Hybrid prediction for reservoir landslide deformation based on multi-source InSAR and deep learning

Abstract

Time series Interferometric Synthetic Aperture Radar (InSAR) technology has been proven to be an effective tool for measuring landslide movements. However, previous research has primarily focused on the innovation and application of InSAR technology, its exploration in the analysis and prediction of slope displacement remains to be explored. Analyzing the coupling relationship between InSAR derived displacement and triggering factors, and applying these into landslide displacement prediction, can provide valuable insights for landslide disaster prevention and mitigation early warning systems. In this study, multi-source InSAR data were collected to obtain the displacement of the Shuping landslide in the Three Gorges Reservoir area. We characterized the temporal and spatial displacement of the Shuping landslide and discussed the response mechanism between landslide movement and triggering factors. Subsequently, the landslide displacement was decomposed into trend and periodic term by the wavelet analysis (WA) algorithm. Long short-term memory (LSTM) and Bidirectional-LSTM (Bi-LSTM) algorithm were employed for time series prediction modeling, and parameter optimization was conducted using the grey wolf optimization (GWO) algorithm. Finally, combining InSAR data with displacement prediction models, we conducted InSAR-assisted displacement prediction research and confirmed its effectiveness in improving prediction accuracy. The findings demonstrate the feasibility of applying InSAR technology in landslide displacement prediction, offering a reference for the prediction and prevention of reservoir-induced landslides.