hazard maps for all 15 PICs can be found in Annex F (Country Risk Profiles). 2.1 Tropical Cyclone Event Generation model validation. The catalog of historical storms was assembled starting with the dataset of the Interna- Tropical cyclones are usually accompanied by tional Best Tracks Archive for Climate Stewardship proj- damaging winds, rains, and storm surge. Ar- ect (IBTrACS). This dataset is endorsed by the World eas both North and South of the equator are known for the frequent occurrence of tropical cyclones, all Meteorological Organization, containing data from throughout the year in the North Pacific and between meteorological agencies across the region, including the months of October and May in the South Pacific. the Joint Typhoon Warning Center (JTWC), the Aus- A review of all available tropical cyclone data was per- tralia Bureau of Meteorology (BoM) and the Fiji Me- formed to create the historical tropical cyclone catalog teorological Service. It is the most comprehensive of (or dataset), which is required to simulate the project’s the available datasets and is current through 2008. The stochastic event set. This dataset was also needed for best track dataset contains the most complete global CATASTROPHE RISK ASSESSMENT METHODOLOGY 33 set of historical tropical cyclones available, combines The tracks of these historical tropical cyclones information from numerous tropical cyclone datasets, are shown Figure 18. Many of these storms simplifies inter-agency comparisons by providing storm have impacted one or more of the PICs, causing data from multiple sources in one place and checks widespread destruction, high economic losses, the quality of storm inventories, positions, pressures, and many casualties (injuries and fatalities). The and wind speeds, and passes the information on to number of storms in the catalog classified by Saffir- the user. The IBTrACS file was used as the basis for the Simpson Category is shown in Table 13. In the last historical tropical cyclone event catalog in this study. 60 years, the Pacific Region from Taiwan (25°N) to Because of data quality for the earliest records, as well New Zealand (35°S) and from Indonesia (120°E) to as consideration of inter-basin data consistency, data east of Hawaii (120°W) has experienced more than collected from 1948 onwards was used. Time peri- 2,400 tropical cyclones, about 41 per year. More than ods when storm intensity information was not present 1,400 formed in North West Pacific (24.8 events/year) were identified and checked to determine if they could and almost 1,000 formed in the South Pacific (16.17 be supplemented with estimated values based on other events/year). The catalog includes also tropical storms known storm parameters. The replacement of missing with winds below hurricane strength. These weaker data was achieved based on regression formulations storms have been included in the catalog because of derived from the distributions of all fully-known data their capability of producing torrential precipitation points. The final historical track dataset was compiled and, consequently, devastating floods. using the quality-controlled and augmented dataset, interpolated to hourly track points. FIGURE 18: Tracks of the approximately 2,400 historical tropical cyclones in the Pacific Islands Region in the last 60 years Marshall Islands Palau Kiribati Federated States of Micronesia Nauru Solomon Islands Tuvalu Fiji Timor-Leste Samoa Papua New Guinea Cook Islands Vanuatu Niue Historical TC Catalog Tonga Saffir-Simpson Category 5 4 3 2 1 Modeled Countries Note: The maximum wind speeds generated by these events range from 74-95 mph for a Category 1 storm to greater than 155 mph for a Category 5 storm. 34 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) TABLE 13. Number of storms in the catalog with Saffir-Simpson classification Saffir-Simpson Wind Speed (mph) Storm Surge (ft) Central Pressure North South Total Category 1mph=1.6kmph 1ft=30.48cm (mbar) mbar=hPa No Intensity       0 286 286 0 (Tropical     144 47 191 Storm) 1 74-95 4-5 980 556 395 951 2 96-110 6-8 965-979 183 106 289 3 111-130 9-12 945-964 193 82 275 4 131-155 13-18 920-944 227 51 278 5 >155 >18 <920 135 17 152 Total     1438 984 2422 TABLE 14. Example of historical hourly track file showing all relevant storm parameters NUM HR LON LAT DIR CP RMAX FS VMAX NAME 1746 53 -164.6 -13.13 325.61 991 26.5 4.07 64.94 NANCY 1746 54 -164.55 -13.17 325.6 991 26.91 4.07 64.91 NANCY 1746 55 -164.5 -13.2 306.13 991 27.33 5.7 65.75 NANCY 1746 56 -164.45 -13.27 306.13 990.33 26.31 5.7 66.75 NANCY 1746 57 -164.4 -13.33 306.11 989.67 25.29 5.7 67.75 NANCY 1746 58 -164.35 -13.4 306.11 989 24.27 5.7 68.73 NANCY 1746 59 -164.3 -14.47 306.1 988.33 23.24 5.7 69.7 NANCY 1746 60 -164.25 -13.53 306.1 987.67 22.22 5.7 70.67 NANCY 1756 61 -164.2 -13.6 302.94 987 21.2 4.11 70.78 NANCY 1746 62 -164.17 -13.65 302.93 979.83 22.66 4.11 79.75 NANCY 1746 63 -164.13 -13.7 302.92 972.67 24.13 4.11 87.88 NANCY 1746 64 -164.1 -13.75 302.93 965.5 25.59 4.11 95.39 NANCY 1746 65 -164.07 -13.8 302.92 958.33 27.06 4.11 102.42 NANCY 1746 66 -164.03 -13.85 302.91 951.17 28.52 4.11 109.08 NANCY 1746 67 -164 -13.9 350.26 944 29.99 6.8 116.8 NANCY 1746 68 -163.9 -13.92 350.26 944 28.88 6.8 116.87 NANCY 1746 69 -163.8 -13.93 350.26 944 27.78 6.8 116.93 NANCY 1746 70 -163.7 -13.95 350.26 944 26.68 6.8 117 NANCY 1746 71 -163.6 -13.97 350.25 944 25.57 6.8 117.06 NANCY 1746 72 -163.5 -13.98 350.25 944 24.47 6.8 117.13 NANCY 1746 73 -163.4 -14 345.55 944 23.36 13.84 120.05 NANCY A tabular example of the historical dataset is at hour 68, Tropical Cyclone Nancy had a maximum provided in Table 14. For every hour, the dataset in- sustained wind of 116.9 mph and a central pressure cludes the storm position in longitude and latitude, the of 944 mb. direction of storm motion (in degrees, measured coun- terclockwise from east), the central pressure (in mb), The spatial and temporal occurrence and se- the radius of maximum wind (in miles), the forward verity of past events have been used as a guide speed of the storm (in mph), the maximum wind speed to simulate potential tropical cyclones and earth- (in mph) and the event name (if known). For example, quakes in the PICs in the future. These simulated CATASTROPHE RISK ASSESSMENT METHODOLOGY 35 events are not necessarily identical to those that oc- cated in the figure is the maximum along the track, curred in the past but are statistically consistent and regardless of whether that category was reached close representative. In general terms, this means that the to or far from any of the PICs. location and severity of all the simulated events may not have been observed in the relatively short histori- 2.2 Tropical Cyclone Intensity Calculation cal records, but such events are possible and the likeli- hood of their occurrence has been derived based on a. Induced Winds the empirical data collected in the region. More rig- Using storm characteristics along the cyclone orously, the statistical consistency is evident in Figure track, the wind model calculates and retains the 19 which shows a good agreement between simulated maximum wind speed at each exposure location and observed event frequencies for tropical cyclones. for wind damage and loss estimation. The genera- The same comparison for tropical cyclones of different tion of local wind fields is a complex procedure requir- severities results also in a good agreement (figure not ing the use of many variables, including the cyclone shown). forward motion (also often referred to as translational The catalog of simulated tropical cyclones, speed), radial distance from storm center to the loca- which spans the Pacific basin, contains more than tion of interest, angle between track direction and sur- 400,000 tropical cyclones grouped in 10,000 po- face wind direction and inflow angle. The maximum tential realizations of what is possible to occur at over-water wind speed is calculated deterministically as any one point in time. The hazard model computes a function of distance from the eye using the “Holland the fields for wind speed, precipitation and storm B” wind formulation, which uses the central pressure surge. For illustration purposes, the tracks of Category to define a maximum wind speed. Then, for a given 5 tropical cyclones in the first 1,000 years of simulat- radius of maximum wind and radial profile shape factor ed activity are shown in Figure 20. The category of a “B”, the wind at a location relative to the center of the storm can change along the track. The category indi- storm can be determined. In the Southern Hemisphere, tropical cyclone winds rotate clockwise. The combined FIGURE 19. Comparison of annual rates of historical and simulated tropical cyclones in several locations in the Pacific Region Annual frequency of storms w/in 300 mile radius of island region 8 n Observed (IBTrk 1948-2008 7 n Observed (JMA/STI 1951-2008) n Simulated 10K 6 Annual frequency 5 4 3 2 1 0 ds i a oa ue u M SM M M M s u Ki i i i u on ea ds e u Fij t at at nd ng st ur la ba al at FS FS FS FS an an n m Ni rib rib Le Pa v ,F nu Na sla ui To iri u Sa Isl Isl a, o, ir, e, Ki T G ao ,K or Va lI lik ni a en ok i, al w m sr ar ba lo ak Pa W sh Ko Ne Ti Co m lim Co ra am ar lo Ta O a M So pu rit h Ki ut Pa So 36 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) FIGURE 20. Tracks of simulated Category 5 tropical cyclones in the first 1,000 years of activity effects of tropical cyclone winds and forward motion tional effects on wind speeds. In general, the rougher (or translational speed) will produce higher wind speeds the terrain, the more quickly wind speeds dissipate and on the left-hand side of the storm (Figure 21). A mir- smoothing algorithms are applied to provide realistic ror image, with highest wind speeds on the right-hand wind speed transitions between adjacent locations. side of the storm, applies to storms above the equator. FIGURE 21. Wind field cross section Differences in surface terrain also affect wind Hurricane Path speeds. Winds travel more slowly at ground level be- cause of the horizontal drag force of the earth’s sur- Eye face, or surface friction (Figure 22). The addition of Left Side Right Side obstacles such as buildings or trees will further de- grade wind speed. The model captures this effect by using a friction coefficient for each location of interest. LULC data for the PICs was processed and aggregated Rmax to the grid domain. Each land cover type has a dif- Stronger Winds Weaker Winds ferent “roughness value” that leads to different fric- Storm Center CATASTROPHE RISK ASSESSMENT METHODOLOGY 37 FIGURE 22. Terrain effects on wind velocity profiles however, a similar separation bubble manifests down- wind and counteracts, to some degree, the protection Vg Vg provided by the hill or escarpment. The surface rough- ness, elevation, and terrain slope along each of eight compass directions at each location are derived from high-resolution elevation data from the USGS and the LULC data developed for this project (see 1.4). hg (smooth) hg (rough) The validation of the modeled wind field was performed for nine tropical cyclones with the larg- est numbers of wind speed recordings available. Smooth terrain Rough terrain These are tropical cyclones Bob 1978 (Cat 4), Meli 1979 (flat and open) (urban or suburban) (Cat 3), Hina 1985 (Cat 4), Kina 1992 (Cat 3), Gavin 1997 (Cat 4), Ami 2003 (Cat 3) and Gene 2008 (Cat 3) In addition to these effects, the wind mod- in FJ and Betsy 1992 (Cat 2) and Ivy 2004 (Cat 3) in VU. el takes into account island topography. Wind Figure 24 shows an example of the validation exercise speeds increase on the windward slopes of mountains, for a) tropical cyclone Meli and b) Kina that hit FJ. The hills and escarpments because of amplification. Such simulated wind fields in miles per hour (mph) as well as features restrict the passage of wind causing the wind the observations (inside the circles where the stations to accelerate as it moves uphill. The slope of the incline were located) are color coded. A similarity of colors in- side and outside the circles implies a good agreement determines the degree of the amplification effect. If between simulated and observed values. the angle of incline is sharp, wind flow separates be- cause momentum near the ground is insufficient to overcome the pressure gradient at the top. A turbulent FIGURE 24. Tropical Cyclone Meli and Kina “separation bubble” develops, which increases dam- (a) Tropical Cyclone Meli hit FJ as a Category 3 in March ageability. The effect of topography on wind is shown 1979 in Figure 23. In the case of downhill winds, the lee- ward slope provides protection. If the slope is sharp, FIGURE 23. The effects of topography on wind WIND Decelerates e Accelerat Vd Vc Decelerates Va Vb a (b) Tropical Cyclone Kina hit FJ as a Category 3 in January WIND Vc Decelerates 1992 es erat Vc el Acc tes Decelera V Downwind separation bubble Va b Upwind separation bubble a WIND No change Separation bubble a Note: The blue line indicated the cyclone path (center). 38 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) A good agreement between the simulated amounts of rainfall with differences in rainfall in differ- and observed wind speeds is found for the nine ent quadrants of the storm and radial profiles of rain- storms considered, as indicated by the red circles fall rates outward from the center of the storm. These in Figure 25. This comprehensive view of the wind characteristics were captured and used as the base for speed validation also shows that simulated wind the rainfall model adopted in this study. Hourly rainfall speeds below 40 mph (green circles), which are gener- rates were computed for each grid cell based on the ally not damaging for structures, were artificially set central pressure of the storm and the relative location to zero. Values in blue were unreasonably low and of the grid to the center of the storm. An additional suggest possible malfunction of the instruments. One scaling factor adjustment was applied to the hourly such example, Kina in 1992, is shown in Figure 24 rainfall to account for basin-specific differences in rain- (b). Given the strength of the storm, maximum wind fall. This factor accounts for observed differences in speed observations at the three stations indicated by peak rainfall rates relative to storm intensity, with peak the arrows of 40 mph or less are unrealistic, given their rainfall rates in the PICs being some 30 percent greater locations with respect to the storm path. than those observed in the Atlantic. Further adjustment to computed rainfall was made to account for the im- FIGURE 25. Comparison of simulated versus observed pact of topography. Rainfall amounts were increased maximum 1’-sustained wind speed observations in upslope areas and decreased in down slope areas at a rate that corresponds to observed terrain effects with elevations greater than 100m. After computing the hourly terrain adjusted rainfall for each hour in the storm track, the values were integrated to calculate the accumulated total rainfall during the storm. A slow moving storm will impact a given location for a longer duration and hence result in higher total rainfall than if the same storm was moving at a faster speed (and shorter duration). The simulated total rainfall values were com- pared with observed values for several storms. The estimated precipitation amount at each ex- posure location was translated into flood depth (i.e., the water depth above the ground level) us- Note: Simulated maximum wind speeds below 40 mph are not ing the topographical information in the catch- retained and appear as zeroes on this plot. ment area. Figure 26 (a) shows an example of the drainage basins on the island of Viti Levu in FJ. Most of the drainage basin areas are small within each is- b. Rainfall-Induced Inland Flood land, and the difference of accumulated precipitation The storm information from the probabilistically amount within the basin is considered to be negligible. generated catalog of tropical cyclones described Within the basin, the flow accumulation number was for the wind field was also used to compute rain- computed at a 15 arc-second grid (around 500m), the fall patterns for each of the simulated storms. amount of upstream area (in number of cells) draining Cyclone-induced flooding results when heavy rainfall into each cell. The values range from 1 at topographic accumulates over the duration of the storm, which highs (river source) to very large numbers (on the order depends on a variety of factors including the intensi- of millions of cells) at the mouths of large rivers. Fig- ty of the storm, the forward speed of the storm, and ure 26 (b) shows the flow accumulation number at impacts of terrain on the wind precipitation. In gen- every 15 arc-second grid in a drainage basin near the eral, strong, slow moving storms produce the greatest Navua River on the south side of the island of Viti Levu. CATASTROPHE RISK ASSESSMENT METHODOLOGY 39 Most of the hydrological information was obtained in the absence of the storm. Storm surge is estimated from HydroSHEDS (http://hydrosheds.cr.usgs.gov). Us- by subtracting the normal or astronomic high tide from ing the flow accumulation numbers in the surround- the observed or simulated storm tide and illustrated in ing cells from each exposure location, a flood factor Figure 27. The largest value of storm surge ever record- was developed and calibrated with the observed flood ed worldwide was produced by Cyclone Mahina, which events. This flood factor and the accumulated precipi- caused a 43 foot (13 meter) storm surge at Bathurst tation were combined to produce flood depth at each Bay, Australia in 1899. exposure location. FIGURE 27. Illustration of storm surge FIGURE 26. Example of drainage basins FJ The storm surge model relies on primary me- (a) Drainage basins on the island of Viti Levu in FJ with the explicit river network teorological variables, including central pressure, forward speed, and radius of maximum wind. Other quantities considered are wind profile, location of the site with respect to the storm track and ba- thymetry (the seafloor topography). In general, gentle sloping bathymetry and wide coastlines are more con- (b) Flow Accumulation number in the 15 arc-second grid cells in drainage basin near Navua River on the island ducive to surge while steeper sloping bathymetry and of Viti Levu coastlines that are common in the PICs do experience 178’0°E less severe surge levels. Like the wind field, the profile Flow Accumulation Value High: 100 of the storm surge field is not symmetrical around the High: 1 storm track. All else equal, in the Southern Hemisphere the surge will be higher at a site to the left of the tropi- 18’0°S 18’0°S cal cyclone track than to the right, while the opposite holds for the surge in the Northern Hemisphere. Fig- ure 28 shows the comparisons between the observed and simulated storm surge heights at the observation location for historical events. In general, there is a good agreement shown. 