Assessment and Mapping of Soil Erosion Hotspots in the Himalayan Region Using RUSLE and Machine Learning Technique

Soil erosion is a major driver of land degradation, particularly in fragile and ecologically sensitive mountainous regions such as the Himalayas. This study assessed soil erosion risk and mapped erosion hotspots in the Bhagirathi watershed, Uttarkashi District, Uttarakhand, by integrating the Revised Universal Soil Loss Equation (RUSLE), remote sensing, GIS, field observations, and Random Forest (RF) machine learning. The assessment utilized 80 spatially distributed surface soil samples for estimating soil erodibility, ALOS PALSAR DEM (12.5 m) for terrain characterization and derivation of topographic factors, and Sentinel-2A multispectral data (10 m) for land use/land cover and spectral information. Rainfall, soil, topographic, vegetation, and management-related factors were integrated within the RUSLE framework to estimate soil loss, while terrain and spectral variables were subsequently used to develop the RF regression model. Areas experiencing estimated soil loss exceeding 40 t ha⁻¹ yr⁻¹ were classified as erosion hotspots. The RUSLE-based erosion risk assessment showed good agreement with field observations, achieving an overall validation accuracy of 85%. The RF regression model demonstrated good predictive performance, with an R² of 0.72, RMSE of 0.31, and Nash–Sutcliffe Efficiency of 0.67. Slope, aspect and NDVI were found to be the most important predictors governing the model performance as well as spatial distribution of erosion hotspots. Erosion hotspots occupied approximately 9.34% of the watershed and were predominantly associated with steep slopes and areas with sparse or absent vegetation. These locations were also clearly identifiable in high-resolution satellite imagery. The findings demonstrate that integrating empirical erosion modelling with field observations, geospatial data, and machine learning can effectively assess erosion risk and identify priority hotspots in complex and inaccessible Himalayan terrain. The approach can support targeted soil and water conservation planning and sustainable land management in fragile mountain ecosystems.