The existing sediment disaster warning systems in Taiwan and Japan are established using a rainfall index based on a statistical model. However, because of the low warning hit rate and lack of definite warning information, they are not fully trusted. This study proposes a new landslide prediction model which adopts slope units as stability analysis targets and integrates a physically-based model and multiple regression analysis as well as used water content as an index for landslide prediction in a basin. The new model not only predicts the location, occurrence time, and scale of landslides but also offers high performance calculation. Moreover, it can also estimate the rainfall-runoff on a slope. If other sediment-related models are incorporated in the new model, it can serve as the foundation for developing a multi-hazard warning system.