RANDOM FOREST PREDICTIONS OF LITHIUM IN BRINES OF THE SMACKOVER FORMATION

The Smackover Formation is part of a regionally important petroleum and brine system in the Gulf Coast region of the southern United States and contains high dissolved lithium concentrations (> 400 mg/L). Economic demand for lithium is expected to increase as the world transitions away from hydrocarbon energy sources, and southern Arkansas has received increased attention as an area with both high lithium brines and existing infrastructure necessary for commercial extraction. In this study, the U.S. Geological Survey (USGS) and the Arkansas Department of Energy and Environment Office of the State Geologist, trained a random forest (RF) machine-learning model to predict lithium concentration in brines using geologic, geochemical, and temperature explanatory variables. Lithium data used to train the RF model were from the USGS Produced Waters Geochemical Database (v. 3.0) and newly collected brine samples throughout southern Arkansas. Predicted lithium concentrations at depths of approximately 1,000 to 3,000 meters across the Smackover Formation ranged from 13 to 389 mg/L. Uncertainty in the mapped RF predictions—based on the 90^th percentile prediction interval across the lithium map—was used with formation thickness and porosity information to calculate the possible range of lithium mass in Smackover Formation brines. Uncertainty estimates captured both the variability associated with RF predictions and the unknown spatial variability of porosity throughout the Smackover Formation. Based on the RF model, dissolved hydrogen sulfide, depth of the perforated interval from which the brine sample was collected, and altitude of the top of the Smackover Formation were the top three most important predictors of lithium concentrations in brines.