Policy Research Working Paper 8746 B eyond the G ap How Countries Can Afford the Infrastructure They Need While Protecting the Planet Background Paper Assessing Rural Accessibility and Rural Roads Investment Needs Using Open Source Data Mehdi Mikou Julie Rozenberg Elco Koks Charles Fox Tatiana Peralta Quiros Sustainable Development Practice Group Office of the Chief Economist February 2019 Policy Research Working Paper 8746 Abstract Rural accessibility is the only metric used in the Sustain- many countries by 2030. If countries spent 1 percent of able Development Goals to track progress toward better their gross domestic product annually on the upgrade of transport services in low- and middle-income countries. rural roads, even under optimistic assumptions on growth This paper estimates the rural accessibility index, defined of gross domestic product, rural accessibility would only as the proportion of the rural population who live within increase from 39 to 52 percent by 2030 across all developing 2 kilometers of an all-season road, in 166 countries using countries. Alternative solutions to rural integration must open data. It then explores the cost of increasing the rural thus be implemented in the short run until countries can accessibility index in 19 countries, using an algorithm that afford to increase significantly access to all weather roads. prioritizes rural roads investments based on their impact For example, drones that supply regular food and medicine on rural access and connectivity. Investment costs quickly supply to remote communities are much more affordable balloon as the rural accessibility index increases, question- than roads in the short term. ing the affordability of universal access to paved roads for This paper was commissioned by the World Bank Group’s Chief Economist office for Sustainable Development Practice Group and is a background paper for the World Bank Group’s report: “Beyond the Gap: How Countries Can Afford the Infrastructure They Need While Protecting the Planet.”  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/research. The authors may be contacted at jrozenberg@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 Assessing Rural Accessibility and Rural Roads Investment Needs Using Open Source Data Mehdi Mikou1, Julie Rozenberg1, Elco Koks2, Charles Fox1, Tatiana Peralta Quiros1 1 World Bank, Washington DC, USA; 2Environmental Change Institute, University of Oxford, Oxford, United Kingdom JEL: O1; O2; R0; R4 Keywords: Rural accessibility index; Open Street Map; WorldPop; Sustainable Development Goal 9; transport investments; drones. 1. Introduction Worldwide, billions of people are unable to access work and educational opportunities because transport services remain either unavailable or unaffordable. The Sustainable Development Goals (SDGs) include a goal to increase rural accessibility, but they do not set a target. Specifically, SDG indicator 9.1.1 refers to the proportion of the rural population who live within 2 kilometers of an all-season road. An "all season road" refers to "a road that is motorable all year round by the prevailing means of rural transport." In countries with a significant monsoon season, this implies paved roads; but in drier countries, all-season roads can also include gravel roads. Access to an all-season road is believed to significantly increase households’ welfare if it can help rural populations access new job markets and social services. For example, in Bangladesh, a road paving project implemented from 1997–2001 increased household expenditure by 9-10 percent on average (Khandker and Koolwal, 2011). Similarly, in Ethiopia, access to an all-season road reduced poverty by 7 percent, and increased household consumption by 16 percent (Dercon et al., 2008). However, a road may bring little economic benefit in areas with no market to sustain non-agricultural jobs. One measure of rural accessibility is the Rural Access Index (RAI), defined by (Roberts et al., 2006) as "the number of rural people who live within two kilometers (typically equivalent to a walk of 20-25 minutes) of an all-season road as a proportion of the total rural population". (Roberts et al., 2006) assessed the RAI in a small number of countries using household survey data. However, this approach was costly, the surveys lacked spatial representativeness, and the resulting RAIs were not necessarily comparable between countries. Ten years later, (Iimi et al., 2016) proposed a new method based on GIS data and applied it in eight countries. They assessed the RAI using WorldPop data for rural population distribution and digitized road network data, including road condition, provided by national road agencies. However, to be able to inform global policy for improving road access in rural areas, global-wide estimates are required. This is the first global study that attempts to provide such estimates. We go on to explore how rural access might be optimally increased via a sequential road upgrading strategy. In