178’0°E 2.3 Earthquake Event Generation The Pacific Region is one of the most actively c. Coastal Flood seismic regions in the world. It is surrounded by the Pacific “ring of fire,” where approximately 90 Tropical cyclone-induced surge is an abnormal percent of the world’s earthquakes and 80 per- rise in sea level accompanying intense storms. T he cent of the world’s largest earthquakes occur. surge height is the difference between the observed level of the sea surface and the level that would have occurred 40 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) FIGURE 28. Comparison of observed and simulated storm surge heights for historical events 6 Simulated surge height (m) 5 4 3 2 1 0 0 1 2 3 4 5 6 Observed surge height (m) 1967 Sally 1979 Meli 1982 Isaac 1983 Oscar 1985 Hina 1986 Sally 1990 Ofa 1992 Betsy 1997 Hina 1998 Cora In order to obtain a relatively complete and CAT catalog, 48,529 are from the Engdahl catalog, homogeneous catalog for the entire region, 58,355 are from the QUAKES catalog and 42 are from three historical earthquake catalogs were uni- literature. The catalog is considered to be complete formly processed and merged. They are the PAGER- from 1964 for events of magnitude 5.3 or greater. For CAT catalog (1900 – 2009), the Engdahl et al. relo- FJ and its vicinity, the list of events of magnitude 6.5 cated global earthquake catalog (1964 – 2004) and and greater can be considered complete since 1910. the QUAKES Data catalog by Geoscience Australia For other regions, events of magnitude 7.0 and greater (1958 – 2009). Before the merging could take place are complete since 1900. the magnitude values assigned to each event needed to be converted to a common scale, in this case mo- Due to the remoteness of the region and poor ment magnitude (Mw). As is customarily done in haz- coverage of local seismographs, about 20 percent ard assessment studies, aftershocks and foreshocks of earthquakes in the historical earthquake cata- were removed from the merged catalog. Given their log did not have an assigned hypocenter depth time limitation the three historical earthquake cata- (the point where the fault begins to rupture). Fig- logs mentioned above did not contain 42 well-known ure 30 (a) shows the distribution of depths for earth- events (some pre-1900) that were identified in various quakes (Mw≥ 5 only) whose depth values are known publications. These additional events were incorporat- in the region that spans from PG to TO. From this da- ed into the final catalog. taset, empirical probability density functions of depth Figure 29 shows the epicenters of the 32,569 for different areas in the region were derived and used historical earthquakes with M≥5.0 that occurred to probabilistically simulate the depth of events whose from 1768 to 2009 and are included in the final real focal depth is unknown. Figure 31 shows a com- combined historical catalog. The oldest event re- parison of empirical distribution of focal depths for corded was a Mw7.5 earthquake that occurred on events in a region of PG vis-à-vis the simulated distribu- June 22, 1768 approximately 50km to the south of tion for events with originally unknown depth. Figure Latangai Island of PG. The catalog includes 114,131 30 (b) shows the simulated depths of all these events earthquakes, out of which 7,205 are from the PAGER- for the PG to TO region. CATASTROPHE RISK ASSESSMENT METHODOLOGY 41 FIGURE 29. Epicenters of the more than 32,000 significant earthquakes that occurred from 1768 to 2009 included in the final historical earthquake catalog developed for the Pacific Region Note: The size of the circles is proportional to the event magnitude FIGURE 30. Distribution of depths for earthquakes (a) Depth distribution of events (Mw≥5 only) with (b) Simulated depth distribution of all of the events with reported depth in the PG and SB region unassigned depth in the original catalogs Note: The numbers in red refer to the different seismotectonic areas included in the hazard model. 42 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) FIGURE 31. Focal depths of earthquakes 0.25 0.2 0.15 Density 0.1 0.05 0 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 220 230 240 250 260 270 280 Depth (km) Note: Empirical distribution of known focal depths of earthquakes in the shaded area of PG is shown in green while the simulated distribution of focal depths of events with unknown depth in the same area is shown in blue. The numbers in red (insert) refer to the different seismotectonic areas included in the hazard model. More than 90 percent of all earthquakes in tion of the subduction zones is a constraint on size the world occur at boundaries of tectonic plates. of mega-thrust earthquakes which cause most of the Among the 15 PICs studied in this project, PW, PG, seismic hazards in this region. Subduction zones were WS, SB, VU, TL, and TO are located on or close to plate segmented according to previous studies by Nishenko, boundaries, which in this area take the form of sub- 1991 with adjustments based on the most recent his- duction zones (Figure 32). Very large events have and torical earthquake ruptures. will occur only on subduction zones. The segmenta- FIGURE 32. Subduction zones (a) Subduction zone in the South Pacific Region (b) Subduction zone segmentation developed by Nishenko (1991) Note: Abbreviations are: NB-New Britain subduction zone; S-SB subduction zone; VA-VU subduction zone; and TK-TO-Kermadec subduction zone. CATASTROPHE RISK ASSESSMENT METHODOLOGY 43 In addition to the subduction of plates, crust- on known faults, b) large interface earthquakes al faults also pose a threat. For example, the 1953 on subduction zones, c) other large subduction Suva earthquake, the most damaging earthquake to earthquakes, such as normal faulting intraplate occur in FJ, took place on a near-shore crustal fault and outer rise events and d) shallow background (Figure 33). Unfortunately, not many active crustal and deep events. The shallow background and deep faults are known in this region and for many of the seismicity were modeled with a gridded seismicity ap- known faults there is insufficient information to esti- proach. Gutenberg-Richter a- and b-values were de- mate their seismicity parameters. Faults that are poten- termined using historical earthquake data and the up- tially active have been identified in Viti Levu (FJ). per bound magnitudes were determined from regional tectonics and the largest magnitudes in the historical Geodetic data was used to identify the direc- catalogs. The rest of the regional seismicity was mod- tion of movement of the tectonic plates and to eled based on seismotectonic setting and historical constrain the plate velocities. A total of 254 geo- seismicity of the region by means of 60 source zones of detic Global Positioning System (GPS) velocity vectors homogeneous activity (Figure 34). These sources in- from various sources were collected. The spatial cov- clude subduction zones, fore arc and back arc regions, erage of GPS data used was very irregular. Along the transforming zones and background area. The seismic- subduction zone, PG and VU had the best coverage, ity in each source zone is modeled at two depth lay- whereas SB did not have any GPS measurements. The ers: a shallow layer featuring subduction interface and 254 GPS velocity vectors and available active faults shallow crustal events, and a deep layer featuring sub- data were used to invert the strain rate field for the duction intra-plate and deep earthquakes. Character- region and to develop a kinematic model of the re- istic earthquake magnitudes and the mean recurrence gion. The modeled velocities were consistent with the intervals of earthquakes on the subduction segments observed values. were determined using historical rupture information The total seismicity in the South Pacific Re- and plate tectonic data. The seismicity on the known gion can be attributed to a) crustal earthquakes faults was modeled by estimating the recurrence rates FIGURE 33. Traces of the potentially active faults on the island of Viti Levu (FJ)2 and location of the 1953 Suva Earthquake Excerpted from Rahiman, 2006. 2 44 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) FIGURE 34. The 60 seismogenic source zones along with the background seismicity constitute the entire seismicity model for the region of characteristic earthquakes from available slip rates earthquake at any subduction zone. Large mega-thrust and historical earthquake data. Magnitude uncertain- earthquakes on the New Britain and SB, the VU, the ties were considered for characteristic earthquakes on TO and the Kermadec section of the subduction zone faults and on subduction zones. For the subduction were modelled. The estimated mean recurrence inter- zones, a three-dimensional model was developed to vals of Mw ≥ 9 earthquakes on those sections was characterize the spatial distribution of subduction re- approximately 600 years, 600 years, 1400 years and lated interface and intra-plate earthquakes. 2700 years, respectively. It is possible that large, yet infrequent me- Figure 35 shows the comparisons of simu- ga-thrust events will break a large portion of lated and observed event frequencies for earth- the subduction zone. The 2004 Andaman-Sumatra quakes. The top left panel, for example, shows the Earthquake broke an estimated 1,600km-long section region around PG. Events of magnitude close to 8 or of the subduction zone where the India Plate slides larger occur, on average, once every 10 years (i.e., an- under the overriding Burma Plate. The largest record- nual rate of about 0.1). The blue line represents simu- ed historical earthquake in the studied region had a lated events and extends beyond the green line, which magnitude of about 8.5. Because the limited length represents observed data. This is because earthquakes of the historical catalog, the possibility that a magni- of that magnitude have not been observed in the his- tude 8.5-9.0 mega-thrust event could occur in this re- torical record but the subduction zones in the Pacific gion cannot be ruled out, especially in the trench from are capable of generating them, albeit very rarely. For New Britain to the SB where the subduction plate is example, until 2011, magnitude 9 events such as the young and fast moving. Further scientific work sug- Tohoku earthquakes in Japan had never previously gests that present evidence cannot rule out a Mw ≥ 9 been observed on that part of the trench. CATASTROPHE RISK ASSESSMENT METHODOLOGY 45 FIGURE 35. Comparison of annual rates of historical and simulated earthquakes in the PG region (top left), SB region (top right), FJ region (bottom left), and TO region (bottom right) Events Events 1000 1000 100 100 of events of events Simulated Simulated 100 100 10 10 Historical Historical Numberof Numberof 10 10 annual number annual number 1 1 1 1 0.11 0.1 Cumulative Annual Cumulative Annual 0.1 0.1 0.01 0.01 0.01 0.01 Cumulative Cumulative 0.001 0.001 0.001 0.001 0.0001 0.0001 0.0001 0.0001 5 5 6 7 8 9 10 10 5 5 6 6 7 7 8 9 10 Magnitude Magnitude Magnitude Magnitude Events Events 100 100 100 100 ofevents ofevents Simulated Simulated 10 10 10 10 Historical Historical Numberof Numberof annual number annual number 1 1 1 0.11 0.1 0.11 0.1 Cumulative Annual Cumulative Annual 0.01 0.01 0.01 0.01 Cumulative Cumulative 0.001 0.001 0.001 0.001 0.0001 0.0001 0.0001 0.0001 5 5.5 5.5 6 6.5 7 7.5 7.5 8 8.5 8.5 5 5 6 6 7 7 88 9 9 10 10 Magnitude Magnitude Magnitude Magnitude Large earthquakes that occur on the “ring of and tsunami waves for the 7.6 million earthquakes fire” are capable, under certain circumstances, of were computed in the hazard module. To account for generating major tsunamis that can travel great their large uncertainty, the ground motion fields for distances. Some of the PICs, such as PG, TO, the SB each earthquake were generated 100 times and then and VU are located on top of or close to the sources grouped in 10,000 potential realizations of what is of these earthquakes. Others, such as CK, MH, and KI possible to occur at any one point in time. Epicenters are more distant (Figure 36). No PIC, however, is com- of events with a magnitude of 7 or larger are shown in pletely immune to the far reaching effects of earth- Figure 36. In addition, the catalog of simulated earth- quake induced tsunamis. quakes also includes selected large magnitude events in South and North America, Japan and the Philippines. The catalog of simulated earthquake events While these events are too far to cause a significant spans the entire Pacific Region from Taiwan to level of ground shaking in the Pacific Region, they New Zealand and from Indonesia to east of Ha- could potentially generate tsunamis capable of causing waii. The fields of ground motion intensity measures damage in the PICs. 46 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) FIGURE 36. Epicenters of simulated events considered as potential sources of tele-tsunamis (a) Epicenters of 50,722 simulated events with magnitude between 7 and 8 in the 10,000 year catalog (b) Epicenters of 2,686 simulated events with magnitude larger than 8 along the Pacific Rim CATASTROPHE RISK ASSESSMENT METHODOLOGY 47 2.4 Earthquake Intensity Calculation The effects of local soil conditions on the ground motion characterization were accounted a. Ground Shaking for using shear wave velocity maps. Given the lack The ground motion that earthquakes generate in of available recordings, the predicted median values of the region is dependent on the location of the rup- horizontal peak ground acceleration (PGA) from the ture with respect to the site, the dynamic of the GMPEs were compared with the few inferred values ex- rupture, the traveling path of the waves from the tracted from the United States Geological Survey (USGS) source to site and the soil conditions at the site. “Did You Feel It?” maps (http://earthquake.usgs.gov/ The use of empirical data extracted from records of past earthquakes/dyfi/). Observed reports of ground shaking earthquakes of similar characteristics, supplemented by were available only for the seven earthquakes in Table science and analytical simulations, is the ideal procedure 15. A comparison between observed ground shaking to shed some light in those areas where data are scarce with predicted values for two of these earthquakes is (e.g., ground motion generated by large earthquakes at shown in Figure 37. The circles in these maps represent short distance from the rupture). Unlike in the Western the observations. A match between the color within the United States or Japan, where records from past earth- circles with the color of the surrounding map indicates quakes are plentiful, they are very rare in the Pacific Re- that the value of the PGA is close to the median value gion and nonexistent in many PICs. The ground motion predicted by the GMPEs. The agreement is generally is calculated under the generally tenable assumption very good. Given the large uncertainty in the ground that the attenuation of seismic waves in different regions motions predicted by GMPEs for any given earthquake, of the world with the same tectonic setting is very simi- some discrepancy between observed and predicted lar. This means, for example, that large subduction zone ground motion values is to be expected. earthquakes that occur in the Pacific regional trenches generate ground motion fields similar to those gener- TABLE 15. Regional earthquakes for which observed ated by events along the plate boundaries elsewhere in reports of ground shaking are available the world (e.g., Nazca Plate in South America, or Casca- Year Magnitude Country dia in North America). In the absence of regional data, it 2006 7.9 Tonga is customary to use ground motion prediction equations 2007 8.1 Solomon Islands (GMPEs) based on data from other parts of the world, as 2007 7.2 Vanuatu was was done in this study. 