People and infrastructure: multi-scale assessment of coastal and fluvial flood exposure in India

India is one of the world’s most flood-prone countries, with present-day risks likely to be exacerbated by climate change in the coming decades. The type of risk varies by location, with the lives, homes, and livelihoods of residents of India’s coastal megacities threatened by coastal floods and storm surges while village-dwellers residing in rural flood plains may additionally lose both crops and livestock. Schools and health facilities throughout the country are also at risk. This multi-scale study employs several datasets, from multiple domains, to generate high-resolution estimates of potential exposure to fluvial and coastal floods for (1) urban and rural populations, (2) health facilities, and (3) educational facilities. Our results, presented at the state level, suggest high exposure to fluvial flooding with about 184 million or more than 1:7 of India’s population at risk. This proportion is somewhat higher for rural dwellers (15.8%) compared with urban residents (14.2%). Urban residents, however, are much more likely to be affected by coastal floods, likely due to the high population densities of India’s coastal megacities. In total, around 19,218 (15%) of health and 34,519 (18%) of educational facilities are exposed to either coastal or fluvial flood risks. A spatially detailed, locally refined, comprehensive flood risk assessment such as this is critical to inform and target public policy and guide disaster risk reduction plans. By improving infrastructure, increasing awareness, and developing proactive, targeted, and inclusive flood plans, communities can build resilience.


Introduction
Floods are the most common extreme weather event and the leading cause of fatalities worldwide, resulting in significant damage to infrastructure and threatening the livelihoods of millions of people every year [1,2]. Floodplains and deltaic regions have growing populations that are increasingly impacted by floods and related agricultural impacts and reduced food availability [3][4][5][6][7][8]. Meanwhile, low-lying cities and towns near the coast in the low elevation coastal zone (LECZ) face additional risks [9][10][11][12][13]. The estimated 750 to 1.1 billion persons globally in this zone are exposed to sea level rise and related coastal ecological harms, causing damage to key infrastructure and disruption of services, particularly in urban areas [8,[14][15][16][17].
India is one of the most flood-exposed and vulnerable countries, with large numbers of people residing in the Ganges-Brahmaputra-Meghna GBM) deltaic system and a large coastal population [12,13]. The GBM is home to a growing population, and due to anthropogenic climate change, is increasingly flood-prone. On the coasts, sea-level rise and coastal flooding are a major concern as urbanization and migration drive high population density in coastal cities; two of India's megacities, Mumbai, and Kolkata, already rank among the most exposed cities in the world [9]. The population of India exposed to fluvial or coastal flooding is already high and projected to remain high or even increase further due to natural growth and migration [12,18].
In addition to people, key infrastructure including schools and health facilities across the country are exposed. Detailed climate change adaptation plans are required to protect this critical infrastructure, by informing investments in both existing and future infrastructure. If schools are flooded and forced to close or are used for post-disaster response, then floods may negatively affect enrollment and attendance, which studies suggest will disproportionately impact girls [19]. Relatedly, if health facilities are damaged or destroyed during extreme weather events, or health resources and staffing are disrupted, this will impact the health and well-being of the population. Additionally, flood events themselves increase the need for medical care, by directly causing physical harm or increasing the risk of water-borne, vector-borne, and other communicable diseases [20].
According to the Intergovernmental Panel on Climate Change's sixth assessment report (IPCC AR6), risk is defined as the potential for adverse consequences for human or ecological systems, including those on lives, livelihoods, health and well-being, economic, social and cultural assets and investments, infrastructure, services, ecosystems, and species [8]. Risk is shaped by the intersection of hazard, exposure, and vulnerability, informed by adaptation-related responses and the limits to adaptation. The risk propeller framework is used to highlight these relationships, the diagram along with detaileddescriptions are provided in the supplementary document (figure S1).
There is a growing literature on the effects of flooding in countries around the world [7,[21][22][23], including in India [24][25][26][27]. However, a comprehensive, multi-scale, and locally refined approach for India consistent with the risk propeller framework is lacking. The emergence of global remotely sensed products to map environmental hazards and human settlements with plausible spatial detail, coupled with increasing availability of higher quality data from census, crowd-sourced platforms, and other governmental programs, offer an opportunity to address these gaps. This study is comprehensive in that it leverages multiple datasets to maximize the inclusion of people and settlements with urban versus rural differentiation, multi-scale in that it allows for analysis from national down to sub-national units, and locally refined meaning that it is spatially detailed, down to the smallest possible census units (settlements) and built-up locations from high resolution remote sensing imagery. This employs a standardized approach for exposure analysis for all of India, down to 30 meter grid cell detail.
This study responds to the need fora flood risk assessment within India. For each subnational territory, we estimate different flood hazards and generate exposure accounts of population and infrastructure by urban and rural areas. Whenever possible, we also discuss population vulnerability. This is the first step for a complete subnational flood risk assessment in India that can be later enhancedby more in-depth vulnerability differentiation. Our analysis leverages multiple datasets from different domains to overlay fine scale estimates of fluvial and coastal flood zones with the locations of urban and rural settlements, and health and educational facilities. We use a remotely sensed measure of potential flooding because it is standardized and globally available, allowing for replication of our approach in other settings. We added the elevation-based LECZ to better capture coastal flood hazard because hydrological models may underestimate the flood hazard in coastal areas [28]. We create and present state level estimates as state authorities (in collaboration with national level entities) are responsible for the development of legislation and policy to address flooding, via infrastructure improvements, early warning systems, and deployment of resources or aid [29].

