Policy Research Working Paper 9262 Analyzing Flooding Impacts on Rural Access to Hospitals and Other Critical Services in Rural Cambodia Using Geo-Spatial Information and Network Analysis Xavier Espinet Alegre Zuzana Stanton-Geddes Sadig Aliyev Veasna Bun Transport Global Practice & Urban, Disaster Risk Management, Resilience and Land Global Practice May 2020 Policy Research Working Paper 9262 Abstract Transport connectivity in Cambodia is challenged by its in an environment with high flood risk. The method was geography and exposure to recurrent flooding. Flood events piloted in three provinces in rural Cambodia, estimating create severe disruptions in segments of the transport net- that for 37 percent of the people on those provinces, it work that undermine access to health, education, and work takes more than 60 minutes to reach an emergency health opportunities as well as create barriers to economic growth. facility. During floods, 27 percent lose all access and 18 Rural accessibility to emergency health facilities and deliv- percent experience an increase of 30 minutes in travel time. ery of medicines and basic food supplies is particularly In conclusion, this method introduces transparency and critical in times of major health crises, such as the ongoing evidence-based support for prioritization of rural transport COVID-19 outbreak. This paper provides a method to investment, identifies the social benefits (health and edu- quantify the impact of flooding on hospital access and other cation) of rural infrastructure investments, and supports critical facilities, aiming to support governments on setting policy dialogue on rural development and resilience. up health emergency mitigation plans for rural transport This paper is a joint product of the Transport Global Practice and the Urban, Disaster Risk Management, Resilience and Land Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at xespinetalegre@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Analyzing Flooding Impacts on Rural Access to Hospitals and Other Critical Services in Rural Cambodia Using Geo-Spatial Information and Network Analysis Xavier Espinet Alegre, 1 Zuzana Stanton-Geddes, 2 Sadig Aliyev, 3 and Veasna Bun 4 JEL: I18, I29, O13, O18, O19, O22, Q54, R42 Keywords: transport planning; climate resilience; disaster risk management; rural roads; rural health access; geospatial analysis; rural development; Cambodia 1 Senior Transport Consultant, SURLEA, World Bank Group, xespinetalegre@worldbank.org, +12024732994 2 Disaster Risk Management Specialist, SCAUR, World Bank Group. 3 Senior Transport Specialist, IEAT1, World Bank Group. 4 Senior Infrastructure Specialist, IEAT1, World Bank Group 1. Introduction Road connectivity in Cambodia is challenged by its geography, which includes the low-lying central plains of the Mekong and mountains in the northeast and southwest, and its exposure to recurrent flooding, particularly in the rainy season, which is between May and September, with regional variability. Cambodia is one of the most vulnerable countries worldwide to disaster risk and climate change, ranked the 12th most disaster-prone country in the 2018 World Risk Index of 172 countries by the Institute for International Law of Peace and Armed Conflict 2018. With 80 percent of the country within the Mekong River and Tonle Sap basins, Cambodia is especially vulnerable to floods, storms, and droughts. The infrastructure sector has seen significant damages due to disasters. For example, in 2014, the infrastructure sector bore some 86 percent of the total damages, predominantly roads and water and irrigation infrastructure (Royal Government of Cambodia 2014) with a cascading effect on local and regional economies due to disruption of infrastructure connectivity and supply chains and loss of market access for key economic goods. Looking ahead, climate change projections indicate temperature increase of 0.7°C–2.7°C by 2060 and increased intensity and frequency of extreme precipitation in the monsoon season and flooding (World Bank undated) which would only exacerbate current vulnerabilities. The impacts of climate change are more noticeable in the rural parts of Cambodia. Higher levels of poverty compounded by poor infrastructure maintenance and limited access to all-season roads (only 5 percent of rural roads are paved) make the 79 percent of the Cambodian population living in rural areas (2016) highly vulnerable to the frequent disruptions of flooding events. The disrupted connectivity during flooding and other disasters greatly threatens the livelihood of those rural households and their access to basic services. In a context of health emergencies and pandemics, such as COVID-19, it is critical that the population has reliable and resilient access to health facilities to ensure rapid medical response and delivery of critical medical supplies. Rural access to hospitals has been at the center of the discussion amid the COVID-19 health crisis (New York Times, 5 Washington Post, 6 and ABC News 7). While most cases and outbreaks have taken place in densely populated areas (such as Northern Italy, New York, or Madrid), those areas tend to have good access to hospitals or have the availability to set up temporary hospitals. Rural areas are much worse equipped to face the impact of a health crisis, in terms of human, operational, and financial resources and capacity. Rural access to health care centers is low; the lengthy travel times to the nearest hospital with necessary emergency 5 https://www.nytimes.com/2020/04/16/health/coronavirus-rural-hospitals.html. 6 https://www.washingtonpost.com/opinions/2020/04/06/covid-19-crisis-is-going-get-much-worse-when-it-hits-rural-areas/. 7 https://abcnews.go.com/Health/surprising-covid-19-hot-spots-coronavirus-threatens-rural/story?id=69940146. 