Abstract: The hydrological model is an effective approach for flood forecasting and water resource management, significantly impacting the determination of critical rainfall in flash flood warnings. This study focuses on three small watersheds in Hebei province: Liulin, Xitaiyu, and Podi. It utilizes the Soil Conservation Service Curve Number hydrological model, generalized likelihood uncertainty estimation method, and Sobol method to investigate the variation range of critical rainfall under different sets of hydrological model parameters, proposing a probability forecast method for flash floods. The results indicate the SCS-CN hydrological model performs well in simulating the three study areas, with an average NSE exceeding 0.7 and the relative error of most flood events' peak discharge below 12%. The uncertainty of critical rainfall gradually increases with the duration of rainfall and the warning level, especially under dry soil moisture conditions. The curve number and initial loss in the hydrological model are the primary parameters affecting the uncertainty of critical rainfall, with average contribution rates of 46.23% and 14.72%, respectively. Compared to the critical rainfall method, the probability forecast method enhances the comprehensive evaluation index by 9.8%, offering additional risk information for flash flood warnings.