Using Machine Learning to Characterize the Genesis of Lithium-Bearing Sedimentary Rocks in the Western United States

Abstract

The lithium used in batteries is currently sourced exclusively from high-altitude saline brines and pegmatites in orogenic metamorphic belts. In the coming years, lithium-bearing volcano-sedimentary deposits are expected to become an increasingly important source of lithium, with several volcano-sedimentary deposits under varying stages of development throughout the Basin and Range province (United States and Mexico) and Serbia. Foremost among these is the Thacker Pass project hosted in lacustrine moat sediments of the ~16.3 Ma McDermitt caldera, Nevada, United States. Because this type of deposit is critical to obtaining a secure domestic supply of lithium in the United States, we initiated a study on more than 50 basins in 21 states. Through observational field work, detailed sampling, and geochemical analyses on the stratigraphic sequences of interest, we created a database of over 1,500 samples characterizing sedimentary basins in the western United States to ascertain the occurrences and formation of volcano-sedimentary lithium deposits. We use multinomial logistic regression and principal component analysis in addition to basic statistical analysis to create a predictive model for lithium concentration utilizing key features of each basin. Basins containing economic (>1,000 ppm) lithium are predominantly characterized by Miocene-aged, small (<1,500 km2) closed lacustrine systems, with volcanic input, containing fine-grained clay material, often deposited under reducing conditions. The model can be used to predict the concentration of lithium in sedimentary databases without lithium data and can be applied as a modern tool for volcano-sedimentary lithium deposit exploration in basins globally.