Abstract: In mountainous rivers, flash floods triggered by intense rainfall pose significant threats to human lives and property. Consequently, timely and accurate flood forecasting is critically important. However, due to the sudden occurrence of river floods in mountainous areas and the complex causes affecting river floods in mountainous areas, the accuracy of flood forecasts in these areas remains limited. This study focuses on the Shouxi River Basin in Wenchuan County, a typical mountainous small watershed in Sichuan Province. It employs the three-water-sources Xin'anjiang model and the vertical mixed runoff model, incorporating the SCE-UA algorithm for parameter optimization, to conduct flood forecasting model research. The results show that the SCE-UA algorithm is suitable for river flood prediction in mountainous areas, which significantly improves the prediction effect of the two models, and increases the qualification rate of the three-source Xin 'an River model and the vertical mixed flow production model by 20% and 35% respectively. Comprehensive analysis of various indicators shows that the forecasting results of the vertical mixed runoff model are better than Xin'anjiang model. The findings of this study can be applied to flash flood disaster warnings in the Shouxi River Basin and provide a reference for flood forecasting in other mountainous rivers.