Data-driven classification of landforms and subsurface structural mapping using ASTER DEM and random forest in the Musonoi copper-cobalt District, Democratic Republic of Congo

Accurate characterisation of landforms and subsurface geological structures is critical for mineral exploration and land-use planning in tectonically complex, data-scarce regions. The Musonoi Copper-Cobalt District (MCC), Lualaba Province, Democratic Republic of Congo (DRC), presents such a challenge given its intricate Neoproterozoic stratigraphy and limited geophysical coverage. Using freely available ASTER GDEM V003 data (30 m resolution) and a digitised 1:100,000 geological map, this study applies a Random Forest (RF) machine learning framework to classify surface landform types and predict subsurface geological structures in the MCC District. Ten terrain and geological predictor variables were used to train three RF models: a landform classification model (Model 1: AUC = 0.707; accuracy = 76%), a subsurface structure prediction model (Model 2: AUC = 0.741; accuracy = 81.5%), and a unified multivariate model combining both variable sets (Unified Model: AUC = 0.7819; accuracy = 78.75%). The results reveal strong correlations between geomorphological features and subsurface structures, showing that RF is a reliable tool for understanding surface–subsurface interactions. These findings demonstrate the utility of ASTER-DEM-based RF modelling as a cost-effective reconnaissance tool for geological characterisation.