Lithium observations, machine-learning predictions, and mass estimates from the Smackover Formation brines in southern Arkansas

Global demand for lithium, the primary component of lithium-ion batteries, greatly exceeds known supplies and this imbalance is expected to increase as the world transitions away from fossil fuel energy sources. The goal of this work was to calculate the total lithium mass in brines of the Reynolds oolite unit of the Smackover Formation in southern Arkansas using predicted lithium concentrations from a machine-learning model. This research was completed collaboratively between the U.S. Geological Survey and the Arkansas Department of Energy and Environment—Office of the State Geologist.  

The Smackover Formation is a laterally extensive petroleum and brine system in the Gulf Coast region that includes locally high concentrations of bromide and lithium in southern Arkansas. This data release contains input files, Python scripts, and an R script used to prepare input files, create a random forest (RF) machine-learning model to predict lithium concentrations, and compute uncertainty in brines of the Reynolds oolite unit of the Smackover Formation in southern Arkansas. This data release also contains a Python script to calculate the total mass of lithium in brines of the Reynolds oolite unit of the Smackover Formation in southern Arkansas based on porosity. Knowledge of data-science and Python and R programming languages is a prerequisite for executing the workflow associated with this product. Users can execute the scripts to prepare input data, train a RF machine-learning model, compute uncertainty, and calculate lithium mass. Explanatory variables used to train the RF model included geologic, geochemical, and temperature data from either published datasets or created and documented in this data release and the associated companion publication (Knierim and others, 2024). See the associated metadata for details.

This data release also includes output files (csvs [comma-delimited, plain-text] and rasters [geospatial grids]) of lithium concentration predictions from the RF model, uncertainty ranges, and lithium mass. The depth of prediction of lithium concentration represents the mid-point depth of the Reynolds oolite unit which varies between approximately 3,500 and 11,300 feet deep (below land-surface datum) and 0 and 400 feet thick across the model domain. For a full explanation of methods and results, see the companion manuscript Knierim and others (2024).