A BACKGROUND PAPER >> INFRASTRUCTURE RESILIENCE Resilience in the Caribbean: Natural Hazards Exposure Assessment and Areas for Future Work AE Schweikert, GL L’her, LG Nield, SW Kerber, RR Flanagan, MR Deinert 12 June 2020 This work is a product of the staff of The World Bank and the Global Facility for Disaster Reduction and Recovery (GFDRR) with external contributions. The sole responsibility of this publications lies with the authors. The findings, analysis and conclusions expressed in this document do not necessarily reflect the views of any individual partner organization of The World Bank (including the European Union), its Board of Directors, or the governments they represent, and therefore they are not responsible for any use that may be made of the information contained therein. Although the World Bank and GFDRR make reasonable efforts to ensure all the information presented in this document is correct, its accuracy and integrity cannot be guaranteed. Use of any data or information from this document is at the user’s own risk and under no circumstances shall the World Bank, GFDRR or any of its partners be liable for any loss, damage, liability or expense incurred or suffered which is claimed to result from reliance on the data contained in this document. The boundaries, colors, denomination, and other information shown in 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. Executive Summary Evaluating the resilience of critical infrastructure systems (power, transport and water sectors) in the Caribbean requires geospatial data on infrastructure, hazard frequency, fragility, and downstream impact modeling. Ideally, resilience analyses can inform decision makers about their most vulnerable assets, losses that would have the largest impacts, and how to prioritize funding in the reality of limited budgets. The assessment of multiple assets within each sector is important to understand interdependencies and how loss of one system (e.g. transmission and distribution) can affect others (e.g. business operations). The interdependency between systems is also important to inform holistic planning. For example, roads are a key asset for water delivery in times of drought or pipe failure and for travel to hospitals or other critical facilities. Taking a systems perspective can help resiliency planning by identifying the most vulnerable and important assets to reduce repeated failure of critical assets. In many countries with high exposure to natural hazards (such as small island states in the Caribbean), a recovery trap is caused by continued repair of repeatedly damaged assets. Climate change may cause this to worsen as the intensity and/or frequency of events including hurricanes, coastal surge, flood and heatwaves changes. Summary of Findings In total, the available infrastructure for 16 countries in the Caribbean were analyzed (Figure ES1). A variety of sources including OpenStreetMap, CHARIM, country-specific data, reports, GoogleMap queries and others were used to compile as much publicly available data as possible. Road infrastructure, power plants, bridges, hospitals, water treatment facilities, ports and airports were assessed because some inventory was available for at least a few countries. Due to limited data on the design of most infrastructure, this report focuses on hazard exposure, which is the first step in risk assessments but stops short of damage estimates. When better information on specific infrastructure assets is available, the methodology and findings in this report can be used as a foundation for future work. The geospatial methods used for this analysis allow for detailed assessment of overlapping hazards, and co-located infrastructure. As more data is available, these methods can be used to understand the risk of cascading failures and broader impacts from infrastructure damages from natural hazard events. Figure ES1: All countries considered in this analysis (purple) For transportation infrastructure, the analysis shows that hurricanes, landslide susceptibility and seismic risk are the highest for road infrastructure. Bridges see a high exposure to pluvial flooding (26% overall, but with some countries above 1/3 of assets including Grenada and Jamaica). Hurricane, landslide and seismic risk exceed 50% of bridge assets for the Caribbean. Airports, especially those built for general aviation (“large” airports), have very high exposure to hurricane risk (100% of facilities). Ports see high exposure (>50%) to hurricanes and seismic risk throughout the Caribbean, although this varies by harbor size, which is an important consideration for analyzing the operability of the most critical facilities. Understanding the vulnerability of these facilities, and the implications of failure in transport of goods and people, is an important area of future work. Power systems were assessed and showed the greatest exposure to hurricanes (64%). In the Dominican Republic, a case study of transmission line vulnerability and criticality showed that loss of different sections of transmission line has a vastly different impact on how many end users lose power. Analysis of the supply chain impact from road failure between ports and diesel fired power plants was assessed for Dominica and showed that location, exposure, distance and redundancy all have impacts on reliable fuel delivery. For water systems, the major limitation for analyses of resiliency in the Caribbean is a lack of data on the physical location, age, and characteristics of the underlaying assets. Public information on water transmission pipelines is notably absent, or incomplete, for all locations. Some islands, including Dominica, have long-term planning programs to upgrade their water systems. Combining this local expertise with data on pipelines, reservoirs, catchments, and other assets could inform a resilience analysis. The largest limitation for this study was data availability, especially for infrastructure assets. Therefore, results are provided at different levels for use in planning even where specific asset data might be missing. Supplemental Information File 1 contains exposure maps for every hazard/asset combination where data existed (over 90 files). Appendix A contains high resolution multi-hazard maps for every country that can be used for identifying high risk regions within each island. A case study for Dominica is done because higher resolution data is available for hazards and some infrastructure assets. However, when speaking with local experts from the Climate Resilience Execution Agency for Dominica (CREAD), it was noted that much of the available data from OpenStreet Maps and other sources was lacking critical context, such as hospital facilities currently closed for repairs, or ports that were too shallow for large ships to deliver fuels. These findings, far from limiting the applicability of the results, highlight the importance of local input and data. Several metrics were developed for this study to assess the impact of infrastructure failure. These include hospital impact from road network failure, water distribution impact from road network failure, power plant impact from interrupted fuel delivery from port and road failures, and impact to electricity delivery from transmission line failure. Each metric is assessed using a flexible network analysis tool designed to determine impacts from a loss of network (e.g. road, transmission) connectivity. The tool allows for a straightforward measure of how connectivity loss will affect travel cost/time between points, or the number of goods/persons who can reach a destination. This tool can integrate hazard-defined failure probabilities of infrastructure segments. Monte Carlo assessments (MC) can be done on the network using these failure probabilities to help inform possible outcomes. Currently, this tool has been developed for internal use, but could be further developed to be used by local administrators. Depending on level of user interface development, this could be for geospatial specialists with some coding background or further developed with a GUI interface to allow non- expert users to perform analyses. The network tool could also be useful in planning exercises when considering the development of new infrastructure or the impacts of climate change. Future Work This study highlights several key areas where future work can focus: 1) Fill data gaps for most critical infrastructure assets, and, when data collection is done, standardize collection and reporting processes for use in multiple contexts. Water data (distribution, water and wastewater treatment facilities, dam location and design) and power data (especially transmission and distribution infrastructure) was particularly difficult to find in collated form for most countries. Inadequate data for some countries (Guyana and Suriname for most data) precluded their inclusion in this assessment but is a good indication of areas where future data collection can further inform vulnerability assessments; 2) Where data was available, it was often in disparate formats with limited metadata. One best practice moving forward could be to report all data in a standardized geographic format, such as WGS84 – EPSG4326 projection. One example in this study was hazard data which, for one set of data, included a different projection for every island. In the Bahamas, this translated to over 100 unique geographic file formats that had to be individually transformed to a common geospatial perspective. This practice can easily lead to incorrect results if not accounted for and takes considerable time to identify and adjust. The results from this study address Recommendation #2, “Building Institutional Resilience” from the Lifelines Report, including the “identification of critical infrastructure and defining acceptable and intolerable risk levels” (Action Item 2.2). The network analysis tool described could also be extended to aid in local planning capabilities. By integrating natural hazard and climate data, providing analysis designed to aid in decision making and increase capacity through specific training it could address