Debris flow forecast is an important means of disaster mitigation. However, the accuracy of the statistics-based debris flow forecast is unsatisfied while the mechanism-based forecast is unavailable at the watershed scale because most of existing researches on the initiation mechanism of debris flow took a single slope as the main object. In order to solve this problem, this paper developed a model of debris flow forecast based on the water-soil coupling mechanism at the watershed scale. In this model, the runoff and the instable soil caused by the rainfall in a watershed is estimated by the distributed hydrological model(GBHM) and an instable identification model of the unsaturated soil. Because the debris flow is a special fluid composed of soil and water and has a bigger density, the density estimated by the runoff and instable soil mass in a watershed under the action of a rainfall is employed as a key factor to identify the formation probability of debris flow in the forecast model. The Jiangjia Gulley, a typical debris flow valley with a several debris flow events each year, is selected as a case study watershed to test this forecast model of debris flow. According the observation data of Dongchuan Debris Flow Observation and Research Station, CAS located in Jiangjia Gulley, there were 4 debris flow events in 2006. The test results show that the accuracy of the model is satisfied. However, the accuracy of the statistics-based debris flow forecast is unsatisfied while the mechanism-based forecast is unavailable at the watershed scale because most of researches on the initiation mechanism of debris flow took a single slope as the main object. In order to solve this problem, this paper developed a model of debris flow forecast based on the water-soil coupling mechanism at the watershed scale. In this model, the runoff and the instable soil caused by the rainfall in a watershed is estimated by the distributed hydrological model (GBHM) and an instable identification model of the unsaturated soil. instable soil mass in a watershed under the action of a rainfall is employed as a key factor to identify the formation probability of debris flow in the forecast mode l. The Jiangjia Gulley, a typical debris flow valley with several debris flow events each year, is selected as a case study watershed to test this forecast model of debris flows. According to the observation data of Dongchuan Debris Flow Observation and Research Station, CAS located in Jiangjia Gulley, there were 4 debris flow events in 2006. The test results show that the accuracy of the model is satisfied.