2007 6.8 Papua New Guinea 2009 5.5 Vanuatu 2009 6.1 Vanuatu 2009 8.1 Samoa FIGURE 37. Comparison of observed PGA values at selected locations (circles) with median values extracted from the GMPE used in this study for subduction zone events (a) Magnitude 8.1 earthquake Sept. 2009, WS Perceived shaking Not felt Weak Light Moderate Strong Very strong Severe Violent Extreme Potential damage None None None Very light Light Moderate Moderate/Heavy Heavy Very heavy Peak AOC (%g) < .17 .17 - 1.4 1.4 - 3.9 3.9 - 9.2 9.2 - 18 18 - 34 34 - 65 65 - 124 > 124 Instrumental intensity I II - III IV V VI VII VIII IX X+ Observation location PGA value inferred from shaking report (circle filled with inferred PGA value) 48 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) (b) Magnitude 7.9 earthquake May 2006, TO Tongatapu Vava’u Observation location PGA value inferred from shaking report (circle filled with inferred PGA value) Perceived shaking Not felt Weak Light Moderate Strong Very strong Severe Violent Extreme Potential damage None None None Very light Light Moderate Moderate/Heavy Heavy Very heavy Peak AOC (%g) < .17 .17 - 1.4 1.4 - 3.9 3.9 - 9.2 9.2 - 18 18 - 34 34 - 65 65 - 124 > 124 Instrumental intensity I II - III IV V VI VII VIII IX X+ b. Tsunami Waves on thousands of islands, the tsunami waves of 111 representative events were explicitly modeled (Fig- 1) For the purpose of tsunami hazard and risk ure 38). These 111 events comprise all local events modeling, the earthquakes in the stochastic with M ≥ 8 (57 events) and a selected set of 54 dis- catalog were divided into two categories: Lo- tant events with M ≥ 8 along the Pacific Rim. The cal earthquakes with M ≥ 7.7 (4,984 events) and dis- tsunami waves of the remaining 5,456 events were tant earthquakes with M ≥ 8.0 (583 events) along implicitly modeled via a wave amplitude and veloc- the Pacific Rim. These are considered significant ity scaling technique discussed below. These 5,456 tsunamigenic events. Given the intensive computa- earthquakes include all local and distant events with tions required to model the propagation of tsunami 7.7 ≤ M < 8 and all other distant events with M ≥ 8 waves across the Pacific Ocean and their inundation whose waves were not explicitly modeled. FIGURE 38. Epicenters of the 111 significant tsunamigenic earthquakes with M ≥ 8.0 on local and circum-Pacific subduction zones whose waves were explicitly modeled CATASTROPHE RISK ASSESSMENT METHODOLOGY 49 2) Events with M < 7.7 whose potential for extensive modeling efforts or theoretical considerations and run- tsunami is relatively minor and, therefore, whose up found amounts to about 2-3 times the near-shore tsunami waves were not modeled. wave-height for a range of realistic slopes, such as wide and narrow bays. An exception is the location The explicit tsunami wave modeling tech- at the apex of the bay geometries, with localized am- nique adopts a hybrid approach. Near-shore wave plifications of 4-6. The adopted sloping beach run-up heights (or wave amplitudes, see Figure 39 ) are com- model yields amplification ranging from 1 for low-lying puted using linear long-wave approximations and flat areas (topography < 5 meters), to 3 for areas with semi-empirical relations are used to infer the inunda- higher topography. For the vast majority of locations, tion run-up distance from the offshore wave heights. the amplification is close to 3. The M9.1 earthquake off the shore of Peru was extracted from the stochastic For the simple geometry of a sloping beach and catalog. Using the hybrid technique discussed above, simple wave characteristics, the run-up can be com- wave heights (in cm) were then generated in several of puted accurately using a simple relationship, where the the PICs. An example of the wave heights generated amplification is primarily dependent on the slope of the around SB are shown in Figure 40. beach. In more complex environments, other relation- ships have been derived from wave tank experiments, FIGURE 39. Definitions of tsunami wave characteristics Inundation depth FIGURE 40. Wave heights around the SB -117 – 25 26 – 35 36 – 45 46 – 55 56 – 66 67 – 76 77 – 86 87 – 106 107 – 157 158 – 2,469 50 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) FIGURE 41. Comparison of simulated and observed wave heights and marigrams (graphic record of the tide levels at a particular coastal station) for the M8.1 2008 SB earthquake. The comparison is computed at Honiara Observed Sunday April 1, 21:06:59, UTC 2007 0.2 Modeled 2 earthquakes on this map 155” 160” 165” 0.1 -5” -5” 0.0 -0.1 -10” -10” 0 5 10 15 20 25 30 35 Time (x 1000 sec) Latitude Longitude Observed (m) Modeled (m) -7.65 156.5 1.9 - 4.4 3.5 -7.7 156.5 2.9 2.8 -7.72 156.53 2.8 2.4 -8.25 156.53 5.2 - 9 2.3 -15” -15” -8.26 156.55 2.8 2.8 155” 160” 165” -8.09 156.84 1.7 1.2 -8.06 156.76 2.3 1.8 -8.3 156.85 3.7 2.1 -8.34 157.28 1.1 1 In cases where the bathymetry and the digital the waves generated by, for example, a given M7.7 terrain models are accurate, this approach leads earthquake were estimated by first searching the clos- to accurate representations of tsunami waves. For est event for which waves were explicitly modeled and example, Figure 41 shows a comparison between mod- by scaling the wave amplitude by an appropriate factor eled and observed wave heights at several locations in depending on the magnitude difference. For example, the SB generated by the 2008 M8.1 earthquake. if the closest event that was explicitly modeled was a A magnitude scaling technique was devised M8.7 earthquake, then its wave heights were down- for estimating the wave amplitude and velocity scaled by a factor of 3.5 and its wave velocity by a fac- of the remaining significant tsunamigenic events tor of 1.87, the square root of 3.5. whose waves were not explicitly modeled. This technique is based on computing the ratio of these 2.5 Ancillary GIS Data wave parameters at the same sites generated by earth- To enable an accurate estimate of hazard and risk quakes of similar rupture location but different mag- assessment for both earthquakes (ground shak- nitude. ing and tsunamis) and tropical cyclones (wind, The simulated datasets have shown that wave precipitation and storm surge) several geo-refer- heights are sensitive, for example, to the event enced datasets were needed. A suit of GIS maps for parameters, slip model along the rupture, wave each country was generated. Discussed here are: direction and coastal shape. Two earthquakes with similar locations but with magnitudes that differ by ■■ Bathymetry maps, needed for the computation of one unit generate wave heights at any given location tsunami-induced waves and of storm surge due to that differ, on average, by a factor of 3.5. Therefore, tropical cyclones; CATASTROPHE RISK ASSESSMENT METHODOLOGY 51 ■■ Topographic maps; The global bathymetry data for the Pacific Ocean is based on the SRTM30 PLUS database ■■ Surface geology maps (when available, since they and consists of a 30 arc second grid (approxi- are needed to develop soil maps); mately 1km) that covers the region of interest ■■ Soil maps, to determine the amplitude and the fre- (Figure 42). In addition, local higher resolution data quency content of earthquake ground shaking; and was also made available for several countries (FJ, PG, WS, SB, TO, TV, and VU) with a resolution that varies ■■ LULC maps to compute surface roughness, which from 3 arc seconds to 8 arc seconds (approximately is influential in estimating wind speed at surface 90m to 250m). There are also local bathymetry con- generated by tropical cyclones and the amount of tour data for parts of PG, PW, FM, and NU and high- precipitation runoff which is necessary for estimat- resolution bathymetry raster data for small areas in VU ing runoff flood risk. and FJ. High-resolution bathymetry contour data was available for TL. FIGURE 42. Bathymetry map based on SRTM30 Plus dataset 52 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) Topography data was provided by NASA’s should be noted that in some cases discrepancies may Shuttle Radar Topography Mission. This dataset be found between the values of Vs30 estimated by this covers the entire domain of the 15 covered nations and method and those that may be measured in the field. If had a resolution of 3 arc seconds (roughly 90 meters). detailed surface geology maps at a regional scale such Geologic maps show the distribution of geologic fea- as those customarily developed for micro-zonation tures, including different types of rocks of a given age studies were to become available, they should be used range. They are useful to infer the depth, and stiffness for earthquake ground motion assessment in lieu of of soil sediments that may be present and to estimate those developed here. the amount of amplifications that seismic waves may LULC maps are used to determine roughness be subject to when filtered by these soil units. The factors and precipitation runoff percentages. geologic maps that were collected during the course Land cover refers to the physical and biological cover of the project are limited in number, coverage, and in over the land surface, including water, vegetation, bare some cases quality, which was expected. No country or soil and artificial structures. Land use usually refers to regional scale site-condition maps based on detailed signs of human activities such as agriculture, forestry surface geology maps are available for the 15 PICs. In and building construction that altered the original land this case, it is customary to use the method developed surface processes. The LULC maps were developed us- by Allen and Wald, which uses topographic data as a ing remotely sensed data (i.e., satellite imagery) of dif- proxy for site conditions. The methodology circumvents ferent resolution and vintage, validated with the aid data inconsistencies and is widely used for hazard and of some ground truthing, virtual truthing (using high- risk assessment purposes around the globe. The soil resolution imagery of more recent vintage and other maps derived using this methodology show the shear internet resources), agriculture census and other ancil- wave velocity of seismic waves in the top 30m of soil, lary data collected during the course of the project. which is denoted as Vs30. High values of Vs30 (e.g., Given the methodology adopted, it is to be expected greater than 760m/s) refer to soft and hard rock site that in some instances the information included in the conditions, which show no significant amplification of LULC maps may be obsolete and inaccurate. The LULC incipient seismic waves. Very low values of Vs30 (e.g., maps developed here are, however, perfectly suitable lower than 180m/s) refer to very soft soil sites where for the scope of assessing wind and flood hazard and significant amplification is expected. Average medium were also developed for establishing a crop exposure to stiff soil conditions have Vs30 values in the 300 to database (see Section 1.4). 500m/s range. Although these maps are developed according to the current state-of-the-art approach, it CATASTROPHE RISK ASSESSMENT METHODOLOGY 53 3. Damage Estimation The third step in the risk assessment procedure displayed in Figure 2 deals with dam- age estimation. This required knowing the vulnerability of crops, structures and probable casu- alty rates for occupied structures that are damaged by the impacts of earthquakes and tropical cyclones. The risk profiles were developed as the final step of the risk modeling. The adverse consequences were measured in terms of economic losses to buildings, infrastructure and crops and by the number of casualties among the affected population. 3.1 Consequence Database a. Data Sources Consequence data from historical natural disasters impacting the 15 PICs was collected from a large variety of sources. Most of these countries are prone to multiple hazards, al- though not all at the same level of severity. A summary of potential natural hazards extracted from the CIA World Factbook3 for these PICs is listed in Table 16. TABLE 16. Potential Natural Hazards of the 15 PICs according to the CIA World Factbook Country Potential Natural Hazards CK Tropical cyclones (November to March) FJ Tropical Cyclones (November to January) FM Typhoons (June to December) Typhoons can occur any time, but usually November to March KI Occasional tornadoes Low level of some of the islands make them sensitive to changes in sea level MH Infrequent typhoons NR Periodic droughts NU Typhoons Active volcanism Frequent and sometimes severe earthquakes PG Mud slides Tsunamis PW Typhoons (June to December) Typhoons, but rarely destructive SB Geologically active region with frequent earthquakes, tremors, and volcanic activity Tsunamis Floods and landslides are common Earthquakes TL Tsunamis Tropical cyclones Tropical Cyclones (October to April) TO Earthquakes and volcanic activity on Fonuafo’ou Severe tropical storms, usually rare TV Low level of islands make them sensitive to changes in sea level Tropical cyclones or typhoons (January to April) Volcanic eruption on Aoba (Ambae) island began on 27 November 2005 VU Volcanism also causes minor earthquakes Tsunamis Occasional typhoons WS Active volcanism The World Factbook, Washington DC: Central Intelligence Agency (CIA), retrieved October 1, 2010, https://www.cia.gov/library/ 3 publications/the-world-factbook. 54 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) FIGURE 43. Distribution of natural hazards by peril and country Peryl type Country Storm surge 1% Landslide 4% Cook Islands 7% Severe local Vanuatu 13% Micronesia 3% storm 2% Flood Tsunami 1% 11% Tuvalu 2% Tonga 7% Timor Earthquake 31% Leste 1% Fiji 21% Solomon Marshall Islands 11% Islands 1% Tropical cyclone 50% Niue 1% Samoa 5% Palau 1% Kiribati 0% Naunu 0% Papua New Guinea 27% The natural hazards considered in the conse- Epicenters database (PDE), the U.S. Geological Survey quence database included tropical cyclones and (USGS) PAGER-CAT earthquake catalog (which is a col- earthquakes (and resulting tsunamis). Other relat- lection of existing data mainly from NGDC, UTSU-CAT, ed hazards such as severe storms (i.e., torrential rains, and PDE), the dataset “Natural Disasters in the Pacific” strong winds), floods, storm surges, landslides/mud- (maintained by the Australian Government agency Au- slides and tsunamis from earthquakes outside of the sAID), the disaster database (maintained by the Pacific region were also considered. The distribution by peril Disaster Network in conjunction with the Pacific Disas- type and country is shown in Figure 43. Volcanic di- ter Risk Management Partnership Network), the Uni- sasters were not explicitly considered for the database. versity of Richmond Disaster Database Project and the Although this type of peril is prevalent in the South Global Active Archive of Large Flood Events maintained Pacific, especially in PG where volcano eruptions have by the Dartmouth Flood Observatory (DFO). Data was proven to be particularly deadly (e.g., the 1951 Mt. also gathered from other sources, including scholarly Lamington eruption, which killed over 3000 people), articles, text books, encyclopedias, reports issued by volcanic hazards are outside the scope of this project. governmental agencies and news articles. Over 600 unique disaster entries have been collected for the Data collection was extensive and the conse- consequence database, each of which had some ac- quence database represents a comprehensive in- count of a notable effect to the population or damage ventory of recorded natural hazards that have had a to the building inventory. In addition, information was significant impact on the population. The majority of captured for around 50 events for tropical cyclones and the data collected for the database was aggregated five for earthquakes, when multiple countries were af- from a number of major disaster databases, both pub- fected by the same event. The EMDAT catalog provides licly and privately available. The databases included data for about 30 percent of the entries in the conse- the Emergency Events Database (EMDAT), the Natu- quence database; likewise, the NGDC, Utsu, NatCat- ral Catastrophe Loss Database (NatCatSERVICE), the Service, and AusAID databases provide data for about National Geophysical Data Center (NGDC) Significant 15, 18, 43, and 37 percent of the entries, respectively. Earthquake Database, the Historical Tsunami Database (HTD), the Catalog of Damaging Earthquakes in the The database assembled is more comprehen- World (UTSU-CAT), the Preliminary Determinations of sive than past databases, as no prior existing da- CATASTROPHE RISK ASSESSMENT METHODOLOGY 55 tabase covers a majority of the entries. However, event, including those that became homeless, in- many entries, especially those from very damaging jured, displaced, evacuated, or disrupted (e.g., af- events, contain data from multiple sources, leading to fected by loss of utilities) by the peril. discrepancies in the quantitative data, particularly eco- ■■ Number of People Homeless – A subset of the nomic losses. By design, the discrepancies have been number of people affected, indicating the number preserved and each piece of data in the consequence of people required to vacate their residence due to database is appropriately referenced. the peril, such as those evacuated or displaced. Great effort was taken to map the events in ■■ Estimated Total Economic Loss – The estimated to- the consequence database to the historical cata- tal economic impact of the event, usually consist- log of tropical cyclones and earthquakes devel- oped in this study and described earlier. Over 80 ing of direct (e.g., damage to infrastructure, crops, percent of the earthquake entries have been mapped housing) and indirect (e.g., loss of revenues, un- to events in the earthquake historical catalog. Likewise, employment, market destabilization) consequences over 85 percent of the tropical cyclone entries have on the local economy. Estimated loss is typically re- been mapped to events in the tropical cyclone histori- ported in U.S. dollars (US$), corresponding to the cal catalog. Most of the events that are not tracked are monetary loss at the time of the event (e.g., cur- older events (i.e., prior to 1900 for earthquakes and rent/nominal US$). Some data were reported in lo- prior to 1948 for cyclones), which are, for the most cal currencies and were converted appropriately by part, are not archived in the historical catalog. using filtered rates for specific (time-of-event) dates based on information supplied by leading market b. Explanation of Data Fields data. Losses are reported as the monetary cost at the time of the event, as well as costs trended to The referenced sources discussed above typi- current values using a macro-economic exposure cally report a brief summary of the disaster con- sequence (e.g., number of people affected and/ growth parameter (as discussed below). Losses are or number of lives lost), and some accounts are typically listed in the consequence database as the strictly qualitative (e.g., “buildings and crops total direct economic loss. Break-down losses are were damaged”). For each entry in the consequence available for those events where detailed assess- database, data from each field is typically an aggre- ment reports were issued, e.g., losses per sector gate account of the total consequence from a particu- (social sector, private sector, infrastructure, etc.), lar disaster event (including related secondary events/ crop losses, and locations for deaths and building effects), with most or all of the damage occurring in damages. the country listed. For earthquake events, the losses are ■■ Total Life Loss – The total number of people re- aggregated for ground shaking and the resulting wave ported dead, missing or presumed dead as a result impact(s) and specific details on the relative losses are of the event, including any resulting deaths from noted for some events. Some events, like those listed starvation, injury, or disease. as floods, landslides, severe storms and storm surges, were not directly linked to reported earthquakes or ■■ Total Injured – The total number of people suffering tropical cyclones. In addition, some tsunami events from physical injuries, trauma or an illness requiring were a result of earthquakes occurring outside the medical treatment as a result the event. South Pacific Region (e.g., the 1877 tsunami in FJ was ■■ Buildings Damaged or Destroyed – The total num- caused by very large earthquake near Chile). ber of buildings (typically listed as “houses”) re- ported to be damaged or destroyed as a result the The main data fields of the consequence database event. While quantitative data of damage is some- are: times reported (e.g., the total number of houses ■■ Total Number of People Affected – A measure of destroyed), much of the data is qualitative (e.g., the estimated number of people affected by the “some houses were damaged.”) 56 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) ■■ Crop Damage – Mainly a qualitative descriptor that 1980s, with the exception of FJ and PG, which had indicates evidence of damage/destruction to the reported values since 1960. Of the 256 entries in the local agriculture, including crops, vegetation and consequence database that report monetary loss, about livestock. one fifth occurred in older years for which economic data was not available. To provide an approximation of The “Significant Earthquake Database” (NGDC) the trended losses for these old events the trend factor, and tthe “Historical Tsunami Database” frequently re- calculated from the above equation, was taken as the port data with qualitative descriptors, given as a four- oldest calculated trend factor for which data is available. level scale which represents estimated ranges of val- This approximation provided a qualitative (and possible ues. This qualitative data was converted to numerical lower-bound) assessment of estimated present-day loss- values based on definitions given by the NGDC. Thus, es since exposure growth typically increases with time. some entries contain multiple reported values, either An example of the consequence database, which lists due to the range of values just mentioned or reports some of the most devastating perils ever recorded for from multiple sources as discussed above. The range of the 15 PICs, is shown in Annex E. values is indicated as high and low estimates. c. Economic Loss Trending TABLE 17. 2009 Economic and population data for the 15 PICs Since monetary loss is usually reported in current (nominal) US$ at the time of event, a macro-eco- GDP per Capita GDP Country Population (US$) (US$, million) nomic approach was used for the estimation of CK 20,0001 10,9072 218.12 present-day losses due to exposure growth. The FJ 849,218 3,573 3,034.4 following equation was used to adjust the historical FM 110,728 2,319 256.8 event losses to the present-day losses: KI 98,045 1,325 129.9 MH 61,026 2,504 152.8 NR 10,000 1 2,396 2 24.02 NU 1,0001 5,8003 5.83 Where, PG 6,732,159 1,172 7,892.8 PW 20,398 9,345 190.6 Lx = Loss at year x in US$ SB 523,170 1,257 657.5 POPx = National population at year x TL 1,133,594 492 558.0 TO 103,967 2,991 311.0 GDPx = Real (constant) GDP per capita at year x TV 10,000 1 3,213 2 32.12 DFLx = GDP deflator at year x VU 239,788 2,713 650.5 x = year when the event occurred WS 178,846 2,776 496.5 Note: All data from the World Bank (2010) unless otherwise noted 1 Estimated value from the United Nations, UNDATA http://data.un.org In the above equation, the population growth ap- 2 2008 value from the UN proximates the increase in the number of assets over 3 CIA World Factbook time. The real GDP per capita growth approximates the wealth increase over time (which is somewhat related d. Database Statistics to the material and labor costs). The GDP deflator (de- fined as the nominal GDP divided by the real GDP) ap- This section outlines key statistics of the conse- proximates inflation over time. Present-day accounts of quence database, with the main intent of provid- these parameters are listed in Table 17. for all countries ing a summary of recorded disaster data and pre- considered. Note that economic data for the nations in- senting a qualitative overview of consequences vestigated is typically reported only as far back as early from natural disasters occurring in the 15 PICs CATASTROPHE RISK ASSESSMENT METHODOLOGY 57 considered. The number of entries recorded for each catastrophic entries occurred in PG and FJ, and many event type and country is presented in Table 18. Over catastrophic entries also occurred in WS, SB, VU and 600 entries have been recorded, of which 150 are de- TO. None or very few disaster events have been report- fined as catastrophic, i.e. reported with at least 10,000 ed for some of the nations considered, especially KI, people affected, 10 million un-trended losses in US$ or NR, NU and PW. Few events have also been reported 10 deaths (their number is displayed in parentheses). for TL, even though this region is subject to multiple Around half of the catastrophic entries are directly re- hazards (Table 18). One possible reason for the lack lated to tropical cyclones and over a quarter to earth- of reported damage is this nation recently gained in- quakes, of which over half were reported with an as- dependence from Portugal and Indonesia, and explicit sociated wave impact. More than half of the recorded records for this nation are not readily available. TABLE 18. Number of database entries for each disaster type and country. Tropical Severe local Peril/Country Earthquake cyclone Tsunami storm Flood Storm surge Landslide Total CK 1 (0) 42 (4) 1 (0) 1 (0) 1 (0) 1 (0) 0 47 (5) FJ 13 (1) 71 (32) 0 10 (2) 30 (7) 0 5 (0) 129 (42) FM 2 (1) 16 (3) 0 0 0 1 (0) 1 (1) 20 (5) KI 1 (0) 1 (0) 0 0 0 3 (0) 0 5 (0) MH 0 10 (1) 0 1 (0) 2 (0) 1 (0) 0 14 (1) NR 0 0 0 0 0 0 0 0 NU 0 7 (1) 0 0 0 0 0 7 (1) PG 78 (16) 6 (4) 3 (1) 7 (0) 32 (13) 2 (1) 17 (10) 145 (45) PW 1 (0) 4 (1) 0 0 0 0 0 5 (1) SB 28 (9) 25 (5) 0 3 (1) 3 (2) 1 (0) 1 (0) 61 (17) TL 2 (1) 1 (0) 0 0 6 (1) 0 0 9 (2) TO 8 (1) 33 (11) 0 3 (0) 1 (0) 0 0 45 (12) TV 1 (0) 11 (0) 0 0 0 0 0 12 (0) VU 26 (4) 50 (8) 0 2 (1) 3 (0) 0 1 (0) 82 (13) WS 9 (3) 12 (7) 1 (0) 2 (2) 1 (0) 1 (0) 0 26 (12) Total 170 (36) 289 (77) 5 (1) 29 (6) 79 (23) 10 (1) 25 (11) 607 (155) Note: Catastrophic events (i.e. reported with at least 10,000 people affected, 10 million untrended US$ in losses, or 10 deaths) are in parentheses. A histogram of the number of entries re- FIGURE 44. Number of consequence database entries corded for each decade is presented in Figure 44. for each decade The figure indicates the number of reported of entries (events), as well as “catastrophic” entries has increased over the recent decades, most likely due to the increase in population. While the data search for the consequence database was exhaustive, data for each entry may not be entirely comprehensive, as some ac- counts of consequence may not have been re- corded or reported. For example, quantitative data for economic loss and loss of life is reported for about 63 percent and 49 percent of the database entries, re- 58 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) spectively. Nevertheless, the consequence database is the 1990s, which had single events that caused very a valuable tool as it provides details for specific signifi- high casualties. For example, the 1998 Earthquake and cant events. These may be used to inform case studies Tsunami in PG killed a reported 2,183 people, by far or used for catastrophe model validation purposes and the most devastating event listed in the consequence offer a qualitative assessment of natural hazard conse- database in terms of lives lost. The NGDC reports that quence in the South Pacific Region. Figure 45 shows 101 to 1,000 people perished in the FM from an earth- that the total economic loss for all 15 PICs per decade quake/tsunami event in 1899 and 101 to 1,000 people is on average 0.70 to 1.47 billion U.S. dollars (trended perished in PG from an earthquake event in 1906. to 2009 US$ using the loss trending as per the discus- Tropical cyclones reportedly have, by far, the sion above). The 1980s and 1990s saw a large amount greatest effect on the population in the South of costly disasters. The low and high estimates of data Pacific (Figure 47). Floods, which are typically due to represent the range of estimates due to different re- non-cyclone related severe storms, also have a large ports from multiple sources or the ranges of data given effect on the population. Over history, approximately by the NGDC databases. 3.5 to 4.1 million people have been reportedly affected by disasters in the 15 PICs; a significant number given FIGURE 45. Economic loss due to natural disasters that the total population was approximately 10 million in the 15 PICs in 2009. FIGURE 47. People affected by natural disasters in all 15 PICs The total number of fatalities for all 15 PICs is on the order of 500 per decade (Figure 46). Exceptions include the decades in the pre-1900 and FIGURE 46. Life loss due to natural disasters Figure 48 shows that tropical cyclones are in the 15 PICs reportedly the most damaging peril in terms of economic loss and earthquakes losses are com- parable. All disasters reportedly caused at least 7.9 bil- lion US dollars in economic losses (trended to 2009 as per the discussions above). To put this in perspective, the total GDP for all PICs in 2009 was about 14 billion current U.S. dollars. FIgure 49 displays the total number of people affected over the entire time, normalized by the respective population of each country in 2009. This figure is a good indicator of the relative disaster impact for each country. FJ, NU, WS, TO and VU have CATASTROPHE RISK ASSESSMENT METHODOLOGY 59 FIGURE 48. All time economic loss due to natural FIGURE 50. Total all time economic loss (trended to disasters in the 15 PICs 2009) divided by the respective 2009 national GDP FIGURE 49. Total number of people affected by disasters FIGURE 51. Economic loss in WS due to disasters each over the entire time divided by the respective population year with respect to the national GDP of the country in 2009 been affected by disasters with a significant impact on ter indicated in Figure 51, which shows the economic the population. Note that these counties are relatively loss (current US$) with the current national GDP versus small and may become completely devastated by a time for WS. Significant economic losses are report- single disaster event, especially tropical cyclones. For ed for certain years, which are typically due to single example, Tropical Cyclone Corine in 1960 reportedly devastating disasters. Cyclone Ofa in 1990 and then affected 4,000 people in NU, which had a population Cyclone Val in 1991 completely devastated WS (see of 5,000 at the time. Likewise, Cyclone Val of 1991 Annex E) reportedly causing economic losses well in affected about 54 percent of the population (88,000 excess of the yearly national GDP. people) of WS. 3.2 Damage Functions Figure 50 plots the total economic loss over the entire time (with trended values to 2009 as The severity of the physical damage is represent- per the discussions above) normalized by the re- ed by damage functions (DFs), which are statisti- spective nominal GDP of each country in 2009. cal relationships that estimate the loss an asset is This figure is a good indicator of the relative economic expected to suffer when subject to different lev- impact of disasters for each country. It indicates that els of intensity (or intensities) induced by a natu- economic losses are significant for most countries, es- ral event. The loss, which reflects the cost of repairing pecially NU, WS and VU. The severity of losses is bet- the damaged asset, is usually expressed as a percent- 60 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) age of the replacement cost of the asset, the damage Two types of natural hazards were explicitly ratio (DR). For example, in the DF shown in Figure 52, considered in this risk analysis: tropical cyclones a 100-mph wind is expected to cause moderate to ma- (inducing wind, precipitation and coastal flood- jor damage that will take about 20 percent of the total ing due to surge of the sea level) and earthquakes replacement cost of the asset to repair. The vulnerabil- (inducing both ground shaking and tsunami). The ity relationship (“vulnerability curve”) links the intensity effects of these hazards were measured by the intensity of an event to building, infrastructure and crop losses. measures (IMs) described below and were used as in- put to the DFs. Other effects of these hazards, such as landslides, liquefaction, and fire-following earthquake FIGURE 52. Vulnerability curve in a typical building were not explicitly considered, but the losses induced (a) Ground shaking by such phenomena were included in the empirical data from historical events used to calibrate the DFs. Timber ■■ Wind speeds (for tropical cyclones) are defined as Masonry the maximum one-minute sustained wind speed at Traditional 10 meters above the ground surface at the expo- Reinforced concrete sure location. Damage ratio ■■ Flood height (for tropical cyclones) is the height of the standing water at the exposure location caused by either tropical cyclone induced precipitation (fresh water) or by storm surge (salt water). ■■ Ground motion intensity (for earthquakes) is gauged by the horizontal PGA or by the 5 percent- damped elastic spectral acceleration (Sa) at oscilla- Ground motion intensity tor periods of 0.3 and 1.0 seconds at the exposure location. ■■ Wave height and velocity (for earthquake-induced (b) Wind speed tsunamis) is defined as the salt water (ocean) peak wave height above ground level at the exposure lo- 100 cation and the wave velocity is defined as the maxi- 90 mum velocity of the wave at the exposure location. 