this paper, we build on the approach developed by (Iimi et al., 2016) and use road networks extracted from OpenStreetMap (OSM) for 166 countries, to assess road access globally. Population data for these countries is, similar to Limi et al. (2016), taken from WorldPop.1 We find that while most European countries have an RAI close to 100 percent, most developing countries are below 60 percent, all Sub- Saharan countries are below 51 percent, and 24 countries are below 20 percent. We then explore the cost of increasing rural access in 19 countries with good quality data, using an algorithm that prioritizes investment in individual rural road segments based on their impact on rural access and connectivity. The results show that costs quickly balloon as the RAI increases, but also that they depend on geography, road network connectivity, population distribution, and road unit costs. Finally, as a thought experiment, we assess the cost of providing regular food and medicine supply by drones to remote areas in Sierra Leone and compare the cost of doing so with the cost of paving roads. We find that even though drones cannot generate the same socio-economic benefits as roads, they can be a good short-term solution to increase rural integration until roads become affordable. 1 www.worldpop.org 2 2. Methods and data The RAI is calculated by overlaying a road network with a population layer, removing the urban areas, drawing a buffer around the roads, and counting the number of people in the buffer (World Bank, 2016). The code and documentation are available at http://rai-wb.readthedocs.io/index.html. We construct a GIS composed of a model of the distribution of rural population, and a geospatial model of rural roads (including their location and type). This first section describes the data that were used and the assumptions that were made to estimate rural accessibility with open source data. 2.1. Population and urban data We calculate rural accessibility for a subset of countries using different data sets for population distribution (WorldPop, LandScan, GHS-Pop) and urban area boundaries (Grump, EC LandScan and EC GHS). The results, displayed in Table 1, show a wide spread of values (RAI ranges from 29% to 51% for the Lao People’s Democratic Republic, depending on the choice of data sets, for example), and different rankings of countries depending on the data sets used. Overall, the RAI is lower when using WorldPop, and higher when using GHS-Pop. This is because WorldPop is the population data set with the least concentrated population in rural areas, while GHS-Pop is the most concentrated of those tested. The impact of the urban/rural area definition is very dependent on the country, though overall EC LandScan tends to lead to higher RAIs. Table 1 RAI calculated using the same roads but different population distribution datasets for a selection of countries with diverse geographies and population distribution Population WorldPop LandScan GHS-Pop Urban/Rural limit Grump EC LandScan EC GHS Grump EC LandScan EC GHS Grump EC LandScan EC GHS Belize 27% 45% 41% 41% 53% 48% 35% 62% 51% Lao PDR 31% 31% 29% 39% 33% 36% 51% 49% 45% Panama 23% 43% 35% 34% 37% 35% 32% 49% 35% Sierra Leone 28% 29% 23% 31% 29% 25% 44% 46% 33% Nicaragua 30% 41% 37% 42% 41% 41% 55% 62% 54% Lesotho 20% 21% 21% 46% 43% 45% 43% 43% 40% Togo 30% 26% 23% 33% 27% 30% 54% 49% 43% Guinea 26% 29% 25% 27% 25% 23% 45% 44% 37% Benin 36% 31% 27% 45% 36% 37% 60% 59% 59% Kyrgyz Republic 25% 30% 31% 45% 45% 48% 54% 62% 66% We eventually chose WorldPop for our population layer. The computational process underlying the WorldPop data is fully transparent (Stevens et al., 2015), and the model is considered to be the most accurate and robust among the currently available data sets (World Bank, 2016). In addition, to maintain global comparability and methodological simplicity, we rely on the GRUMP (Global Rural-Urban Mapping Project) data set for urban-rural delimitation, as in (World Bank, 2016). Note that WorldPop uses roads as a factor to model population distribution, so there is an endogeneity in our analysis which could lead to an overestimation of rural accessibility. However, WorldPop is still the data set with the lowest concentration of population in rural areas and thus the one that gives the lowest rural accessibility in most countries (Table 1). These caveats call for high caution in the interpretation of the results presented here, and for further 3 research to understand if and how these data sets can be used to systematically assess progress towards rural accessibility. Furthermore, these data sets are a snapshot of the population distribution in a given year. In the next decades, population will grow and migrate. To explore the potential consequences of population growth and migration on the RAI, we used spatial population projections by 2030 consistent with the Shared Socioeconomic Pathways (SSPs) (Gao, 2017). We maintained the road network constant (as given by Open Street Map in 