The triggering threshold estimation of Majiagou landslide based on data mining

Abstract

The deformation of the reservoir landslide is mainly governed by the combined action of rainfall and the fluctuation of the reservoir water level. The determination of the threshold of triggering factors is of great significant in the stability analysis and evaluation of the potential landslide. Existing empirical threshold model is mainly based on statistical analysis to fit the explicit function between triggering factors and displacement, which is widely used in rainfall-triggered landslides. However, for reservoir landslides, the relationship among rainfall intensity, reservoir water level, fluctuation rate and landslide displacement is highly nonlinear, which hindered the application of existing empirical threshold model. To tackle the scientific challenge, a novel data mining-based threshold estimation method is proposed in this study. The Majiagou landslide, located in the Three Gorges Reservoir (TGR) region, is selected as the study stie. Firstly, the Distributed Fiber Optical Sensing (DFOS) technology has been adopted to record the rainfall, reservoir water level fluctuations, and deformation information for two years in real time; Then, the evolution pattern of Majiagou landslide was analyzed in depth; Finally, the cluster analysis and decision tree algorithm are used to determine the threshold value of the rainfall and the fluctuation of the reservoir water level. Among which, 80% of the data set is used for training model, and the remaining 20% is used for validation. The study here provides a new and effective method to estimate the triggering threshold and contribute to the prediction and early warning of reservoir landslides.