DEPTH-CLASSIFIED GROUNDWATER ARSENIC MODELING IN THE GANGES DELTA USING MACHINE LEARNING

Arsenic, a class I carcinogen, has been widely detected in the groundwater of the Ganges River Delta, raising significant public health concerns for both India and Bangladesh. Previous local to sub-regional studies have indicated that deep aquifers (greater than 150 meters) were arsenic-free, leading both nations to invest in drilling deep wells despite the higher installation costs and efforts involved. However, recent research has found high arsenic levels in deeper wells, which challenges the notion that deep aquifers are a ubiquitously safe alternative source of drinking water. This study employs the random forest machine learning algorithm to model the depth distribution of arsenic within the delta aquifers by probabilistic mapping of arsenic occurrence above the WHO guideline of 10 μg/L at shallow (< 70 meters below MSL), intermediate (70-150 meters below MSL), and deep (> 150 meters below MSL) aquifer zones, followed by the predictive binary delineation of arsenic-contaminated zones. Our findings indicate that while deeper groundwater generally has lower arsenic concentrations, elevated arsenic levels persist across all depths in the western and northern-most boundary of the delta. In contrast, the southern regions, particularly near the active delta mouth, show significantly lower hazards. Overall, the results from this study identify regions with potential arsenic contamination in the deep aquifers, necessitating local-scale studies to delineate the contamination extent before allocating resources to deep well drilling for arsenic-safe water. To assist policymakers and stakeholders, we propose developing arsenic hazard maps for individual districts and sub-districts, coupled with vulnerable population data to guide effective prevention, mitigation, and remediation efforts.