Optimizing Rainfall-Triggered Landslide Thresholds to Warning Daily Landslide Hazard in Three Gorges Reservoir Area

Abstract. Rainfall is intrinsically connected to the incidence of landslide catastrophes. Exploring the ideal rainfall threshold model (RTM) for an area in order to determine the rainfall warning level (RWL) for the region for daily landslide hazard warning (LHW) is critical for precise prevention and management of local landslides. In this paper, a method for calculating rainfall thresholds using multilayer perceptron (MLP) regression is proposed for 453 rainfall-induced landslides. First, the study area was divided into subareas based on topography and climate conditions. Then, two methods, MLP and ordinary least squares (OLS), were utilized to explore the optimal RTM for each subregion. Subsequently, 11 factors along with three models were selected to predict landslide susceptibility (LS). Finally, to obtain daily LHW result for the study area, a superposition matrix was employed to overlay the daily RWL with the ideal LS prediction results. The following are the study's findings: (1) The optimal RTMs and calculation methods are different for different subregions. (2) The Three-dimensional convolutional neural network model produces more accurate LS prediction results. (3) The daily LHW was validated using anticipated rainfall data for July 19, 2020, and the validation results proved the correctness of the LHW results and RTM.