Abstract
Study region: This study investigates two typical semi-arid loess watersheds in China, the Yiluo River Basin and Upper Wei River Basin, where terrace renovation, vegetation restoration, and urbanization have substantially altered watershed hydrological processes. Study focus: Conventional SWAT-based land-use studies often apply extreme land-use scenarios and oversimplify landscape heterogeneity. Terraces are treated as static BMPs with fixed Curve Number (CN), ignoring rainfall-driven CN variations and lowering simulation accuracy in terrace-dominated loess regions. This study uses realistic spatial land-use configurations instead of extreme scenarios and improves SWAT via slope-based terrace delineation and rainfall-intensity-dependent dynamic-CN calibration to evaluate land-use hydrological impacts and the conditional applicability of dynamic CN. New hydrological insights for the region: The 2000 land-use dataset yields the best model performance (NSE = 0.8837, R² = 0.8865, PBIAS = −7.83%), superior to 2010 and 2020 datasets. Grassland expansion reduces runoff by 12.97%, much higher than the 2.49% from forest expansion, and urban expansion produces unstable interannual runoff responses. The dynamic-CN method improves daily-runoff simulation under drought and heavy-rainfall conditions, whereas static CN works better under mild stable rainfall. Terraces redistribute surface runoff into subsurface flow and baseflow with little impact on annual total runoff.