The accuracy of flood simulations in semi-arid regions has long been a pressing issue that has necessitated the development of effective solutions. To address the limitations in simulation precision in these areas, this study proposes an innovative methodology to enhance the performance of distributed hydrological models. This study proposes the Hydrologically Sensitive Fill-Spill Zone (HSFSZ) concept-a process-based hypothesis positing that depression-storage and threshold-activated connectivity govern runoff generation in semi-arid regions. To examine this hypothesis, we develop the CASC2D-HSFSZ model-a structural variant of CASC2D-by embedding an HSFSZ identification system that dynamically delineates ponding-prone areas and governs them under a fillspill runoff regime. This system integrates four key factors: natural depression distribution, topographic wetness index, Euclidean distance to the nearest stream channel, and land use types. This integration facilitates accurate delineation of HSFSZs within a watershed. This study focuses on the Huangjiahe River basin. First, multi-scale grid resolution data were employed to establish the HSFSZs, eliminating the impact of varying grid resolutions on the identification of HSFSZs extents. Based on this identification system, three HSFSZs extents, accounting for 13%, 30%, and 53% of the watershed area, were identified. Seven flood events occurring between 1981 and 2010 were selected, and the CASC2D-HSFSZ model was applied for each of the three HSFSZs extents for simulation. This allowed for the identification of the most suitable HSFSZs extent for the watershed. A comparative analysis was conducted, which identified the 30% HSFSZ configuration as the most effective. Compared to the original CASC2D model, the optimized CASC2D-HSFSZ model achieved notable improvements: the Nash-Sutcliffe Efficiency (NSE) increased from 0.75 to 0.78; the relative error in runoff depth decreased from -21.18% to -8.22%; and the peak flow relative error remained stable (-36.01% to -36.58%). These findings demonstrate that the CASC2D-HSFSZ model, by explicitly accounting for spatial heterogeneity in hydrological processes, significantly improves flood simulation accuracy in semi-arid regions. This study presents a scientifically robust and technically sound approach to regional water resource management and flood mitigation.