Abstract: Sudden and frequent mountain torrent disasters in small watershed feature short time and severe hazard, hence posing severe threats to regional economic growth and the safety of people's lives and property. Limited by the lack of rainfall and flood data in mountainous areas as well as the imperfect flow generation and confluence mechanism, the storm and flood simulation for small watershed is subjected to poor reliability and low accuracy. With Guanshan small watershed, a typical basin with mountain flood disaster in the Qinling-Daba Mountains in central China as a case study, we collected the data of ten representative flood events in recent years and simulated the floods with Xin'anjiang model,TUWMODEL,and TOPMODEL,respectively.On the basis of the modelling results, we calculated the weights of the aforementioned models in different intervals by using the BMA (Bayesian model averaging) method and presented a combined hydrological model suitable for small watersheds. Results demonstrated that compared with single hydrological model, the BMA method could enhance the pass rate of simulating flood peak discharge, flood volume and time of peak by 4.5%,39.7%,and 48.9%,respectively.The present research offers an effective approach for simulating the floods in Guanshan small watershed, and meanwhile is of theoretical and practical values for improving the accuracy of flood forecasting in small watershed in mountainous area.

Key words: BMA method for dividing flow interval, Xin'anjiang model, TUWMODEL, TOPMODEL, torrential rain, flood simulation, Guanshan small watershed