Shallow groundwater in the Bengal Basin aquifer system of Bangladesh and India is severely contaminated with dissolved arsenic, leading to potentially devastating health impacts for the estimated 80 million people drinking shallow groundwater in the arsenic-affected region. A potential solution is installation of deeper wells to produce water from depths that are currently arsenic-free, but pumping may induce downward migration of arsenic, contaminating the entire aquifer above the wells. Evaluation of management alternatives through large-scale hydrogeologic analysis and numerical groundwater modeling is necessary to ensure a sustainable low-arsenic water supply, though reliable hydrogeologic data is not widely available.

Possible model structures and regional parameters are estimated using three independent data sets and approaches. These include: driller logs, measurements of hydraulic head, and isotopic composition (interpreted as groundwater age). The result is a simply-structured regional groundwater model that is applied to evaluate possible patterns of groundwater flow, and to elucidate the controls on spatial scale of flowpaths for both predevelopment and developed conditions.

Capture zones for alternative pumping patterns are evaluated, focusing on locating recharge areas for domestic wells. Three pumping configurations are considered (all shallow wells, all deep wells, and shallow irrigation wells with deep domestic wells) and tested for sensitivity to model features and parameters. The shallow configuration only provides high-arsenic water. The deep configuration draws high-arsenic water downwards, eventually to the wells. The third split configuration is most advantageous, providing either arsenic-free source areas or long transport times of high-arsenic water to domestic wells in more than 90% of the arsenic-impacted area. Though data is sparse, this approach based on analyses of available data, establishment of a simply-structured regional numerical groundwater model, and analysis of the sensitivity of model response to relevant features and parameters produces robust results that have important implications for water management policy.