Recommendation #4, “Improve Decision Making”, and Actions 4.1-4.3 [1]. Background This study is focused on assessing the vulnerabilities of critical infrastructure (power, water, transport) in the Caribbean to natural hazards and climate change impacts. This work builds on ongoing initiatives aimed at resilience as well as past studies highlighting the importance of identifying the most vulnerable assets, the impacts they may have when damaged, and how to prioritize limited investment funding. A lot of existing literature and programs highlight the importance of more robust consideration of natural hazards in infrastructure planning as well as the disproportionate impact they have on small islands developing states (SIDS). The United Nations ECOSOC has a ‘Pathways to Resilience’ resource page focused on resilience and risk-informed background information, cooperative meetings and next steps highlighting the high risks from climate change and cyclones [2]. The International Institute for Sustainable Development (IISD) discusses the Caribbean Resilience Fund and the risk posed to the region from climate and hazard events [3]. The Inter-American Development Bank (IDB) highlights the benefit of investments in climate resilience in the region, although treats climate and hazards separately [4]. The World Bank has several initiatives focused on reducing risk from natural hazards. These include the Small Island State Resilience Initiative Program (SISRI) [5] which highlights that more than half of the countries most vulnerable to natural hazards are SIDS [6]. The 2019 document, Lifelines: The Resilient Infrastructure Opportunity, is a global review of critical issues related to resilience in telecommunications, water, transport and electricity sectors [1]. The report shows that every $1 spent in more resilient infrastructure yields up to $4 in benefits. Readers are referred to this report for robust background in the topic for each sector, including some work on exposure and vulnerability of assets, but also discussion of the important policy, funding and capacity building needs related to more resilient and sustainable institutions, processes and infrastructure. A 2019 article by Koks et al. [7] looks at multi-hazard risks to transport infrastructure (road and rail). The authors find expected annual damages globally of $3.1-22 billion USD from direct impacts of natural hazards, the majority coming from flooding. Wealthier countries invest more in resilient infrastructure design and hazard mitigation. This is beneficial considering that more than half of assets are likely to have positive returns on flood protection investments. However, the findings illustrate the geographic variation in risk, as well as the economic context, with countries that have higher exposure and lower income bearing a high proportion of risk relative to inventory and GDP. Network vulnerability of infrastructure to hazards has been investigated for seismic risk to a historic transmission grid in the US [8], to gas networks in Europe [9] as well as a small set of water and power infrastructure [10]. Vulnerability as a function of the dependency between systems is highlighted as an important consideration in [9], [10]. The impact of storm surge and sea level rise on roadways and accessibility of regions was investigated using simulated risks for Spain [11]. Two recent case studies focused on network vulnerability highlight the critical role that transport plays in vulnerability assessments and resilient planning [12] [13]. In Vietnam, rapid growth has led to expansion of road networks, yet much of the infrastructure has not been designed for resilience against climate and natural hazards [12]. This study showed that the impact of hazards on the transportation networks is high and likely to increase under climate change. The most critical and vulnerable segments are identified to help prioritize planning investments. In Mozambique, network analysis methods inform the relative value of road segments (most critical segments) based on origin-destination modeling for agricultural goods and fisheries (time/access to market). The transportation network is assessed based on risk from hazards and roads are prioritized based upon economic and risk considerations as well as social metrics such as poverty. Resilience interventions are also assessed under a range of climate scenarios and a cost-benefit analysis is done to identify where investments may be most beneficial overall [13]. Data and Methods Infrastructure and Hazard Data The data used in this analysis comes from a variety of sources. Table 1A lists the data available for each country and its resolution. In Table 1A, hazards for which very little or no data was available were not included, such as volcanic hazard or windstorms not resulting from hurricanes. Some data was only available in specific return intervals (flooding, coastal surge, seismic) or semi- quantitative data (landslide susceptibility). Where possible, return interval data was chosen for similarity across data sets (such as the 0.02 annual probability of occurring). For flooding and coastal surge data, inundation (depth) was the information available, excluding important considerations for impact and damage assessments such as velocity. For hurricane data, only windspeed was available. This excludes important additional hazard stress such as rainfall and surge inundation. In some countries (such as Dominica) specific, post-hazard surveys and data collection was done (e.g. [14]). While no concurrent hazard impact modeling was done for this analysis, multi-hazard maps are presented in Appendix A and can inform locations that, for example, are high risk for multiple hazards such as coastal surge and hurricanes. For seismic data, the data gives the expected peak ground acceleration (PGA) for a 1/475-year event. This is equivalent to a 10% probability of occurrence in 50 years and is a common output for earthquake modeling and can be used to help determine building standards in different geographies [15], [16]. Landslide susceptibility data was available on a qualitative basis (“high”, “medium” and “low” risk). It should be noted that susceptibility is not equal to landslide risk, as the probability of landslides occurring depends on a number of factors, including water content of soil, soil type, slope, vegetation, mitigation activities, seismic activity and other factors. Different seasons, compounding and concurrent hazards (flooding, rainfall, seismic) and other factors are important considerations for future work in landslide risk modeling. Table 1B lists the available data for infrastructure assets and their resolution. It should be noted that where infrastructure data is available, it is likely incomplete, as most data sources are not from country-specific studies or governing bodies. This is an area for future work and more in-depth analyses. Tables 2A and 2B summarize the data sources for hazard and infrastructure data, respectively. In many cases, the infrastructure datasets were augmented with research from reports, GoogleMap queries, and infrastructure management agency sources. For some countries and infrastructure types, very limited data was available and therefore excluded from the analysis. Where the sources classified infrastructure by size or type, these classifications were used. For ports, data is classified by the harbor size (“very small”, “small”, “medium” and “large”) based upon a combination of factors including area, facilities, and wharf space as defined in [17]. Airports are similarly classified provided for “small” airports, “medium” airports, and “large” airports, defined based on the data classifications, from Megginson (2020) [18]: Small airports have little to no scheduled service and light general aviation traffic; Medium airports have scheduled regional airline service, or regular general aviation or military traffic; Large airports have major airline scheduled services with millions of passengers per year, or denote major military bases. For roads, classifications were provided for road types 1-5. Because specific design considerations (surface types, number of lanes) are not always included in this data, classifications were done based upon author expertise and discussions with subject area experts. “Primary” roads are defined as classifications 1 and 2 road type, “secondary” roads are type 3, and “tertiary” roads encompass roads types 4 and 5 [19]. These classifications can be adjusted when local information on design standards is available and can be matched with appropriate fragility curve information to determine hazard impacts. An important note here is that while an island state may consider the major road on the island their ‘primary’ road, that does not necessarily correspond to ‘primary’ road design considerations in other locations. For example, a primary road in the United States or Western Europe may be made of cement and be 4-8 lanes wide or more. The fragility and damage curves that could be applied to this primary road would not correspond to a primary road made of asphalt that is 2-4 lanes wide. This is a noted area for local expertise and future considerations and an important component of data collection and modeling estimates. In general, water infrastructure data was extremely limited and not publicly available. Wastewater treatment through the Caribbean is limited with significant direct discharge and use of septic systems and latrines [20]. Septic systems are prone to failure when soils become saturated, which can happen during periods of heavy precipitation if soil drainage capacity is exceeded. Data on water transmission and distribution lines was available only for Dominica and then excluded location information. The total length of each (250km and 450 km respectively) as well as their general construction material (e.g. old pipes mainly consist of a galvanized steel, while newer lines consist mainly of cement-lined ductile iron, PVC, and HDPE) is insufficient for hazard exposure modeling [21]. However, some Caribbean island states have detailed plans for upgrades to their water