Landslides often occur in steep mountain areas during heavy rainstorms. Establishing landslide prediction models is one of the essential strategies for disaster prevention in mountain areas. The kinematic subsurface-flow approximation and the infinite-slope instability analysis were used to develop a rainfall-induced shallow landslide prediction model. Firstly, the runoff hydrograph and the temporal variation of soil water storage were obtained by calculating the runoff yield and concentration of slope according to the theory of the kinematic subsurface-flow approximation. Then the temporal variation characteristics of saturated water level was studied. Finally, based on the theory of the infinite-slope instability analysis, and the analyzed slope stability the temporal variation of factor of safety was calculated. The Namasia District in Kaohsiung of Taiwan was chosen as a studied area to test the applicability of the model. It was found that the predicted location of landslide occurrence during Typhoon Morakot is consistent with those obtained from satellite images, and the values of the calibrated model parameters are consistent with physical meanings, which shows that the physically based model has good reliability. Moreover, the variation of the slope factor of safety was analyzed by applying double-peak design hyetographs with different rainfall peaks. The result showed that when the rainfall increases, the subsurface flow, and the saturated water level raise quickly to result in the decreasing of factor of safety value. On the contrary, while the rainfall decreases, the rate of subsurface outflow is higher than rainfall intensity, the saturated water level would drop slowly. Hence, the slope factor of safety is increasing and gradually returns to its natural state. Moreover, the influence of soil thickness on slope stability was further studied during the rainstorms. The results showed that factor of safety is reduced as the increasing of the soil thickness. It also reveals that if a low-peak rainfall occurs and followed by a high-peak rainfall, it would result in a higher possibility of landslides. It was expected that this study can give a clear physical explanation for the landslide occurrence and provide a useful tool for landslide prediction.