Tracing copper mineralisation anomalies using multi-element geochemistry and machine learning in the Okiep Copper District, South Africa

Abstract

Copper (Cu) remains a critical component of the global energy transition, underpinning technologies from renewable power infrastructure to electric vehicles. As demand continues to increase, the discovery of new Cu deposits becomes increasingly urgent. However, greenfield exploration is becoming riskier and costlier, prompting a renewed focus on brownfield regions, especially those with underutilised historical datasets and proven mineral systems. The Okiep Copper District, located in the Northern Cape Province of South Africa, hosts one of the world’s oldest known Cu districts. It represents a geologically complex, multistage mineralised system within the Mesoproterozoic anorthosite–charnockite–dominated Koperberg Suite. Despite extensive historical mining, the region remains underexplored using modern tools, offering a compelling opportunity to revisit its prospectivity through data-driven approaches. This study aims to refine exploration targeting in the Okiep District by integrating machine learning (ML) based anomaly detection with multivariate geochemical interpretations. In this study, we applied three unsupervised outlier detection algorithms, namely (a) Isolation Forest (IF), (b) Local Outlier Factor (LOF), and (c) Angle-Based Outlier Detection (ABOD), to regional stream sediment geochemical data. Each algorithm provides details of potential Cu hotspots for target exploration, i.e., IF and LOF showed concentration of geochemical anomalies closer to structural boundaries, while ABOD provided spatial zonation patterns. Key findings include (a) a significant spatial correlation (22–30% within 5 km) between Cu occurrences and major faults, particularly the Skelmfontein Thrust Zone; (b) a diagnostic Cu-Fe₂O₃T-Zn association consistent with intermediate sulfidation conditions; and (c) the identification of novel high-temperature pathfinder elements (Sb, U, Th, Co, Y), suggesting a metasomatic overprint potentially linked to hybrid crustal-mantle fluids. Based on these findings, a mineral prospectivity model is proposed for Okiep-type deposits, integrating structural architecture (e.g., fault intersections and dyke margins), geochemical vectors, and metallogenic analogues. This holistic framework underscores the value of coupling domain knowledge with machine learning for copper prospectivity mapping in structurally complex terranes. The results have direct implications for brownfield exploration strategy and highlight the need to incorporate multi-variate/source data.