Abstract: Accurate prediction of floods in semi-humid and semi-arid basins relies on a precise representation of runoff generation mechanisms. This study applies the physically based Representative Elementary Watershed (REW) model to simulate flood processes in the Yangquan (semi-humid) and Chabagou (semi-arid) basins. By analyzing the dynamics of saturation-excess and infiltration-excess runoff components directly derived from the model, we quantitatively identify the runoff generation patterns and their dominant controlling factors. Results show: ① The Yangquan Basin achieves higher simulation accuracy (E_NS = 0.73 for both calibration and validation periods), with runoff generation varying between mixed saturation–infiltration mechanisms and infiltration-dominated single mechanisms. ② The Chabagou Basin exhibits lower accuracy (E_NS = 0.68 and 0.51), where runoff is consistently dominated by infiltration-excess mechanisms controlled by rainfall intensity. ③ The runoff mechanism is more diverse and dynamically modulated by both soil moisture and rainfall in semi-humid regions, whereas it is singular and intensity-driven in semi-arid regions. The REW model effectively captures these processes, supporting improved flood forecasting across different hydroclimatic settings.