Supporting documentation for predicting groundwater hydrochemical facies in three dimensions across the conterminous United States with random forest classification

This data release provides workflow documentation and spatial predictions of groundwater hydrochemical facies (HCFs) across the conterminous United States at multiple depths. Predictions were generated using a Random Forest Classification model trained on a comprehensive groundwater chemistry dataset (>150,000 sites). Samples were classified based on major ion composition into one of six hydrochemical facies—Calcium-Magnesium Bicarbonate (CaMg-HCO₃), Calcium-Magnesium Sulfate (CaMg-SO₄), Chloride (Cl), Mixed, Sodium-Potassium Bicarbonate (NaK-HCO₃), and Sodium-Potassium Sulfate (NaK-SO₄). Sixteen features were selected or engineered to represent processes and(or) conditions that might affect the ionic composition of groundwater.
Raster outputs at 1 km2 resolution include (1) categorical surfaces showing the most probable facies per pixel and (2) probability surfaces providing classification confidence. Predictions are available for multiple depths including the water table, the estimated bottom of drinking supplies, and 200 meters and 400 meters below the estimated bottom of drinking supplies. All rasters are georeferenced with consistent resolution and extent, enabling integration with other hydrogeologic and geospatial datasets. The workflow documentation details model training, validation, and prediction steps to ensure reproducibility and transparency.
These results support the management of water resources by identifying spatial water-quality patterns, potential salinity sources, and baseflow compositions in unsampled areas. Model predictions could be used in subsequent modeling efforts for the mapping of salinity or other groundwater characteristics.