Controls and predictions of geogenic redox-sensitive contaminants in Danish groundwater

Groundwater is a crucial drinking water source, and maintaining its quality becomes increasingly important as global demand rises. This study examines the spatial distribution and controlling factors of five geogenic groundwater contaminants—arsenic (As), manganese (Mn), iron (Fe), ammonium (NH₄), and total phosphorus (P)—across Denmark using classification machine learning models. We trained the models using data from over 7100 well intakes and 34 spatial covariates related to soil, geology, groundwater level, and recharge. The models achieved area under the curve scores ranging from 0.85 to 0.90. Predicted areas exceeding guideline thresholds for drinking water (As, Mn, Fe, NH₄) or ecological ecosystem status (P) cover 14% (As), 78% (Mn), 74% (Fe), 80% (NH₄), and 49% (P) of Denmark's area; more than 60% are classified with high confidence, with uncertainty varying by monitoring density, among other factors. The 2D spatial predictions integrate multi-depth observations and reflect geological and hydrogeological controls. Elevated As levels were mainly associated with glacial meltwater sand and clay aquifers overlying pre-Quaternary marine clays. Feature importance analysis confirms that complex Quaternary layering, confining clay units, and redox-sensitive conditions—particularly in the presence of reactive Fe/Mn oxides and organic matter—drive As mobilization. All five contaminants tend to co-occur spatially under reducing conditions, indicating shared redox-driven release mechanisms. This study shows how national-scale, high-resolution datasets combined with machine learning can predict groundwater quality, aiding well siting and water treatment decisions, and provides transferable insights for regions with similar hydrogeological conditions.