2 equipment to treat cases such as COVID-19 may become life-threatening for some of the most critical patients (Forbes 8). Additionally, rural accessibility may become critical to ensure social and economic recovery of rural areas post- crisis. Rural accessibility gains have a strong correlation with growth in agricultural production (Limi, You, and Wood-Sichra 2017), school attendance rate (Vasconcellos 1997), and child mortality rates (Gage and Calixte 2006). Strong evidence has motivated governments and financing institutions to prioritize interventions to enhance rural accessibility. In 2006, the World Bank Group developed the Rural Accessibility Index (Roberts, Shyam, and Rastogi 2006) as a metric to support the prioritization of rural roads investment with the aim to close the accessibility gap that most rural areas suffer from. The methods and calculations used to determine the Rural Accessibility Index have evolved quickly since its inception, and in recent years, driven also with advances in analytics, availability of geo-reference data, and computing power, several new methods have been developed (Transport & ICT 2016; Mikou et al. 2019). To keep these methods simple and straightforward, the calculations of this index and the prioritization of interventions have been focusing on the origin of population, for example, where rural people live, but they do not consider the destination of that access and the facilities to which they are getting access. A few recent studies have demonstrated that calculating accessibility to specific facilities (for example, hospital, water sources, schools, or employment opportunities) using geo-spatial analysis is feasible to support the prioritization of rural investments (Andres et al. 2018; Krambeck et al. 2019; Quiros, Kerzhner, and Avner 2019). Currently, the World Bank is updating some of the methodologies for the Rural Accessibility Index and developing a toolbox to allow road agencies to calculate this index without the need of a geo-referenced Road Asset Management System. There is evidence that disruptions of road infrastructure due to climate impacts have a tremendous impact on rural accessibility especially when communities lack alternative roads and accessibility is very poor even under non-disaster event conditions (Jenelius and Mattsson 2012; Taylor 2012). Transport disruptions are estimated to produce US$107 billion economic impacts globally, not only on the loss of assets but also mostly because of disruption of trade and loss of business continuities (Rozenberg et al. 2019). Recent studies have scaled down this estimate to rural communities and have developed methods to prioritize interventions to reduce the impact and increase the resilience of the rural transport network (Alegre et al. 2018; Rozenberg et al. 2017). These methods are based on traffic flow and have not fully considered nonmotorized transport and the dramatic effect of disruption to the level of accessibility. In rural areas, where motorized traffic is low (for example, fewer 8https://www.forbes.com/sites/claryestes/2020/04/03/calm-before-the-storm-rural-hospital-workforce-pushed-to-the-breaking- point-when-treating-covid-19/#549dad573501. 3 than 100 vehicles per day), new metrics and methods need to be developed to capture the social gains that an increase in resilient accessibility can bring. The Institute for Applied Economic Research of Brazil has recently published (in April 2020) a technical note presenting the use of geo-spatial modeling to evaluate transport accessibility to hospitals for critical COVID- 19 patients in the largest urban area in Brazil. The authors prove the relevance of evaluating the transport gaps on access to hospitals so that the government can target mitigation measures. Hallegatte, Rentschler, and Rozenberg (2019) already connected the concepts of accessibility, transport disruption, and climate impacts. A new method was proposed for the city of Kampala in Uganda (Rentschler et al. 2019) to quantify the loss of accessibility to referral hospitals due to the disruption of the urban roads network. This method serves as the foundation for the analysis described in this paper. The analysis in this paper, piloted in three provinces in Cambodia (Kampong Cham, Kratie, and Tboung Khmum), quantifies the loss of accessibility to referral hospitals as well as calculates accessibility to schools and markets using methods similar to those described by Rentschler et al. (2019) but goes one step beyond to quantify the benefits of rural roads improvements and to develop a prioritization framework to select those investments that have the largest gains in resilience of rural accessibility. This analytical model was applied for the prioritization of rural roads under the Cambodia Road Connectivity Improvement Project (P169930). The analysis proposed in this paper comprises three focus areas: 1. Health emergency preparedness. The analysis used data on location of hospitals to assess the accessibility of rural people to social services, as a key indicator to ensure effective response to health crisis of rural areas. This analysis detected which areas have lower rural accessibility and what would be the accessibility gains owing to the improvement of the rural road network. Accessibility was measured as the ability of people to reach hospitals in a given time frame. This analysis also includes schools as a critical point of interest for the broader human capital development interventions. 2. Economic growth. The analysis used agricultural production data (source: International Food Policy Research Institute [IFPRI] 9) and location of regional markets (source: OpenDevelopmentCambodia website) to assess the accessibility of rural farmers to regional markets, as a key indicator to foster economic recovery of rural areas post-crisis. This analysis detected which areas have lower rural accessibility and what would be the accessibility gains owing 9 https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/PRFF8V. 4 to the improvements in the rural roads network. Accessibility was measured as the ability of agricultural products and farmers to reach local markets in a given time frame. 3. Climate resilience. The analysis for economic growth and human development was combined with climate resilience data, using flood risk maps already available through the Southeast Asia Disaster Risk Insurance Facility (SEADRIF) to assess the vulnerability of the roads to floods and the impact of such floods to accessibility to economic growth and human capital development. Using the water levels from the flood maps and the type and condition of roads, the analysis determined which roads would be impassable during flood events and then assessed which roads would require interventions to enhance climate resilient accessibility to economic growth and human development opportunities. 2. Methods 2.1. Accessibility Model The underlying accessibility model 10 follows different assumptions: 1. Two travel modes are modeled: vehicle and bicycle Each mode has different travel speeds as reflected in Table 1. Based on local consultation, each mode is assigned to a different destination type. 2. Bicycles are used to access schools, high schools, and health centers. Vehicles are used to access markets and referral hospitals. 