80 a. Buildings 70 Damage ratio (%) 60 Building DFs were developed to estimate the 50 vulnerability of different construction classes to 40 the effects of earthquakes and tropical cyclones. 30 These DFs refer to typical buildings in each construction 20 class. In addition, several secondary modifiers for dif- 10 ferent perils were considered in the damage estimation 0 of buildings to differentiate the within class building- 40 60 80 100 120 140 160 180 200 to-building vulnerability (Table 19). More precisely, Maximum one-minute sustained winds (mph) these secondary modifiers refer to characteristics of the building which tend to increase or decrease the vulner- ability with respect to that of the typical building in its respective construction class. For example, the pres- ence of window shutters is likely to reduce the vulner- CATASTROPHE RISK ASSESSMENT METHODOLOGY 61 ability of wind damage as compared to the vulnerability FIGURE 53. Example of the validity and accuracy of the of a similar building with no shutters. Likewise, a build- tropical cyclone DFs developed in this study ing with a tall, unbraced, stilt-like foundation would be 100% more vulnerable to ground shaking than a similar build- 90% Historical Observations Average Damage Function ing with a slab foundation. The effects on the expected 80% losses for buildings that have characteristics related to 70% Damage Ratio more than one modifier are cumulative. The extent of 60% the increase or decrease in the vulnerability due to each 50% modifier is based on extensive analytical and empirical 40% analyses. 30% 20% 10% TABLE 19. List of secondary modifiers affecting the DFs 0% for typical buildings per each construction class 40 60 80 100 120 140 160 180 200 Maximim One-Minute Sustained Winds (mph) Earthquake Tropical cyclone Ground Note: Average damage function refers to a weighted average of the Secondary modifier shaking Wind Flood Surge DFs for different types of buildings. Building deflect x x x x Foundation type x The DFs for tsunamis, which consider both the Foundation bracing x height and velocity of the wave, are almost entirely type empirical and were developed initially from sourc- Roof shape x es in the literature and post-event reconnaissance Roof pitch x data acquired for this project. For example, tsunami Roof material x x building damage in the TO island of Niuatoputapu due Shutter type x to the 2009 WS earthquake and tsunami). The dam- Wall opening type x age level induced by tsunami waves depends primar- Wall material x x x x ily on the maximum wave height at the site; accord- Minimum floor height x x ingly, it was selected as the primary intensity measure Note: The absence of a cross indicates that the corresponding for the tsunami DFs. Faster-propagating waves tend to characteristic does not trigger any change in the DF for that peril. cause greater levels of damage than slower-propagating waves of the same height. The wave velocity is positively The DFs for typical buildings in each construc- correlated with wave height and their correlation was tion class for earthquake ground shaking, wind, accounted for in this study. The wave height was used and flood were developed using vulnerability first to estimate a damage ratio expected in a structure models for building structures. The DFs were cali- of given characteristics and if the maximum wave veloc- brated using historic building damage data collected ity was above a given threshold then the structure was from various sources, including a damage reconnais- considered a total collapse. The DFs developed for this sance study for the 2009 M7.6 Padang Earthquake study were validated with analyses of data from past that struck offshore West Sumatra, building damage tsunami events in the region (e.g., the 2009 M8.1 WS earthquake – see Figure 54 ). data from the 2007 M8.1 earthquake in the SB, and photographs collected after the 2010 Tropical Cyclone The losses estimated using the resulting DFs Pat that devastated the island of Aitutaki in the CK. were then compared against the observed losses Figure 53 illustrates the validity and accuracy of the for several historical events in the region (see pre- DFs by comparing a weighted average of the DFs from vious section). Comparisons of the modeled and ob- the construction types in the regions affected by his- served ground-up losses are shown for tropical cyclone torical tropical cyclones with observed damage data events, earthquake events and earthquake events causing from various sources. both ground shaking and tsunami damage in Figure 55. Tropical cyclone events 62 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) FIGURE 54. Example of the validity and accuracy of the FIGURE 55. Comparison of simulated and observed tsunami developed in this study ground-up losses to buildings, infrastructure, and crops due to natural disasters (a) wind, flood and storm surge for select historical tropical cyclone events n Minimum reported losses n Simulated losses n Maximum reported losses Note: The DF from Reese et. al. (2011) is inferred from fragility curves from actual data for the 2009 WS Tsunami. The DF from AIR was developed independently before this reference was published. (b) ground shaking for select historical earthquake In general, there was good agreement between the events (buildings and infrastructure) modeled losses and the observed losses. The compari- son of modeled versus observed losses is, however, more n Minimum reported losses favorable for tropical cyclones than it is for earthquakes. n Simulated losses This was to be expected for two reasons: firstly the lo- n Maximum reported losses cation of the fault rupture, which may be hundreds of kilometers long, is highly uncertain, given that for most events only the epicenter is known; secondly, unlike tropical cyclones where observations of wind speed and precipitation at selected locations are available, there are no recordings of ground shaking at any of the sites hit by these earthquakes. Finally there is also a significant degree of uncertainty in the observed losses as reported (c) ground shaking and tsunami wave for select historical by different agencies. earthquake events (buildings, infrastructure and crops) b. Emergency Losses Earthquake ground shaking and tsunami events Total ground-up losses (million US$) The losses above reflect both the cost needed n Minimum reported losses n Simulated losses to repair or replace the damaged assets and the n Maximum reported losses emergency losses that local governments may sustain as a result of providing necessary relief and undertaking recovery efforts. Such efforts include debris removal, setting up shelters for those made homeless, or supplying medicine and food. In this study, emergency losses were estimated as a frac- tion of the direct losses. Research on historical tropical cyclones and earthquakes indicates that an “average” estimate of the emergency losses as a percentage of Note: The minimum and maximum reported values were reported by the direct losses suffered by residential dwellings, com- different sources. The observed losses are trended to 2010 US$ values. CATASTROPHE RISK ASSESSMENT METHODOLOGY 63 mercial establishments, public buildings, schools and and crop economic losses, models were devel- hospitals is about 16 percent for earthquake ground oped to estimate the number of fatalities and in- shaking and 23 percent for tropical cyclones and flood. juries (casualties) for each specific peril. In general, Those percentages were applied in this study. Similar- estimating the number of human casualties from di- ly, a factor of 23 percent was applied to direct losses sasters with reasonable accuracy is more difficult than caused by tsunamis. estimating economic losses. The number of casualties is dependent on several aspects, such as human be- c. Infrastructure Assets havior, time of the event, efficiency of communication to the affected population (e.g., notice of an incum- The development of DFs for infrastructure assets bent tsunami or tropical cyclone), the occurrence of followed a similar approach, except that the DFs non-modeled effects (e.g., landslides, fire following were developed for typical assets in each vulner- earthquakes) or the destruction of critical assets (e.g., ability class and secondary modifiers were not hospitals, dams, lifelines). These conditions are often considered. Infrastructure DFs used the same input unrelated to the severity of the event and generally epi- intensity measures as mentioned for buildings. The sodic, making them difficult to predict. For example, an development of the infrastructure DF was based on earthquake or tsunami that occurs at night time may AIR’s proprietary vulnerability model and, therefore, a cause more casualties since more people are in build- detailed description was omitted in this report. These ings and are not as alert as in the day time. Likewise, proprietary DFs were validated and calibrated based on the total number of casualties for a hurricane would historic loss data of the PICs. be much less if the storm is well forecasted and people decide to evacuate the area. d. Crops Three casualty models were developed for Significant damage to crops in the PICs has been observed from wind and rain effects as well as earthquake ground shaking and tsunami and one flood caused by tsunami waves and storm surge. for tropical cyclones. These models were primarily Table 20 outlines the relative vulnerability of different empirical and relied mostly on historical data in the crop types for different natural hazards. A crop damage region. For simplicity, these casualty models assumed model for tropical cyclones was specifically developed that all the population resides in residential dwellings for this project. This model considered interacting ef- (i.e. it is implicitly assumed that the events strike in the fects of wind and precipitation damage at the crop lo- middle of the night) and that casualties were not ex- cation. Empirically based bivariate DFs were developed plicitly dependent on human decisions. based on the observed losses from historic tropical cy- ■■ The earthquake fatality model was based primar- clones in PICs for three different crop types (root crops, ily on USGS’s PAGER system, which uses empirical tree crops and annual crops). The supporting data was methods to estimate casualties as a function of the not sufficient to differentiate the vulnerability of spe- shaking intensity and the number of people ex- cific crops within the same crop class (e.g., banana and posed to such intensities. The model uses empiri- papaya belonging to the tree crops). Figure 56 shows cal parameters that are specific to the PIC region. the DFs of tree crops and root crops. Since most crops At each location, the fatality rate was estimated as do not tolerate high salt environments (see Figure 57 a function of the ground shaking intensity (here a-c. ) a 100 percent damage level was assumed for all measured by PGA). crops submerged by at least 20 cm of salt water due to storm surge and tsunami waves. ■■ The tropical cyclone fatality model was developed specifically for this project. Estimating fatalities from tropical cyclones is extremely complex and e. Fatalities and Injuries essentially no sources from the literature were In addition to developing DFs that relate the in- available. The model developed for this study tensity of an event to building, infrastructure, was strictly empirical and is based on fatalities for 64 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) TABLE 20. Relative vulnerability of certain crops to natural hazards Indicators of damage to crops1 Cyclone Flood Drought Leave Salwater Saltwater Water Crops Uprooted Defoliation shredded spray Lodged inundation logged Lodged Sliced Willing Taro-inmature x o o o o x o o o x Taro-mature • o o o o • • • • • Sweet potatoes • o o o – • • • • • Yams-inmature x o o o o x x x x x Yams-mature • o o o o • • • • • Cassava-inmature x o o o o x x o x x Cassava-mature x • • o • • • • • • Sago palm x o o o x o o x o o Coconut x o o o x – o x o o Cocoa x o o o o x o x o o Citrus x o o o o x o o o o Mango x o o o x o o x o o Coffee x o o o x x o x o o Breadfruit x o o o o o o o – o Bananas x o o o x x o x o o Pawpaw x o o o x x x x o x Rice-wetland x o o o o x o o o x Rice-dryland x o o o o x o o o x Mixed vegetables x x x x o x x x o x Sugarcane x o o o o x o x o o Pineapple x o o o o x o o o o Passionfruit x o o x o x x o o o Kava • • o o o • • • o o Ginger x o o x o x x o o o Vanilla o o o o o o o o o o Maze x x o x o x o o o o Note 1: The sriousness of the damage depends on the severity of the event. Indicated here are the possible effects of a severe event. Other natural occurrences such as earthquakes and volcanic eruptions with their accompanying landslide, tsunami, ashfall, lava flow, etc., often result in destruction to crops. Keys: x = Destroyed. Crops will not survive and have to be replanted. Damage incurred renders the crops useless for consumption or for sale. • = Destroyed but salvageable. The crop has matured, and although it will not recover, it can be salvage for consumption or for sale, or to be stored or preserved if immediatelly harvested following the event. – = Not applicable. The indicators of damage do not apply due to the characteristics of the crops. The crops may not be too badly affected. FIGURE 56. Crop DFs for (a) root crops and (b) tree crops (a) (b) 100 100 Precipitation (inch) 90 90 2 80 80 8 14 70 70 20 Damage ratio Damage ratio 60 60 26 Precipitation (inch) 32 50 2 50 38 8 40 40 14 30 20 30 26 20 20 32 10 38 10 0 0 0 20 40 60 80 100 120 0 20 40 60 80 100 120 Wind speed (mph) Wind speed (mph) CATASTROPHE RISK ASSESSMENT METHODOLOGY 65 historical cyclones in the region. The model esti- FIGURE 57. Comparison of simulated and reported mated the number of fatalities as a function of fatalities for selected historical a) earthquake ground the total economic losses, which was used as a shaking events, b) tropical cyclone events, and proxy of the damage to buildings. This approach c) earthquake ground shaking and tsunami events. was used since the data from the economic loss (a) Earthquake ground shaking events estimates ware more robust than the estimated damage level incurred by single buildings subject n Minimum reported to the effects of storms. The model used a three- n Simulated parameter power curve fit with a threshold value n Maximum reported of 0.1 percent of the country’s population. Thus, it was assumed that the total number of fatali- ties will never exceed 0.1 percent of the country’s population, regardless of the severity and path of the storm. This assumption, along with the val- ues of the other two parameters in the power curve-fit model, was verified by historical data on a country-specific level. ■■ A tsunami fatality model was also developed spe- cifically for this project. This model is also empirical (b) Tropical cyclone events but it relates the number of fatalities to building damage (more specifically, the number of build- n Minimum reported ings damaged beyond a certain threshold). The n Simulated empirical parameters were calibrated and validated n Maximum reported through simulated damage estimation of historic events and data from the literature. The simulated number of fatalities is com- pared to reported numbers for some historical earthquakes (ground shaking only) in Figure 57 a) and historical earthquakes (ground shaking and tsunami) in c). b) compares the simulated and the observed number of fatalities for some historical tropical cyclones. In some events, such as tropical cy- clone Namu in 1986, the simulated fatality estimate (c) Earthquake ground shaking and tsunami events falls short of the observed value since some secondary effects of the cyclone (such as landslides) were not ex- plicitly modeled. n Minimum reported n Simulated Estimating injuries is even more of a vola- n Maximum reported tile exercise than estimating fatalities. Most ca- sualty models available in the literature, in fact, do not estimate the number of injuries. Also, injuries caused by historic events are not as well reported as fatalities, and therefore provide a less robust empiri- cal basis to support a model. Empirical injury models were developed specifically for this project. These models assume that the number of injuries is