2018) and calculated the RAI in a selection of countries in Africa where population changes are expected to be significant. We used two extreme SSPs for rural population: SSP1, in which population growth is slow and the share of people living in urban areas increases rapidly; and SSP3, which has a higher population growth and a higher share of people living in rural areas. The results displayed in Figure 1 show that changes in population distribution, as modeled by (Gao, 2017), have a limited impact on the RAI. In SSP3 the RAI remains almost constant in most countries, while it tends to decrease over time in SSP1. This is because the people migrating from rural to urban areas in (Gao, 2017) are the ones living within 2 kilometers of a paved road, thus increasing the relative share of people who live outside the 2 kilometer buffer over time. Figure 1 Evolution of the RAI when keeping the road network constant but using spatial population projections consistent with the Shared Socioeconomic Pathways by 2050 provided by (Gao, 2017). Note: the RAI in 2010 is different from the one calculated with WordPop data since the initial population distribution is different. For all the results presented in this paper, we used the static population distribution of WorldPop. It is however important to keep in mind that increasing the RAI might be even more difficult than what we find if the people who live close to roads in rural areas migrate to urban areas. 4 2.2. Road data In the original formulation of the Rural Accessibility Index, the “all-seasonality” of rural roads is based on the surface type and roughness of each road. The RAI only considers the roads serviceable if they are paved roads in ‘good’ or ‘fair’ condition, and unpaved roads in ‘good’ condition. National road agencies typically have a database of their road network with some information on paving and condition. However, in many low-income countries, these databases are incomplete and rarely updated (Espinet Alegre et al., 2018). An alternative to these databases is OpenStreetMap (OSM), a collaborative project to create a free map of the world. OSM allows any user to edit the map anywhere in the world, without restriction. It represents perhaps the largest map in the world, though completeness varies from region to region. OSM users classify roads based on their function, using their knowledge of the local area or satellite imagery. We reduce this classification down to four categories (primary, secondary, tertiary, and tracks, see Annex 2 for the detailed classification). Unfortunately, OSM does not allow users to provide information on the type of pavement or the quality of the road; and yet information on road quality is essential for the calculation of rural accessibility. To overcome this issue, the RAI is calculated three times, assuming that all-season roads can be loosely defined from the OSM ‘highway’ tag as (i) primary or secondary roads only; (ii) primary, secondary and tertiary roads; (iii) all roads. We then compare the results with previous RAI estimates (Figure 2) and with the fraction of paved roads provided by the International Roads Federation2 and conclude that the best proxy for all-season roads are primary and secondary roads. 2 https://worldroadstatistics.org/ 5 Figure 2 Comparison between RAI based on primary and secondary roads and the one previously calculated by (Iimi et al., 2016). Note: Numbers below the countries’ names are the level of completeness of OSM data. Bangladesh is an outlier because of the poor quality of OSM road data. Burundi, conversely, is an outlier because OSM ranks some roads as primary even if they are of poor quality or unpaved. 2.3. Investment costs Investment costs for upgrading earth roads to paved roads or gravel roads are taken from the Doing Business ROCKs update (World Bank, 2018). This database is a compilation of road-related projects completed by multilateral development banks. It gives the actual and estimated unit costs of road works in various countries, in order to track cost overruns. For this paper, costs were averaged by region, as not all countries were represented in the ROCKs database (Table 2). Maintenance costs are the same for all regions. Table 2 Cost per kilometer of road interventions from ROCKS averaged by region (USD) Paving 4 Paving 2 Upgrade Gravel Routine Routine Periodic Periodic lanes lanes to Paved Maintenance Maintenance Maintenance Maintenance Paved Gravel Gravel Paved South Asia 3,570,000 843,000 420,000 19,000 4,000 2,000 15,000 23,000 Sub-Saharan 3,800,000 933,000 616,000 23,000 4,000 2,000 15,000 23,000 Africa Middle East and 2,333,000 665,000 413,000 19,000 4,000 2,000 15,000 23,000 North Africa East Asia and 4,597,000 1,200,000 703,000 39,000 4,000 2,000 15,000 23,000 Pacific Latin America and 4,154,000 1,395,000 695,000 37,000 4,000 2,000 15,000 23,000 Caribbean 6 Eastern Europe 1,718,000 1,588,000 567,000 27,000 4,000 2,000 15,000 23,000 and Central Asia 2.4. Prioritization of rural investments We built