Machine learning predicted redox conditions in the glacial aquifer system, northern continental United States

Groundwater supplies 50% of drinking water worldwide and 30% in the United States. Geogenic and anthropogenic contaminants can, however, compromise water quality, thus limiting groundwater availability. Reduction/oxidation (redox) processes and redox conditions affect groundwater quality by influencing the mobility and transport of common geogenic and anthropogenic contaminants. In the glacial aquifer system, northern United States (GLAC, 1.87 million km²), groundwater with high arsenic or manganese concentration is associated with reducing conditions and high nitrate with oxidizing conditions. This study uses machine learning to identify the relative influence of drivers of redox conditions (e.g., residence time vs. reactivity) across the glacial landscape. We developed three‐dimensional boosted regression tree models to predict redox conditions using the likelihood of low dissolved oxygen or high iron as indicators of anoxic conditions. Results indicate that variation in redox condition is controlled primarily by residence time (e.g., groundwater age and relative depth) and to a lesser extent by geochemical reactivity (e.g., subsurface contact time, soil carbon). Older water and deeper wells, along with more water storage or slower water movement was associated with higher probability of anoxic conditions. Mapped model results illustrate regions where anoxic redox conditions may mobilize geogenic contaminants or oxic conditions may limit denitrification potential. Results may also provide simplified redox input for process or predictive models of, for example, arsenic, manganese, or nitrate. Machine learning modeling methods can lead to improved understanding of contaminant occurrence and what drives redox conditions, and the methods may be transferable to other settings.