systems [22]. Incorporation of detailed data would allow a much more detailed resiliency assessment for water systems using the tools presented in this analysis. An important consideration for this assessment is that limited (if any) information on design parameters was available for most infrastructure. Without basic information such as the size of a facility, pipe material (steel, PVC), transmission pole structure (wood, composite, cement, monopole or lattice structure) or road surface (cement, bitumen/asphalt, gravel, dirt), impact and damage estimates are impossible to assess. Therefore, this study focuses on exposure analysis for all country/hazard combinations. An illustrative damage estimate is done for the road network based in Dominica for flooding inundation. Future work on water systems could include extending data for water systems. The location of treatment plants (including their capacities and elevation) and water transmission lines can sometimes be obtained through extensive searches of local media that cover construction and maintenance projects. For some countries, including Dominica, detailed, local data was available at high resolution. This includes both hazards (e.g. [23]) and infrastructure data (e.g. [24], [25]). With local consultation, better incorporation of additional hazards, such as volcanic risk, could be included (for example, maps are available of qualitative risk (“very high”, “high”, etc.)). For infrastructure data, Dominica had many recent reports and assessments that assessed existing infrastructure (such as DOWASCO’s water infrastructure data and the Army Corps of Engineers recent assessments [26], [27]). Additional work was done to identify additional infrastructure resources. For this reason, Dominica is chosen as a case study in the results to illustrate the impacts of hazards on infrastructure. Additional future work in this area would include more detailed data from additional countries (location of infrastructure) for exposure assessments. If design information was also available (such as transmission line voltage, power plant fuel types and size, wastewater treatment facility capacity and technology, bridge span and width, road drainage design and surface type, etc.) additional impacts assessments can be evaluated. Impact assessments, particularly those focused on damage costs, downtime and repair estimates, require some design information to assess. For climate change impact, an energy metric, Cooling Degree Days (CDD), was chosen for analysis. Cooling degree days are a common measure of energy demand that measures the difference between daily minimum and maximum ambient temperature relative to a set indoor temperature (ie: [28]). These data were developed by the authors, using downscaled daily temperature data from [29]. The results are shown for the 2050 decade, which is the average annual value of years 2050-2059. The change from a 30-year historical baseline was calculated using average annual data from 1970-1999. Table 2A: Hazard Exposure Data Hazard [metric] Source International Best Track Archive, NOAA [30]. Hurricanes [windspeed] (historical) Historical storm track data processed and gridded to 0.125-degree resolution for annual probability by authors Landslide [qualitative risk ranking] World Bank [31] Authors calculation, Data from NEX-GDDP [29] Daily temperature values used to calculate cooling degree Climate Change: Temperature [CDD] day metrics for historical and future annual averages for 0.25-degree resolution global data Flooding [pluvial, fluvial, 0.02 annual event] World Bank (SSBN) [32] Seismic [PGA] [1/475] Local Data: World Bank [33] Global Data: CHRR [34] Coastal surge [0.02 annual event] High resolution modeling data for each island [35] Table 2B: Infrastructure Assets Data and Sources Asset Source Road network GRIP4 [19] Ports National Geospatial-Intelligence Agency [36] Airports ourairports.com [18] World Bank (CHARIM) [25] Bridges OpenStreetMap (2019) Global Power Plant Database [37] Power plants Augmented with specific research data Water and Wastewater Treatment Plants Local data sources, country reports, OpenStreetMap, (WWTP) Google queries. (e.g.: [20, 26, 38]) World Bank (CHARIM) [25], Buildings (hospitals, other) OpenStreetMap (2019) Analysis Methods Infrastructure Exposure to Hazards For the infrastructure and hazard data sets described in Tables 1 and 2, exposure analyses were performed for every country and set of infrastructure/hazard combination. Input data was processed using Python, uploaded to an SQL database and geographical queries were used to identify overlaps, aggregate data, and produce spatial outputs. Figure 1 provides an example of analysis for single asset/single hazard exposure looking at road network exposure to the 0.02 annual flood probability (“50-year” flooding event) in panels A-C, while Panel D shows the probability of failure based on fragility curves. This last step (Panel D) is explained in greater detail in the Dominica Case Study section below. As a note, only road classification and location data was available, excluding important considerations such as drainage infrastructure (culverts were not considered). Figure 1 - Exposure and damage impact example (Dominica). Road infrastructure impact from 0.02 annual probability (“50-year flood”) flood (combined pluvial, fluvial and coastal surge) event. Panel A is the road network in Dominica. Panel B shows the 0.02 flood water depth. Panel C is road exposure to flood depth. Panel D is road failure probability by segment. Many of the hazard datasets have different time periods of return probability, resolution and some datasets are only qualitative or semi-quantitative. Therefore, a simple framework was developed to allow for simultaneous consideration of each of the datasets. Following the methodology employed by Koks et al. [7], a reclassification scheme was used to assign levels of intensity for each hazard (0-5, with “0” being no risk or unknown risk, “1” being very low risk and “5” being extremely high risk). Specifically included in the multi-hazard model mapping are hurricanes, fluvial flooding, pluvial flooding, coastal surge, landslide, and seismic risks. The highest resolution data available in each location for each hazard was used. A full description of the data used and how reclassification was done for each hazard can be found in Table 3. Where return periods were available, the 0.02 annual probability was used (1/50), which included flooding and coastal surge. For seismic risk, the available data was expected Peak Ground Acceleration (PGA) based on the 1/475 return period event, which is equivalent to a 10% exceedance in a 50-year timeframe. PGA is used in seismic design standards for structures and increases with higher intensity earthquakes (for example fragility curves using PGA values, see [39]). For hazards where data is qualitative or semi-quantitative (such as landslide susceptibility, ranked from “low” to “high”), ‘exposure’ was determined by selecting all geographies where the hazard exceeded an intensity level of “2”. For seismic and flood data, the classification from [7] was used. For hurricane windspeed data, the highest category of storm expected within a 50-year period with at least 20% probability of occurrence was used. Storms of intensity at Category 2 or higher from the Saffir-Simpson scale were considered ‘exposed’. This is detailed in Table 3 below and results can be found for each country in Appendices A and B. Multi-Hazard Assessment Using the classifications described in Table 3, ranking hazard intensity from 0 (none) to 5 (extremely high risk), a multi-hazard assessment was done. This enables an assessment of risks in aggregate with the objective of identifying areas that have relatively higher risks when multiple hazards are considered. There is no dependence between hazards assessed here due to data limitations, only the geographic overlap in hazard occurrence. Classification for each hazard is shown in Table 3. In addition to the multi-hazard risk analysis, an exposure assessment for each infrastructure asset (Table 2B) was completed for every country. Because of the extremely high amount of data processed (16 countries, 8 infrastructure asset categories, 6 hazards and 6 potential classifications of hazard intensity), results are given for all assets, in each country, for hazards based on a binary exposure threshold of hazard intensity. In Table 3, boxes in blue indicate inclusion in exposure analysis for infrastructure assets (full results in Appendix B). By overlaying each hazard risk layer in each location, a single map is generated with a scoring matrix of 0-30, with “0” being no risk/unknown risk for all hazards and “30” being extremely high risk for all hazards assessed. As a note, there is no location with a score >25 (which would indicate all 6 hazards have the highest exposure factor). Appendix A details these findings for all countries excluding Suriname and Guyana, where hazard input datasets were insufficient and/or incongruent for the analysis of multi-hazard risk. Below, Figure 3 details the multi-hazard maps for Dominica and Dominican Republic, given the focus on these locations in later sections. It should be noted that these results are most useful in identifying areas of higher relative risk and not as absolute risk assessment profiles. In all locations, more specific modeling and data will be useful in informing decision making. Asset Exposure and Fragility Modeling For the infrastructure damage case studies in Dominica and the Dominican Republic, fragility (impact) curves were used when available to estimate the impact that exposure to a hazard might have on the infrastructure asset. These curves are not specific to either island. If local data on design, past damages and other engineering impacts are available, it would improve accuracy. The curves used for this study could be applied to other locations for similar estimates on damage from natural hazard exposure. For flooding, road impact curves were used from [7, 13]. For thermal power plants (non-nuclear), transmission lines and substations, data