3. People can walk up to 2 km to road and/or to destination, modeling last-mile connectivity in addition to the main mode of transportation. 4. Roads become impossible to use with more than 0.5 m of water above road design levels for vehicles and 0.25 m for bicycles. 11,12 Design standards are based on local construction practices (Table 1). Walking becomes impossible for any road under the flood zone regardless of water depths. The model does not consider specific design or location of bridges or culverts. Rather, it assumes that a water depth equal to or less than the hydraulic design would not damage the roads. 10 The model, written on python ArcGIS algorithms, uses raster-based Cost Distance function to estimate travel time to the closest destination – see Annex 1 for a sample code. 11 These values reflect a conservative approach to accommodate uncertainties and lack of information on the road network attributes such as road vertical alignment, embankments or culverts (usual values are 20 to 30 cm). 12 The model does not consider reduction of travel speeds for flood levels between 0 to 50cm. It only models passable or impassable. 5 The rural road network follows the assumptions described in Table 1. With the project intervention, travel time will be reduced (increase speed) and the road will become more resilient as drainage capacity (hydraulic design) will be larger (fourth and fifth columns in Table 1). Table 1: Road Network Characteristics and Assumptions Road type Before Project After Project Speed Hydraulic Design Speed Hydraulic Design (km/hour) (years) (km/hour) (years) Primary 50 20 80 50 Secondary 40 10 60 20 Tertiary 30 10 60 20 Rural roads (cars) 10 5 40 10 Rural roads (bicycle) 10 5 20 10 Walking 4 — 4 — The accessibility model has three main steps: 1. The travel time to the closest point of interest (schools, hospitals, or markets) is calculated for each cell in the project area. 13 The cell resolution is 500 m. 2. The population and value of agricultural production are aggregated at the 500 m resolution. 3. The population and value of agricultural production are combined with travel time and aggregated for three thresholds (30, 60, and over 60 minutes). In some instances, some population or agriculture cells may not have road access to a certain point of interest (for example, some villages in the river islands); those populations are considered ‘isolated’ and would be counted as such. Isolation would increase in cases of flood impacts as some feeder roads, lacking alternative routes, would be disrupted. 2.2. Climate Resilience Analysis To quantify the climate vulnerability, the accessibility model was used imposing the disruption of certain roads based on the flood levels and their hydraulic designs. The climate analysis is done as follows: 1. Intersect floods map with road network. Flood event model is a 50-year return event. 2. Disrupt road network based on hydraulic designs. For example, a main road will be disrupted if the difference between the 50-year flood and the 20-year flood is more than 0.5 m. This is calculated for each segment based on the design assumption. 13In rural context, density of critical facilities is very low, and the population normally has only one choice based on distance, as it is the case in the pilot provinces. In urban context, this assumption may not be valid as people’s choices for critical facilities may depend on a wide range of factors (e.g. facility reputation, insurance, social network...). 6 3. Recalculate travel times for each cell based on this new scenario. 4. Compare baseline with flooding scenarios and determine increased travel time (or loss of access) for non-isolated cells. 5. Combine loss of access with population/value of agricultural production and identify isolated cells. 2.3. Input Data The model uses a set of databases combining open source data with government official data. Description and sources are detailed in Table 2 and displayed in Figure 1 and Figure 2. Table 2: Data Inputs Description Source Population United Nations (UN) estimates at 100 m resolution from AsiaPop 2015 Points of Geo-spatial location of schools (2015), hospitals (2018), Ministry of Rural Development, interest and markets Ministry of Education, and Ministry of Health Road OpenStreetMap + project-specific shapefile (Ministry of OpenStreetMap and Ministry of Rural network Rural Development) + additional roads to access all Development points of interest Agriculture Value of Crop Production (2010) at 10 km resolution IFPRI Spatial Production Allocation Model (SPAM) Floods Flood water depth maps (5–100 years return period) Deltares for SEADRIF Figure 1: Input Data - Point of Interest in Project Figure 2: Input Data - Flood Maps (source: author’s Area (source: author’s calculations) calculations) 2.4. Evaluation and Prioritization of Interventions Based on the assumption of speed and hydraulic design for roads after project interventions, new accessibility results were calculated. First, the road network was updated with new speeds to calculate gains in accessibility. 7 The climate resilience analysis was updated with new hydraulic design to calculate reduction of loss of accessibility during flood events. Four main scenarios are modeled: pre-project (non-flooded and flooded) and post-project (non-flooded and flooded). To prioritize the interventions, the individual contribution of each road in terms of gains on accessibility and reduction of climate vulnerability (loss of accessibility) was calculated. The accessibility model was run for both non-flooded and flooded scenarios for each road under evaluation individually and assuming all other roads are not intervened. The pilot application considered 133 road segments, so in total, the accessibility analysis was computed for 134 (Baseline + 133 roads) × 3 points of interest × 2 (baseline and flooding) = 804 travel time scenarios. 3. Results The model was used in the evaluation of more than 1,600 km of new investment in three flood-prone provinces in Cambodia: Kampong Cham, Kratie, and Tboung Khmum. The total population across these three provinces is over 2 million, Kampong Cham being the most populated province with the largest number of high schools and hospitals, Kratie being the least populated province, and Tboung Khmum being the province with the largest value of agricultural production among the three provinces. 