Geologically constrained data-driven modeling for mineral prospectivity mapping

Abstract The discovery and development of mineral resources are critical for sustaining modern energy demands. However, the geological processes that control mineralization are inherently complex, introducing considerable spatial variability that presents significant challenges for predictive modeling. While machine learning approaches have been increasingly applied to mineral prospectivity, many fail to explicitly incorporate key geological constraints, limiting their capacity to resolve the nonlinear and directionally dependent nature of mineralizing systems. Here we present a geologically constrained data-driven method that explicitly accounts for the spatial non-stationarity and anisotropy in ore-forming processes. In the benchmark case study from Canada, our method demonstrates a 7.4% improvement in recall performance compared with existing models. This robust performance is also observed in applications to the southern Cordillera region. Furthermore, the method elucidates regional ore-forming controls and quantifies spatial anisotropy in porphyry copper systems. Our findings demonstrate that incorporating geological constraints into data-driven models enhances both the accuracy and interpretability of mineral prospectivity assessments, offering a robust path forward in resource exploration.