was used from [39]. Additional analyses could include seismic impact on water transmission and distribution is also possible if the length lines, their diameter, construction material and location is known. Seismic impact on water treatment facilities can be estimated from [39] and impact from hurricane winds on transmission lines can be estimated using data from [39, 40]. Facility Impact Modeling Two scenarios were used to evaluate the impact that infrastructure disruption or failure would have on critical facilities. Both scenarios rely on user-defined origin and destination pairings (such as travel from houses (“origins”) to hospitals (“destinations”) on roads (“network”)). Three distinct origin and destination pairs evaluated in this report used roads as the network in Dominica: houses to hospitals, ports to power plants, and water treatment facilities to towns. For these pairs, the network connection was determined using speed limit estimates to calculate travel cost, measured in travel time. An additional analysis for the Dominican Republic was done using the available transmission and distribution data as the network, while origin-destination pairing was comprised of power plants (origin) and homes (destination). For the Dominican Republic, some gaps in data affected the analysis and is best used for illustrative purposes for what the tool can do when better data is available. Drop-Link Analysis The first scenario is a drop-link analysis, where every segment of networked infrastructure is dropped individually and an analysis of travel time between origin(s) and destination(s) is calculated. This analysis is designed to highlight the relative importance of each network segment relative to all other segments, answering questions such as “which areas of my network should I prioritize for hardening?” The computed travel time values between every origin and destination pair is calculated for the entire network each time a segment is removed. These values are compared to a ‘normal’ scenario where travel times between origin and destination are calculated with no disruptions to the network. The difference of change between normal and perturbed scenarios is recorded for each network segment. In the detailed case studies below for Dominica, the road network was divided into segments of length with a maximum length of 100 meters. Every road segment was dropped one time, which required over 10,000 runs for Dominica. This provides a relative criticality value for each road segment relative to every other road segment. Roads that are more critical indicate that their loss would increase travel times between origins and destinations more than roads with lower criticality values. The output of this analysis is a map, where criticality values are shown (see, for example, Figure 4, which illustrates the criticality of each road segment relative to time increases between origins (ports) and destinations (power plants)). Monte Carlo Analysis The second scenario is a Monte Carlo (MC) analysis where multiple networked infrastructure segments are dropped based upon failure probabilities. These failure probabilities are informed by the hazards. For example, some portion of the road network is exposed to the 0.02 annual flood probability (see Figure 1). Using fragility curves to estimate the likelihood of damage to the network based upon flood depth, road surface and classification, a probability of segment failure is calculated. This probability is used in MC runs to analyze different scenarios of networked infrastructure performance under each hazard scenario. Similar to the drop-link analysis above, the change between ‘normal’ and ‘perturbed’ scenarios is recorded for each run. The major difference between the drop-link and MC scenarios is that the first analysis drops a user-defined set of infrastructure links for each run (in this analysis, 1 link at a time), while the MC scenario may have several links disturbed in a single scenario due to a hazard event. Both analyses produce quantitative output in the form of a matrix of travel time differences between the perturbed and normal networks. For visualization purposes, the output of the drop-link analysis is a map (e.g. Figures 4, 6 and 9) that shows the relative criticality of each link in relation to travel between origin-destination pairs. The visual output of the MC analysis is a histogram (e.g. Figures 5 and 7) that show aggregate increases in travel time based on different scenarios of network failures. The latter can help identify the likelihood (how often) substantial time increases are likely. This second scenario is used to calculate the impact of segment failure on the travel of goods/persons between origin and destination. For example, the delivery of fuel from a port to a power plant can be analyzed for the increase in time required to transport fuel under different hazard scenarios. Another application of this metric is the ability to estimate the change in destination based upon network functionality. This latter application is used to estimate the change in number of persons expected at hospitals. As a note, due to limitations in data (such as Average Annual Daily Traffic), no traffic or congestion considerations were included. Therefore, the findings in travel cost (time) are based on road speed limits. The results likely indicate best- case scenarios and underestimate areas where congestion might be higher from re-routing. This is a parameter that could be included in simulations if data was available for different road segments. Using these two analysis methods, three metrics are produced: Segment Drop-Link criticality: impact of dropping a road (or other networked infrastructure) segment on travel cost between all origin and destination pairs (every segment is dropped once, and travel cost is calculated over entire network routing). The impact is defined by the mathematical distance between ‘normal’ and ‘perturbed’ network matrices of travel cost between all origin and destination pairs, calculated over the entire drop-link segment analysis. Impact on Service Delivery/Supply Chain: importance of the segment or set of segments defined by increased travel cost. Importance is defined by the increase in the cost of delivery of good/person from origin to destination. Impact on Facility Access: critical facility impacts are identified based upon their increase or decrease in expected access based on least-cost travel routes if a link or set of links is lost relative to the overall network. The outcome of this analysis is the number of persons, goods, or other flow on a network that are rerouted to a different destination based on segment disruption. More information on this methodology can be found in L’her et al. [42]. Results Exposure Modeling: All Countries/All Infrastructure Individual country results for single hazard/single asset exposure can be found in Supplementary Information File1. These resulted in over 90 unique individual maps. Summary results for exposure of assets by hazard are shown in Table 4. Appendix B contains individual tables (B1-B40) for infrastructure exposure to each hazard at a level “2” or above (as defined in Table 3). These results are highlighted below for each infrastructure type. All data used for analysis is listed in Tables 2A and 2B. Road infrastructure is categorized by primary, secondary and tertiary/other roads. Primary roads see the highest exposures to hurricane, landslide susceptibility and earthquakes. Overall, primary roads have high exposure to hurricane, landslide and seismic risk, although this varies greatly by country (Table B1-B2). All primary roads are exposed to hurricanes in Antigua and Barbuda, Barbados, Jamaica and St. Kitts and Nevis, while no inventory is exposed in Trinidad and Tobago or Guyana. Most countries have at least 50% of primary roads in locations with medium or high landslide susceptibility, excluding Guyana, Barbados and Belize. Very few (less than 2%) of primary roads exist in locations exposed to the 0.02 annual event coastal surge inundation. Similar patterns of exposure hold across secondary and tertiary road infrastructure. Overall, the highest road hazard exposures are to hurricanes and landslide-susceptible locations (67%), earthquakes (47%), and lower amounts for flooding (7% for pluvial flooding, 3% for fluvial flooding) and coastal surge (less than 1%). Full data for every country and hazard available for both total kilometers exposed and percentage of inventory for primary, secondary, tertiary and all roads can be found in Tables B1-B8. Power plant exposure is highest to earthquakes, with 100% of assets exposed in Antigua, Barbados, Dominica, Dominican Republic, St. Kitts and Nevis, Trinidad and Tobago and St. Vincent. Fossil fuel (and unknown fuel type) plants see 66% exposure to hurricanes and 59% exposure to earthquakes, though these vary greatly by geography. For example, all fossil power plants in The Bahamas are exposed to hurricanes, while none are in seismic risk areas. All (5) of Jamaica’s fossil fuel plants are exposed to hurricanes, while 80% are exposed to seismic risk. All (4) of Jamaica’s renewable facilities are exposed to hurricanes, earthquakes and in areas with medium-high landslide susceptibility. All plants (100%) in Antigua, Dominica, Grenada, St. Kitts and Nevis and St. Vincent are located in regions with medium or high-risk landslide susceptibility. Power plants in Suriname, Haiti and Dominica see high exposure to flooding as a percentage of total inventory. Full data for every country and hazard available for both total number of facilities exposed and percentage of inventory for fossil fuel, renewable (solar, wind, biodiesel, hydroelectric) and all facilities can be found in Tables B9-B14. Throughout the Caribbean, a high percentage (>50%) of bridge inventory is exposed in nearly all regions to hurricane, landslide and seismic risk. Antigua and Barbuda, Barbados, Dominica, Grenada, Jamaica, St. Kitts and Nevis and St. Vincent have 100% of bridges exposed to hurricanes. All bridges (100%) in Antigua and Barbuda, Barbados, Dominica, St. Kitts and