Metallogenic prediction based on multi-source remote sensing and machine learning: A case of lithium ore in Jiajika, China

• High-precision mineralization prediction using multi-source remote sensing and spectral data. • Integrated extraction of alteration using SAM and PCA techniques. • SHAP-based interpretation of mineralization factor importance. • ML and DL applied to lithium prediction; MLP shows optimal performance. The Jiajika Mine, once the largest lithium deposit in Asia, is targeted for further exploration to identify additional mineral resources in the region. In this study, we developed a comprehensive metallogenic prediction framework that integrates multi-source remote sensing data—including GF-5, ASTER, Landsat-8, and spectral data of terrestrial objects—with machine learning models: Random Forest (RF), Support Vector Machine (SVM), and Multi-layer Perceptron (MLP). The main results are as follows: (1) By combining the results of all three models, we identified 14 favorable mineralization zones, including three known ore sites. (2) Among the three models, the MLP achieved the highest performance, with an Area Under the Curve (AUC) value of 0.788, while RF had the lowest at 0.742. (3) Iron anomalies, structural density, and tourmaline alteration were key factors influencing metallogenic prediction. These were followed by alteration features related to spodumene mineralization, muscovite alteration, silicification, andalusite and beryl-associated spectral signatures. In contrast, hydroxyl anomalies and cordierite alteration had the least impact on the identification of favorable metallogenic zones. This method demonstrates the effectiveness of integrating the multi-source remote sensing data with machine learning techniques in mineral exploration and provides valuable insights for metallogenic prediction in regions with complex geological structures.