Machine Learning as a Tool to Predict Lithium in Brines Webinar

Speaker: Katherine Knierim, Hydrologist, US Geological Survey, Lower Mississippi – Gulf Water Science Center, Little Rock, Arkansas

The chemical composition of high salinity groundwater, or brine, is important to understand for quantifying the availability of both water and critical mineral resources. For example, brines may have high concentrations of dissolved critical minerals, such as lithium (Li), and provide an important economic resource as the world transitions to a greater reliance on Li-ion batteries. The Smackover Formation is part of a regionally important petroleum and brine system in the Gulf Coast region of the southern United States (U.S.) and includes high Li concentrations (> 400 mg/L) in the brines. In this study, the U.S. Geological Survey and the Arkansas Department of Energy and Environment—Office of the State Geologist used published and newly (2022) collected brine Li concentration data to train a random-forest (RF) machine-learning model using geologic, geochemical, and temperature explanatory variables. Predicted brine Li concentrations from the RF model at approximately 1,000 to 3,000 meters depth across the Smackover Formation ranged from 3 to 420 mg/L. Uncertainty in the mapped RF model predictions—based on the 90th percentile prediction interval across the Li map—were used with formation thickness and porosity information to calculate the range of the Li mass in Smackover Formation brines. This study provides an example of using machine learning to predict deep brine chemistry for a critical mineral resource evaluation. For more info see: https://pubs.usgs.gov/publication/fs20243052/full