IISER MOHALI

Abstract: Lithium exploration in Himalayan–Tibetan leucogranites is complicated by heterogeneous enrichment and the limited availability of measured Li concentrations in regional geochemical datasets. This talk presents an integrated approach combining compositional data analysis and explainable machine learning to identify geochemical patterns associated with Li fertility. Using a curated dataset of whole-rock analyses, the study evaluates the ability of multivariate geochemical signatures to distinguish Li-enriched from Li-poor leucogranites, with particular emphasis on validation across independent publications and external validation using samples from an adjacent region. Explainable machine learning highlights the relative contributions of trace elements and elemental ratios to model predictions, helping identify geochemical associations linked to Li enrichment. The resulting HimLithium screening tool (https://himlithium.streamlit.app/) translates these insights into a practical framework for prioritising samples and localities for further investigation. The approach does not replace direct Li measurement or establish causal controls; rather, it demonstrates how routinely available geochemical data can support regional-scale screening and improve our understanding of lithium fertility in evolved granitic systems.