a model to prioritize rural road investments based on two simple criteria: (i) maximizing the RAI increase per kilometer paved; and (ii) candidate roads for upgrade must be connected to the existing primary and secondary network. The only investment option available in our model is to upgrade existing roads labeled as either ‘tertiary’ or ‘track’ in OSM to an all-season road where those roads are directly adjacent to the network of current primary and secondary roads. The analysis was done in 19 countries in which the potential for increasing the RAI is high, and for which the data were complete enough.3 The algorithm for investment prioritization takes a two-step approach, which can be described as follows: Step 1: The OSM road data are processed to identify the largest current connected network of primary, secondary, tertiary roads and tracks (the largest connected sub-graph of the network). In many countries, a significant number of road segments are disconnected or misaligned in OSM. A quick fix is to generate a buffer of 200 meters around each road and remove the roads that are not connected to the network outside the buffer. In addition, all roads with a length of less than 500 meters were omitted, as well as the roads serving less than 0.01% of the total rural population. Step 2: The network of connected roads is then overlaid with the population layer, and the number of people living within 2km of each tertiary road or track is calculated. Tertiary roads and tracks are then ranked according to the number of people per kilometer of road. The first road from this list is recorded and marked as upgraded to all-season, the new RAI is calculated, and the algorithm starts at Step 2 again, until no more improvement to accessibility can be achieved given the restrictions outlined in Step 1. 3. Results and discussion 3.1. Global Rural Accessibility Index estimates Results for rural accessibility are displayed in Figure 3. While most European countries have an RAI close to 100 percent, most developing countries are below 60 percent, all Sub-Saharan countries are below 51 percent, and 24 countries are below 20 percent. 3 Data were considered good enough if the completeness score on Open Street Map was higher than 75 percent. 7 Figure 3 Rural accessibility index using OSM primary and secondary road networks As explained in section 2.2, the RAI was initially calculated for different definitions of “all-season roads”, using either only those roads tagged as ‘primary’ or ‘secondary’ in OSM, those tagged as ‘primary’, ‘secondary’ or ‘tertiary’, or those tagged as ‘primary’, ‘secondary’, ‘tertiary’ or ‘track’. By comparing the RAI measured with primary and secondary roads to the RAI measured with primary, secondary, tertiary and track roads, we find that a number of countries could more than double their RAI by upgrading their tertiary roads and tracks to all-season (Bolivia could go from to 20 to 70 percent, Sierra Leone from 27 to 90 percent)—but many others would not see a significant benefit (Mauritania and Turkmenistan would stay below 30 percent), even if all these countries start from similar shares of primary and secondary roads (Table 3). The low potential to increase rural access in some countries is because these countries are too vast, their population too scattered, or their existing tertiary road network too scant (Annex 2). Table 3 Road types in selected countries primary secondary tertiary track other RAI (calculated with primary and secondary roads) Bolivia 7% 5% 63% 26% 1% 20% Sierra Leone 3% 4% 53% 39% 1% 27% Mauritania 10% 4% 49% 36% 0% 9% Turkmenistan 18% 10% 58% 13% 1% 5% 3.2. Prioritizing rural investments based on rural accessibility and connectivity The algorithm described in section 2.3 was applied to 19 countries with a high OSM completion index and low rural accessibility. This section presents results for a few countries with contrasted results, while Annex 2 displays all results. Figure 4 (a) and (b) display the cost of upgrading tertiary roads, in the order identified by our algorithm, against the corresponding increase in access. Figure 4 (a) shows the cumulative cost of upgrading the roads, 8 while Figure 4 (b) shows the cost of each road upgrade per unit of RAI gained, in Sierra Leone. Results show that in Sierra Leone, paving tertiary roads would increase the RAI from 28 to 70 percent, but at a cost of $4 billion—more than the country’s GDP in 2017. Costs quickly balloon as more and more kilometers of road need to be upgraded to reach the most remote populations: improving the RAI by 1 percentage point would cost $30 million (about 1 percent of GDP) when the RAI is 30 percent, but $200 million when the RAI is 70 percent. (a) (b) (c) (d) Figure 4. Results of the algorithm for Sierra Leone. Initial roads: primary and secondary roads in red, tertiary roads in cyan. Panel A presents the Cumulative cost of increasing access in Sierra Leone from 28 to 70 percent. Panel B 9 presents the Marginal cost of increasing access in