3.1. Baseline Results Table 3 and Table 4 show the results of the baseline accessibility and climate disruption analysis for the population across the three provinces (2.2 million people). Each row is the population and percentage of total population that can reach a certain destination in 0–30 minutes, 30–60 minutes, and more than 60 minutes. Value of agricultural production is not measured in population but rather in the value of agricultural production that can reach the regional market. In terms of accessibility to education, constraints are high for high schools with only 55 percent of the population 14 able to reach high schools in 30 minutes, making it suitable for an intervention (Table 3 and Figure 3). Accessibility constraints are high for referral and emergency medical facilitates with only 32 percent of the population able to reach them in 30 minutes and about 37 percent at more than 60 minutes, therefore making it suitable for an intervention (Figure 4). Finally, in terms of accessibility to markets, about 28 percent of the value of agricultural production can reach a market within 60 minutes (Table 3). 14The access to high-schools considers the whole population disregarding school-age. This assumption is made due to lack of undelaying georeferenced age population data. 8 Table 3: Results – Accessibility Baseline Scenario Only High Schools Only Referral Value of Agricultural Travel Time (×1,000 people) and Hospitals (×1,000 Production (US$, % people) and % millions) and % All population 2,220 — At 2 km of any road 2,167 1,700 0–30 minutes 1,199 55% 702 32% 166 10% 30–60 minutes 563 26% 609 28% 310 18% More than 60 minutes 343 16% 794 37% 1,090 64% Figure 3: Travel Time to High Schools in Project Figure 4: Travel Time to Referral Hospitals for Areas (source: author’s calculations) Project Area (source: author’s calculations) The climate vulnerability scenario assumes flooded roads are impassable and recalculates accessibility analysis (Table 4) and loss of access (additional travel time required relative to baseline scenario - Table 5). The last row indicates the number of people who are either isolated (they cannot reach the facility) or flooded (located in the flood plain themselves). The impact of floods is high. During heavy rains and floods, about 26 percent of the population lose access to critical facilities, health or education centers (Table 4), and 21 percent of the value of agricultural production loses access to markets (Table 4, Figure 5 and Figure 6). 9 Table 4: Results – Climate Vulnerability Scenario Value of Agricultural Travel Time High Schools Referral Hospitals Production (×1,000 Diff. (×1,000 Diff. (US$, Diff. % of Total % of Total % of Total people) people) millions) All population 2,220 — At 2 km of any 2,167 1,700 road 0–30 minutes 834 39% -16% 513 24% -8% 129 8% -2% 30–60 minutes 409 19% -7% 435 20% -8% 238 14% -4% More than 60 minutes 299 14% -2% 579 27% -10% 838 49% -15% Isolated 625 29% 15 26% 639 30% 27% 495 29% 21% Looking at results from this climate disruption in more detail, Table 5 shows the losses in accessibility (measured as an increase in travel time) for each facility in terms of people impacted. Aside from people who become isolated because of flood disruptions, about 355,000 (16 percent of total) and 401,000 (18 percent of total) will require at least 30 additional minutes to reach a high school or referral hospital, respectively. For agriculture, about US$210 million of value of production (10 percent of total) will need an additional 60 minutes to reach the closest markets. These values are different than in Table 3, as in this table all people with impacted access are accounted, rather than looking only at people who jump any of the three thresholds (30, 60 and more than 60 minutes). Table 5: Results – Loss of Access Value of Agricultural High Schools Referral Hospitals Loss of Access Production (US$, (×1,000 people) and % (×1,000 people) and % millions) and % 0–30 minutes 659 30% 569 26% 461 22% 30–60 minutes 193 9% 178 8% 153 7% More than 60 minutes 162 7% 223 10% 210 10% Isolated 571 26% 587 27% 441 21% 15 The % of total and difference of isolated trips are not the same as there is approximately 2% of population in those provinces that is isolated in the non-flooded scenarios. The people lives on river islands without connection to road network and inland waterway transport is not modeled in this analysis. 10 Figure 5: Travel Time to Markets in Pilot Area (source: Figure 6: Climate Impact on Travel Time to Markets author’s calculations) (source: author’s calculations) 3.1.1 Province Profiles Accessibility and climate resilience diverge greatly among provinces. Kratie ranks the lowest in accessibility to a hospital with more than 44 percent of its population at more than 60 minutes from a referral hospital (see Table 6). Kratie is the least densely populated province with 30 people/sq. km compared to 170 and 227 in TBK and Kampong Cham, respectively. The low population density linked with the fact that the eligibility criteria to install health facilities depend on population (with an estimated average of 120,000 people served by a referral hospital), Kratie has the lowest number of referral hospitals with lowest population density leading to high travel time for its population to reach the closest facility. On the other side, Kampong Cham has fair accessibility to hospitals, with almost 70 percent of the population in the 60 minutes range. Table 6: Accessibility to Referral Hospital by province Kampong Cham Kratie Tboung Khmum Population Density (pop/sq km) 227 30 170 Referral Hospital Density (pop/hospital) 115,000 118,000 140,000 Travel Time Population % Population % Population % 0–30 minutes 364,497 36% 93,266 31% 274,861 33% Access to Hospitals 30–60 minutes 307,854 31% 54,693 18% 247,664 30% More than 60 minutes 306,297 31% 135,295 44% 297,839 36% In terms of accessibility to markets and economic potential, Kratie has very low accessibility to markets, with more than 77 percent of its value of agricultural production at more than 60 minutes from a regional market (see Table 7). 