Predictability of redox potential and oxygen status in managed aquifer recharge sites based on sensor data and regression techniques

Managed Aquifer Recharge (MAR) systems are increasingly utilized for rainwater harvesting, water treatment, and groundwater remediation. Studies focus on selecting effective MAR sites by evaluating the quantity and quality of recharged water, using soil maps and topographical data to assess physical characteristics and infiltration capacity. However, determining the impact of a specific MAR site on water quality is complex and requires extensive sampling across various times and locations. This study addresses these challenges using large datasets of physicochemical variables from soil sensors from four large-scale MAR experiments in California’s Central Valley. The four MAR sites were flooded for two to four weeks with about 1.32 m³ m⁻² of water. Before, during, and after the flooding, soil redox potential (E_h), volumetric water content (θ), soil temperature (ST), and gaseous oxygen (O₂) were measured continuously in the subsurface at various depths and locations. E_h and O₂ show a decline after wetting events and an increase once flooding ends and the dry cycle starts. The dynamics of E_h and O₂ under MAR were described using an exponential model. In this model, the constants for increase or decrease, defined as temporal coefficients, were determined. Soil dynamics were categorized based on clay content, distinguishing between soils with less than 5 % clay and those with more than 5 % clay. Regression models, multi linear regression, and Support Vector Machine Regression were employed to predict the temporal coefficients of E_h and O₂ using readily measurable variables, such as soil texture, changes in θ, and the initial concentrations of nitrate and dissolved organic carbon. These field-validated models are essential for predicting the development of anoxic conditions and can be used to identify optimal temporal criteria for maintaining water quality during MAR operations.