Abstract: The rainfall-triggered landslide disaster early warning model GRAPES-Landslide is a physical deterministic model that couples the GRAPES model (Global/Regional Assimilation and PrEdiction System) and the TRIGRS model (Transient Rainfall Infiltration and Grid-based Regional Slope-stability). The input parameters, such as cohesion and friction angle, used in the TRIGRS model have been identified as a major source of uncertainty, because of their spatial variability. Such uncertainty of numerical weather prediction also has an impact on landslide forecasting. The authors propose an ensemble GRAPES-Landslide model for landslide prediction, taking into consideration the uncertainty of the input parameters and rainfall prediction. There are five rainfall predicting members in the ensemble model including the GRAPES model, the warm latent heat nudging method, the warm latent heat nudging method with a nine-point moving average filter, the simple averaging method of the first three members, and the averaging method of the probability matching of the first three members. Using the cumulative distribution for each random variable and a random number generator, 100 sets of parameter values were randomly generated. The ensemble model was applied to forecast the landslide occurrences in Min-San-Jiao, Fujian Province, during a typhoon rainfall process in 2013. Results showed that the observed landslide areas were located in the high risk areas. Compared with the operational landslide forecasting, the prediction result of the ensemble GRAPES-Landslide model was more accurate. The ensemble GRAPES-Landslide model provides a new probability prediction method for landslides.