Nevis, St. Lucia and Trinidad and Tobago are in locations with earthquake risk. Flooding risk varies by location, but overall, 26% of bridges are exposed to pluvial flooding. Full data for every country and hazard available for both total bridges exposed, and percentage of inventory can be found in Tables B15- 16. Ports, in this analysis, are grouped into four categories based on classification from The World Port Index [36]. No “large” ports exist in the Caribbean, while there are 8 “medium” ports, 21 “small” ports and 39 “very small” ports. For all port classifications, exposure to hurricane and seismic risk is greater than 50%. For this study, no data on port traffic or volume was available. This could be an important area for future work, especially in islands with few larger ports where the loss of a single location could cripple supply import during a hazard response phase. For example, Jamaica has 11 ports, but only 1 ‘medium’ sized port. This location is exposed to both hurricane and earthquake risk. Assessment of whether the loss of this location could be mitigated by utilization of small ports (2) would be useful in hazard planning. It is unlikely, based upon local feedback (e.g. [43]) that the ‘very small’ ports, of which there are 8 in Jamaica, have deep enough harbors to accommodate larger ships. Full data for every country and hazard available for each port size and total ports exposed to each hazard by the number of facilities and percentage of inventory can be found in Tables B17-B24. Similar to ports, the airports data used in this study are grouped into “small”, “medium” and “large” categories. Small airports have only light aviation passenger traffic, while Medium and Large airports see high general aviation traffic. For small airports, there are over 170 locations throughout the Caribbean, with the Bahamas, Suriname and Dominican Republic with more than twenty each. Overall, these structures see relatively low exposure to most hazards, with hurricane being the greatest (33%). Antigua is one exception, with both of their small airports exposed to hurricane, landslide and seismic risk. For medium and large airports, where damages might have a greater impact on tourism accessibility, the highest exposure is seen in the Bahamas. 27 out of 28 are exposed to hurricane risk and 11% to pluvial flooding. All large airports in the Caribbean (Bahamas, Belize, Dominican Republic and Jamaica) are exposed to hurricane risk. Full data for every country and hazard available for each airport size and total airports exposed to each hazard by the number of facilities and percentage of inventory can be found in Tables B25-B32. Water infrastructure was generally very difficult to obtain throughout the Caribbean. Many of the assets in this report were found through local agency websites, Google Maps and other queries. The structures with the greatest data available were water and wastewater treatment facilities, although this spanned only 6 countries and is likely incomplete: Antigua and Barbuda, the Dominican Republic, Grenada, Jamaica and Trinidad and Tobago. For regions with data, over 80% are in areas with landslide susceptibility and seismic risk. None are exposed to coastal surge, and flooding is relatively low (6% fluvial, 10% pluvial). The Dominican Republic has 50 facilities and nearly all are in areas with medium-high landslide susceptibility and all facilities are in areas of seismic risk. This is an example of areas for future work with possible interactions between hazards (seismically induced landslides). Full data for every country and hazard available for total number of locations and percentage of inventory exposed is available in B33-B34. Dams had very limited data availability, although those with hydropower plants were considered in the renewable power plants analysis (TablesB11-B12). Data on the interaction of water systems with transportation networks is assessed for a water service area in Dominica using the network analysis tool described above (see Figure 6). Hospitals and health centers see highest exposure risk to hurricanes (49%), landslide susceptibility (67%) and earthquakes (65%). These are much higher than flooding (1% for fluvial and 5% for pluvial) and coastal surge (less than 1%). The Dominican Republic has the largest number of facilities (162) and sees approximately 90% of facilities exposed to landslide and earthquake risks. Flooding (pluvial) has the highest exposure in Jamaica, St. Lucia and St. Kitts and Nevis with approximately 15% of facilities. Antigua and Jamaica have nearly all facilities exposed to hurricanes and earthquakes. Coastal surge and flooding (fluvial) is very low, with less than 1% of facilities in all locations except for Haiti (3% for fluvial flooding, 4% for pluvial flooding) and Suriname (13% facilities exposed to fluvial flooding). Full data for every country and hazard is available for the total number of facilities and overall percentage of inventory in Tables B35-B36. Table 4: Infrastructure exposure by hazard, all Caribbean, for which data on infrastructure and hazard exists. Results are presented as percentage of total infrastructure assessed. Individual data for each country can be found in Appendix A. Table 2 details data availability for infrastructure and hazards. "Exposure" is defined for each hazard based on a reclassification scale intensity level of "2" or higher. Individual results for each country and hazard and asset can be found in Appendix B. Exposure to Hazard Total Amount (Infrastructure as % of total for which data is available) Infrastructure (km or facility/ Asset Flood Flood Coastal Hurricane Landslide asset count) Seismic (Fluvial) (Pluvial) Surge (Wind) (Susceptibility) Primary Roads 6195 4% 9% 0.0% 74% 63% 56% Secondary Roads 18790 4% 9% 0.3% 63% 60% 46% Tertiary/ Other Roads 49378 3% 7% 0.2% 67% 70% 47% Bridges 6866 10% 26% 1% 55% 60% 56% Power Plants 53 2% 13% 2% 66% 47% 59% (Fossil Fuel/ Other) Power Plants 19 16% 21% 0% 58% 79% 63% (Renewables) Water/Wastewater 56 7% 13% 0% 11% 84% 96% treatment facilities Ports 60 3% 2% 3% 63% 32% 58% (Small & Very Small) Ports (Medium) 8 0% 0% 0% 63% 25% 63% Airports (Small) 171 2% 8% 1% 33% 21% 19% Airports 82 1% 7% 0% 82% 27% 32% (Medium/Large) Hospitals/ 490 1% 5% 0.2% 49% 67% 65% Health Centers For climate change impacts, the metric of cooling degree day (CDD) in the 2050 decade was chosen, Figure 2. This is because it has direct implications for energy demand and can affect the design of buildings including windows, air conditioning and other components. Findings show that RCP 8.5 indicates at least an average annual increase of 10% in CDD above the historical baseline for all countries and results exceed 20% in parts of Haiti, the Dominican Republic, and Guyana. The bottom right shows the results for Guyana and Suriname. Table 5 details the impacts on the buildings as a percent of total inventory assessed. Consideration of other climate risks, specific to the impact they have on infrastructure, is a next step. For example, heatwaves can soften and damage paved road infrastructure, while increased precipitation may erode earth or gravel roads. Figure 2 - Percent change in average annual cooling degree days (CDD) in the 2050 decade compared to 30-year historical baseline. Results shown for change indicated by at least half of the General Circulation Models (GCMs) assessed. Left graphic (A) is RCP 4.5 results and right graphic (B) is RCP 8.5 Results. Source: Authors calculations, daily climate data from [29] Table 5: Percent of buildings impacted by climate change in 2050 decade. Climate change impact is measured in change in annual Cooling Degree Days (CDD). CDD is measured in the average annual maximum percent change in 2050 decade (2050-2059) when compared to average annual historical baseline (1970-1999). Values computed for maximum change predicted by at least half (>10) of the GCMs for RCP 4.5 and 8.5. >5% CDD >10% CDD >15% CDD >20% CDD Increase Increase Increase Increase Buildings impacted by RCP 4.5 100% 11.8% 0.1% 0.0% CDD Increase (Percent of Total Building Inventory Assessed) RCP 8.5 100% 100% 4.8% 0.1% Multi-Hazard Modeling Multi-hazard maps for all countries were produced based on the hazard reclassification scheme (0-5 intensity ratings for all 6 hazards) and can be found in Appendix A. Figure 3 details these findings for Dominica and Dominican Republic. Red and orange colors indicate regions where high levels of hazard intensity for multiple hazards intersect. For example, regions with a score of 20 and higher see high levels of exposure to multiple hazards. This could be exposure to the highest (intensity level 5) level of hazard for 4 hazards, ‘very high’ (intensity level 4) exposure for 5 hazards, or similar combinations. Generally, these occur along rivers because of the combined risk of flooding, landslide and other hazards. Areas in blue show low or no exposure for all hazards. Figure 3 - Multi-Hazard Map for Dominica (left) and Dominican Republic (right). Areas in orange and red indicate high risk for multiple hazards. Maps for all countries can be found in Appendix A. Facility Impact As described above, three metrics are used to assess the impact of infrastructure failure on critical assets. Below, “criticality” refers to segment drop-link criticality, which is the impact of the failure of a single segment of the network and calculating the impact its loss has on flow between origin and destination. This is done for every segment in the network and provides a relative value for each segment when the normal (no disruption) and perturbed analyses are compared. Impact on facility is defined by both the change in expected goods or persons at a specific destination and the change in ‘traffic’ along the network. Where data on voltage of transmission lines, traffic/congestion, pipe size and other metrics are available, this analysis could provide robust planning information including how disruption would be expected to impact both the network and the destination facilities. Several assessments are done. For Dominica, imported fuel delivery to power plants, potable water