directly proportional to the number of fatalities. These rela- 66 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) tionships were derived from historical data in the re- are injured for every one fatality due to earthquake gion and are peril-specific (Figure 58). For example, ground shaking. it is assumed that, on average, about nine people FIGURE 58. Empirical relationships of deaths versus injuries for different perils based on historic data in the region Ground shaking Tsunami 2,500 450 Regression Regression 400 I = 8.72° D I = 1.00° D 2,000 R2 = 0.50 350 R2 = 0.28 300 1,500 Injuries Injuries 250 200 1,000 150 500 100 50 0 0 0 50 100 150 200 250 0 50 100 150 200 Deaths Deaths Tropical cyclone 90 Regression 80 I = 11.70° D 70 R2 = 0.33 60 Injuries 50 40 30 20 10 0 0 1 2 3 4 5 Deaths CATASTROPHE RISK ASSESSMENT METHODOLOGY 67 4. Country Catastrophe Risk Profiles State of the art catastrophe risk models were developed to assess the economic and fiscal impact of natural hazards like tropical cyclones, earthquakes and tsunamis in 15 PICs (see Catastrophe Risk Modeling Framework). The results of the extensive simula- tions are assembled in the country catastrophe risk profiles. The risk profile for each country expresses the likelihood that adverse consequences of events (with different severities) will occur within a certain time frame (e. g., once in the next year or once within the next 50 years, etc.). This section highlights some results. It also shows some inter-country comparisons including the fiscal risk and budget envelopes of the 15 PICs. Specific details and the compre- hensive risk profiles of all individual 15 PICs can be found in Annex F (Country Risk Profiles). Risk profiles were derived from the impact estimated for all the simulated fu- ture events. For each event of a given severity and location (i.e., a magnitude 8 earthquake offshore PG), the intensity in the nearby region (e.g., the peak horizontal acceleration of the ground predicted at each location) was calculated using the mathematical models mentioned earlier. The level of damage and direct losses for any given asset at any given location in the affected PIC are estimated based on the characteristics of the asset (e.g., timber frame build- ing) and on the level of intensity predicted at that location (e.g., a peak horizontal accelera- tion equal to 30 percent of gravity). The total losses for any simulated event are equal to the sum of the losses at all locations affected by that event. The estimation of casualties caused by any event was done in a similar fashion. The loss and casualty calculations were repeated for all the simulated 400,000 tropical cyclones and 7.6 million earthquakes. The risk profiles were obtained by ranking the losses and the casualties of all the simulated events. The 10,000 simulations of potential future annual tropical cyclone and earthquake activity captured in the stochastic catalogs shows that some years will see no significant tropical cyclones or earth- quakes affecting any of the PICs, while other years may see one or more devastating events affecting the islands, similar to what was observed historically. The entire set of simulations enabled an assessment of the average risk that the pool of the 15 PICs (or each PIC separately) faces due to these natural perils. Figure 59 shows the average annual loss (AAL), for all 15 PICs, which is the aver- age loss that can be expected to occur in any given year and their contributions from the different perils. The same set of simulations indicates also that the average annual number of injuries and fatalities that is expected every year in these 15 PICs due to earth- quakes and tropical cyclones combined is about 2,000. The risk assessment revealed that, every year on average, all 15 PICs combined experience damage caused by natural hazards estimated at US$284 million, or 1.7 percent of the regional GDP. The regional average hides a high disparity among the PICs due to their exposure to natural hazards and the size of their economies. It is estimated that, as a percent of their national GDP, VU, NU and TO experience the largest AAL with 6.6, 5.8 and 4.4 percent, respectively (Figure 60). These countries are among those countries ranked highest globally, assessing average annual disaster losses scaled by GDP. An adverse year, which occurs once every 75 years, would impact the PICs dif- ferently. The most affected PIC by an estimated 75-year loss in terms of their national GDP 68 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) FIGURE 59. AAL for each of the 15 PICs 100 n Tropical cyclone 10 n Earthquake ground motion 8 n Tsunami Average annual loss (million US$) 80 6 4 2 60 0 or M Co te sh s. s. u ue Tu i lu u at la ur I lI va es FS Ni rib k Pa al Na o -L Ki ar 40 m M 20 Ti 0 M e ds s u ue i lu u G Va Fiji on atu ds a oa at nd ng st la ur PN va FS an sh lan Ni rib m Co r-Le Pa nu Na sla To Tu Sa Isl Ki Is lI o k m al o Ti m ar lo M So FIGURE 60. Estimated AAL for the 15 PICs, as percentage of national GDP 7.0 6.0 5.0 4.0 % GDP 3.0 2/0 1.0 0.0 u ue a M ds i s ds u oa G te i lu u Fij at nd ng t la ur PN va es ua FS an an Ni m rib Pa Na sla To -L Tu Sa n Isl Isl Ki or Va lI on ok al m sh Co Ti m ar lo So M would be TO, followed by VU and FM. For more ex- ceeded. Figure 62 shows the annual rate of exceed- treme losses, such as a 250-year loss, FM would be the ance versus total ground-up losses for a 50 percent most affected, with an estimated loss in excess of 95 chance of a 50 year event in FJ with losses of US$641 percent of national GDP (Figure 61). These estimates million caused by a category 3 Tropical cyclone affect- only capture the direct losses, i.e., physical damage on ing the area. Displayed in Figure 63 is the exceedance buildings, major infrastructure and cash crops. Total probability as percentage of national GDP versus dif- losses, including indirect losses (e.g., business interrup- ferent return periods (e.g., frequency of occurrence) tion) can be several times higher than the direct losses. for all individual 15 PIC Disaster Risk Profiles. Curves The exceedance probability describes the display the combined risk of earthquake, tsunami and probability that various levels of loss will be ex- tropical cyclone. CATASTROPHE RISK ASSESSMENT METHODOLOGY 69 FIGURE 61. Estimated 75-year loss and 250-year loss, as percentage of national GDP 100 n 250 year loss n 25 year loss 90 80 70 60 % GDP 50 40 30 20 10 0 FSM TON PAL MAR COO VAN SOL TIM SAM FIJ TUV TUV KIR FIGURE 62. Example of exceedance probability of a Cat 3 tropical cyclone affecting FJ TC total ground-up loss – Fiji 1 50%/50 years Annual rate of exceedance 10%/50 years 5%/50 years 0.1 50% in 50 years (73 MPR) 0.01 641 M 0.001 0 500 1,000 1,500 2,000 Total ground-up loss (million US$) Storm Wind speed Storm surge Central pressure category (mph) (ft) (mbar) 1 74-95 4-5 960 2 96-110 6-8 965-979 3 111-130 9-12 945-964 4 131-155 13-18 920-944 5 >155 >18 <920 70 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) FIGURE 63. PIC Disaster Risk Profile 120 NIU 100 TON COO MAR 80 PAL FSM TIM 60 SOL % SAM VAN 40 FIJ 20 TUV PNG KIR 0 NAU 0 50 100 150 200 250 Return period (years) The AAL for all 15 PICs averaged over the butions to the AAL from the different area councils are many realizations of annual activity are shown displayed in absolute terms in Figure 66 and normal- in Figure 64 (separately for tropical cyclones and ized by the total asset values in each area council in earthquakes). The contribution to the AAL due to Figure 67, which shows how the relative risk varies by buildings, infrastructure and crops is also shown. Fig- area council across the country. ure 65 shows the AAL for VU only. For VU, the contri- FIGURE 64. AAL caused by tropical cyclones and earthquakes in all the 15 PICs combined Tropical cyclone Earthquake Average annual loss = 178 million US$ Average annual loss = 106 million US$ 5.9% 0.1% 16.3% 31.7% 62.4% 83.7% n Buildings n Cash crops n Infrastructure CATASTROPHE RISK ASSESSMENT METHODOLOGY 71 FIGURE 65. AAL due to tropical cyclones and earthquakes and its contribution from the three types of assets for VU. The individual charts for all 15 PICs can be found in Annex F (Country Risk Profiles) Tropical cyclone Earthquake Average annual loss = 36.8 million US$ Average annual loss = 11.2 million US$ 3.9% 0.0% 13.0% 29.3% 66.8% 87.0% n Buildings n Cash crops n Infrastructure FIGURE 66. Contribution from the different villages and islands to the AAL for tropical cyclone and earthquake in VU. Individual maps for all 15 PICs can be found in Annex F (Country Risk Profiles) 72 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) FIGURE 67. Contribution from the different villages and islands to the tropical cyclone and earthquake AAL divided by the replacement cost of the assets in each location in VU. Individual maps for all 15 PICs can be found in Annex F (Country Risk Profiles) Figure 68 shows the direct loss risk profile There is a 50 percent chance that the Pacific Region for earthquakes, tropical cyclones and for both will face disaster losses in excess of US$1.3 billion in earthquakes and tropical cyclones for all 15 PICs any 50 year period. Similarly, Figure 69 shows the ca- combined. For example, the 15 PICs are expected to sualty risk profile for earthquakes, tropical cyclone and collectively observe an annual loss due to earthquakes for both earthquakes and tropical cyclones for all 15 exceeding about US$1 billion (or 6 percent of the to- PICs combined. tal regional GDP), on average, once every 200 years. CATASTROPHE RISK ASSESSMENT METHODOLOGY 73 FIGURE 68. Direct loss risk profiles by peril for all 15 PICs combined (a) (b) Note: The direct losses in the vertical axis are expected to be exceeded, on average, once in the time indicated in the horizontal axis. The losses are expressed in a) absolute terms and b) normalized by the total GDP of the PICs. FIGURE 69. Casualty (deaths plus injuries) risk profiles by peril for all 15 PICs combined Note: The number of casualties in the vertical axis is expected to be exceeded, on average, once in the time indicated in the horizontal axis. 74 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) 5. The Pacific Risk Information System (PacRIS) The Pacific Risk Information System (PacRIS) houses the most comprehensive re- gional database of baseline exposure (buildings, infrastructure and crops) and prob- abilistic risk assessment results for all 15 PICs. The exposure database leverages remote sensing analyses, field visits, and country specific datasets to characterize buildings (residen- tial, commercial, and industrial), major infrastructure (such as roads, bridges, airports, and electricity), major crops and population. More than 530,000 buildings were digitized from very-high-resolution satellite images, representing 15 percent (or 36 percent without PG) of the estimated total number of buildings in the PICs. About 80,000 buildings and major in- frastructure were physically inspected. In addition, about 3 million buildings and other assets, mostly in rural areas, were inferred from satellite imagery. PacRIS includes the most comprehensive regional historical hazard catalogue (115,000 earthquake and 2,500 tropical cyclone events) and historical loss data- base for major disasters, as well as country-specific hazard models that simulate earthquakes, tsunamis) and tropical cyclones. PacRIS contains risk maps showing the geographic distribution of potential losses for each PIC as well as other visualization prod- ucts of the risk assessments, which can be accessed, with appropriate authorization, through an open-source web-based platform. PacRIS enables proactive regional integration with the newly launched Open Data for Resilience Initiative (OpenDRI). OpenDRI is a ‘World Bank- wide’ initiative that seeks to support decision making by facilitating the sharing and use of information for building resilience to natural hazards in a changing climate. PCRAFI developed the Pacific GeoNode as a central web-based data and information sharing platform enabling PICs and development partners’ access to the risk information, tools and knowledge products developed under the initiative. As such, it is one of the flagships for good practice under OpenDRI. In the next phase, PacRIS will be further built and refined by: (I) Strengthening the capacity of technical agencies at national and regional level in its use and maintenance; (II) Strengthening the risk information and underlying data, including the update of risk exposure databases; (III) Strengthening the data-sharing platform at SPC/SOPAC in order to achieve expanded reach and allow access of PCRAFI data and information to the wider Pacific community; and by (IV) Developing country risk atlases and other knowledge products to effectively communi- cate risk information to policy and decision makers. The data and models of the PacRIS will evolve to suit the needs of its applica- tions. Therefore, the applications will drive the System’s evolution and manage- ment. CATASTROPHE RISK ASSESSMENT METHODOLOGY 75 6. Applications Poor populations tend to live in higher-risk areas, making them more likely to be af- fected by adverse natural events. The vulnerability of the poor to natural disasters and the effects of climate change are expected to increase due to increased population pressure push- ing the poor to live in more marginal areas. This has led to widespread acceptance of the need for mainstreaming disaster risk and climate change in development planning and financing. PCRAFI is an innovative initiative providing for the first time quantitative, proba- bilistic risk information and tools for risk assessment, disaster risk financing and insurance solutions for PICs. The knowledge products derived from this initiative provide unique and relevant information for multiple sectors. There is a focus under the next phase of PCRAFI to make the risk information and tools available to policy and decision makers. An emphasis is on developing capacity of selected PICs to strengthen and mainstream climate and disaster risk information into urban and infrastructure planning and macro-economic planning. The next phase of PCRAFI will continue to develop PacRIS, the GIS platform used as the database infrastructure to develop selected applications for smarter DRM in- vestments. Three applications will be supported building on PacRIS (Figure 70): 1) Macroeconomic planning and Disaster Risk Financing; 2) Mainstreaming of risk information into urban and infrastructure planning; and 3) Rapid post-disaster impact estimation. Integrating climate change projections, further professional and institutional capacity building and other applications could be added or developed over time. FIGURE 70. Pacific Risk Information System and its applications 76 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) 6.1 Post Disaster Response Capacity The domestic property catastrophe risk insur- and Disaster Risk Financing ance markets are currently under-developed in the South Pacific. This initiative assists in the design Access to liquidity in the aftermath of a disaster of disaster risk insurance products, both sovereign di- is essential for governments to ensure immediate saster risk insurance for governments and disaster mi- and effective post-disaster response. While do- cro-insurance for households and small and medium nor partners have been responsive in the aftermath of sized enterprises (SMEs). It provides insurance com- disasters, several PICs have faced post-disaster liquid- panies and other financial institutions with technical ity shortages that have limited their ability to respond knowledge to enable and implement parametric disas- quickly. Some PICs have already engaged in proactive ter risk insurance mechanisms in the PICs. The Pacific financial management of natural hazards as part of Catastrophe Risk Insurance Pilot assists MH, WS, SB, their national DRM plan. For example VU, MH and CK TO, VU via a risk pooling mechanism to enhance their have established national DRM funds. financial response capacity. Financed by the Govern- ment of Japan, the 5 participating countries purchased The development objective of the Pacific Di- earthquake and/or tropical cyclone coverage for the saster Risk Financing and Insurance Program is to 2012-2013 and 2013-2014 pilot period. increase the financial resilience of the PICs against natural hazards and to improve their capacity to meet post-disaster funding needs without com- promising their fiscal balances and development 6.2 Disaster Risk Reduction and Urban/ objectives. It aims to assist the PICs in the improve- Infrastructure Spatial Planning ment of their macroeconomic planning against natu- PacRIS ensures that disaster risk and climate ral hazards, including ex ante budget planning. PacRIS change information and considerations form an helps the PICs to design and implement an integrated integral part of the urban and infrastructure plan- national disaster risk financing strategy relying on an ning process. The