Abstract:
Rainfall is one of the most critical triggering factors for landslides in mountainous regions of China. This paper proposes a monthly-scale rainfall–landslide probability assessment framework that integrates rainfall forecasting with landslide occurrence probability modeling. First, a nationwide monthly rainfall forecasting model at 1 km resolution was developed using a Patch-based Long Short-Term Memory (LSTM) network. The results demonstrate that the model performs well at the monthly scale. Second, a rainfall–landslide database was constructed based on four typical heavy rainfall events (Dehong, Yunnan in 2020; Daguan, Zhaotong in 2021; and Gongshan in 2020 and 2022), containing more than 8, 500 landslide records along with multiple environmental factors such as elevation, slope, curvature, relief, and stratigraphy. This database was then used to train and validate a rainfall–landslide probability model, which effectively captured the spatial distribution characteristics of landslides under varying rainfall conditions. Finally, by coupling the predicted monthly rainfall data with the probability model, nationwide monthly landslide probability distribution maps were generated. The results show that this approach achieves an effective linkage from typical event cases to long-term monthly prediction, providing a scientific tool for disaster preparedness, early warning, and risk management. It holds important significance for enhancing China’s capacity in geological disaster risk governance.
下载:
