INCORPORATING AIRBORNE ELECTROMAGNETIC SURVEY DATA INTO MACHINE-LEARNING MODELS TO PREDICT GROUNDWATER SALINITY IN THE MISSISSIPPI RIVER VALLEY ALLUVIAL AQUIFER

The Mississippi Alluvial Plain, located in the south-central United States, is experiencing long-term groundwater-level declines; this has raised concerns about future groundwater availability within the surficial Mississippi River Valley alluvial aquifer (MRVA). Areas of high salinity in the MRVA limit where groundwater can be used for irrigation, public supply, and domestic use. A boosted regression trees (BRT) machine-learning model was trained on available specific conductance data to predict spatially continuous surfaces of groundwater salinity at multiple depths across the MRVA. BRT model explanatory variables included well construction information, water levels, and surficial variables such as geomorphology and soils. The BRT model also incorporated subsurface resistivity data from one of the largest regional Airborne Electromagnetic (AEM) surveys completed of a principal aquifer. This work expands on prior BRT salinity predictions of the MRVA, extending predictions to the Gulf of Mexico where salinity is high. AEM data are not yet available over the southern extent of the MRVA. The BRT model was trained with and without AEM variables to test the ability of the BRT model to predict salinity where explanatory data are missing, and response data are sparse. Additionally, model sensitivity to AEM data was evaluated to better understand how AEM variables influence salinity predictions. Groundwater salinity predictions will help inform future AEM surveys of the southern extent of the study area where high groundwater salinity can obfuscate changes in aquifer sediment resistivity.