Predicting Groundwater Arsenic Contamination in the Central Valley, California

Access to information on the presence and severity of groundwater contamination is necessary for the safe use of groundwater. Geogenic contaminants such as arsenic, hexavalent chromium, and uranium, are particularly onerous because of their impacts on human and ecosystem health. However, the complexity of the biogeochemistry that governs the fate and transport of geogenic contaminants makes controlling their mobilization especially problematic. Implementing reactive management solutions after problems have exceeded regulatory thresholds has proven to be costly both in human, environmental, and economic terms. Therefore, greater importance should be placed on proactive management strategies to prevent adverse effects on groundwater quality. To build these strategies, however, requires an understanding of the factors that influence ultimate contaminant concentrations. Using publicly available data, we have created models to predict the probability of arsenic exceeding its MCL value (10 ug/L) and threat maps that correspond with those exceedances across the Central Valley of California. The primary modeling tools used were the machine learning techniques of Random Forest Regression (RFR) and Random Forest Classification (RFC). Arsenic concentrations were organized into two datasets according to well depth. These were used to create shallow and deep models fitted with 12 and 13 predictor variables, respectively. Models were also built to give managers preemptive capabilities by understanding probabilities of exceedance of half the MCL value (5 ug/L), which can help determine at-risk locations, which may then mitigate arsenic concentrations before they reach the MCL threshold. The importance of variables in the aquifer system varied by depth. For example, pH was more important in the shallow model for predicting arsenic, while dissolved oxygen was more important in the deep model. Further refinement of predictor variables and improved model sensitivity may prove useful in developing management side actions that prevent or mitigate arsenic contamination in the Central Valley.