potential cost of damage to build- optimal combination of reserves, contingent credit, in- ings and infrastructure in urban areas contributes to a surance, and donor grants. The program supports the large proportion to the total AAL in the PICs. Accord- following activities: ing to PCRAFI, the AAL comprises predominately of ■■ Capacity building on integrated disaster risk financ- potential damage to buildings and infrastructure with ing and insurance; 70.2 percent and 26.2 percent respectively, and only a minor part (less than 4 percent) is attributed to poten- ■■ Development of Private Disaster Risk Insurance tial damage of cash crops. Although over 80 percent of Markets; and buildings in PICs are located in rural areas, two thirds ■■ Piloting of Pacific Disaster Risk Insurance Program of the total asset values (in terms of replacement costs) for governments. are concentrated in urban areas. Therefore strengthen- ing urban/infrastructure planning and design will assist The disaster risk financing strategy is an inte- to reduce disaster losses of countries. gral part of the national DRM and CCA agenda. The financial management of natural hazards comple- ments the ongoing disaster risk reduction activities 6.3 Post-Disaster Assistance and undertaken by the PICs. The approach builds on an Assessment integrated, three-tier financial strategy against natural hazards. It includes self-retention, such as a contin- The aim of the Rapid Disaster Impact Estima- gency budget and national reserves, to finance small tion application is to provide disaster managers but recurrent disasters, a liquidity mechanism for less and first responders with tools and information frequent but more severe events, such as contingent to quickly gain an overview following a disaster. credit and disaster risk insurance to cover major natural Vital information on areas and population affected and disasters. the likely severity of the event in terms of potential fa- CATASTROPHE RISK ASSESSMENT METHODOLOGY 77 talities, injuries and building, infrastructure and crop 6.5 Reporting and Monitoring damage in a timely fashion will aid more targeted re- Agencies sponse and early recovery. For the implementation of some risk mitiga- The application builds on the information and tion strategies (e.g., catastrophe bonds) it is nec- tools developed under PCRAFI and provide a first essary that reputable organizations are selected damage estimate based on modeled losses within to report the occurrence and the characteristics hours after the event. Following severe disasters field of large natural events that may impact the coun- teams will be deployed to carry out detailed damage tries at stake. Decision making criteria should include assessments in selected areas and provide field verified independency (i.e. the organization should not have model updates. Apart from providing crucial disaster any real or perceived conflict of interest with the eco- impact information in a standardized, timely and accu- nomic or political environment in the region), depend- rate fashion, this initiative will systematically collect new ability, accuracy and uniformity across regions of the validation information following future disaster events world. There are several organizations in the world at to refine the vulnerability/loss models used by PCRAFI. large and in the region which could serve as official This application will support the use of exposure data reporting agencies. For example, some organizations as baseline for Damage and Loss Assessments (DALA) with in-depth knowledge of the tropical cyclone ac- and will be closely linked to a SPC/SOPAC initiative to tivity in the Pacific (both hemispheres) are the Joint strengthen the regions capacity in DALA via the Post Di- Typhoon Warning Center (JTWC), Australia Bureau of saster Needs Assessment framework. Meteorology (BoM), Shanghai Typhoon Institute, Japan Meteorological Agency (JMA) and the Fiji Meteorologi- cal Service. For earthquakes they are the United States 6.4 Early Warning Systems and DRR Geological Service (USGS), Geoscience Australia (GA) Communication and GNS Science in New Zealand. PCRAFI is providing access to critical informa- tion on population and assets at risk, which can assist in the preparation and response to a disas- ter. As part of developing rapid disaster impact esti- mation tools and services PCRAFI will provide disaster managers with timely information on disaster impacts. It is anticipated that these services and tools will be extended with additional funding to provide pre-event damage forecasts for approaching tropical cyclones (and tele-tsunamis) and hence strengthen regional ear- ly warning products. 78 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) 7. References Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI), Component 1: Hazard Data and Loss Data Collection and Management, Technical Report Submitted to the World Bank by AIR Worldwide, December 2010. Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI), Component 2: Expo- sure Data Collection and Management, Technical Report Submitted to the World Bank by AIR Worldwide, September 2011. Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI), Component 3: Country Catastrophe Risk Profiles, Technical Report Submitted to the World Bank by AIR Worldwide, December 2011. Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI), Component 4: Port- folio Risk Analysis Report, Technical Report Submitted to the World Bank by AIR Worldwide, November 2011. Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI), Catastrophic Loss Da- tabase Tool, User’s Manual Submitted to the World Bank by AIR Worldwide, December 2011. Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI), Risk Assessment Meth- odology, September 2011. Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI), Progress Brief, August 2011. Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI), Pacific Disaster Risk Financing and Insurance Program (PDRFIS), August 2011. Asian Development Bank (ADB), I, Final report, TA 6496-RE: Regional Partnership for Climate Change: Adaptation and Disaster Preparedness, July 2011. Pacific Catastrophe Risk Financing Initiative, Catastrophe Risk Assessment and Options for Regional Risk Financing; The World Bank; September 2008. CATASTROPHE RISK ASSESSMENT METHODOLOGY 79 Annex A Field Survey Locations The following locations were field surveyed in the 11 countries visited by the project teams: CK ■■ Rartonga Island – CK’s most populous island and home to the national capital of Avarua. ■■ Aitutaki Island – popular tourist destination; prone to cyclone hazard. FJ ■■ Suva, including the suburban areas of Nausori and Lami – a very large urban area, FJ capital and largest city. ■■ Nadi – a large city with significant tourism and sugar cane industries. FM ■■ Yap proper – population center of the Yap State; relatively high exposure to tropical cyclone and earthquake-induced tsunami. ■■ Weno in Chuuk State – FM’s largest city; susceptible to tropical cyclones, floods and landslides. KI ■■ South Tarawa – KI capital and largest city. PG ■■ Note: PG’s capital and largest city, Port Moresby, was excluded because of relatively low exposure to natural hazards, extremely high costs and severe security concerns. ■■ Lae – the capital of Morobe Province and second largest city in PG; a major port, prone to earthquake, tsunami and storm surge hazard. ■■ Madang – large town and capital of the Madang Province; a major tourist destination, prone to earthquake, tsunami and storm surge hazard. ■■ Ramu Sugar Facility – on route from Madang to Lae; a large agriculture project, prone to earthquake and flood hazard. ■■ Rabaul/Kokopo – former provincial capital of East New Britain (devastated by a vol- cano in 1994); large port facilities, prone to earthquake, tsunami, and volcano hazard. PW ■■ Koror – PW’s largest city and major touristic center. SB ■■ Honiara – SB’s capital and largest city. ■■ Guadalcanal plains – rural area with major agriculture use, close to Honiara. ■■ Auki – provincial capital of Malaita; highest density of rural population in SB and rela- tively easy to access. 80 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) ■■ Noro – large town in the Western Province; a major sea port and fish cannery. ■■ Munda –largest settlement on the island of New Georgia in the Western Province. ■■ Gizo –capital of the Western Province and SB’s second largest city. ■■ Ringgi – settlement on Kolombangara Island in the New Georgia Island group. TO ■■ Nuku’alofa, including other areas in Tongatapu – Nuku’alofa, TO’s capital and largest city; located on TO’s main island of Tongatapu, prone to earthquake and tropical cyclone hazard. TV ■■ Funafuti – TV’s capital and most populated atoll. VU ■■ Port Vila, including suburbs and surrounding villages – VU’s capital and largest city. ■■ Luganville and surrounding villages – VU’s second largest city and only other large settlement in VU be- sides Port Vila. ■■ South-West Tanna Island – a major tourism/resort destination. ■■ North-East Ambae Island (Lolowaï) – a rural settlement on an outer island. WS ■■ Apia – WS’s capital and largest city. CATASTROPHE RISK ASSESSMENT METHODOLOGY 81 Annex B Building Locations (Level 4 Methodology) The level 4 methodology of extracting the spatial distributions dealt with buildings that are mostly located in rural areas. They were inferred using image processing techniques from low to moderate-resolution satellite imagery and/or census data. This methodology does not involve manually digitizing footprints and was applied mainly in rural areas of PG, TL, SB, VU, FJ, FM, MH and KI (and to a lesser extent CK, TO, TV), where high-resolution satellite imagery was limited or not existent. Buildings with non-residential occupancy type were inferred us- ing different techniques. For clarity, the estimation of residential buildings and non-residential buildings was treated separately. The residential building inference was conducted using a three-step process. Low- and moderate-resolution satellite imagery was used with computer-aided detection, which is based on the brightness and specific color of a specified location with respect to neighbor- ing areas, to identify rural settlements. Trended 2010 population counts from the population database were used to estimate the number of people within detected settlements. Then the average number of persons per dwelling (household), collected from census data, was used to estimate the number of dwellings and consequently the number of residential buildings (by using the average number of dwellings per building). Once settlements were detected, the population residing in respective census areas was distributed using the average number of people per dwelling available from the country-level average of each respective country (Table 21). Constraints were applied to the number of dwellings assigned to prevent unrealistic scenarios. For example, no more than 50 dwellings were allowed per 100-meter grid cells. TABLE 21. Country-specific average number of people per dwelling based on census data Country Census Year Average Household Size CK 2006 3.7 FJ 2007 4.8 FM 2000 6.7 KI 2005 6.3 MH 1999 8.7 NR 2006 5.9 NU 2006 3.2 PG 2000 5.5 PW 2005 3.9 SB 1999 6.3 TL 2010 4.7 TO 2006 5.8 TV 2002 6.0 VU 2009 4.8 WS 2006 4.4 82 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) In some cases, such as when imagery was not avail- one adopted in this study, is a standard application, able or when images were obscured by excessive cloud there are limitations inherent to this approach. The cover, image-processing techniques were not possible. In interference of cloud coverage and the existence of addition, in certain areas such as heavy forests, the tech- settlements under the canopy of trees are two obvious nique did not always detect cells because the settlements examples. In order to minimize the impact of these lim- were either hidden under tree canopy or the dwellings itations, every reasonable effort was expended to ac- were of the same color as the surrounding terrain. For curately detect the existence of identified settlements these situations, settlement locations were assumed to be and minimize errors (e.g., false positives). For example, located at the centroid of the respective enumeration dis- satellite imagery data was supplemented with ancillary trict, and the number of dwellings was inferred from the data; including LULC maps, data sets that indicate vil- population counts. While this centroid approach is the lage locations, census data, the population database, least accurate technique of the building detection meth- and DIGO4 data. For example, in PG, the population odologies adopted here, the enumeration districts typi- database contains over 20,000 locations of known hu- cally have a very fine resolution as to not overly misrep- man settlements, which was collected from the Uni- resent the building locations. For example, about 20,000 versity of PG. For this country, the image detection dwellings – 15,000 in PG – were aggregated to the cen- technique was performed only within a two-kilometer troids of about 3,000 enumeration districts, resulting in radius around these known settlements. only about seven dwellings per centroid point location. Table 2, which shows the total number of build- Although identifying human settlements with low- ings identified by this image processing technique, it to moderate-resolution satellite imagery, such as the is apparent that PG has by far the largest number of FIGURE 71. Number of households per cell in PG Rural households per grid <1 1-2 2-5 5 - 10 10 - 50 >50 Level 2 or 3 data available DIGO data refers to data collected by the Australian Air Force that was provided to AIR for use in this specific project. 4 CATASTROPHE RISK ASSESSMENT METHODOLOGY 83 buildings identified using the Level 4 analysis. Given eas of these PICs, a dwelling may include several build- the size, population and rural nature of the country, ings, such as the main house, kitchen and toilet build- this conforms to expectations. Figure 71 provides a ings and a work-shop, storage shed or farm building. thematic map of the number of households in PG as Thus, in terms of enumerating the value of assets for an example. Note that the grayed out areas in this map the purpose of a risk analysis, it is assumed that in ru- were manually digitized and therefore excluded from ral areas there is more than one building per dwelling. the Level 4 procedure. Although most individual cells An investigation to estimate of the average number of are not visible, the patterns of colors give an indica- buildings per dwelling was carried out in rural areas of tion of how households are spread. In PG the patterns a number of countries and it was determined that one of light color indicate higher household density for dwelling comprises, on average, two buildings. regions surrounding urban areas. There are large re- This issue of dwellings versus buildings has been gions with regular settlements in the form of farmland verified by visual inspection (Figure 72), analyses of or development on transportation corridors or water- fully digitized areas, census data and discussions with ways. This is validated by visual inspection with high- experts. For example, the 2006 population and housing resolution imagery (e.g., Google Earth). For some lo- census report for WS indicates that the total number of cations, particularly in southeastern PG, there are vast private households is 23,813, while a total of 46,048 areas with a low concentration of households or no buildings were counted as either owned or rented by households at all. the private households, giving a ratio of 1.93 buildings The census data provided information on the num- to one household. Thus, for the Level 4 building extrac- ber of people per dwelling (household), which in turn tion methodology (i.e., residential building detection in was used to estimate the number of buildings. It is im- rural areas), which relies on the dwelling count, each portant to point out the distinction between a dwelling dwelling detected was counted as two buildings. This and a building. A dwelling is typically defined as the assumption is reflected in the building count reported total living space asset of a family unit. In the rural ar- in Table 2. FIGURE 72. Example of the distinction between dwellings and residential buildings in a rural area near the village of Lembinwen on VU Note: Approximately 20 buildings are visible, while nine households are enumerated from the census (shown as black/white targets) 84 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) The Level 4 building inference methodology was dential purposes. For rural areas, the ratio is slightly based on census data which only enumerates residen- higher, with 88 to 97 percent of the total buildings tial dwellings. Location and counts of non-residential used for residential purposes. Thus, for the Level 4 buildings, including commercial, industrial and public building detection methodology in rural areas, extra buildings were inferred from those of residential build- buildings were added at random among the locations ings. Data from the building field surveys, building of the detected settlement cells, so that approximately counts from completely digitized countries (e.g., NR 95 percent of the buildings were designated as resi- and NU), as well as ancillary sources (A 2007 Water- dential. For example, if 95 buildings were detected in works Division Survey in the CK and the Pacific Cities ten settlement cells in a certain area, five non-residen- Database) indicate that 80 to 97 percent of the total tial buildings were added in one randomly chosen cell buildings in a given area in the PICs are used for resi- among the ten identified. CATASTROPHE RISK ASSESSMENT METHODOLOGY 85 Annex C Construction Type The construction type and general condition of buildings in the PICs varies greatly, from mid-rise re- inforced concrete office buildings and modern wood frame or masonry houses (mainly in urban cen- ters) to traditional style houses and poorly built dwellings. Table 22 and Table 23 present the distri- bution of building construction type for all PICs including statistics for urban areas, extracted directly from the exposure database. Traditional style houses are still very common in the PICs, especially in rural areas. 