This study develops a semi-distributed daily runoff simulation framework for the Yuetan Basin by integrating river-network correction, watershed discretization, and hydrological routing within a three-source Xinanjiang modeling system. First, the actual river system was identified from remote sensing imagery and compared with the river network extracted from the original DEM. Locations with clear deviations were identified, and the DEM was corrected by lowering elevations along the main channel to improve drainage representation. The optimal catchment area threshold was then determined using the river network density method. Combined with China’s first-order small watershed area standard and the principle that sub-basin outlets should be located on the main stream, the basin was discretized into sub-basins. Areal precipitation for each sub-basin was calculated using the Thiessen polygon method. Runoff generation was simulated at the sub-basin scale, the separated flow components were routed to sub-basin outlets using the linear reservoir method, and the outlet discharges were further routed to the basin outlet using a time-segmented Muskingum method, forming a semi-distributed three-source Xinanjiang model. Using data from 1988–2007 for calibration and 2008–2015 for validation, the model achieved Nash–Sutcliffe efficiency coefficients of 0.93 and 0.92, respectively. The results indicate that DEM-corrected river networks and physically consistent watershed discretization can effectively improve daily runoff simulation and provide a useful framework for basin-scale hydrological analysis and water-risk management.