Integrating Revised Universal Soil Loss Equation (RUSLE) Modeling and Random Forest-Based Feature Importance Ranking to Assess Soil Erosion Risk and Spatial Variability of Pb in a Tropical Post-Tin Mining Landscape

Open-pit tin mining transforms tropical landscapes into heterogeneous Technosols that are vulnerable to soil degradation and heavy-metal redistribution. This study integrated the Revised Universal Soil Loss Equation (RUSLE) and Random Forest-Based Feature Importance Ranking (RF) to assess spatial erosion patterns and identify key environmental predictors of Pb concentration in the Diniang Watershed, Bangka Island, Indonesia. Soil erosion was estimated using spatial layers of rainfall erosivity, soil erodibility, topography, vegetation cover, and support practices. Five predictors (erosion rate, soil pH, soil organic carbon (Corg), and distance to roads and rivers) were integrated with spatial Pb concentration data and analyzed using a Random Forest Regressor. The results revealed pronounced spatial heterogeneity in tropical technosol properties and Pb concentration. RUSLE indicated that 92.32% of the watershed experienced low potential soil loss (<15 t ha⁻¹ yr⁻¹), while localized areas were classified as high to very severe erosion, forming potential erosion hotspots. Random Forest ranking identified soil pH and Corg as the dominant predictors of Pb concentration, with substantially higher importance than erosion rate, distance to roads, and distance to rivers. These findings suggest that Pb concentration in the study landscape is more strongly associated with soil chemical properties than with modeled erosion under the conditions examined. The integrated RUSLE-RF framework provides a spatially explicit diagnostic approach for identifying soil erosion risk and potential heavy-metal hotspots and can support targeted restoration and monitoring of tropical post-tin mining landscapes. Field-based validation and denser sampling are recommended to strengthen model reliability.