Predicting Arsenic in Drinking Water Wells in Glacial Aquifer in Western and Central Minnesota, USA

Approximately 40% of available arsenic data for groundwater in western and central regions of Minnesota exceed the 10 μg/L drinking water standard. However, arsenic concentrations vary considerably over short distances and regionally across the state. A boosted regression tree (BRT) model was developed to predict the probability of arsenic occurring above the drinking water standard in groundwater at typical depths used for drinking water supply in glacial aquifers in western and central Minnesota. The BRT model, using about 75 predictive factors such as well construction characteristics, glacial material characteristics, and surficial characteristics (such as soil texture, soil chemistry, or land use), predicted probabilities of elevated arsenic in well water with about 65% total accuracy. Predictive factors determined to be influential for predicted probabilities included clay gap (distance from top of screen to overlying confining unit), nearest major river (a proxy for hydrological position in the landscape), horizontal hydraulic conductivity, and distance to the top of the bedrock from the bottom of the well. For example, smaller clay gaps were typically related to higher probability of elevated arsenic concentrations. The BRT model results were then used to generate maps illustrating elevated arsenic probabilities at the depth of a typical domestic drinking water well across the modeled regions. This a first application of BRT to model elevated arsenic probabilities in a glacial aquifer system.