Census data
In this paper, we use the most recent India census data, that is for 2011. India's office of Registrar General and Census Commissioner has published a very large collection of detailed, settlement-specific tabulations of 2011 population census data, termed primary census abstracts (PCAs). Each settlement is assigned a category, which reflects the official way of India's census to define urban and rural areas based on jurisdictional criteria [30]. Accordingly, the category of urban settlements is either 'statutory town', 'census town', 'ward', or'outgrowth' whereas 'village' settlements constitute rural areas. PCAs are linked with their spatial boundaries, using the proprietary 'Village Map' data products from ML Infomap LLC 5 . The resulting spatial population dataset contains more than 650,000 settlement records, which presents the highest spatial detail of administrative units. While more recent spatial data on village boundaries in 2021 [31] are available, the population records of these new boundaries are still from 2011. Moreover, the 2011 census data offers the most fine scale resolution and complete spatial data available, particularly for urban areas where we have boundaries for statutory towns, ward boundaries of all large towns, outgrowths, and census towns. For example, the city of Mumbai is represented by only 2 settlements in the settlement boundary dataset of 2021, compared to 102 in the 2011 dataset. Similarly, while Kolkata and its surrounding areas are represented by 212 settlements in 2011, this number is reduced to 44 in 2021. To minimize uncertainty and maintain consistency in population allocation from settlements to their constituent built-up cells across both urban and rural areas, we decided to use the 2011 dataset, which is more spatially detailed across both landscapes.

Satellite derived built-up products
The Global Human Settlement Layer (GHSL) uses remotely sensed imagery at various spatial resolutions to quantify human presence on the Earth [32]. One GHSL product is the spatial footprint of human settlements, referred to as built-up areas. To align with the census data (collected in 2011) we use the Landsat-based GHSL built-up area grid (imagery collected in 2014/2016) at 30m resolution [33,34]. While GHSL is considered robust in delineating urban built-up areas, it is limited in identifying rural built-up areas [35,36]. To alleviate this issue, we combine GHSL with the High-Resolution Settlement Layer (HRSL) [37]. HRSL also provides estimates of population distribution and built-up locations at approximately 30 m resolution but can better identify and delineate rural settlements because it relies on very high-resolution satellite imagery at around 2 m resolution. Upon visual inspection, we found some issues with HRSL in urban areas. By combining these two products, we generate a synergic layer at 30 m resolution that can more reliably identify built-up areas across both urban and rural landscapes.

