Predicting Groundwater Hydrochemical Facies in Three Dimensions with Random Forest Classification, USA

Area of Science:

• Hydrogeology
• Geochemistry
• Machine Learning

Background:

• Understanding groundwater hydrochemical facies (HCFs) is crucial for managing water resources.
• Existing methods for HCF mapping are often limited in spatial and vertical resolution.

Purpose of the Study:

• To develop and apply a three-dimensional random forest classification (RFC) model to predict groundwater HCFs across the conterminous United States (CONUS).
• To map HCFs at a 1-km² resolution to depths of 400 m below the base of drinking water.

Main Methods:

• Utilized major-ion data from 152,673 sites to define six HCFs (CaMg-HCO₃, NaK-HCO₃, CaMg-SO₄, NaK-SO₄, Cl, Mixed).
• Engineered model features including elevation relative to the base of drinking water (ERDW) and geologic unit flags.
• Employed RFC modeling, identifying ERDW as the most significant feature.

Main Results:

• Successfully mapped HCFs across CONUS at a 1-km² resolution and to significant depths.
• Predicted CaMg-HCO₃ facies near the water table in humid regions and carbonate/crystalline rock areas.
• Observed a rapid transition from HCO₃ to Cl facies at depths below drinking water supplies.

Conclusions:

• The RFC model provides accurate, high-resolution 3D predictions of groundwater HCFs.
• Model predictions align with geochemical expectations and are validated by point data and regional averages.
• The HCF maps can inform groundwater resource management, including salinity and characteristic mapping.