InSAR Deformation Time-series Reconstruction For Rainfall-induced Landslides Based On Gaussian Process Regression

Landslides is a sliding geological phenomenon,which will destroy buildings,block rivers,destroy roads,farmland,agricultural machinery and hydropower stations,and pose a great threat to the daily production and life of human.Due to the special landform and climate conditions,China is one of the countries with the most frequent landslide disasters in the world,thus,it is very necessary to strengthen the monitoring of landslides.Conventional landslide monitoring methods have many limitations,such as low monitoring density,small monitoring range and inability to continuously monitor in real time.Interferometric Synthetic Aperture Radar(InSAR),as a new remote sensing technology,can obtain large area and continuous ground deformation field,which not only provides a new monitoring method for landslides,but also provides more data support for government decision-making.In recent years,InSAR technology has developed rapidly,but how to accurately obtain the time series of ground deformation remains a difficult problem..Traditional filtering technology is hard to find the balance between time and space filtering,which leads to the direct filtering of some important deformation signals.In order to better suppress the non-deformation signal and retain the deformation signal as much as possible,this paper proposes a reconstruction method of InSAR deformation time series for raininduced landslide.The basic idea of this method is to establish a Gaussian process regression(GPR)model,and integrate the historical rainfall data into the modeling process as a priori knowledge of the surface deformation series.In this case,the highfrequency deformation signal can be preserved well.The main work and contributions of this paper are as follows:(1)Taking Tongjia Daping Landslide in Fengjie county,Chongqing as the research object,based on the timing InSAR technology,the corresponding data processing flow was designed,and 108 synthetic aperture radar(SAR)images collected by Sentinel-1A satellite covering the study area are processed by interference and the time series InSAR deformation field in the research area was obtained.(2)Three representative InSAR observation points were selected in the leading edge,middle edge and posterior edge of the landslide,and combined analysis with rainfall data in the dimension of time/frequency proved that Tongjia Daping landslide is a typical raininduced landslide.(3)Based on the analysis of the characteristics of rain-induced landslide,a kernel function,linear periodic combination function(LCF),which is suitable for describing the surface motion mode of rainfall-type landslide was proposed.Compared with the traditional "time filter",the time series recovered by GPR model using LCF as the kernel function can better maintain the details of InSAR observations and reconstruct the time series of InSAR deformation.(4)The time series of InSAR observation points were divided into training set and test set,and two experimental schemes were designed to verify the prediction ability of the GPR method.Compared with the radial basis function(RBF),the results predicted by the GPR model using LCF as the kernel function can better maintain the overall trend of InSAR observation curve,which further proves that LCF can better describe the surface movement mode of rain-induced landslides.