delivery to town holding tanks during drought conditions, and household routing to hospital destinations are shown below. For the Dominican Republic, power delivery to buildings along the transmission network is illustrated. Critical Facility Access, Supply Chain: Ports and Power Plants The criticality of the network and impact on fuel delivery to power plants is shown, Figure 4. The origins (seaports) and destinations (diesel power plants) are shown on the far left. The middle graphic details criticality of the road segments as defined above. The zoomed in regions of Roseau and Portsmouth are shown on the far right and display the supply chain importance of links in the region. Red and black roads indicate that these segments are critical to the delivery of fuel to the power plant, without regard for the origin point (port where fuel is imported). Figure 4 – Single drop-link analysis between seaports and diesel power plants (left). The criticality of each road segment is detailed in the middle (travel cost criticality). The far-right maps detail zoomed-in regions of service delivery/supply chain based on network performance under 0.02 annual flood exposure damages (fluvial, pluvial and coastal surge). Two diesel power plant locations, Figure 5, were further assessed to understand the impact of simulated hazard failure on multiple road segments using a Monte Carlo simulation for 1,000 runs. A comparison was run to determine the impact of hazard data informing failure probabilities. The base (‘normal’) travel time from ports to power plants was found using the Dijkstra algorithm [44] and segment-specific average traffic velocities (based on speed limits of roads). Failure rate is based on 0.02 annual event exposure for combined pluvial, fluvial and coastal surge. The Dijkstra algorithm was used to determine the minimum travel time given each scenario of segment failures. When a segment failure resulted in no route being possible from ports to a power plant, it was assumed that it would take 1 day to fix each failed route segment along the path that minimizes the aggregate travel time (this is a user-defined value and can be adjusted where data is available). Figure 5 shows the histogram for travel cost (change in travel time relative to baseline) for each of the 1,000 simulations and shows that Portsmouth is most vulnerable to fuel loss during road segment failure. Results in grey indicate that supply was completely cut off from road failure until those roads were repaired. Results in green indicate that there was an alternate, though longer, route available. Figure 5 – Travel cost for fuel delivery to power plant. Results in grey indicate that supply access was completely cut off by road failure until those roads were repaired. Results in green indicate that there was an alternate, though longer, route available. Critical Facility Access, Supply Chain: Potable Water Delivery During Drought Due to very limited water infrastructure data for Dominica, no assessment of water delivery during normal operations was feasible for this report. However, data on potable water delivery via trucks to towns was done as an illustration, using data from [26], [45]. This is particularly relevant should existing water transmission (250km) and distribution lines (450km) fail due to seismic events or landslides. In Dominica, there are several water areas. Water Area 1 (WA1) encompasses Roseau and surrounding towns. A treatment facility in Antrim noted 16 storage tanks along the route. According to [26], [45], the distribution system stretches from Kingshill in the South to Mero in the North. Comprised of 16 storage tanks, the system has a total storage capacity of 2.2 million gallons. This includes 300,000 gallons in bulk storage used for supplying cruise ships. The criticality of road segments can be seen in Figure 6. Figure 6 - Road criticality for potable water delivery in Water Area 1, near Roseau, Dominica Critical Facility Access and Failure Impact: Hospitals The impact of link failure on critical facility access (hospitals) is assessed using several metrics. Households are used as origin points, hospitals are the destination points, and the road network is used for travel. Travel cost is measured in seconds, although no congestion or traffic information was available for this assessment and is based only on road speed limits. Therefore, the travel cost is likely an underestimate. Figure 7 shows the expected change in hospital destination. As many as 2,000 households change their destination based on perturbed road conditions. These values are based upon travel to the hospital facility is the closest location under a perturbed network when compared to the normal network operation. These values have an impact on the road network, including the expected increase in flow along each road segment (Figure 8). Figure 7 - Hospital facility impact measured in change in expected number of persons traveling to destination using 3-hospital destination OSM data. These figures show the results from 1,000 Monte Carlo simulations to calculate change in hospital destination based on road network failure. Road network failure is calculated by segment probability of failure using 0.02 annual (1/50 year) maximum water depth flooding event exposure and road fragility curves for combined pluvial, fluvial and coastal surge. Figure 8 - Change in traffic flow based on travel to closest hospital for every household based on road perturbation and least-cost travel defined by time Finally, the criticality of each segment is shown in Fig. 9. Network criticality is shown for every road segment based on drop-link analysis, with orange and red segments indicating higher importance as defined above. The original data used in this analysis (CHARIM data) lists 3 hospitals, located in Portsmouth, Roseau and Marigot. However, based on feedback from local experts, only 1 hospital (Roseau) is operational for critical care [43]. Therefore, to highlight the importance of considering local data in assessment, Figure 9 shows the difference in road criticality for both sets of data. The left graphic is based on local input, whereas the right graphic indicates findings based on publicly available data. While criticality in some segments is similar, the two input data sets do show different results. Figure 9 - Road segment criticality for household (origin) travel to hospitals (destination). Drop-link analysis is used to compute travel time based on speed limits of roads. No congestion or traffic metric is included in the analysis. Figure A (left) shows results with only one critical care facility (Roseau Hospital) open, while Figure B (right) shows results based on OSM data. The criticality metric is a measure of overall importance to the network. The graphics illustrate the importance of local input in the data collection and analysis process. Dominican Republic: Transmission Line Criticality and Impact on Power Delivery to Buildings Analyzing the criticality of transmission lines as the network for power delivery was done in the Dominican Republic, Figure 10. This illustrative analysis was performed for power delivery from power plants (origins) to substations (destinations) along the transmission line network. Substations were weighted for relative importance based on the population each serves. Power plants were weighted based on their power generation capacity, measured in megawatts for each generation facility. Limited data availability and incomplete data sets make this assessment one that will benefit from local expertise and better data sources. For this analysis, Voronoi polygons were used to identify the population served by each substation. This assumption predicates that a local distribution network line serving that population relies on power from the substation it is closest to. For the transmission network, no voltage data was available, so a homogenous, bi- directional network was assumed. There are also many gaps in connection of the transmission network which were manually adjusted to ensure a connected infrastructure network. Lines in red indicate that these segments are more critical to power delivery and their loss/inoperability would result in lost power delivery to substations serving the most people. These are situated, in the current analysis, on lines that connect multiple power plants or are very near dense population centers. For planning purposes, these results could help prioritize which lines are hardened or upgraded through pole type (for example, wood poles are more likely to fail in high winds than composite structures) and the most important substations to protect against flooding and other hazards. Lines in blue indicate that loss of these segments would have a lower overall effect on power delivery. Inclusion of line voltage, directionality, cascading failure assessments and distribution infrastructure could improve this analysis. Figure 10 - Power delivery loss from transmission line failure in the Dominican Republic. Origin locations are power plants, destinations are each substation, with power traveling along a homogenous, bi-directional transmission line network. Each substation is weighted based on the population it serves (calculated through the use of Voronoi polygons) and every power plant is weighted based on the Megawatts of power it produces. The criticality of each transmission line segment indicates the relative importance from least critical (blue) to most critical (red). More critical segments indicate that a greater population is served and/or power could fail to be delivered if that segment fails. Conclusions Building resilient systems requires a multi-faceted approach and long-term planning. A primary requirement is data about the systems, the vulnerabilities they are exposed to, and the interconnections between them. The Caribbean region is comprised of many small island states that are highly exposed to natural hazard events. This analysis builds on recent work on critical infrastructure systems by looking at the exposure of assets in the Caribbean countries. A multi- hazard assessment was done using available