Machine learning uncovers provenance of source rocks for volcano-sedimentary lithium mineralization in South China

Abstract

The Early to Middle Triassic sedimentary units in South China, belonging to the so-called “Green bean rock” (hereafter “GBR”), host significant volumes of potentially economic clay-type volcano-sedimentary lithium (Li) mineralization. However, the source material and the processes that led to the enrichment of Li in these clay deposits remain unclear. This is especially true of the uncertain provenance of the igneous material that eventually forms this Li mineralization. In this study we apply machine learning to geochemical data from igneous rocks and GBR samples to determine the nature of the source rock, the type and source of the magma associated with the GBR, and the initial Li contents of these protoliths. The results of this Random Forest (RF) modeling indicate that the GBR protolith was entirely derived from a dacitic magma, whereas the petrology of these samples indicate that the source magma for the GBR protolith was derived from an intermediate to acidic dacite-rhyolite magma. The RF modeling suggests the protolith volcanic ash was primarily derived from the Sanjiang Orogenic Belt and the Shiwandashan Belt in South China. The location and distribution of the GBR relative to the Sanjiang Orogenic Belt and the Shiwandashan Belt indicates that the GBR has a significant directionality with preferential NE-SW and N-S orientations, indicating the likely influence of paleomonsoon conditions during GBR formation. The Li content of the GBR protolith is <50 ppm, with 68.75% of the data generated during this study having a Li concentration of <20 ppm, indicating that the Li within the GBR was primarily derived from water–rock interactions during the deposition period. This study provides new insights into the process involved in the formation of the GBR and the associated Li enrichments in this region as well as outlining the value in integrating machine learning models with big data in mineral deposit research.