Sierra Leone for RAI between 28 and 70 percent. Panel C presents the Initial road network RAI, which is 27.6 percent. Panel D presents the road network after 1350 tertiary roads have been upgraded, resulting in a RAI of 70 percent. Comparing several countries shows that the costs of improving rural access depend on geography, road network connectivity, population distributions, and road unit costs. While Bolivia has the same rural population distribution as Morocco and a similar one to Togo, mountainous Bolivia would need to spend $2 billion to increase the RAI from 20 to 30 percent, Morocco could go from 30 to 47 percent with the same amount, while smaller Togo could go from 30 to 65 percent (Figure 5). Figure 5 Cost of greater accessibility depends on many country-specific factors. Note: Rural population on the y axis excludes people who live within 2km of a currently existing primary or secondary road. Moreover, many countries will not be able to achieve universal access to all weather roads any time soon. In Kyrgyzstan, where the road network is very sparse and where small pockets of population are scattered throughout the country, the marginal cost of rural accessibility goes from $10 million when the RAI is 28 percent to $120 million (about 2 percent of the country’s GDP) when the RAI is 32 percent. In Kyrgyzstan, increasing the RAI to 100 percent might thus not be a viable objective. Similarly, in Burkina Faso, the cost of improving the RAI from 32 to 33 percent goes up to $500 million because hundreds of kilometers of roads need to be paved to reach the remaining population (Annex 3). Countries aiming to improve social integration through improved rural access might therefore find that they cannot afford to build a large rural road network, let alone maintain it. In Togo and Sierra Leone, a network that provides access to about 70 percent of the rural population would cost around 2 percent of their current GDP annually to maintain (Table 4). Given that in the past, it is estimated that Sub-Saharan countries have been spending between 1.9 and 3.5 percent of their GDP on infrastructure (Fay et al., 2019), the cost of maintaining such a network appears very high. 10 Table 4 Capital and maintenance costs of reaching a given RAI objective Country RAI objective Total cost of upgrading Tertiary Total annual maintenance costs (paved (percent) Roads (percent of current GDP) roads) (percent of GDP) Sierra Leone 51 30 1.1 70 85 1.8 Togo 56 25 2 69 65 2.2 Burkina Faso 24 16 1.7 32 42 2.1 Guinea 40 17 1.1 47 39 1.4 Given that goals and costs are so country-dependent that it is impossible to cost overall rural access, we can reverse the question and ask: How much access could countries achieve by 2030 if they each spent 1 percent of their GDP on new rural roads every year? Our results show that with optimistic assumptions on GDP growth, the increase in access could go from 9 percentage points on average in East Asia to 18 percentage points on average in Sub-Saharan Africa (Table 5). Table 5 also shows that most of the gains can be obtained by spending only 0.5 percent of GDP every year, since the cost of incrementally improving access quickly escalates as the served population increases. Table 5 Universal access to paved roads is not within countries’ reach by 2030. Share of rural population within 2km of a primary or secondary road (percent) RAI if all countries spend 0.5 RAI if all countries spend 1 Current RAI percent of their GDP per year to percent of their GDP per 2030 year to 2030 East Asia & Pacific 52 59 61 Europe & Central Asia 29 37 40 Latin America & Caribbean 34 42 45 Middle East & North Africa 39 49 51 South Asia 43 54 57 Sub-Saharan Africa 29 42 46 Note: GDPs for each country grows according to the Shared Socio-Economic Pathway 5, which has the highest growth rate (Dellink et al., 2017). 3.3. Environmental impact of rural roads A growing body of literature is sounding the alarm on the potential negative impacts of opening or upgrading roads on forest cover and biodiversity, although the degree of impact strongly depends on the local context and road type. (Asher et al., 2018) do not find a significant impact of rural roads on forests in India, while (Pfaff et al., 2018) and (Damania et al., 2018) find that impacts can be significant in the Amazon rainforest and in Africa. To understand the possible order of magnitude of the impact, we run three scenarios: (i) newly paved rural roads lead to deforestation in an area of 25 meters on each side of the road (minimum impact, mainly due to road works), (ii) the impact spreads to 2 kilometers on each side of the road; and (iii) the impact extends to 10 kilometers on each side of the road (this is not the most extreme scenario we could consider, since Desbureaux and Damania (forthcoming) find that in Kenya the blast of a road on biodiversity can go as far as 25km). If the impact is 2km or more of deforestation, as Figure 6 shows, the consequences for 11 biodiversity could be dramatic in