hazard data and infrastructure asset exposure was quantified. The impacts of asset failure on supply chains, power delivery, and access to critical facilities was assessed using case study data from the Dominican Republic and the Commonwealth of Dominica. While data was limited for many regions assessed, future work can build on these flexible tools and methods as data becomes available. The methodology developed for assessing these impacts is specifically designed to apply equally to transmission and distribution infrastructure, water distribution, railways, shipping routes or other similar data. The choice of road infrastructure as the network of analysis for most assessments in this report is due to very limited data availability of other networked infrastructure types, and/or that the networks themselves are very sparse with limited redundancy due to the small geographies. The latter is true, for example, with transmission infrastructure in small islands, where a single ring of transmission encircles the island. In this scenario, loss of a link likely limits transmission connectivity for the entire network. An important next step also includes analysis of the interdependencies within and between systems as well as the operational lifecycle of critical assets. As shown for fuel delivery and power plant operation, hospital access and potable water distribution, the sectoral silos that define traditional analysis limit the understanding of the true impacts, including the potential for cascading failures. This is also true for repair of damaged assets, which rely on passable roads and operational airports and ports for supply delivery and transportation. Past work has shown that inclusion of long-term resilience thinking into asset design can potentially reduce damages from hazards occurring over the lifetime at lower costs than repairing after damages occur (e.g. [12, 45]). This longer-term perspective, including coordinated data collection and reporting, is an important component in resiliency planning. References [1] S. Hallegatte, J. Rentschler, and J. Rozenberg, Lifelines: The Resilient Infrastructure Opportunity, Sustainable Infrastructure. Washington, DC: The World Bank, 2019. [2] United Nations Economic and Social Council (ECOSOC), “Pathways to resilience in climate- affected Small Island Developing States (SIDS)– A Forward-Looking Resilience Building Agenda: Promises, results and next steps,” Nov. 13, 2018. https://www.un.org/ecosoc/en/events/2018/pathways-resilience-climate-affected-small- island-developing-states-sids%E2%80%93-forward (accessed Apr. 30, 2020). [3] C. B. Wahlen, “Caribbean Leaders Discuss Debt for Climate Adaptation Swap, Caribbean Resilience Fund | News | SDG Knowledge Hub | IISD,” IISD’s SDG Knowledge Hub, Oct. 10, 2019. https://sdg.iisd.org:443/news/caribbean-leaders-discuss-debt-for-climate- adaptation-swap-caribbean-resilience-fund/ (accessed Apr. 30, 2020). [4] Inter-American Development Bank (IDB), “Climate Change in the Caribbean Small Island States,” Inter-American Development Bank, 2020. Accessed: Apr. 30, 2020. [Online]. Available: https://www.iadb.org/en/ove/climate-change-caribbean-small-island-states. [5] Global Facility for Disaster Reduction and Recovery (GFDRR), “Small Island States Resilience Initiative.” https://www.gfdrr.org/en/sisri (accessed Apr. 30, 2020). [6] N. Carrasco, C. P. Quijada Robles, and L. Piccio, “Standing stronger together: Reflections on building resilience for small island states,” Jun. 03, 2019. https://blogs.worldbank.org/sustainablecities/standing-stronger-together-reflections- building-resilience-small-island-states (accessed Apr. 30, 2020). [7] E. E. Koks et al., “A global multi-hazard risk analysis of road and railway infrastructure assets,” Nat. Commun., vol. 10, no. 1, pp. 1–11, Jun. 2019, doi: 10.1038/s41467-019-10442- 3. [8] F. Cavalieri, P. Franchin, J. A. M. B. Cortés, and S. Tesfamariam, “Models for Seismic Vulnerability Analysis of Power Networks: Comparative Assessment,” Comput.-Aided Civ. Infrastruct. Eng., vol. 29, no. 8, pp. 590–607, 2014, doi: 10.1111/mice.12064. [9] K. Poljanšek, F. Bono, and E. Gutiérrez, “Seismic risk assessment of interdependent critical infrastructure systems: The case of European gas and electricity networks,” Earthq. Eng. Struct. Dyn., vol. 41, no. 1, pp. 61–79, 2012, doi: 10.1002/eqe.1118. [10] L. Dueñas‐Osorio, J. I. Craig, and B. J. Goodno, “Seismic response of critical interdependent networks,” Earthq. Eng. Struct. Dyn., vol. 36, no. 2, pp. 285–306, 2007, doi: 10.1002/eqe.626. [11] H. Demirel, M. Kompil, and F. Nemry, “A framework to analyze the vulnerability of European road networks due to Sea-Level Rise (SLR) and sea storm surges,” Transp. Res. Part Policy Pract., vol. 81, pp. 62–76, Nov. 2015, doi: 10.1016/j.tra.2015.05.002. [12] J. Eun Oh et al., “Addressing Climate Change in Transport. Volume 2: Pathway to Resilient Transport,” The World Bank, Australia-World Bank Group Strategic Partnership in Vietnam and NDC Partnership Support Facility, Sep. 2019. Accessed: Oct. 21, 2019. [Online]. Available: http://documents.worldbank.org/curated/en/438551568123119419/pdf/Volume-2- Pathway-to-Resilient-Transport.pdf. [13] X. Espinet, J. Rozenberg, K. Singh Rao, and S. Ogita, “Piloting the use of network analysis and decision-making under uncertainty in transport operations,” World Bank Group, Policy Research Working Paper WPS8490, Jun. 2018. [14] M. Heidarzadeh, R. Teeuw, D. Simon, and C. Solana, “Storm wave runups and sea level variations for the September 2017 Hurricane Maria along the coast of Dominica, eastern Caribbean sea: evidence from field surveys and sea-level data analysis,” Coast. Eng. J., vol. 60, no. 3, 2018, [Online]. Available: https://www.tandfonline.com/doi/full/10.1080/21664250.2018.1546269?scroll=top&need Access=true. [15] American Society of Civil Engineers, “ASCE 7: Minimum Design Loads for Buildings and Other Structures,” The Executive Director Office of the Federal Register, Washington, D.C., Second Edition SEI/ASCE 7-02. [16] K. S. Rukstales and A. M. Shumway, “Data Release for 2018 Update of the U.S. National Seismic Hazard Model.” U.S. Geological Survey, 2019, doi: 10.5066/P9WT5OVB. [17] National Geospatial Intelligence Agency, “World Port Index, 27th Edition,” United States Government, Springfield, Virginia, Pub.150, 2019. Accessed: Jun. 08, 2020. [Online]. Available: https://msi.nga.mil/api/publications/download?key=16694622/SFH00000/Pub150bk.pdf&t ype=view. [18] D. Megginson, “Our Airports,” Jan. 2020. https://ourairports.com/about.html#credits (accessed May 02, 2020). [19] J. R. Meijer, M. A. J. Huijbregts, K. C. G. J. Schotten, and A. M. Schipper, “Global patterns of current and future road infrastructure,” Environ. Res. Lett., vol. 13, no. 6, p. 064006, May 2018, doi: 10.1088/1748-9326/aabd42. [20] Government of the Commonwealth of Dominica, “Rapid Damage and Impact Assessment, Tropical Storm Erika,” ACP-EU Natural Disaster Risk Reduction Program, Sep. 2015. Accessed: May 05, 2020. [Online]. Available: https://www.gfdrr.org/sites/default/files/publication/Dominica-Rapid-Damage-and-Needs- Assessment.pdf. [21] Dominica Water and Sewerage Company Limited (DOWASCO), “Personal Communication,” May 2020. [22] Caribbean Water and Wastewater Association (CWWA), “CDB funds Dominica water project,” Sep. 04, 2019. https://cwwa.net/news/cdb-funds-dominica-water-project/ (accessed May 05, 2020). [23] Government of the Commonwealth of Dominica, Maps. Physical Planning Division. [24] V. Jetten, “CHaRIM Project, Dominica National Flood Hazard Map Methodology and Validation Report,” University of Twente, Draft 1.1, May 2016. [25] ACP-EU Natural Disaster Risk Reduction Program, “Caribbean Handbook on Risk Management | CHARIM.” http://charim.net/ (accessed May 05, 2020). [26] Dominica Water and Sewerage Company Limited (DOWASCO), “Water System Areas,” Jan. 2020. https://www.dowasco.dm/index.php/our-services/waterservice?id=53. [27] US Army Corps of Engineers, “Water resourcse assessment of Dominica, Antigua, Barbuda, St. Kitts and Nevis,” Dec. 2004. [28] K. P. Moustris, P. T. Nastos, A. Bartzokas, I. K. Larissi, P. T. Zacharia, and A. G. Paliatsos, “Energy consumption based on heating/cooling degree days within the urban environment of Athens, Greece,” Theor. Appl. Climatol., vol. 122, no. 3–4, pp. 517–529, Nov. 2015, doi: 10.1007/s00704-014-1308-7. [29] NASA Center for Climate Simulation, “Nasa Earth Exchange Global Daily Downscaled Projections (NEX-GDDP),” Accessed 2019. https://www.nccs.nasa.gov/services/data- collections/land-based-products/nex-gddp (accessed May 04, 2020). [30] National Centers for Environmental Information, “IBTrACS - International Best Track Archive for Climate Stewardship,” National Oceanic and Atmospheric Adminstration, 2018. Accessed: Nov. 29, 2018. [Online]. Available: https://www.ncdc.noaa.gov/ibtracs/. [31] The World Bank, “Global landslide hazard | Data Catalog,” Nov. 25, 2009. https://datacatalog.worldbank.org/dataset/global-landslide-hazard-1 (accessed May 05, 2020). [32] C. C. Sampson, A. M. Smith, P. D. Bates, J. C. Neal, L. Alfieri, and J. E. Freer, “A high‐ resolution global flood hazard model,” Water Resour. Res., vol. 51, no. 9, pp. 7358–7381, Sep. 2015, doi: 10.1002/2015WR016954. [33] The World Bank, “EQ-LEA-GEM-475,” GFDRR Innovation Lab, Jan. 16, 2017. http://45.55.174.20/layers/hazard:lea02_475yr (accessed May 05, 2020). [34] Center For Hazards And Risk Research-CHRR-Columbia University and Center For International Earth Science Information Network-CIESIN-Columbia University, “Global Earthquake Hazard Distribution - Peak Ground Acceleration.” Palisades, NY: NASA Socioeconomic Data and Applications Center (SEDAC), 2005, doi: 10.7927/H4BZ63ZS. [35] Deltares, “Assessing the impact of sea level rise potential in the Caribbean,” Delft, The Netherlands, Technical Report, 2020. [36] National Geospatial Intelligence Agency, “Maritime Safety Information,” World Port Index, 2020. https://msi.nga.mil/Publications/WPI (accessed May 02, 2020). [37] World Resources Institute, “Global Power Plant Database - Data | World Resources Institute,” Jun. 2019. https://datasets.wri.org/dataset/globalpowerplantdatabase (accessed May 05, 2020). [38] Government of Grenada, GeoServer, Water Facilities. 