Fluvial flood hazard-Fathom Global model
The Fathom Global model is a high-resolution hydrological flood hazard model that delineates geographic areas at risk of fluvial flooding. It maps maximum water depth across the entire global domain for 10 specified return periods at 3 arc second spatial resolution (∼90 m) [28]. The model employs several inputs and innovations to generate a global flood model, such as a global terrain model corrected for elevation biases in forested and urban areas, catchment descriptors of climate class, upstream annual rainfall, global runoff data, catchment delineations, and river networks. It has been validated against high-quality local benchmark flood hazard information in parts of the world where such data are available. It is shown to capture between two-thirds and three-quarters of the area determined to be at risk in the benchmark data without generating excessive false positive predictions and has already been used in global flood risk assessments [28,38].
Fluvial flood hazard can be summarized by flood depth, the extent of the flood or area inundated, and frequency. We created a binary measure of depth for any flooding over 0 meters, to avoid using arbitrary cut-offs that would exclude certain floods, and to be more inclusive of potential flood risk. Floods may impact areas differently based on many characteristics, for example, construction of embankments or type of land use, so for this analysis, we chose to include any level of flooding above 0 meters. Our approach creates three levels for flood hazards based on the extent and frequency of floods based on return years of 10, 50, or 100 years. Grid cells with a return period of 10 years are categorized as level 3; these areas have a higher likelihood of flooding (10% chance per year) and cover a small extent. Level 2 areas have a return period of 50 years, meaning a 2% chance of flooding per year. For level 1, these are 100-year floodplains (1% chance per year) and extend beyond the reach of the other two levels. Level 3 areas are shared between 10-year, 50-year, and 100-year return intervals, Level 2 areas are shared between 50-year and 100-year intervals, and level 3 only includes areas inundated once a century. We assume that level 1 floods are the most severe because they occur rarely and reach a broader geographic area outside the usual fluvial flood footprint. Level 3 is the areas at the most imminent and frequent risk of flooding.

Coastal flood hazard-Low Elevation Coastal Zone
The LECZ [13] is a global product used to model areas susceptible to coastal floods [12,39]. LECZ is constructed based on coastally contiguous elevation data. The most recent and enhanced generation of LECZ is based on the Multi-Error-Removal Improved-Terrain DEM (MERIT-DEM) [40] at 250 m spatial resolution. The LECZ is defined as 10 meters or less coastally contiguous elevation. We categorize 0-5 meters above sea level as risk level 2, or the highest risk, and 5-10 meters as risk level 1.

Infrastructure: health and educational facilities
We accessed the most recent OpenStreetMap (OSM)'s spatial datasets of health and educational facilities for India in 2022 from the Humanitarian Data Exchange (HDX) portal [41,42]. Because India's population is mostly rural, and therefore it is likely that more facilities are in rural areas, we used an additional dataset in 2019 from the Pradham Mantri Gram Sadak Yojana (PMGSY) III to ensure a complete list of facilities in rural areas. PMGSY is a program by the Government of India to improve access to economic and social services such as rural markets, schools, and health facilities [43,44]. This dataset includes location information on the health and educational facilities in rural areas, where data based on OSM may be missing. Because our main objective was to create a complementary dataset of facilities with sufficient inclusion in both rural and urban areas and not a comparison between the two, we proceeded with combining these two datasets although they were published about two years apart.
To generate complete spatial datasets of facilities, we processed and integrated the OSM and PMGSY III files. We extracted OSM health facilities if their attribute was listed as 'Hospital', 'Clinic', 'Medical', and 'Pharmacy' under 'amenity'. If amenity was missing for a record, we extracted facilities with 'Health Centre' or 'Hospital' in their 'name' attribute. We also extracted PMGSY III records listed as 'Primary Health Centre', 'Bedded Hospital', and 'Community Health Centre'. We retrieved OSM educational facilities with 'school', 'university', 'college', or 'kindergarten' in their amenity attribute. From PMGSY III, we extracted those identified as 'High School -General', 'High School -Girls', 'ITI', 'Higher Secondary School', and 'Degree College'. To remove potential internal duplicates in each dataset, we only kept one record from those that were within a 100 m proximity. To generate complementary datasets for health and educational facilities, we linked OSM and PMGSY records and removed potential duplicates across the datasets by dropping any PMGSY-based facility within a 1 km buffer around an OSM-based facility to compensate for the inconsistencies in name-matching, which was significant across the datasets. The final two collective datasets contain unique records for health (∼125,000) and educational facilities (∼187,000), respectively.
We also created a crude measure of population served by each facility by dividing the number of people by the number of each kind of facility per state. This was calculated to estimate the potential average number of people who would lose access if a facility was inaccessible or destroyed due to flooding.

