A Study On Landslide Deformation Prediction Based On Time-series InSAR Inversion Results

Landslides and their secondary hazards occur frequently and have a wide distribution.They directly damage various types of buildings,infrastructure,and ecological environments,posing a significant threat to people’s production and life.Therefore,landslide monitoring and prediction are of utmost importance.Currently,traditional surveying techniques are often used for on-site observations,but they suffer from shortcomings such as high observation costs and limited coverage.With the rapid development of remote sensing technology,the use of InSAR to conduct large-scale and high-precision ground observation has become mature.Effectively combining InSAR with deep learning technology for landslide and other geological disaster surface deformation monitoring and prediction has significant theoretical significance and engineering application value.Based on this,this thesis selects the Outang and Xinpu landslides adjacent to the Yangtze River Basin in Fengjie County,Chongqing,as the research objects.The timeseries InSAR technology is used to monitor the surface deformation of the landslide area.Using the InSAR deformation inversion results as the data source,a deep learning network-based landslide deformation prediction study is conducted.The main research contents and achievements are as follows:(1)The surface deformation of the landslide area was inverted based on time-series InSAR technology to analyze the stability of the landslide.Using 60 Sentinel-1A radar satellite images from May 2019 to May 2021 and SBAS-InSAR technology,the surface deformation information of the study area was inverted.Based on relevant geological data and surface deformation results,the time-series deformation information of 8,799 highly coherent points within the landslide-affected area was extracted.Combining the external influencing factors of the landslide,the deformation characteristics and stability of the landslide were evaluated.The results showed that although the landslide deformation continued,the overall deformation was stable,and there was no risk of disaster.The inversion and analysis results provide data support for subsequent deformation prediction.(2)A gated recurrent unit(GRU)neural network model landslide deformation prediction method based on InSAR surface deformation inversion results is designed.Considering the shortcomings of existing landslide deformation prediction methods that are mainly focused on individual ground observation points,this thesis used InSAR deformation inversion results as the data source and applied the GRU neural network to predict the deformation of the landslide body for the first time.The large coverage of InSAR deformation inversion results can take into account the spatial correlation of landslide deformation.The unique time memory of the GRU network can establish the correlation between deformation at different times and further explore the temporal regularity of landslide deformation to better reveal the future surface deformation changes of the landslide body.Experimental comparative analysis showed that the proposed method has high reliability and can effectively achieve short-term prediction of global surface deformation of the landslide body.(3)A landslide surface deformation prediction method considering multiple influencing factors was proposed.The deformation of landslides is controlled by multiple influencing factors,and only by comprehensively considering these factors can more scientific prediction results be obtained.This thesis selected two points in the landslide body with larger accumulated deformation as the experimental objects and proposed a surface deformation prediction method for landslides(WT-GRU)that considers multiple influencing factors.The method decomposes the time-series deformation of the point into periodic deformation and trend deformation using wavelet transform,and constructs a GRU neural network with multiple landslide dynamic influencing factors to learn the nonlinear features of trend deformation and periodic deformation.Finally,the total deformation prediction value for each point is obtained.Experimental results showed that compared with existing machine learning prediction methods,the proposed method has better performance in various evaluation indicators.In summary,the thesis focuses on the research of landslide surface deformation monitoring and prediction method based on InSAR technology and deep learning network.InSAR surface deformation inversion was carried out,and based on this,the thesis achieved short-term prediction of the global surface deformation of the landslide from the perspective of global deformation prediction.From the perspective of considering multiple deformation influencing factors,the thesis combined multiple dynamic influencing factors of the landslide and proposed a fine-grained prediction method for single-point deformation based on wavelet transform and GRU neural network,which learned the non-linear features of trend deformation and periodic deformation of each point,and obtained the total deformation prediction value for each point.The related results extend the application scope of InSAR landslide monitoring with high potential for engineering applications,and can provide a reference for the research of InSAR combined with deep learning for landslide deformation prediction.The thesis selected multiple influencing factors based on the deformation characteristics and mechanism of the landslide,but due to the different influencing factors and their importance for different landslide deformations,the experiments did not exhaust all the combinations of influencing factors.In addition,the prediction of landslide surface deformation considering multiple influencing factors has only been tested on a single point,and future research could be conducted on the prediction of large-scale global deformation considering multiple deformation influencing factors.