countries like Togo, Sierra Leone, Belize, and Morocco. Large damage to biodiversity can however be avoided by prioritizing roads that limit harm to biodiversity. Two recent studies show how rural roads can be prioritized based on both their economic and social benefits and their potential negative environmental impact by new types of analysis (Damania et al., 2018; Laurance et al., 2014). The same algorithm as the one used here could indeed be adapted to avoid the roads that come too close to a biodiversity hotspot. Figure 6 Road paving can have major impacts on forest cover 3.4. Solutions for rural integration if paved roads are not affordable Even though universal rural access to all weather roads cannot be provided in all countries by 2030, other solutions exist in the meantime to improve social integration. When rural paved roads are not affordable, other options could include investing in cabotage in coastal areas (Iimi and Rao, 2018), or smaller roads better suited for bicycles and motorcycle traffic (Raballand et al., 2010). Alternatives to paving In countries that have a dry climate and no rainy season, gravel roads can be sufficient for rural access at a significantly lower cost than paved roads. In Morocco for example, the RAI can be increased from 32 to 65 percent with gravel roads for less than 0.5 percent of GDP in total (Figure 7). In practice, investment costs for rural roads can be somewhere between the cost of paved roads and the cost of gravel roads, if drainage structures (like culverts) are put in place but the surface of the road is not paved. 12 (a) Morocco (b) Lao Figure 7 Cumulative cost of increasing access in Morocco with paved roads, gravel roads or a mix of both. Panel A presents the cumulative cost of increasing access in Morocco. Roads are paved if they get more than 100mm of rain in at least one month of the year. Panel b presents the cumulative cost of increasing access in LAO, PDR. Roads are paved if they get more than 400mm of rain in at least one month of the year Drones More recently, some countries have been experimenting with drones for medical supply delivery in remote rural areas (USAID Global Health Supply Chain Program-Procurement and Supply Management, 2017). While drones do not generate the same benefits as transport infrastructure for access to economic opportunities, they can significantly improve people’s lives by carrying medical supplies and school supplies, especially if people have access to the internet for consulting doctors. Figure 8 Rural areas accessible by drone in Sierra Leone 13 As a thought experiment, the cost of supplying low-density rural areas by drones on a weekly basis was estimated for Sierra Leone. New high-end commercial drones have a range of 40 - 60km, but cheaper ones, like the ones used by nongovernmental organizations, have a range of 20km (Raptopoulos, 2013; USAID Global Health Supply Chain Program-Procurement and Supply Management, 2017). In Sierra Leone, 75 percent of the rural population that lives more than 2km away from a primary or secondary road is within 10km of a primary or secondary road, 16 percent are between 10 and 20km, and 8 percent are between 20 and 40km (only 1 percent lives further than 40km from a primary or secondary road). If drone deliveries target only low-density areas (less than 150 people per square kilometer), the cost of delivering 1kg of supplies for 10 people every week would be between $1 per person per year and $26 per person per year, depending on their distance to the closest all-season road and the drone technology used. Table 6 presents the results based on the assumptions that a $3,000 drone flies at 40 kilometers per hour, can carry 1 kilogram, and can fly for 8 hours. A $50,000 drone flies at 100 kilometers per hour, can carry 2 kilograms, and flies for 8 hours. Costs include the capital cost of the drones plus operating costs (including labor). Isolated population is the share of the rural population living further than 2 kilometers away from an all-season road. This results in total estimated cost of $5 million to $9 million per year. Table 6 Estimated annual cost of delivering weekly supplies to low density areas in Sierra Leone for different maximum population densities and different drone technologies Max density served Share of isolated rural Total annual cost in million $ (average over 5 years) (Population/km²) population served (percent) 3 000$ Drones 50 000$ Drones 150 69 5 9 100 51 4 7 75 37 3 5 Based on the most conservative assumptions in this analysis for Sierra Leone, bringing weekly supplies to remote areas by drone costs 3 times less every year than increasing rural accessibility from 30 to 31 percent; and 22 times less than increasing accessibility from 69 to 70 percent. Moreover, these costs are likely to go down quickly, as drone markets mature, and the number of service providers increases. Thus, at least in the short term, drone delivery could help increase social integration in remote rural areas. 