Combination of InSAR and neural networks for the deformation monitoring and prediction of Fanjiaping landslide

Traditional methods for monitoring surface deformation of landslides have significant limitations, including small monitoring coverage, difficulty in acquiring information in complex terrains, and high economic costs. Furthermore, the nonlinear and uncertain characteristics of deformation time series for large and complex landslides remain a critical challenge in landslide deformation monitoring and prediction research. Taking the Fanjiaping landslide in the Three Gorges Reservoir Area as the study area, small baseline subset InSAR (SBAS-InSAR) combined with surface GPS monitoring data was used for landslide deformation monitoring. Based on the SBAS-InSAR time series data and long short term memory (LSTM), a study on landslide deformation prediction was conducted. The results show that during the study period, the SBAS-InSAR deformation monitoring results of the Fanjiaping landslide were largely consistent with the deformation areas and magnitude levels indicated by surface GPS monitoring data, aligning with on-site investigation findings. The displacement deformation of the Fanjiaping landslide was found to be closely related to the elevation distribution of the slope and reservoir water level conditions. When the reservoir water level exceeds 160 m, the influence of seepage pressure on slope displacement deformation is minimal. However, when the reservoir water level falls below 160 m, seepage pressure becomes the dominant factor, with displacement deformation during the water level decline phase significantly exceeding that during the water level rise phase. Additionally, the rate of reservoir water level decline has a substantial impact on the displacement deformation of the Fanjiaping landslide, with the Muyubao landslide area showing a more pronounced deformation response to the rate of reservoir water level decline compared to the Tanjiahe landslide area. By comparing the predictive performance, confidence interval estimation, and correlation tests of the LSTM model and traditional neural network models, the results indicate that the LSTM model consistently maintains high predictive accuracy. This validates that the combination of InSAR and neural networks for landslide monitoring and prediction can provide critical data references and information support for geological disaster prevention and mitigation in the Three Gorges Reservoir Area.