Abstract: Hydrological modeling is an effective approach to flood forecasting and water resources management, significantly impacting the determination of critical rainfall in flash flood warnings. Focusing on three small watersheds in Hebei province, namely Liulin, Xitaiyu and Podi, this study examines the ranges of variations in the critical rainfalls under different sets of hydrological model parameters, using the Soil Conservation Service Curve Number (SCS-CN) hydrological model, a generalized likelihood uncertainty estimation method, and the Sobol method. A probability forecast method for flash flood events is developed. The results indicate this SCS-CN model performs well in three case studies, achieving an average Nash-Sutcliffe efficiency coefficient exceeding 0.7 and a relative error below 12% in its peak discharge calculations for most of the flood events. The uncertainty of critical rainfalls gradually increases with rainfall duration or warning level, especially under the conditions of low soil moisture. The curve number and the initial loss ratio in the hydrological model are two primary parameters that affect critical rainfall uncertainty and contribute 46.2% and 14.7% on average, respectively. Compared to the critical rainfall method, our probability forecast method enhances the comprehensive evaluation index by 9.8%, offering additional risk information for flash flood warnings.

Key words: runoff simulation, probability forecast of flash floods, generalized likelihood uncertainty estimation, critical rainfall, uncertainty analysis