4. Conclusion This study presented first estimates of global road access in rural areas, combined with road upgrade cost and strategies to explore how rural access can be increased. To do so, we used the latest state-of-the-art publicly available data sets: road network data from OpenStreetMap and global population estimates from WorldPop. The results showed that for some countries, universal access to paved roads is not a realistic goal in the short to medium term. We thus reversed the question and asked how much incremental access can be achieved if countries spend 1% of their GDP annually on the upgrade of rural roads, beginning with those that have the highest impact. We found that, under optimistic assumptions on GDP growth, rural accessibility would only increase from 39 to 52 percent by 2030 across all developing countries. We also found that the deforestation effect that accompanies paving these roads can be significant in many countries. 14 Finally, as a thought experiment, we also assessed the cost of providing regular food and medicine supply by drones to remote areas in Sierra Leone. We found the cost is dramatically lower than the cost of paving the required roads in the same areas. While we do not claim that drones can replace roads, we argue they can be a useful short-term solution to improve the welfare of remote populations. For future developments, measures of rural accessibility must be used with caution when it comes to planning investment in rural roads at the country level. Upgrading rural roads to all weather roads does not guarantee welfare returns if connectivity with the rest of the network and with economic opportunities is not ensured. Second, the investments can have significant negative impacts on biodiversity. This means that the RAI alone cannot be used in the absence of a broader network model to prioritize investments based on a broader list of criteria. Avenues for improvement include using more sophisticated measures of accessibility based on travel time to nodes of interest for social or economic goals, using criticality analysis that will prioritize roads with the highest impact on travel cost or economic activity (Espinet Alegre et al., 2018), and adding penalties when roads have negative impacts on forests and biodiversity. References Asher, S.E., Garg, T., and Novosad, P.M. (2018). The ecological impact of transportation infrastructure (The World Bank). Damania, R., Russ, J., Wheeler, D., and Barra, A.F. (2018). The Road to Growth: Measuring the Tradeoffs between Economic Growth and Ecological Destruction. World Development 101, 351–376. Dellink, R., Chateau, J., Lanzi, E., and Magné, B. (2017). Long-term economic growth projections in the Shared Socioeconomic Pathways. Global Environmental Change 42, 200–214. Dercon, S., Gilligan, D., Hoddinott, J., and Woldehanna, T. (2008). The Impact of Agricultural Extension and Roads on Poverty and Consumption Growth in Fifteen Ethiopian Villages. American Journal of Agricultural Economics 91 (2009) 4 91. Espinet Alegre, X., Rozenberg, J., Rao, K.S., and Ogita, S. (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 (The World Bank). Fay, M., Han, S., Lee, H.I., Mastruzzi, M., and Cho, M. (2019). Hitting the Trillion Mark -- A Look at How Much Countries Are Spending on Infrastructure (The World Bank). Gao, J. (2017). Downscaling Global Spatial Population Projections from 1/8-degree to 1-km Grid Cells. Iimi, A., and Rao, K. (2018). Spatial Analysis of Liberia’s Transport Connectivity and Potential Growth (Washington DC, USA: The World Bank). Iimi, A., Ahmed, A.K.F., Anderson, E.C., Diehl, A.S., Maiyo, L., Peralta Quiros, T., and Rao, K.S. (2016). New rural access index : main determinants and correlation to poverty (The World Bank). Khandker, S.R., and Koolwal, G.B. (2011). Estimating the Long-Term Impacts of Rural Roads: A Dynamic Panel Approach (Rochester, NY: Social Science Research Network). Laurance, W.F., Clements, G.R., Sloan, S., O’Connell, C.S., Mueller, N.D., Goosem, M., Venter, O., Edwards, D.P., Phalan, B., Balmford, A., et al. (2014). A global strategy for road building. Nature 513, 229–232. 15 Pfaff, A.S.P., Robalino, J., Reis, E.J., Walker, R., Perz, S., Laurance, W., Bohrer, C., Aldrich, S., Arima, E., Caldas, M., et al. (2018). Roads & SDGs, tradeoffs and synergies: Learning from Brazil’s Amazon in distinguishing frontiers. Economics: The Open-Access, Open-Assessment E-Journal 12, 1–26. Raballand, G., Macchi, P., and Petracco, C. (2010). Rural road investment efficiency : lessons from Burkina Faso, Cameroon, and Uganda (The World Bank). Raptopoulos, A. (2013). No roads? There’s a drone for that. Roberts, P., KC, S., and Rastogi, C. (2006). Rural Access Index: A Key Development Indicator (Washington DC, USA: World Bank). Stevens, F.R., Gaughan, A.E., Linard, C., and Tatem, A.J. (2015). Disaggregating Census Data for Population Mapping Using Random Forests with Remotely-Sensed and Ancillary Data. PLOS ONE 10, e0107042. USAID Global Health Supply Chain Program-Procurement