| dc.description.abstract | Groundwater is a critical resource supporting millions of domestic and municipal users across the country. As pressures on this resource intensify, concerns about groundwater quality are only increasing. Large-scale monitoring efforts generate extensive datasets that capture spatial and temporal trends in groundwater chemistry, offering an opportunity to better understand and manage water quality challenges. Machine learning techniques, particularly decision tree models like random forests (RF), are well-suited to detect patterns in complex, high-dimensional environmental data. When paired with interpretability tools such as SHapley Additive exPlanations (SHAP), these models can insights into underlying geochemical processes. This approach enables the identification of potential risk zones and supports informed decision-making without the need for extensive, costly sampling. Our analysis demonstrates that machine learning models can accurately predict the presence of certain naturally occurring contaminants and highlight geochemical conditions—such as redox dynamics and groundwater mixing—that may influence their mobility. The integration of machine learning and geochemical interpretation holds promise for improving groundwater protection efforts, particularly for communities reliant on private wells and for regions where monitoring is limited. More broadly, this framework can be applied to aquifer systems worldwide, advancing both scientific understanding and management of groundwater quality. | en_US |
