Coal mining in arid regions has profoundly altered landscapes, exacerbating soil erosion and threatening fragile ecosystems. Accurate quantification of spatiotemporal erosion patterns and identification of dominant contributing factors are critical for sustainable land management. This study employed the Revised Universal Soil Loss Equation (RUSLE) integrated with remote sensing data to evaluate soil erosion in the Shendong mining area, one of China's largest coal production bases, from 1990 to 2023. RUSLE estimates were validated against sediment yields from check-dam monitoring, achieving R2 = 0.65 and NSE = 0.61, indicating satisfactory model performance. Erosion rates were high before 2012 but declined thereafter, with 80.39% of the area exhibiting reduced soil loss. Medium-intensity mining zones showed the highest erosion. Geographical detector analysis indicated precipitation (q = 0.35) and NDVI (q = 0.27) as dominant natural factors. Furthermore, strong interaction effects-particularly between precipitation and NDVI (q = 0.59), and between mining intensity and either precipitation (q = 0.57) or NDVI (q = 0.55)-highlighted that human disturbance significantly amplified the vulnerability of the study area to natural erosive forces. These findings offered a useful reference for designing and implementing spatially targeted ecological restoration strategy in the coal mining areas of arid regions.