11 Table 7: Accessibility to markets by province Kampong Cham Kratie Tboung Khmum Travel Time Population and % Population and % Population and % 0–30 minutes 59 15% 36 4% 72 15% Access to 30–60 minutes 106 27% 70 9% 134 28% markets More than 60 minutes 207 52% 619 77% 257 54% 3.2. Evaluation of the Benefits of Pipeline Project Investments The proposed pilot area considers rehabilitation of some main roads and a large portion of the rural road network. There are more than 1,300 km of roads under intervention by the Cambodian government and other development partners/donors in those same provinces (Table 8 and Figure 7). For that reason, including the current roads under ongoing or planned interventions is critical to maximize the benefits in the area and avoid overlaps. Table 8: Project Roads Length under Consideration Figure 7: Roads under Intervention in the Project Provinces (source: author’s calculations) Project Roads Total Kampong Tboung Kratie (km) (km) Cham Khmum Main roads 173 47 97 29 Rural roads 1,688 465 500 723 Roads financed by 1,312 487 319 506 various donors The accessibility values to Referral Hospital, High Schools and Markets were recalculated based on improved travel speeds from project interventions. Results in Table 9 suggest that the number of people who needed more than 1 hour to reach a referral hospital will decrease 12 percent compared with baseline scenarios. In term of access to education, people with a high school at 30 minutes away will increase 6 percent up to 61 percent of total population and improvements of market access will see 10 percent reduction on agriculture values that need more than 60 minutes to reach markets. 12 Table 9: Results – Accessibility After Project Scenario Value of Agricultural Travel Time High Schools Referral Hospitals Production (×1,000 Diff. (×1,000 Diff. (US$, Diff. % of Total % of Total % of Total people) people) millions) All population 2,220 — At 2 km of any 2,167 1,700 road 0–30 minutes 1,318 61% 6% 892 41% 9% 234 14% 4% 30–60 minutes 549 25% -1% 674 31% 3% 417 25% 7% More than 60 minutes 238 11% -5% 540 25% -12% 916 54% -10% Looking at results from this scenario in more detail, Table 10, shows the gain in accessibility (measured as a reduction in travel time) for each facility in terms of people benefitted. Larger gains are seen in accessibility to high schools and referral hospitals with more than 541,000 and 911,000 people, respectively, experiencing a reduction of travel time thanks to project interventions (Table 10). More than US$342 million of the value of agricultural production would see more than 45 minutes reduction in travel time to regional markets. All these values are larger than in Table 10, as in this case, all people with benefited access are accounted, rather than looking at only at people who jump any of the three thresholds (30, 60 and more than 60) in Table 9. Table 10: Improvement of Accessibility from Project Interventions Value of Reduction in Travel High Schools Referral Hospitals Agricultural Time (×1,000 people) (×1,000 people) Production (US$, millions) 5–20 minutes 372 358 210 20–45 minutes 142 308 265 > 45 minutes 27 245 342 Total 541 911 817 In terms of resilience, more than 152,000 and 393,000 people, respectively, will experience a reduction of more than 20 minutes travel time during rainfall events to access high schools and hospitals (sum of second and third rows in Table 11). 13 Table 11: Improvements in Resilient Accessibility from Project Interventions (during flooding events) Value of Reduction in Travel High Schools Referral Hospitals Agricultural Time (×1,000 people) (×1,000 people) Production (US$, millions) 5–20 minutes 275 301 134 20–45 minutes 122 180 203 > 45 minutes 30 212 290 Total 427 693 627 3.3. Prioritization of Road Segments The prioritization uses a multicriteria approach using the outputs of the accessibility analysis and the climate vulnerability analysis. Each criterion is based on absolute gains of accessibility by each individual road segment, calculated as the sum of the reduction in travel time in minutes for every person or the value of agricultural production. The values for each of the criteria are then normalized from 0 to 100 and then averaged out using equal weights. A value of 100 would indicate that the segment is the top segment in each of the criteria. Each road segment is then given a score from 0 to 100. These criteria and approach were discussed and agreed with the main stakeholders. After prioritization is done using this analysis, additional selection is done to select the final list of road candidates for improvement. Social and environmental safeguards screening further screens the list of identified roads to ensure that the proposed investment does not cause major negative environmental and social impacts and maximizes development gains for the beneficiaries. Finally, on this, a final short list of lifeline roads can be identified which are eligible for investment with a balanced representation of the targeted provinces (Figure 8). 14 Figure 8: Prioritization Framework The highest scoring 300 km are shown in Table 12. Each row is a road segment. Columns 3–8 are the absolute gains in accessibly in minutes × persons or minutes × US$, millions. Columns 9–12 are the normalized combined scores. The top road segment has a combined priority score of 55. Table 12: Summary of Road Prioritization Results Accessibility Climate Resilience Combined Length Map Rank Length Province Hospita Hospita Rank Rank Cumulative ID School Market School Market Climate (km) l l Access Total (km) Resilience (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) Kratie 7 72,695 86,048 457 29,814 720,883 9,748 44 67 55 42 42 Kratie 6 4,722 7,335 367 142,295 845,934 10,196 7 97 52 14 56 Kratie 8 55,094 270,671 1,540 50,224 176,029 549 72 20 46 30 86 Kratie 10 83,802 72,925 2,854 35,239 49,598 1,409 75 14 45 47 133 Kampong 13 62,782 168,176 116 99,195 90,269 245 46 26 36 22 155 Cham Tboung 5 29,620 282,494 396 40,786 205,293 489 49 18 34 22 177 Khmum Tboung 8 48,477 176,374 163 95,403 77,460 471 42 25 33 28 204 Khmum Kratie 5 17,257 50,865 2,510 5,550 365,594 465 42 17 29 74 279 Kampong 17 37,253 14,771 6 155,278 2,035 3 16 33 25 5 284 Cham Kratie 9 53,732 131,931 392 24,142 44,007 83 41 7 24 23 307 Figure 9 is a representation of the values of Table 12, where the scores are plotted on the x- and y-axes. The size of each point is related to the length of the segment. The four highest scoring roads are all in Kratie province. Kratie road segment 6 ranks the highest in resilience but low on accessibility, while Kratie road segments 8 and 10 rank the highest in accessibility gains but low on resilience. Kampong Cham roads rank lower compared to the other two provinces. 