Data integration and analysis
To evaluate population exposure to potential fluvial and coastal floods, we first generated a high-resolution spatial population grid (30 m). Dasymetric modeling was implemented to allocate each settlement's population to its constituent built-up grid cells, assuming that a settlement's population is more likely to reside in its builtup locations [45][46][47][48]. While global grid-based population products, such as GHS-Pop [49], and GPW [50], are already available, these allocate population from larger administrative units (e.g., sub-districts) [51], employ one built-up layer in their approach, and their resolution is coarse. Our approach uses the smallest possible settlement units and the synergic built-up layer to minimize inherent uncertainty in population allocation from large units [52] and missing rural population [35] to generate a more locally sensitive representation of population distribution essential for plausible exposure evaluation [38,53]. The unique outcome is a highresolution population grid (30 m) by downscaling populations of settlements to their constituent grid cells. We also generated a second grid to quantify the urban/rural status of each cell, assuming all cells within a settlement inherit its urban/rural status. The resulting dataset is the urban/rural grid at 30 m spatial resolution, in the Mollweide projection.
Finally, we integrated the population and urban/rural grids of each Indian administrative division (27 states and 8 union territories based on census 2011) with flood layers to estimate summaries of urban and rural population exposures to the two risk levels of coastal floods and three risk levels of fluvial floods. Figure 1 illustrates the approach for estimating exposure to the LECZ-based coastal flooding for the city of Mumbai in the Maharashtra state.
To evaluate the exposure of facilities to potential fluvial and coastal floods, we implemented a similar data integration approach. We integrated facility locations with the urban/rural and flood grids for states and union territories to derive their number of facilities exposed to the two flood types and different risk levels. Throughout, we report both the proportion of people and facilities exposed, as well as the absolute number, within each state. The proportion highlights the relative exposure, quantifying in which states, populations and facilities are more likely to be exposed. The absolute number is used because even a small proportion of exposure in a populous state may translate into a high number of population and facilities affected, which is associated with greater financial costs and losses of life.

Results
In total, there are many more rural residents exposed to any type of flood (∼168 rural versus ∼91 million urban people) (table 1). This is particularly the case for fluvial floods (e.g., for risk level 3, 53.5 rural versus 19.1 million urban people are potentially exposed). For coastal flooding, however, there are more urban residents exposed (37.6 versus 36.4 million). Figure 3 highlights fluvial and coastal flood exposures in proportional terms by urban and rural populations at the national level. It shows that 6.4% of rural dwellers are potentially exposed to the most imminent fluvial flooding category compared to 5.1% of urban residents. For coastal flooding this is the opposite, with urban residents more likely to be exposed; they are three times more likely to be exposed to risk level 1 of coastal floods than rural residents (figure 3).
Compared to another Asian flood-prone country, 33% (∼453.3 M) of China's population in 2015 was concentrated in floodplains [6]. Floodplains in that study are defined by a riverine flood depth map with a 100year return period. So, the corresponding estimate in our analysis is ∼184.8 M (15.3%). Figure 4 depicts potential proportional population exposures to each flood type and its levels per subnational territory (i.e., 2011 states and territories), sorted by the absolute number of people exposed to fluvial floods. For example, while both urban and rural residents in Assam are more likely to be affected by fluvial floods, the larger population of West Bengal leads to higher aggregate exposures ( figure 4). Moreover, the maps demonstrate the spatial heterogeneity of urban and rural exposures to all levels of fluvial floods in both proportional and absolute terms. The three discrete colors along the horizontal axis of the bi-scale scheme (in green/blue shades) represent the terciles of absolute exposure, while the vertical axis delineates the three terciles of proportional exposure (purple shades), together by their interactions making nine distinct categories to simultaneously differentiate exposure in absolute and relative forms. For example, dark blue (top right corner of the color scheme) indicates areas of greatest exposure both in the proportion and number of persons in each state or union territory exposed: urban exposures being highest in Punjab and West Bengal, and rural exposures being highest again in Punjab but also in Bihar and Odisha. Urban and rural exposures are also quite high, especially as a fraction of their population, in the northeastern states overall. Table 2 summarizes the counts of health and educational facilities potentially exposed to the different risk levels of the two flood types. In total, 19,218 (15%) health and 34,519 (18%) educational facilities are in areas characterized by at least one of these flood types. Although rural facilities are considerably more exposed to fluvial flood zones (8,504 versus 4,920 for health facilities, and 21,278 versus 3,824 for educational facilities), exposure of urban health facilities to coastal flood zones is more frequent (3,637 versus 2,157), and the gap between rural and urban exposures is significantly narrower for educational facilities (5,564 versus 3,853). In proportional terms, rural facilities are more likely to be affected by fluvial floods, and urban facilities are more likely to be affected by coastal floods, much as with population exposure. Around 12% of rural health and 14% of rural educational facilities are exposed to any of the 3 fluvial flooding risk levels, while these percentages are 9% for urban health and 12% for urban educational facilities. In coastal areas, 6% and 12% of urban health and educational facilities are exposed to any level of coastal flood risk, while these percentages are 3% and 4% for their rural counterparts. Figures 5 and 6 show the percentages of health and educational facilities in India's states and territories potentially exposed to either flood type, sorted by the total absolute number of facilities exposed to fluvial floods. Thus, Maharashtra and Uttar Pradesh have the highest number of health facilities exposed (figure 5), and Uttar Pradesh and Assam have the highest number of educational facilities exposed (figure 6). We created state-level, urban and rural terciles of the percentage of facilities exposed to any type of flood (purple shades on the y-axis), and terciles of the ratio of population per exposed facility (green/blue shades on the x-axis) as an estimate of the population served, which is derived based on assigning an average number of population served by each exposed facility by dividing the population of each state by its number of exposed facilities. The maps with the bi-scale scheme show the state-level overlap of these two indicators. The first dimension highlights the likelihood of exposure and the other the population impacted that rely on these facilities. Note, in both (figures), the absence of the darkest blue color (the top right corner of the scheme) in the maps indicates there is no subnational area with both dimensions falling in the top third of their respective distributions.

