Novel approach to quantitative risk assessment of reservoir landslides using a hybrid CNN-LSTM model

Rational evaluation of slope safety status provides a scientific basis for landslide hazard prevention in practical engineering. The Three Gorges Reservoir Area (TGRA) is a famous landslide-prone area in China, and many reported landslide cases triggered by rainfall and/or reservoir water are distributed in it. Although probabilistic risk assessment provides a rational means to evaluate slope safety quantitatively, most existing studies pay attention to time-independent landslide risk assessment and ignore the influences of time-variant external environments (e.g., rainfall and/or reservoir water). This research aims to develop a novel quantitative landslide risk assessment approach by integrating advanced deep learning (DL) algorithms of CNN and LSTM. Taking Bazimen landslide for example in this study, the efficacy of the hybrid CNN-LSTM model and the other four DL algorithms are systematically investigated. Observations indicate that the hybrid CNN-LSTM reasonably portrays the temporal evolution regularity of time-dependent landslide risk and performs the best among the five candidate models in the Bazimen landslide example. The proposed approach addresses the dilemma of prompt landslide risk assessment under complex environments from the perspective of time-series forecasting and can serve as a reliable tool for engineers in landslide disaster prevention engineering practice.