15 Figure 9: Road Prioritization Rank 3.4. Additional Use: Evaluation of Planned or Completed Projects The analysis also has the potential of evaluating already selected/planned and/or completed road projects with assumptions on speeds and hydraulic designs collected from the field, to better understand the benefits of these investments from climate and accessibility perspectives. This approach was piloted on 130 km for selected roads in three flood-prone provinces (Siem Riep, Kandal, and Stung Teng) as part of a World Bank Group-supported Cambodia Southeast Asia Disaster Risk Management Project (P160929) in rural roads of Cambodia. The methodology was used to evaluate the differences in accessibility between the baseline and the climate vulnerability scenario with and without the project. Based on this, the gains in rural accessibility during flood events were calculated, highlighting the increased resilience to those events. For example, in Siem Riep province, intervention in 51 km of resilient rural roads would mean resilient access to high schools for 23,000 students, resilient access to referral hospitals for 34,000 people, and resilient access to regional markets for more than US$33 million of the value of agricultural production. 3.5. Limitations of the Analysis This analysis has some limitations linked to data available or simplification of methodology that could eventually be addressed in future works. The main three limitations are described as follows: 1. The model only considers road transport. Although the model considers motorized and nonmotorized road transport, it does not model inland waterway nor public transport (like buses 16 or rail). Public transport is not an issue in the pilot provinces in Cambodia as it is rarely present and only serves for long-distance trips (for example, to the capital); however, if this method is to be applied in other regions (especially in the urban setting), this limitation may be important. At the same time, inland waterways are relevant in the pilot areas as there is evidence that villages depend on that mode of transport especially during rainy and flooding season. As data on inland waterways were not available, it was not possible to include it in the model at this point. 2. Accessibility is estimated based on the shortest paths to the point of interest. The model considers accessibility as the availability to reach the closest point of interest. For example, for agricultural production, it uses the provincial market as its destination regardless of the type of crops. This may not be always true as some products are sold locally, and some may be transported to the capital for international exports. This is a simplification as there are no available data on actual origin and destination or household surveys that would allow for that level of detail. Similarly, the model used assumptions of travel speeds rather than actual travel speeds (in the exception of the additional use case study in Section 3.4). This would require an intense data collection campaign that is beyond the scope of this study. 3. Agricultural data have low resolution and have no distinction between seasons. The values of agricultural production are taken from an open-source global model developed by IFPRI. 16 Its resolution is 10 km by 10 km, and it provides an indication of high-productivity areas. The values used in our model are the estimated annual value of production. As such, we did not differentiate between crops that are grown during dry or rainy seasons nor if they are perishable or not. Therefore, the estimation for climate impacts may be over- or undercalculated. With more detailed information about crops, one could model the specific crops that would be affected by disruption during the rainy season, and it would allow for a more accurate sense of economic impacts. 4. Conclusion and Next Steps Focusing on three pilot provinces, Kampong Cham, Kratie, and Tboung Khmum, the study has highlighted the urgent need to invest in resilient accessibility. For example; • Only 32 percent of its population can reach a referral hospital and emergency services in half an hour. Even though there is an extensive network of community health centers in the rural areas, accessibility to a referral hospital with emergency services such as intensive care bed 16 https://www.mapspam.info/. 17 units is low. This may be critical in treating patients with severe and quickly deteriorating conditions such as in COVID-19. Investment in rural roads would bring large benefits with reduction of people who need more than 1 hour to reach a referral hospital from 37 to 25 percent. • Just over half of the rural population in the project provinces can reach a high school in half an hour. That may be one of the factors limiting human development growth in rural areas. Investment in rural roads will increase this number to 61 percent. • Only about 28 percent of the value of agricultural production can reach a market within an hour, and during floods, 21 percent of the value of agricultural production effectively loses access to markets. Lack of access to markets limits growth of income among the rural population, including women, who account for over 44 percent of employment in the agricultural sector, and may slow down economic recovery post a crisis. Investment in rural roads will increase that value by 11 points to 39 percent. • Investing in 1,300 km of rural roads toward climate resilient could bring large gains in accessibility, with more than 390,000 people experiencing a reduction of more than 20 minutes travel time during rainfall events to access referral hospitals. Furthermore, more than US$300 million of the value of agricultural production would see a reduction in travel time to regional markets of over 45 minutes. The methodology presented in this paper aims to bring access to hospitals and schools, climate resilience, and economic growth to the prioritization and evaluation of rural transport projects. This data-driven analysis was able to provide the government with information and concepts that can guide their decision-making processes about new investments as well as an understanding of the benefits of existing interventions. This has various benefits, particularly in the context of limited data on the condition of rural roads, including: • Introduces transparency and evidence-based support for prioritization of new investments. The method provides a framework to support decision making, prioritization, and selection on new investments in rural transport based on analytics and data. The approach can also support policy makers in the preparation of sectoral strategies and financing plans for rural roads investments and maintenance in countries