Geographic variation of exposure to floods
Overall, our findings confirm the significant absolute and proportional number of people and facilities exposed to potential flooding across India. We identify significant geographic heterogeneity by state and by urban or rural classification. Rural dwellers, particularly those in states located within the GBM basin, are highly exposed to fluvial flooding, while states in the LECZ that contain India's megacities are at heightened risk of coastal flooding. The widespread spatial footprint of fluvial floods and the considerable presence of rural communities in India make the exposure of rural populations and facilities considerably higher in absolute terms. Based on the official criteria, despite being the world's second most populous country, India is also one of the least urban, with only 33% classified as urban in 2015 [30,54]. Fluvial flooding disproportionately impacts states within the GBM basin such as Assam, Bihar, Uttar Pradesh, and West Bengal, as well as the most populous states such as Maharashtra. This finding is consistent with previous literature that suggests Bihar is the most flood-prone state in India, with 55% of its area affected by frequent flooding [24], and that Bihar and Uttar Pradesh having the highest population exposures in the GBM basin [55].
Our results on proportional exposures of both population and facilities indicate that LECZ areas with high coastal flood exposure are disproportionately urban. For example, there are more than 15 million urban residents in West Bengal potentially susceptible to coastal floods, accounting for over half (52%) of its total urban population. West Bengal is the state where Kolkata is located (a city with more than 14 M residents), which is likely driving this observation.
The facilities and infrastructure that serve India's growing population follow a similar pattern of flood exposure, as their catchment areas relate to population density. Urban health and educational facilities are twice and four times more likely to be exposed to coastal floods than their rural counterparts, respectively. States such as Kerala, Rajasthan, and Andhra Pradesh have high exposures, in part because these states have a relatively high number of facilities. In contrast, however, Bihar is one of the states with the highest population exposure but not significant exposure to facilities, but this is likely due to a relative lack of facilities. Despite being the third most  populous state of India, it is severely under-resourced and one of the poorest states, with fewer schools and facilities for its population [56]. Conversely, Kerala is a much wealthier state with high fluvial and coastal exposures and a lower population per exposed facility.

Risk assessment
Risk from environmental hazards is a complex and multi-aspect problem, with the exposure of population and facilities and mapping of hazards partly explaining many factors to quantify risk and harms. A comprehensive risk assessment should incorporate information regarding poverty, marginalized groups, occupational status, and additional measures of vulnerability to fully describe exposure, sensitivity and adaptive capacity as floods will not impact all households and communities in the same way [8]. We elucidate some examples of vulnerability differentiation that can improve our subnational risk assessment framework and leave their further incorporation to future work. A rural resident in Assam is more likely to be exposed to fluvial floods than their counterpart in Bihar. However, given the large rural population presence in Bihar (∼92 M versus ∼27 M) and its heavy reliance on agriculture (covering 70% of its area, employing 76% of its workforce, and contributing to 42% of its GDP) [24], fluvial floods could pose a greater risk to livelihoods and well-being in Bihar.
In general, rural communities tend to be more vulnerable in the face of environmental hazards due to limited social, economic, and physical resources [57]. Therefore, it may be conceivable that, for example, higher exposure of rural communities to coastal floods in Andhra Pradesh and Odisha are associated with higher risk than their urban counterparts. Nonetheless, despite all benefits of urbanization in reinforcing resilience, it can also aggravate flood risks. For instance, the rapid growth of urban lands in floodplains, especially in Asia [58,59], places more people and infrastructure in harm's way, especially when it is accompanied by unsustainable patterns such as environmental degradation and destruction of wetlands that buffer against floods [60,61].