74 percent of buildings in rural PG (61 percent of all buildings in the entire country) are in built in the traditional style. Each country or region has its own version of traditional construction, but most of the traditional style buildings in the PICs are generally similar (e.g., timber pole or bamboo frames with twine connections and thatched roofs). These houses generally use untreated materials and typically need to be replaced or repaired every four to five years, but may last up to 20-30 years. There are noticeable regional variations of the traditional buildings in the PICs. For example, the most common style of traditional house in WS is the ‘fale’, which is characterized by a large roof supported with timber poles and no walls. Many traditional houses on the coast of PG are built on long stilts to elevate it from the water, while most in-land dwellings in PG are set directly on the soil. Modern style dwellings, which are more common in urban centers, are usually built in colonial style architecture. In general, the modern style buildings in the PICs have lower construction standards compared to other, more industrialized countries. Construction code enforcement is very limited, but there have been recent efforts (e.g., the Pacific Building Standards Project and the South Pacific Disaster Reduc- tion Programme) to modernize and standardize building construction. For modern commercial/public buildings and high-end residential houses, construction practices and design specifications are usually borrowed from Australian and/or New Zealand standards. TABLE 22. Construction type statistics of urban and rural buildings in the 15 PICs, extracted from the exposure database Total Construction Type CK FJ FM KI MH NU NR PG PW SB TL TO TV VU WS Total Total Count 10,602 266,140 31,988 27,589 12,894 1,108 2,755 2,393,279 5,719 169,112 398,685 34,751 3,018 100,746 48,831 3,507,217 Single story timber frame 29.0% 48.5% 42.6% 13.6% 41.7% 49.8% 26.1% 20.2% 47.0% 19.5% 8.0% 53.7% 15.9% 21.1% 31.7% 21.8% Multi-story timber frame 1.0% 0.4% 0.3% 0.6% 1.0% 0.9% 0.4% 0.4% 0.9% 0.4% 0.2% 1.0% 0.7% 0.1% 0.5% 0.4% with closed-under Multi-story timber frame 0.4% 0.2% 0.1% 0.1% 1.2% 0.0% 0.2% 0.5% 0.0% 1.0% 0.1% 0.3% 0.3% 0.0% 0.1% 0.4% with open-under Single story masonry/concrete 46.7% 29.4% 41.6% 27.3% 27.5% 31.6% 51.0% 2.0% 31.8% 5.0% 39.9% 28.8% 35.9% 27.8% 18.6% 10.7% Multi-story masonry/concrete 2.4% 6.6% 1.7% 0.9% 5.2% 0.6% 2.7% 0.1% 11.0% 0.4% 2.9% 2.5% 1.2% 1.5% 1.2% 1.1% Single story combination 5.0% 1.5% 0.7% 2.0% 8.0% 13.8% 5.3% 1.3% 1.6% 1.0% 0.5% 2.7% 11.0% 1.2% 2.4% 1.3% masonry/concrete & timber frame Multi-story combination 1.3% 0.3% 0.1% 1.0% 2.0% 0.6% 0.8% 0.3% 1.0% 0.8% 0.3% 1.5% 2.8% 0.1% 0.6% 0.3% masonry/concrete & timber frame Single story steel frame 0.7% 0.4% 1.4% 0.9% 1.0% 0.9% 2.1% 1.0% 1.4% 0.5% 0.4% 0.7% 0.5% 0.7% 0.8% 0.9% Multi-story steel frame 0.1% 0.0% 0.1% 0.0% 0.1% 0.0% 0.2% 0.0% 0.3% 0.0% 0.1% 0.1% 0.0% 0.0% 0.0% 0.0% Open walled structure with 0.2% 0.9% 0.6% 2.3% 1.2% 0.0% 1.8% 0.6% 0.6% 0.5% 0.4% 1.5% 0.7% 0.8% 9.0% 0.7% non-wooden pole frame Open walled structure with 0.1% 0.4% 0.2% 0.9% 0.3% 0.0% 0.7% 0.1% 0.2% 0.0% 0.1% 0.3% 0.3% 0.3% 31.4% 0.6% wooden pole frame (fale) Uninhabitable or poor 3.0% 6.8% 0.8% 2.1% 3.9% 0.0% 4.8% 6.6% 1.9% 2.4% 6.3% 4.6% 5.5% 7.2% 1.6% 6.2% construction Traditional 1.1% 3.4% 8.7% 45.1% 5.0% 0.0% 1.0% 66.3% 0.3% 61.4% 38.1% 1.0% 11.0% 36.8% 0.0% 54.3% Other single story 7.9% 0.9% 1.0% 2.6% 1.6% 1.7% 2.5% 0.5% 1.2% 6.8% 2.5% 1.0% 13.7% 2.1% 2.0% 1.2% Other multi-story 1.2% 0.2% 0.1% 0.6% 0.3% 0.0% 0.4% 0.1% 0.6% 0.4% 0.2% 0.2% 0.4% 0.2% 0.2% 0.1% 86 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) In urban areas, the construction type of most buildings indeed masonry. Thus, for the exposure database, ma- is timber frame or masonry/concrete (see Table 23 for sonry and concrete constructions were grouped into a the distribution of construction types). Timber frame single category. Note that this grouping does not imply buildings are typically built with light timber members that the vulnerability of concrete and masonry build- and the walls are generally made from timber planks, ings is the same. Distinct DFs (vulnerability models) of plywood, solid panel (fibre-cement sheets) or metal this grouped concrete/masonry construction category sheets. Masonry buildings are usually lightly reinforced were developed by referencing the occupancy type with slender steel bars or not reinforced at all. Some and/or the building secondary characteristics. masonry buildings are constructed with concrete block walls and beams that confine the block wall. Larger Combination masonry/concrete and timber-frame buildings, such as modern government facilities and buildings are also very common in the PICs. Generally, major commercial buildings are usually constructed for single story buildings of this type, one section of with reinforced concrete under more stringent design the building is constructed out of masonry/concrete standards. In general, it is difficult to determine the while another part (usually an addition or attached quality of masonry or concrete construction from field room) is constructed out of timber. Likewise, for mul- surveys, since the structural system is obstructed by the tistory buildings of this type, the bottom floor is often walls or paint. In fact, one of the only ways to deter- built of concrete/masonry while the top floors are tim- mine the quality of the masonry/concrete construction ber frame. Industrial buildings, which are usually large is to inspect the site during construction or view the warehouse-type structures, are typically made of light- damaged structure during disaster reconnaissance, gage steel and corrugated metal walls. Poorly con- when the interior reinforcement is exposed. Because structed buildings, which generally consist of an im- of this, there is some uncertainty in the true distribu- provised wooden frame with corrugated metal walls, tion of concrete or masonry construction, although it are also common in the PICs. is understood that the majority of these buildings are TABLE 23. Construction type statistics of urban buildings in the 15 PICs, extracted from the exposure database Urban Construction Type CK FJ FM KI MH NU NR PG PW SB TL TO TV VU WS Total Total Count 7,607 99,707 6,904 9,749 7,487 1,108 2,755 187,659 3,991 36,226 139,170 27,794 1,432 33,190 13,250 578,029 Single Story Timber Frame 33.8% 37.4% 43.5% 19.9% 33.3% 49.8% 26.1% 43.6% 38.2% 31.1% 9.4% 51.2% 23.6% 21.2% 39.4% 31.7% Multi-Story Timber Frame 1.4% 0.9% 1.4% 1.6% 1.8% 0.9% 0.4% 5.3% 1.4% 1.7% 0.6% 1.3% 1.5% 0.2% 1.8% 2.3% with Closed-Under Multi-Story Timber Frame 0.6% 0.6% 0.4% 0.2% 2.0% 0.0% 0.2% 5.3% 0.0% 4.5% 0.2% 0.4% 0.7% 0.0% 0.3% 2.2% with Open-Under Single Story Masonry/ 40.5% 28.3% 29.9% 36.0% 34.2% 31.6% 51.0% 6.3% 33.6% 10.8% 58.4% 29.4% 36.0% 38.1% 23.0% 28.4% Concrete Multi-Story Masonry/ 3.4% 17.3% 7.1% 2.6% 9.0% 0.6% 2.7% 1.4% 15.7% 1.9% 8.1% 3.2% 2.6% 4.6% 3.8% 6.4% Concrete Single Story Combination 6.3% 1.9% 1.1% 1.8% 4.5% 13.8% 5.3% 2.0% 1.8% 2.0% 0.6% 3.3% 9.4% 3.1% 3.7% 1.9% Masonry/Concrete & Timber Frame Multi-Story Combination 1.8% 0.8% 0.6% 2.8% 3.4% 0.6% 0.8% 3.2% 1.5% 3.8% 0.9% 1.9% 5.8% 0.3% 1.9% 2.0% Masonry/Concrete & Timber Frame Single Story Steel Frame 0.7% 0.7% 2.4% 1.5% 1.5% 0.9% 2.1% 4.0% 2.0% 1.4% 0.7% 0.8% 0.5% 1.7% 1.0% 1.9% Multi-Story Steel Frame 0.1% 0.1% 0.3% 0.1% 0.2% 0.0% 0.2% 0.5% 0.4% 0.1% 0.2% 0.1% 0.0% 0.0% 0.1% 0.2% Open Walled Structure with 0.2% 1.8% 1.3% 4.0% 1.7% 0.0% 1.8% 3.0% 0.6% 1.6% 0.8% 1.8% 0.8% 2.0% 5.5% 2.0% Non-Wooden Pole Frame Open Walled Structure with 0.1% 0.8% 0.5% 1.9% 0.5% 0.0% 0.7% 0.6% 0.2% 0.0% 0.2% 0.3% 0.3% 0.8% 12.9% 0.8% Wooden Pole Frame (Fale) Uninhabitable or Poor 3.2% 5.8% 2.4% 3.9% 4.1% 0.0% 4.8% 15.4% 2.1% 7.0% 6.7% 4.1% 7.3% 7.3% 3.1% 9.0% Construction Traditional 0.2% 1.6% 7.0% 19.2% 1.7% 0.0% 1.0% 7.4% 0.3% 26.6% 11.4% 0.7% 1.4% 16.4% 0.0% 8.5% Other Single Story 6.0% 1.5% 1.5% 2.7% 1.6% 1.7% 2.5% 1.1% 1.5% 5.9% 1.3% 1.1% 9.3% 3.8% 3.1% 1.9% Other Multi-Story 1.7% 0.5% 0.6% 1.6% 0.5% 0.0% 0.4% 0.8% 0.9% 1.7% 0.7% 0.3% 0.8% 0.5% 0.5% 0.7% CATASTROPHE RISK ASSESSMENT METHODOLOGY 87 Annex D Infrastructure Exposure Database The main types of infrastructure considered, along with total counts for each country and the general scope (and quality) of the data, is presented in Table 24. Most of the assets whose ex- tent is described as “comprehensive” and “extensive” have been completely geo-located. For example, almost all airports (and airstrips), major dams, mines, ports and power plants (wind, solar, fossil fuel, hydroelectric) and most oil/gas plants, docks, and major bridges (concrete, steel, and timber) have been geo-located. Most of the assets listed under “urban zones” refer to data collected in major urban centers, mostly from the field surveys. Only a few of the bus stations and helipads were located, however. Finally, major roads (mostly paved roads) and major railways have been indexed (FJ is the only PIC with major railways). TABLE 24. Inventory of infrastructure indexed for the PICs Type Data extent CK FJ FM KI MH NI NR PG PW SB TL TO TV VU WS Total Airport Comprehensive 11 28 12 23 38 1 1 519 3 41 8 6 1 34 5 731 Bridge Extensive 28 738 51 28 2 – – 685 15 49 316 2 – 27 52 1,993 Bus station Few – 7 – – – – – 1 – – – – – – 1 9 Communications Urban zones 12 25 – 1 5 – 1 38 3 46 – 28 – 31 9 200 Dam Comprehensive – 3 – – – – – 4 – – – – – – – 7 Dock Extensive 27 31 101 9 26 1 6 64 39 39 1 62 4 16 6 432 Generator Urban zones 1 17 – – – – – 56 – 14 – 5 – 5 4 102 Helipad Few – – – – 7 – – – – – – – – – – 7 Mine Urban zones – – – – – – – 7 – 1 1 – – – – 9 Oil & gas Extensive – 2 – – – – – 10 – 5 1 – – 1 1 20 Port Comprehensive 1 11 6 6 8 1 1 21 5 8 2 4 1 4 5 84 Power plant Comprehensive 14 45 22 4 14 1 1 64 2 21 16 4 9 5 11 233 Water intake Urban zones 18 22 – – – 21 – 8 – 9 – 176 – 6 4 164 Storage tank Urban zones 229 456 60 85 173 3 188 349 39 120 41 169 31 115 64 2,122 Sub-station Urban zones 7 5 – – – – – 6 – 1 10 – 16 10 – 55 Water treatment Urban zones – 10 – – 1 – 1 5 1 3 2 – – 2 2 27 88 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) Table 25 indicates how many kilometers of roads and count provided by the CIA World Factbook. This com- railways have been indexed. Also shown is a compari- parison indicates the breadth and accuracy of the infra- son of infrastructure assets indexed in this study with a structure database. TABLE 25. Inventory of roads, rail ways, airports, and port with comparisons to the CIA World Factbook Total road length (km) Rail length (km) Airport count Port count Country Database CIA* Database CIA* Database CIA* Database CIA* CK 130 320 (33) 0 0 11 10 1 1 FJ 3,540 3.440 (1.692) 408 597 28 28 11 3 FM 185 240 (42) 0 0 12 6 6 3 KI 144 670 0 0 23 19 6 3 MH 52 2.028 (75) 0 0 38 15 8 3 NR 22 24 (24) 0 0 1 1 1 1 NU 115 120 (120) 0 0 1 1 1 1 PG 4,692 9,349 (3,000) 0 0 519 562 21 5 PW 167 n/a 0 0 3 3 5 1 SB 14 1,360 (33) 0 0 41 36 8 4 TL 4,241 6,040 (2,600) 0 0 8 6 2 1 TO 367 680 (184) 0 0 6 6 4 3 TV 16 8 (8) 0 0 1 1 1 1 VU 567 1,070 (256) 0 0 34 31 4 3 WS 652 2,337 (332) 0 0 5 4 5 1 *Numbers in parentheses are paved roads CATASTROPHE RISK ASSESSMENT METHODOLOGY 89 Annex E Example of Consequence Database An example of the consequence database, which lists some of the most devastating perils ever recorded for the 15 PICs, is reported in Table 26. This table is condensed for brevity; the actual database lists additional details and data fields. TABLE 26. A short list from the consequence database Number Total Total economic loss of people life (nominal million US$) Event Country Year affected loss Low estimate High estimate Notes Apia, TC WS 1889 – 147b – – Most deaths from shipwrecks TC FJ 1931 – 200a – – – Half of the people killed from land- TC VU 1951 – 100a 0.25a 0.25a slide on Ep Severe wind storm SB 1956 – 200a – – – Severe local storm WS 1964 – 250a – – – TC Bebe FJ 1972 120,000a 3b 22.5b 22.5b 500 buildings damaged Earthquake & tsunami SB 1975 – 200a – – All deaths from tsunami Tremendous crop losses; TC Meli FJ 1979 359,000a 53a 0.4c 0.4c 2-3 m surge height above normal sea level 45,000 homeless; TC Isaac TO 1982 146,512a 6a 20.4b 22d 90% banana crop destroyed; Most buildings destroyed TC Oscar FJ 1983 200,000a 9a 50b 76d 1000 buildings damaged/destroyed Extensive damage to houses & Flood from heavy rain FJ 1986 215,000a 19a 15.4a 30d infrastructure 90,000 homeless; TC Namu SB 1986 150,000a 101a 10b 20d 65 Injured, 5,805 houses damaged, 6,096 houses destroyed TC Tusi WS 1987 2,000d 0 100d 100d 40 injured, most buildings destroyed 15,000 homeless, thousands of hous- es damaged, 5,000 houses destroyed; TC Uma VU 1987 48,000a 48a 25c 150d Loss of life possibly from shipwreck; 95% of buildings in Port Vila dam- aged TC Ofa WS 1990 195,000a 8a 120e 200a Devastated entire island TC Val WS 1991 88,000a 13a 200e 278a 80% of buildings damaged/destroyed Flood from heavy rain PG 1992 90,000a - 12i 15d 100,000 homeless Earthquake PG 1993 20,200a 60f 5a 5a 10,000 homeless, 200 injured TC Kina FJ 1993 160,000a 23d 100a 120e 19,000 homeless Earthquake PG 1993 20,200a 53f 5a 5a 200 injured 4,134 properties damaged; Extensive damage to crops (40% TC Gavin FJ 1997 3,500a 25a 5d 33.4g destroyed) and buildings (90% destroyed in Yasawa region) 12,000 homeless, 30 dead, 10 TC Justin PG 1997 15,000a 8a 150d 150d missing, more than 500 buildings destroyed 90 Pacific Catastrophe Risk Assessment and Financing Initiative (PCRAFI) Number Total Total economic loss of people life (nominal million US$) Event Country Year affected loss Low estimate High estimate Notes 70% of houses damaged on TC Paka MH 1997 – 0l 80l 100b Ailinglaplap Atoll 9,500 homeless; Earthquake & tsunami PG 1998 9,867a 2,183f 5f 24f 1,000 injured, “many” houses (~101-1,000) destroyed Housing, agriculture, schools, and TC Dani VU 1999 100,000c 32c 6.4c 6.4c health facilities damage; Damage to crops (taro, manioc) 100 people injured, “many” houses Earthquake & tsunami VU 1999 14,100c 18d 5f 24f (~101-1,000) destroyed infrastructure severely damaged 5,985 buildings damaged, 2,662 TC Ami FJ 2003 45,000d 17a 30a 65.1b houses destroyed 80% food crops (predominantly TC Ivy VU 2004 54,000a 2a 45c 45c mango and banana) damaged Whole country affected, damage to 90% of buildings (housing, hospital, TC Heta NU 2004 702a 1e 5d 55c commercial buildings), crops, utilities and transport systems; 12 houses, one hospital destroyed, 200 homeless TC Guba PG 2007 162,140a 172a 70d 183i 1,000 houses destroyed Earthquake & tsunami WS 2009 10,000c 149f 147.5j 200k All deaths from tsunami Sources: a EMDAT: The Emergency Events Database b Wikipedia c AusAid: The natural disaster database maintained by AusAID d NatCat: The Natural Catastrophe Loss Database (NatCatSERVICE) issued by Munich Re e Sivakumar et al. (2005) f NOAA g Fijian Government h PDN: The disaster database maintained by the Pacific Disaster Network i DFO j ReliefWeb k Okal et al. l NOAA CATASTROPHE RISK ASSESSMENT METHODOLOGY 91 Annex F Country Risk Profiles Melanesia ■■ Republic of Fiji (FJ) ■■ The Independent State of Papua New Guinea (PG) ■■ Solomon Islands (SB) ■■ Republic of Vanuatu (VU) Micronesia ■■ Federated States of Micronesia (FM) ■■ Republic of Kiribati (KI) ■■ Republic of the Marshall Islands (MH) ■■ Republic of Nauru (NR) ■■ Republic of Palau (PW) Polynesia ■■ Cook Islands (New Zealand) (CK) ■■ Niue (New Zealand) (NU) ■■ Kingdom of Tonga (TO) ■■ Tuvalu (TV) ■■ Samoa (WS) SE Asia ■■ Democratic Republic of Timor-Leste (TL) The country risk profiles are available online on the Web site: http://pacris.sopac.org The World Bank 1818 H Street, N.W. Washington D.C. 20433, USA