that are exposed to floods and climate change uncertainties. • Provides a new set of metrics to evaluation of current investment. The increase in accessibility during the dry season and the reduction in travel time to access human capital development and economic centers provide new insights on the evaluation of interventions already under development. The reduction of accessibility loss during the rainy season can be 18 used as a new metric for the monitoring and evaluation of the efficiency of investment and to better understand the benefits of improving accessibility and resilience of roads. • Identifies socioeconomic benefits of rural infrastructure investments. Economic analysis of rural transport investments has traditionally focused on reduction of road user operating cost by improving road surface conditions. The framework presented here allows to bring social benefits to rural transport investment as gain in accessibility to human capital development centers such as for health and education. At the same time, introducing the gain in accessibility to markets by productive agricultural land, the method identifies a new metric for economic growth that goes one step beyond the more traditional methods focusing on measuring road operating cost. • Supports policy dialogue on rural development and resilience. The data, maps, and results from this model engage decision makers, allowing them to identify areas with lower accessibility and high climate vulnerability. This information can inform strategic goals and action plans for the national and provincial governments. The analysis has also identified the next avenues for further improvement of this model. The first is to include supply chain resilience, based on enterprise and business census. This will bring an additional dimension to the analysis as it will evaluate economic losses of business interruption. This method has already been developed in the context of Tanzania (Colon, Hallegatte, and Rozenberg 2019), and it is another step toward operationalizing some of the analysis included in the report of Hallegatte, Rentschler, and Rozenberg (2019). There is also potential to assess poverty impacts of road disruptions using the methodology by adding the results of the model to poverty maps. Another avenue is to scale up the analysis for all 27 provinces across Cambodia to produce whole-network analysis which can inform government strategies for closing the infrastructure and accessibility gaps. 5. Acknowledgments The authors would like to thank the counterparts at the Ministry of Rural Development of the Royal Government of the Kingdom of Cambodia, especially the management and technical staff of the Ministry of Rural Development, Department of Rural Roads, and General Technical Department for their collaboration and support. We would like to thank World Bank colleagues who provided very useful advice and insights during data collection and analysis, especially Bowen Wang, Transport Analyst; Nareth Ly, Senior Health Specialist; Voleak Van, Health Analyst; Fata No, Education Specialist; Hideaki Hamada, Senior Financial Sector Specialist; Antoine Bavandi, Senior Financial Sector Specialist; and Catherine Ansell, Financial Sector Specialist. The authors also appreciate valuable comments provided by peer reviewers Dung Anh Hoang, Senior 19 Transport Specialist, Jun Rentschler, Economist at GFDRR and Phaloeuk Kong, Operational Analyst. The authors are grateful for the guidance and support received from Inguna Dobraja, World Bank Country Manager for Cambodia; Almud Weitz, Practice Manager for Transport; Abhas K. Jha, Practice Manager for Urban Development and Disaster Risk Management; and Mark Austin, Program Leader for Sustainable Development and Infrastructure. This work was funded by the Global Facility for Disaster Reduction and Recovery (GFDRR) through the Japan Disaster Risk Management Program, technical assistance program titles “Cambodia - Strengthening Financial Planning and Rural Infrastructure for Disaster Resilience (TF0A4487/P160929).” 20 6. Bibliography Alegre, Xavier Espinet, Julie Rozenberg, Kulwinder Singh Rao, and Satoshi Ogita. 2018. “Piloting the Use of Network Analysis and Decision-making under Uncertainty in Transport Operations: Preparation and Appraisal of a Rural Roads Project in Mozambique under Changing Flood Risk and Other Deep Uncertainties.” Policy Research Working Paper WPS 8490, World Bank Group, Washington, DC. Andres, Luis Alberto, Samir Bhatt, Basab Dasgupta, Juan A. Echenique, Peter William Gething, Jonathan Grabinsky Zabludovsky, and George Joseph. 2018. “Geo-spatial Modeling of Access to Water and Sanitation in Nigeria (English).” Policy Research Working Paper WPS 8357, World Bank Group, Washington, DC. Colon, Celian Olivier Victor Florent, Stephane Hallegatte, and Julie Rozenberg. 2019. Transportation and Supply Chain Resilience in the United Republic of Tanzania. Washington, DC: World Bank Group. Gage, A. J., and M. Guirlène Calixte. 2006. “Effects of the Physical Accessibility of Maternal Health Services on Their Use in Rural Haiti.” Population Studies 60 (3): 271–288. Hallegatte, Stephane, Jun Erik Maruyama Rentschler, and Julie Rozenberg. 2019. Lifelines: The Resilient Infrastructure Opportunity. Sustainable Infrastructure. Washington, DC: World Bank Group. Iimi, Atsushi, Liangzhi You, and Ulrike Wood-Sichra. 2017. “Spatial Autocorrelation Panel Regression: Agricultural Production and Transport Connectivity (English).” Policy Research Working Paper WPS 8089, World Bank Group, Washington, DC. Institute for International Law of Peace and Armed Conflict. 2018. World Risk Report 2018. Jenelius, Erik, and Lars-Göran Mattsson. 2012. “Road Network Vulnerability Analysis of Area-covering Disruptions: A Grid-based Approach with Case Study.” Transportation Research Part A: Policy and Practice 46.5 (2012): 746–760. Krambeck, Holly, Li Qu, Sarah Elizabeth Antos, and Charles James Edward Fox. 2019. Rural Accessibility Mapping: Completion Report for the East Asia and Pacific Umbrella Facility for Gender Equality. Washington, DC: World Bank Group. Mikou, Mehdi, Julie Rozenberg, Elco Eduard Koks, Charles James Edward Fox, and Tatiana Peralta Quiros. 2019. “Assessing Rural Accessibility and Rural Roads Investment Needs Using Open Source Data.” Policy Research working Paper No. WPS 8746, World Bank Group, Washington, DC. Nakamura, Shohei, and Paolo Avner. 