42 Geophysics/Geodynamics (Posters)

Added due to amt of discipline abstracts submitted.

Wednesday, 11 March 2026
9:00 AM - 12:00 PM
RCC, Lower Level Hall

Machine Learning Driven Prospectivity Analysis of Li-bearing Pegmatites at Ewoyaa, Southern Ghana, using Multisource Geoscience Data

Paper number: 42-0
Program: Technical Sessions
Time: 9:00 AM – 9:00 AM

Authors

Abstract

The accelerating demand for lithium globally requires the development of exploration approaches that integrate heterogeneous geoscientific datasets in complex geological settings. Ghana’s Ewoyaa lithium deposits occur within the Paleoproterozoic terranes of the Birimian Supergroup, where dense vegetation, structurally controlled mineralization, and limited deposit surface exposure reduce the effectiveness of the traditional mapping techniques.

This study integrates remote sensing, airborne geophysics, and geochemical datasets to develop a machine learning-driven prospective mapping framework that models the likelihood of Li-bearing pegmatites occurrence. High-resolution Landsat-9 and WorldView-3, along with drone imagery at 10 cm pixel resolution, were used for spectral analysis to evaluate the effectiveness in mapping the spodumene-bearing pegmatites and associated alterations. The airborne magnetic and radiometric datasets were analyzed using upward continuation, derivative filtering, and Euler deconvolution, which helped interpret the geology and geological structural patterns contributing to the potential mineralized zones. These datasets were complemented by multi-element geochemical concentrations obtained from soil, rock, and drill hole samples determined using Laser-Induced Breakdown Spectroscopy (LIBS) and portable X-ray fluorescence (pXRF), lithium, and other pathfinder elements.

All datasets were transformed into a uniform feature space suitable for supervised machine learning, with strategies such as SMOTE-Tomek, StandardScaler, and Random Forest-based SelectFromModel implemented to address class imbalance, feature scaling, and dimensionality reduction. Multiple algorithms, including Random Forest, Support Vector Machine, Extreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM), were evaluated with stratified cross-validation and randomized hyperparameter optimization, and ensemble stacking was employed to enhance the robustness. Feature significance and SHAP analyses were used to incorporate model interpretability in order to evaluate the relative impact of geochemical, structural, and geophysical predictors on patterns of lithium mineralization.

From the preliminary model evaluations, important structural and geochemical controls on lithium pegmatite mineralization are captured by tree-based ensemble approaches. The study proposes ML as its central analytical framework for Li-bearing pegmatites mapping, while providing a transferable, data-driven workflow applicable to other structurally complex and deeply weathered critical mineral provinces.

Geological Society of America Abstracts with Programs. Vol. 58, No. 1

doi: 10.1130/abs/2026SE-411553

© Copyright 2026 The Geological Society of America (GSA), all rights reserved. Permission is hereby granted to the author(s) of this abstract to reproduce and distribute it freely, for noncommercial purposes. Permission is hereby granted to any individual scientist to download a single copy of this electronic file and reproduce up to 20 paper copies for noncommercial purposes advancing science and education, including classroom use, providing all reproductions include the complete content shown here, including the author information. All other forms of reproduction and/or transmittal are prohibited without written permission from GSA Copyright Permissions.