Source matters: spatial modelling of drinking water sources and vulnerabilities in LMICs

Safe and reliable drinking water underpins health, livelihoods, and water security. The way in which water is accessed is an integral component of human-water systems, influencing multiple aspects of water security such as reliability, contamination risk, accessibility, and cost, and influencing hydrological dynamics through patterns of abstraction. Understanding the diversity of water sources is particularly important in low and middle income countries (LMICs), where many households obtain water through direct abstraction that is often unregulated and forms an “unseen” demand within hydrological systems.

To enable global progress tracking, international monitoring has largely classified drinking water sources into two broad groups: “improved” and “unimproved.” While useful for benchmarking access and supporting cross-country comparison, this binary categorisation conceals major functional differences in how and where water is accessed, and obscures distinct vulnerabilities that influence both human wellbeing and hydrological systems. For example, piped networks, boreholes, tanker delivery, and rainwater collection exhibit fundamentally different abstraction patterns and vulnerabilities yet are grouped as “improved,” limiting functional understanding of complex human-water interactions. Despite the importance of drinking water access as the most foundational interface between people and hydrological systems, spatially explicit information on water source types remains limited.

Using geolocated household survey data from Demographic and Health Surveys (DHS) for 53 countries (2010–2024), we produce high resolution (5km) maps of nine functionally meaningful water source groups for LMICs. A multivariate random forest model with spatial cross-validation and global biophysical and socioeconomic covariates is then used to generate 5 km gridded probability estimates of water source use. Model performance is evaluated using hold-out validation and national-level comparisons to Joint Monitoring Programme (JMP) estimates to support validation in countries lacking geolocated data.

Our results reveal large subnational variability in drinking water source types that is masked by improved/unimproved metrics. Notably, while the improved/unimproved dichotomy broadly reflects microbial contamination risk, it fails to capture vulnerabilities related to other dimensions of water security such as reliability, accessibility, and affordability. Maps also highlight hotspots of direct groundwater and surface water abstraction, illustrating where household-level abstraction is particularly important for understanding hydrological systems. By moving beyond the improved/unimproved dichotomy, this work provides new evidence to support hydrological modelling, exposure and vulnerability assessment, and strengthens the basis for integrating human water use into water security assessments.