2018. “Spatial Distributions of Job Accessibility, Housing Rents, and Poverty in Nairobi, Kenya.” World Bank Policy Research Working Paper WPS8654, World Bank Group, Washington, DC. Quiros, Tatiana Peralta, Tamara Kerzhner, and Paolo Avner. 2019. “Exploring Accessibility to Employment Opportunities in African Cities: A First Benchmark.” World Bank Policy Research Working Paper 8971, World Bank Group, Washington, DC. 21 Rentschler, Jun Erik Maruyama, Johannes Michael Braese, Nicholas K. W. Jones, and Paolo Avner. 2019. “Three Feet Under: The Impact of Floods on Urban Jobs, Connectivity, and Infrastructure.” Policy Research Working WPS 8898, World Bank Group, Washington, DC. Roberts, Peter, K. C. Shyam, and Cordula Rastogi. 2006. “Rural Access Index: A Key Development Indicator.” Transport Paper Series No. TP-10, World Bank, Washington, DC. Royal Government of Cambodia. 2014. Post-Flood Early Recovery Need Assessment Report. Rozenberg, Julie, Cecilia M. Briceno-Garmendia, Xijie Lu, Laura Bonzanigo, and Harry Edmund Moroz. 2017. Improving the Resilience of Peru's Road Network to Climate Events. Washington, DC: World Bank Group. Rozenberg, Julie, Xavier Espinet Alegre, Paolo Avner, Charles James Edward Fox, Stephane Hallegatte, Elco Eduard Koks, Jun Erik Maruyama Rentschler, and Mersedeh Tariverdi. 2019. From a Rocky Road to Smooth Sailing: Building Transport Resilience to Natural Disasters. Washington, DC: World Bank Group. Taylor, Michael A. P. 2012. “Remoteness and Accessibility in the Vulnerability Analysis of Regional Road Networks.” Transportation Research Part A: Policy and Practice 46.5 (2012): 761–771. Transport & ICT. 2016. Measuring Rural Access: Using New Technologies. Washington DC: World Bank. Vasconcellos, E. A. 1997. “Rural Transport and Access to Education in Developing Countries: Policy Issues.” Journal of Transport Geography 5 (2): 127–136. World Bank. 2018. Investing in Logistics for Sustainable Economic Growth. 22 Annex 1: Accessibility Model - ArcGIS Python Sample Code # Import system modules import arcpy from arcpy import env ## Road Network Input with Attribute with Travel Time per 100 m inFeatures = "C:/Users/wb499012/OneDrive - WBG/Cambodia/GIS/RoadSpeeds_Scenarios_v7.shp" ## Population point shapefile with Attribute on Total Population and Ag Value per point inPopPoint = "C:/Users/wb499012/OneDrive - WBG/Cambodia/GIS/Pop_Point_2km_v4.shp" ## Facilities point shapefile inHSchool = "C:/Users/wb499012/OneDrive - WBG/Cambodia/GIS/MTR/HighSchools_ProjArea.shp" inMarket = "C:/Users/wb499012/OneDrive - WBG/Cambodia/GIS/Market_ProjArea.shp" inRHospital = "C:/Users/wb499012/OneDrive - WBG/Cambodia/GIS/Hospitals_ProjArea_v4.shp" ##Raster with walking travel time (90second per 100m) with waterbodies removed inWalkbuf = "C:/Users/wb499012/OneDrive - WBG/Cambodia/GIS/ttwalk_buff_v3.tif" for valField in ["Base_time","Proj_time"]: outRaster = "RoadTime_"+valField[0:4]+"_vLE3" outCDS = "TTS_"+valField[0:4]+"_vLE3" outCDH = "TTH_"+valField[0:4]+"_vLE3" outCDM = "TTM_"+valField[0:4]+"_vLE3" outCDHS = "TTHS_"+valField[0:4]+"_vLE3" outCDRH = "TTRH_"+valField[0:4]+"_vLE3" outCDS_dr = "TTS_dr_"+valField[0:4]+"_vLE3" outCDH_dr = "TTH_dr_"+valField[0:4]+"_vLE3" outCDM_dr = "TTM_dr_"+valField[0:4]+"_vLE3" outCDHS_dr = "TTHS_dr_"+valField[0:4]+"_vLE3" outCDRH_dr = "TTRH_dr_"+valField[0:4]+"_vLE3" assignmentType = "MAXIMUM_COMBINED_LENGTH" priorityField = "Proj_speed" cellSize = 100 arcpy.env.extent = "MAXOF" if len(arcpy.ListFields(inFeatures, valField))>0: ## By Bike - 20km and 0.25 m Climate Thre # Execute PolylineToRaster arcpy.PolylineToRaster_conversion(inFeatures, valField, outRaster,assignmentType, priorityField, cellSize) outRaster_1=arcpy.sa.Con(arcpy.sa.IsNull(outRaster),inWalkbuf,outRaster) outCostDistance = arcpy.sa.CostDistance(inHSchool,outRaster_1) arcpy.sa.ZonalStatisticsAsTable(inPopPoint,"pointid",outCostDistance,outCDHS,"#","SUM") ## Modeling climate disruption based on 3 types of roads (see assumption in paper) ##rastercals_sp are the difference between RT50 and RT (20, 10 and 5 - depending on road type) waterdepth rasters outRaster_Disr = arcpy.sa.Con(outRaster_1,arcpy.sa.Con("rastercalc_sp36",outRaster_1,"","Value<0.25"),outRaster_1,"Value>35") outRaster_Disr = arcpy.sa.Con(outRaster_Disr,arcpy.sa.Con("rastercalc_sp9",outRaster_Disr,"","Value<0.25"),outRaster_Disr,"Value>8") outRaster_Disr = arcpy.sa.Con(outRaster_Disr,arcpy.sa.Con("rastercalc_sp7",outRaster_Disr,"","Value<0.25"),outRaster_Disr,"Value>5") outCostDistance_dr = arcpy.sa.CostDistance(inHSchool,outRaster_Disr) arcpy.sa.ZonalStatisticsAsTable(inPopPoint,"pointid",outCostDistance_dr,outCDHS_dr,"#","SUM") ## By Car - 40km and 0.5 m Climate Thre # Execute PolylineToRaster ## Model higher speed for cars outRaster_1=arcpy.sa.Con(outRaster_1,"9",outRaster_1,"Value = 18") outCostDistance = arcpy.sa.CostDistance(inRHospital,outRaster_1) arcpy.sa.ZonalStatisticsAsTable(inPopPoint,"pointid",outCostDistance,outCDRH,"#","SUM") outCostDistance = arcpy.sa.CostDistance(inMarket,outRaster_1) arcpy.sa.ZonalStatisticsAsTable(inPopPoint,"pointid",outCostDistance,outCDM,"#","SUM") outRaster_Disr = arcpy.sa.Con(outRaster_1,arcpy.sa.Con("rastercalc_sp36",outRaster_1,"","Value<0.5"),outRaster_1,"Value>35") outRaster_Disr = arcpy.sa.Con(outRaster_Disr,arcpy.sa.Con("rastercalc_sp9",outRaster_Disr,"","Value<0.5"),outRaster_Disr,"Value>8") outRaster_Disr = arcpy.sa.Con(outRaster_Disr,arcpy.sa.Con("rastercalc_sp7",outRaster_Disr,"","Value<0.5"),outRaster_Disr,"Value>5") outCostDistance_dr = arcpy.sa.CostDistance(inRHospital,outRaster_Disr) arcpy.sa.ZonalStatisticsAsTable(inPopPoint,"pointid",outCostDistance_dr,outCDRH_dr,"#","SUM") outCostDistance_dr = arcpy.sa.CostDistance(inMarket,outRaster_Disr) arcpy.sa.ZonalStatisticsAsTable(inPopPoint,"pointid",outCostDistance_dr,outCDM_dr,"#","SUM") inPopPoint="Pop_Point_2km_v4" arcpy.AddJoin_management(inPopPoint,"pointid",outCDRH,"pointid") arcpy.AddJoin_management(inPopPoint,inPopPoint+".pointid",outCDM,"pointid") arcpy.AddJoin_management(inPopPoint,inPopPoint+".pointid",outCDHS,"pointid") arcpy.TableToTable_conversion(inPopPoint,"C:/Users/wb499012/OneDrive - WBG/Cambodia/GIS/Results/V3/","Pop_TTRHM_"+valField+".csv") arcpy.RemoveJoin_management(inPopPoint) arcpy.AddJoin_management(inPopPoint,"pointid",outCDRH_dr,"pointid") arcpy.AddJoin_management(inPopPoint,inPopPoint+".pointid",outCDM_dr,"pointid") arcpy.AddJoin_management(inPopPoint,inPopPoint+".pointid",outCDHS_dr,"pointid") arcpy.TableToTable_conversion(inPopPoint,"C:/Users/wb499012/OneDrive - WBG/Cambodia/GIS/Results/V3/","Pop_TTRHM_dr_"+valField+".csv") arcpy.RemoveJoin_management(inPopPoint) 23