2020. [39] Miyamoto International and World Bank Group, “Increasing Infrastructure Resileince Background Report,” The World Bank, 7189546, Feb. 2019. Accessed: Jan. 31, 2020. [Online]. Available: http://documents.worldbank.org/curated/en/620731560526509220/pdf/Technical- Annex.pdf. [40] Quanta Technology, “Cost-Benefit Analysis of the Deployment of Utility Infrastructure Upgrades and Storm Hardening Programs,” Project No. 36375, Mar. 2009. Accessed: May 04, 2020. [Online]. Available: http://www.puc.texas.gov/industry/electric/reports/infra/utlity_infrastructure_upgrades_r pt.pdf. [41] et. al New York Power Authority, “Build Back Better: Reimagining and Strengthening the Power Grid of Puerto Rico,” Dec. 2017. Accessed: Mar. 12, 2019. [Online]. Available: https://www.governor.ny.gov/sites/governor.ny.gov/files/atoms/files/PRERWG_Report_PR _Grid_Resiliency_Report.pdf. [42] G. L’her, A. Schweikert, R. Flanagan, and M. Deinert, “Modeling Critical Facility and Network Impacts from Extreme Events,” Prep., 2020. [43] Climate Resilience Execution Agency for Dominica (CREAD), “Discussion with CREAD Experts,” Mar. 24, 2020. [44] E. W. Dijkstra, “A note on two problems in connexion with graphs,” Numer. Math., vol. 1, no. 1, pp. 269–271, Dec. 1959, doi: 10.1007/BF01386390. [45] Dominica Water and Sewerage Company Limited (DOWASCO), “Water Area 1,” Jan. 2020. https://www.dowasco.dm/water_areas/waterarea1.pdf (accessed May 04, 2020). [46] Schweikert Amy et al., “Road Infrastructure and Climate Change: Impacts and Adaptations for South Africa,” J. Infrastruct. Syst., vol. 21, no. 3, p. 04014046, Sep. 2015, doi: 10.1061/(ASCE)IS.1943-555X.0000235. Appendix A: Multi-Hazard Risk Maps This appendix contains the multi-hazard risk maps for every country for which sufficient data for all hazards was available. The resolution of the data varies between countries and the scale in the images below varies based on island size. A simple framework was developed to allow for simultaneous consideration of each of the datasets. A reclassification scheme was used to assign levels of intensity for each hazard (0-5, with “0” being no risk or unknown risk, “1” being very low risk and “5” being extremely high risk). Specifically included in the multi-hazard model mapping are hurricanes, fluvial flooding, pluvial flooding, coastal surge, landslide, and seismic risks. Because each hazard has many measurements of data (hurricanes, for example, have 5 categories and a likelihood of occurring each year can range from 0%-100% based on historical data), specific thresholds were chosen to simplify the assessment for visualization and analysis. Consistent with the exposure results presented in Appendix B, below are the multi-hazard maps for each country. The Table A1 details the hazard metric and ranking for the multi-hazard score. By overlaying each risk layer in each location, a single map is generated with a scoring matrix of 0- 30, with “0” being no risk/unknown risk for all hazards and “30” being extremely high risk for all hazards assessed. As a note, there is no location where the multi-hazard combined risk exceeds “25”. It should be noted that these results are most useful in identifying areas of higher relative risk and not as absolute risk assessment profiles. In all locations, more specific modeling and data will be useful in informing decision making. . Appendix B: Country-Level Infrastructure Exposure to Hazards A detailed description of data analysis is provided in the main report, including Tables 2A and 2B, which details data sources for infrastructure and hazards data, but summarized here again for clarity. Table 3 provides a full explanation of hazard reclassification and exposure categories shown here (a multi-hazard risk analysis of score “2” or higher is considered ‘exposed’ to shown in the tables below). An especially important note is that hazards were assessed based on the data available, which is a subset of all hazard risks. For flooding and coastal surge, the analysis looks at the 0.02 annual risk (equivalent to the “50-year event”). Flooding and surge data have the highest resolution data available of all the hazards, for all locations. For hurricane events, 0.25-degree gridded risk was determined based on historical eye-of-the-storm tracks. Exposure was assessed on a binary analysis of 20% or greater likelihood of a Category 2 storm or higher occurring annually. Category 2 storms were chosen based on available design standard information for infrastructure assets such as transmission and distribution infrastructure. Landslide susceptibility data was only available for ‘susceptibility’ which requires additional inputs to obtain risk. Landslide data was available in a ‘low’, ‘medium’ or ‘high’ risk scale, ranked from 0 (no risk) to 3 (high risk). L ocations rated medium (2) or high (3) were considered for this exposure analysis. Seismic data was available for the 1/475-year event, which is equivalent to a 10% likelihood of exceedance of design standards in a 50-year timeframe and is a common metric for seismic risk. Regarding infrastructure, data was assessed on available sources. Most sources have their own classification system (such as ports, airports and roads) which were used in this analysis. An especially important note is that this may be different from local classifications. For example, road classifications were available as 1-5, with 1 meaning primary roads/highways and 5 indicating earth tracks or unclassified roadways. There is no specific design standard attached to each classification in the dataset (such as roadway material or numbers of lanes). Roadways can be made of cement (highways), asphalt/bitumen (primary, secondary and tertiary roads), gravel or dirt. Each material leads to very different responses to stressors such as flooding. For further assessments of hazard impact, understanding the width of roads (number of lanes), drainage capacity and infrastructure and material is imperative for an accurate assessment of the impact of hazards. Therefore, in this report, we have classified ‘primary’ roads as level 1 and 2, ‘secondary’ roads as level 3, and ‘tertiary’ as levels 4 and 5. Based upon local design standards, the impact assessment and fragility curve/response to hazards can inform damage likelihoods and cost metrics and be adjusted for future work. Note: • All tables are given for the number of assets (number of facilities or kilometers) and then for the percentage of those assets relative to the total infrastructure dataset available for that location. • As noted throughout the report, data limitations are an acknowledged area of future work. To clarify where assets exist but are not exposed, compared with locations where no assets exist and/or no data was available (an inventory of zero), the latter are removed from the tables and boxes are blank. • The exposure analysis was performed on available infrastructure data. As noted in the main report, it is likely this data is incomplete or not fully up-to-date. For example, when available data was considered by local experts in Dominica, it was noted that the available data for ports, hospitals and bridges would be improved with local input. Tables B1-B8: Road infrastructure exposure to hazard events. Tables given for primary roads (Tables B1-B2), secondary roads (B3-B4), tertiary roads (B5-B6) and all roads (B7-B8). The first table for each type is the assets (kilometers) exposed, while the second table is the percentage exposed relative to the entire inventory. Primary roads are defined as Level 1 and Level 2 from GRIP4 classification. Secondary roads are defined as Level 3. Tertiary roads are classified by Level4 and Level5 GRIP4 data [19]. Tables B9-B14: Power plant exposure to hazard events. Tables provided for fossil fuel and unknown fuel facilities (B9-B10), renewables including hydropower, solar, wind and/or biofuels (B11-B12) and all facilities (B13-B14). Data from [37] and individual country reports, Google searches and other sources. Tables B15-B16: Bridges exposure to hazard events. Table provided for total number of assets (B15) and percentage of total inventory for which data was available (B16). Data from [25] and OpenStreetMap data as available. Tables B17-B24: Ports exposure to hazard events. Ports data is provided for “very small” ports (B17-B18), “small” ports (B19-B20), “medium” ports (B21-B22) and all port facilities (B23-B24). According to the available data, no “large” ports exist in the countries analyzed. Port definitions in this report align with classifications from the original dataset [36]. Port data is classified by the harbor size, which is determined by a combination of factors including area, facilities, and wharf space as defined in [17]. Tables B25-B32: Airports exposure to hazard events. Airport data is provided for “small” airports (B25-B26), “medium” airports (B27-B28), and “large” airports (B29-B30) and all airports (B31-B32). Airports are defined based on the data classifications, from [18]: Small airports have little to no scheduled service and light general aviation traffic; Medium airports have scheduled regional airline service, or regular general aviation or military traffic; Large airports have major airline scheduled services with millions of passengers per year, or denote major military bases. Tables B33-B34: Water and Wastewater Treatment Plant Facility (WWTP) exposure to hazard events. Tables for total facilities exposed (B33) and percentage of facilities (B34) based on available data. Tables B35-36: Hospital Facility Exposure to hazard events. Tables for total facilities exposed (B35) and percentage of facilities (B36) based on available data Tables B37-B40: Building infrastructure exposed to climate change hazard, defined by average annual change in Cooling Degree Days (CDD) between the 2050 decade and a 30-year historical baseline (1970-1999) where at least 10 models agree (21 total models were assessed). Data is provided for RCP4.5 (Tables B37-B38) and RCP8.5 (Tables B39-40).