Taiwan has unique geographical and geological features. In recent years, climate change has led to the increased frequency of intense rainfall events, leading to sediment disasters on slopelands. In this study, classification
and regression tree models, including a landslide area model and a landslide susceptibility model, were established to
predict landslide areas in a watershed and to estimate the spatial distribution and scale of landslides. To develop a
landslide area predictive model, data were collected on 24 typhoon-triggered landslide inventories in the Shihmen
Reservoir watershed and the landslide area of slope units as well as on 12 physiographic factors and 2 rainfall factors.
The model was then evaluated using four methods, and a comparison of characteristics was conducted to verify the
prediction results. Subsequently, a landslide susceptibility model was established to calculate the landslide susceptibility value of each slope unit and to convert it into a landslide ratio. The predicted landslide area of each slope unit was then combined with the landslide ratio for predicting the landslide area for the entire watershed. The results indicated that among all models, the predicted landslide area model provided the most accurate results, with 􀜴􀬶 values of 0.989and 0.500 for the training and testing sets, respectively. These results indicated that the model not only estimated the landslide area of the entire watershed but also provided information on the spatial distribution of landslides.
Key Words: Shihmen Reservoir watershed, classification and regression tree, landslide area analysis