A geospatial machine learning prediction of arsenic distribution in the groundwater of Murshidabad district, West Bengal, India: spatio-temporal pattern and human health risk

Arsenic (As) contamination of groundwater in parts of South and Southeast Asia is a public health disaster. Millions of people living in these regions could be chronically exposed to drinking water with As concentrations above the World Health Organizations provisional guideline of 10 {micro}g/L. Recent field investigations have shown that the distribution of groundwater As in many shallow aquifers in India and Bangladesh is evolving rapidly due to massive irrigation pumping. This study compares a decade-old dataset of As concentration measurements in groundwater with a dataset of recent measurements using geospatial machine learning techniques. We observed that the probability of As concentrations >10 {micro}g/L was much greater in the regions between two major rivers than in the regions close to the Ganges River on the eastern border of the study area, where As concentrations >10 {micro}g/L had been measured prior to 2005. The greater likelihood that As is present away from the river channel and is found instead in the interfluvial regions could be attributed to the transport and flushing of aquifer As due to intense groundwater pumping for agriculture. We estimated that about 2.8 million people could be chronically exposed to As concentrations >10 g/L. This high population-level exposure to elevated As concentrations could be reduced through targeted well-testing campaigns, promoting well-switching, provisions for safe water access, and developing plans for raising public awareness. Policymakers could use the ternary hazard map to target high-risk localities for priority house connections of piped water supply schemes to help reduce human suffering. Key pointsO_LIA high-resolution predictive analysis was conducted using geospatial machine learning techniques to identify human suffering. C_LIO_LIA comparison of decadal arsenic measurements and a machine learning prediction suggests a shift in hotspot location. C_LIO_LIGroundwater in a region between two major rivers was found to be unsafe for agricultural and drinking purposes. C_LI Plain language summaryWe conducted a high-resolution predictive analysis using geospatial machine learning algorithms to identify the extent and hotspot location of arsenic (As) contamination in the Murshidabad district of West Bengal, India. The predictive analysis identified an area between two adjacent major rivers in which the probability of As concentrations >10 g/L in groundwater is significantly greater than in other areas. There is a shift in As hotspot location from the regions near the river toward the regions between the two adjacent rivers, possibly due to intense groundwater pumping for agriculture. We estimated that about 1.6 million people could be at high-risk from drinking water contaminated by As concentrations >10 g/L. Policymakers could use the hazard map and the analysis of treated piped drinking water networks to provide access to targeted safe water wells for affected households.