Tree-based machine learning algorithms for soil erosion vulnerability (SEV) prediction in Saddang Watershed, south Sulawesi, Indonesia

ABSTRACT Soil erosion represents a significant environmental threat, particularly in watersheds with complex geomorphological characteristics. This study evaluates soil erosion vulnerability (SEV) in the Saddang Watershed, Indonesia, using tree-based machine learning (ML) models: decision tree (DT) and random forest (RF). The research incorporates key hydrological, topographical, and environmental factors, such as drainage density, rainfall erosivity, soil bulk density, and vegetation index, which influence soil erosion dynamics. A dataset of 2,000 locations, validated through 20-fold cross-validation, was utilized for model training and evaluation. Results indicate that the RF model demonstrated superior predictive accuracy (AUCROC: 0.953) compared with the DT model (AUCROC: 0.917), attributed to RF's ensemble methodology that enhances robustness against overfitting. Spatial analysis classified SEV into five categories, highlighting moderate to very high vulnerability near riverbanks and steep terrains. Drainage density and river proximity emerged as the most influential factors, underscoring the necessity for targeted conservation measures in these areas. The integration of ML models with GIS and remote sensing provides a robust framework for real-time SEV assessment, aiding sustainable land-use planning. This approach offers valuable insights for policymakers and practitioners, enabling evidence-based interventions to mitigate soil erosion and enhance environmental resilience in watersheds worldwide.