and Supply Management (2017). Unmanned Aerial Vehicles Landscape Analysis : Applications in the development context (Washington DC, USA: Chemonics International Inc). World Bank (2016). Measuring rural access : using new technologies (The World Bank). World Bank (2018). Road Costs Knowledge System (ROCKS) – Doing Business Update (Washington DC: World Bank). Annex 1: OSM roads classification Below is a list of OSM road labels and classification in primary, secondary, tertiary, and track. All other types of roads are in “other”. ‘highway’ tag in OSM  Meta‐Classification   unsurfaced   services   corridor   escape   emergency_access_point   track  emergency_bay   construction   bridleway   path   track   living_street   residential   service   tertiary  tertiary   tertiary_link   unclassified   16 road   secondary   secondary  secondary_link   primary   primary_link   trunk   primary  motorway   trunk_link   motorway_link       Annex 2: All RAI results OSM  Country   RAI PS    RAI PST    RAI PSTT   completeness   Afghanistan  20.3  38.8  56.8  10%  Albania  66.7  98.5  99.8  75%  Algeria  72.6  86.6  87.8  98%  Angola  19.6  40.9  42.2  60%  Argentina  43.2  73.1  74.7  100%  Armenia  17.7  28.9  29.5  94%  Azerbaijan  27.2  58.1  65  71%  Bangladesh  47  77.3  77.6  19%  Belize  26.7  70  76.2  98%  Benin  36.1  80.7  81.7  82%  Bhutan  34.2  64.1  67.6  17%  Bolivia  20.5  70.1  76.6  100%  Bosnia and Herzegovina  79.2  98.7  99  97%  Botswana  28  56  66.1  61%  Brazil  35.5  65.5  67.5  100%  Bulgaria  37.7  99.8  99.9  97%  Burkina Faso  15.7  51.7  56.3  93%  Burundi  65  93.6  93.8  87%  Cambodia  38.1  89.2  90.8  75%  Cameroon  31.1  66.6  73.5  100%  Cabo Verde  56.8  86  95.2  100%  Central African Republic  22.8  33.6  40.8  99%  Chad  13.9  28.5  35.3  58%  China  57.4  71.4  71.8  24%  Colombia  23.9  52.9  63.5  73%  Congo, Dem. Rep.  21.8  46.3  50.6  34%  17 Congo, Rep.  12.5  39.7  46  92%  Costa Rica  45.5  93  93.8  93%  Côte d'Ivoire  22.7  72.6  74.2  61%  Djibouti  28.3  40.9  53.5  42%  Dominica  53.1  83  97.6  100%  Dominican Republic  49.3  82.4  85.9  76%  Ecuador  49.1  81.8  83.9  98%  Egypt, Arab Rep.  78.8  95.5  95.7  17%  El Salvador  46.1  92.3  92.9  75%  Eritrea  23.3  33  34.5  82%  Ethiopia  14.1  48  53.6  42%  Gabon  19.7  32.3  63.6  100%  Gambia. The  48.8  84.9  96.1  61%  Georgia  37.4  76  85.4  97%  Ghana  43.8  93.9  94.2  45%  Guatemala  30.4  71.8  73.1  47%  Guinea  26.2  71  78.7  100%  Guinea‐Bissau  15.3  53.3  65.2  27%  Honduras  30.7  63.3  65.2  67%  Hungary  83.9  99.8  100  100%  India  42.8  69.2  69.6  36%  Indonesia  47.9  71.6  72.6  52%  Iran, Islamic Rep.  20.3  46  49.3  25%  Iraq  40.8  73.8  74.7  65%  Jamaica  48.1  93.2  94.2  100%  Jordan  47.2  84.9  87.6  96%  Kazakhstan  26.3  50.9  57  29%  Kenya  41.7  83.6  86.2  71%  Korea, Dem. Rep.  57.7  82.4  83.5  66%  Kosovo  82.1  99.9  100  92%  Kyrgyz Republic  25.4  58.2  64.9  85%  Lao PDR  31.5  63.5  70  76%  Lebanon  78  97.5  98.6  79%  Lesotho  20.2  72.6  92.4  99%  Liberia  39.2  77.8  84.3  87%  Libya  26.6  57.6  59  75%  Macedonia. FYR  84.9  99.8  99.8  100%  Malawi  36.5  90.4  91.8  32%  Malaysia  50.1  65.7  68.8  99%  Mali  27.2  64.1  69.5  94%  18 Mauritania  9.4  19.1  26.9  80%  Mexico  40.8  74.7  76.4  80%  Mongolia  18.9  36.1  42.7  38%  Montenegro  64  98.9  99.8  100%  Morocco  31.4  80.3  88.8  80%  Mozambique  17.8  53  55.7  55%  Myanmar  38.3  68.4  69.8  47%  Namibia  52  68.2  73.4  93%  Nepal  51.6  89.5  93.7  100%  Nicaragua  30.1  61.7  68  86%  Niger  20.5  35.6  42.8  91%  Nigeria  41.5  85  85.9  36%  Pakistan  39.9  59.2  60.1  14%  Panama  22.7  58.4  61.2  89%  Paraguay  26.8  79.3  89.1  100%  Peru  28  47  49.9  100%  Philippines  64.1  87.1  89.5  100%  Romania  82.1  99.9  100  100%  Rwanda  44.2  92.3  92.7  44%  São Tomé and Príncipe  46  71.4  80.6  99%  Senegal  24.1  63.1  75.1  70%  Serbia  69  98.9  99.1  100%  Sierra Leone  27.6  90.6  94.2  84%  Somalia  38.2  57.8  92.1  98%  South Africa  26.5  72.2  77  81%  South Sudan  14.4  26.2  34.5  58%  Sri Lanka  68.7  95.5  96.8  57%  Sudan  10.5  24.3  26.9  87%  Eswatini  52.4  97.1  97.9  82%  Syrian Arab Republic  54.4  95.6  96.5  100%  Tajikistan  26.3  55  59  30%  Tanzania  29.9  73.2  74.9  59%  Thailand  46.1  92  93.3  77%  Togo  30.4  85  87.8  77%  Tunisia  36.8  72.9  79  91%  Turkmenistan  5.1  10.8  13.2  77%  Uganda  45  91.2  91.9  55%  Ukraine  47.7  94.1  97.5  95%  Uzbekistan  19.3  57.2  61.4  26%  Venezuela, RB  22.2  42.6  46  68%  19 Vietnam  66.7  88.5  88.8  47%  West Bank and Gaza  71.8  91.8  99.7  100%  Yemen, Rep.  28.4  51.1  53  73%  Zambia  12.5  36.3  38.7  72%  Zimbabwe  15.8  65.9  72.7  68%  20 Figure 9 RAI calculated using WorldPop and GRUMP urban delimitation, using different OSM roads. PS is for Primary and Secondary roads, PST for Primary, Secondary and Tertiary, and PSTT for Primary, Secondary, Tertiary and Tracks. 21 Annex 3: All road prioritization results 22 23 24 25 26 27