Nitrate contamination in groundwater across Switzerland: Spatial prediction and data-driven assessment of...

Excessive nitrate in groundwater, which is primarily caused by anthropogenic activities, is a worldwide problem. Consequently, Goal 6 of the UN Sustainable Development Goals lists nitrate as one of the key indicators of groundwater quality. However, in most countries, the nationwide occurrence of nitrate is unknown, as the monitoring networks only represent small points in space. To bridge this gap, machine learning modelling that predicts nitrate concentrations at a high spatial resolution is a promising tool to identify high-risk areas. Here, we use random forest machine learning to predict nitrate concentrations across Switzerland based on 1336 monitoring sites. The model revealed that approximately 35 % of the Swiss Plateau, Switzerland's most populous region, has a high probability of exceeding the Swiss guideline value of 25 mg/l for groundwater nitrate. We also investigated the individual importance and influence of anthropogenic and environmental variables associated with high nitrate concentrations by combining SHapley Additive exPlanations with expert knowledge of physical and geochemical processes. In addition to well-known influences of anthropogenic features (e.g. land use), we found that other environmental features including high springtime precipitation, low summertime precipitation, low soil organic carbon content, low river density and greater distance to large rivers, were indicative of high nitrate concentrations. These features directly relate to large-scale nitrate transport and attenuation processes (denitrification and dilution), but have received sparse attention in nitrate risk assessment and mitigation measures. Therefore, the approach and results of our study can be useful for nitrate studies around the world.