Key points are not available for this paper at this time.
Abstract Manganese (Mn) is vital for steel and battery production, requiring an improved understanding of ore textures to optimise extraction. This study investigates the Paling Pan ferruginous manganese (Fe-Mn) deposit in the Postmasburg Manganese Field (PMF), where ores, hosted in the Gamagara Formation, unconformably overlie Campbellrand dolostones. We integrate geology, geochemistry (major oxides), mineralogy, and machine learning (ML) to classify ore textures (massive, vuggy, layered, conglomeratic) and predict manganese oxide (MnO) grades. These textures exhibit distinct geochemical signatures that reflect the depositional environment, diagenesis, and supergene processes. Depth profiles show MnO enrichment in karstic depressions, with braunite and bixbyite correlating strongly with Fe2O3 and SiO2, while partridgeite and hollandite influence Mn distribution. ML analysis (Random Forest, XGBoost, Gradient Boosting) of drill-core geochemical data achieved an F1-score of 0.8682 for texture classification and 0.9983 coefficient of determination (CoD) for MnO prediction. Results demonstrate that MnO distribution is controlled by primary sedimentary processes and secondary alteration. This data-driven approach enhances geochemical modelling, enabling better resource evaluation in the PMF. The study advances understanding of Mn mineralisation processes and establishes an integrated ML-geological framework for exploring sedimentary FeMn deposits, with direct applications for optimising exploration and mining strategies in similar deposits globally.
Buthelezi et al. (2026) studied this question.