Policy implications
Policymakers responsible for developing action plans and infrastructural improvements require highly detailed and state-level information on flood risks, such as those presented in this study, to deploy targeted and effective programs. India's flooding situation is dynamic and complex, and flood management practices have only been partially successful, relying on a variety of methods, mainly physical ones such as embankments [62]. Promotion of additional non-structural flood policies, especially in rural areas, by emphasizing early warning systems, livelihood diversification strategies, and community engagement activities to promote collective preparedness and resilience is needed. These policies should also address key factors that differentiate vulnerability such as gender, age, and occupational status [55,63,64].
The growing significance of urban areas as the hub of knowledge, population growth, migration, and economic development is manifested by Goal 11 (sustainable cities and communities) of the United Nation's Sustainable Development Goals (SDGs). India's coastal cities will have to make concerted efforts to achieve sustainability, especially facing climate change and growing environmental hazards. They should invest in infrastructure, urban planning, and early warning systems combined with community-based adaptation strategies rooted in urban equity to accommodate the growing and diverse populations [65].

Limitations
There are limitations to this analysis. First, it is possible that some people were allocated incorrectly if built-up grid cells were missed, which is more likely to occur in rural areas where satellite imagery may miss small settlements. This could result in the over-allocation of population to other grid cells. However, by integrating GHSL and HRSL down to 30 m resolution pixels, we mitigated this issue by detecting built-up grid cells using the two complementary datasets, giving the most complete and accurate dataset on population distribution across urban and rural areas of India. Second, facility databases may not be comprehensive, for example, some rural health posts or schools may be missing from facility listings (or some may be listed that no longer exist). We mitigated this issue by integrating two different and recent facility datasets to allow for the identification and removal of potential duplicates. Third, we rely on satellite-derived and modeled estimates of coastal and fluvial flood risk, which reflects exposure based on the population from the 2011 census; if people were to migrate out to safer areas, and the population distribution is re-structured, then we are assuming an overlap that may not be correct. For example, recent trends of urbanization in India rooted in rural-urban migration over the past decade, population growth and density, and the rapid strengthening of the weight of cities with populations of hundreds of thousands or millions (metropolisation) [66,67], mandates revisiting our exposure assessment, especially in urban areas, once the 2021 census at the settlement level is released. Fourth, although there should not be much misallocation across flood or LECZ categories, this is possible in our data integration and more likely for communities at the margins of these zones. We minimized this issue by using highly granular and fine-level data for both floods and populations. While we took steps to address these limitations, it is not possible to fully avoid the consequent uncertainty.

Conclusions
As the frequency and intensity of flood events increases in India, a comprehensive investigation of the scale and geographic distribution is critical for risk reduction and planning. An essential component of risk assessment is the estimation of population exposure. While some regional and state-level flood assessments exist [55], this study uses different population-oriented, remote sensing, and physical datasets to explore urban and rural population exposures to different levels of coastal and fluvial floods, including estimates of the exposure of key infrastructure communities rely on. Our findings suggest high exposure to both coastal and fluvial flooding. The risk of fluvial hazards is higher for rural (15.8%) versus urban residents (14.2%), however, urban residents are at much higher risk of coastal flooding (10% versus 4.3%). Additionally, we find a total of 15% of health and 18% of educational facilities are exposed to flood hazards, with implications for these key infrastructures. Future research may further disaggregate the total population by key measures of vulnerability, to better understand the distribution of subgroups that are disproportionately at risk (e.g., by age, gender, ethnicity, or sector of employment), and may create additional cut-offs regarding the severity of flood events (e.g., z-scores of flood depth and extent). Our exposure estimation is the most locally sensitive to date, highlighting exposure in both urban and rural landscapes. Our results can inform targeted, state-level planning, early warning systems, and financing in order for policymakers to address flood risks across India with the potential to significantly improve health and economic outcomes.

Data availability statement
Estimates derived from this study and the publicly available datasets that support the findings can be shared upon request, however, the complete 2011 census boundaries and Fathom Global Model, with their references indicated in the text, cannot be made publicly available upon publication because they are proprietary data.