This study innovatively constructs a “Nature-Engineering” dual-driven dynamic prediction system for soil and water loss, pioneering the coupling of natural recovery processes with engineering disturbance effects. Integrating remote sensing monitoring, GIS spatial analysis, mathematical modeling, and field survey techniques, it overcomes the limitations of traditional single-factor erosion research. Taking the Jidayuan Railway as a case study, quantitative results reveal: during the natural recovery period, soil erosion in semi-arid abandoned soil (slag) fields (31,010 tons) significantly exceeds that in semi-humid areas (1,713 tons); during construction, bridge engineering zones (11,713 tons) and construction living areas (10,866 tons) in semi-arid regions emerge as erosion hotspots; newly increased erosion is particularly severe in arid zones (e.g., 8,552 tons in bridge zones), while semi-humid areas exhibit only 6.5% of this magnitude (555 tons). GIS spatial analysis indicates that 73.45% of the 沿线 area experiences slight erosion, but engineering disturbances markedly intensify erosion in high-risk zones, with abandoned soil fields and bridge areas precisely identified as sensitive units. This system provides a scientific basis for dynamic coupled prediction of soil erosion control in ecologically vulnerable railway projects.