Research On Key Technologies For Early Identification,Monitoring And Forecasting Of Wide-Area Landslides With Spaceborne Radar Remote Sensing

Landslide is a type of geohazard with the highest occurrence frequency,the largest damage,and the most extensive distribution worldwide.More importantly,China is one of the countries with the most extensive distribution of landslide hazards in the world,due to complex tectonic movements and geological settings.The large-scale landslides that occurred in China in recent years have caused huge economic losses and casualties,such as the 2017Maoxian landslide,the 2018 Baige landslide,and the 2022 Zhijin landslide,etc.In addition,there are still many active landslide hazards in the mountainous areas of western China,which seriously affect the geological safety of mountain towns and the planning and construction of national key projects.Consequently,in order to achieve the goal of scientific prevention and management of landslide hazards,it is urgent to study three key scientific issues,that is,“where the landslides are?”,“how a landslide displacement develops?”,and“when a landslide likely ruptures?”.Radar remote sensing is composed of InSAR and SAR offset-tracking techniques,with the characteristics of high precision,large spatial coverage,and all-day and all-weather working capabilities,thus having unique advantages in answering the three scientific issues mentioned above.As a result,it has been widely used in the investigation of landslide hazards for various purposes,and has achieved many outstanding research results in terms of early identification,deformation monitoring and forecasting.However,the landslides are usually located in areas with complex geomorphological and environmental conditions,coupled with their broad characteristic spectrum(e.g.,high altitude,high concealment,and strong burstiness),causing the radar remote sensing still faces several scientific puzzles and technical bottlenecks in the early identification,deformation monitoring and forecasting of landslides.First,most of the research related to the early identification of landslides with SAR and InSAR methods mainly are in localized regions with favorable observation conditions,however,80%of landslides occurred in remote rural areas in recent years.This leads to new challenges arising for landslide prevention and management with radar remote sensing.Therefore,it is an urgent task to study the key techniques for high-accuracy and automatic SAR/InSAR processing and early landslide identification in regions with an area of millions of square kilometers.Second,in the investigation of multi-deformation gradient landslides in areas with complex geomorphologic and environmental conditions(e.g.,covers of dense vegetation and ice and snow,and steep terrain),InSAR measurements are inevitably affected by decorrelation noise,DEM error,atmospheric delay,and phase unwrapping error,and SAR offset-tracking measurements are affected by topographic relief,land cover,deformation magnitude,and data processing error.These factors jointly contribute to the reduction of the accuracy of radar remote sensing deformation measurement,thus causing a high probability of omissions and misjudgments for landslide detection.Therefore,it is needed to develop new SAR/InSAR theories and approaches to detect and monitor landslides with high accuracy in complex observation environments.Third,conventional InSAR and SAR offset-tracking methods are challenging to directly invert the three-dimensional(3D)deformation of slopes,and thus,there are several limitations in the fine-scale monitoring and mechanism interpretation of landslide deformation.Fourth,some landslides often experience decades of years from formation to rupture,there is no related research to continuously monitor the deformation of landslides for greater than 10 years through the integration of multi-platform SAR images.Fifth,the forecasting of landslide deformation at present is mainly based on in situ observations.It is extremely rare for research on the prediction and early warning of landslide deformation using radar remote sensing.Along with the increasing revisit period of SAR satellites,especially it will be decreased to one day in the case of integrating multi-source SAR images,the topic of landslide deformation forecasting with radar remote sensing needs to be further investigated.To address the key issues and technical bottlenecks mentioned above,this thesis conducts research on key technologies for early identification,monitoring and forecasting of wide-area landslides with radar remote sensing and its applications.The main contents and the research outcomes obtained in this study are summarized as follows:(1)Study and improve three key technologies in high-precision InSAR processing in wide regions,and obtain the InSAR surface deformation map of the Qinghai-Tibet Plateau and Loess Plateau(3140000 km~2),with a spatial resolution of 20 m.First,in order to improve the processing efficiency of massive SAR images and suppress the transmission of low-frequency error signals,this thesis proposes a rapidly block-based InSAR processing scheme.Compared with the traditional full-scene InSAR processing,the proposed approach improves the processing efficiency by more than 4 times.Second,in view of the low efficiency of the conventional manual interferogram selection method,this thesis studies an automatic/semi-automatic selection method of interferogram based on the graph theory by jointly using prior coherence proxy,real coherence and standard deviation of interferogram.The research results demonstrated that the new approach can reduce the time cost of interferogram selection by dozens of times.Third,the InSAR interferograms are severely contaminated by atmospheric delays in areas with complex geomorphologic and environmental conditions.The external atmospheric products such as GACOS cannot work well in the deformation monitoring of small-scale landslides due to the sparse spatio-temporal resolution;in addition,there is no consider the spatial and seasonal variability of tropospheric properties in the conventional phase-based correction model.To address these issues,a block-based approach for InSAR atmospheric correction is proposed.The research results derived from the C-band Sentinel-1 images in Deqin county,Yunnan province of China and the X-band PAZ images in the city of Alcoy,Spain,illustrated that the performance of the proposed method outperforms that of the GACOS and conventional phase-based model.(2)Study and improve four key technologies in high-precision SAR offset-tracking processing in wide regions,and they are successfully applied to detect and monitor the complex slope movements in extra high mountain regions in the Qinghai-Tibet Plateau.First,aiming at the problem that SAR offset-tracking computation in wide regions is a time-consuming process,this thesis proposes a block-based SAR offset-tracking processing scheme,including the partition of SAR images,block-based SAR offset-tracking computation,and mosaicing of deformation results,etc.Second,the topographic relief in high mountains can introduce large systematic offset errors in the azimuth and slant-range directions in the SAR offset-tracking computation.Thus,a novel approach for ortho-rectification of SAR images based on external DEM is proposed to solve this issue.The research results in rugged terrain areas in the Jinsha River demonstrated that the proposed approach can effectively correct the pixel misalignments between primary and secondary SAR images resulting from topographic relief,thus achieving the high-precision co-registration of SAR images with larger spatio-temporal baselines;on the other hand,it can improve the cross-correlation of offset pairs and the estimation accuracy of landslide deformation.Third,aiming at the problem that the measurement accuracy of SAR offset-tracking is sensitive to the size of reference and search windows in complex observation environments,this thesis designs a new cross-correlation computation method using adaptively varying windows.The research results showed that the new method can not only ensure high spatial resolution of the estimated deformation field,but also improve the estimation accuracy.Fourth,for the first time,an innovative algorithm for the two-dimensional(2D)time series surface deformation inversion using cross‐platform SAR offset-tracking observations is proposed based on singular value decomposition and robust estimation,in which a new combination and optimized selection method of SAR offset pairs is developed.The pre-event deformation of the Baige landslide with a magnitude greater than 60 m between January 2007 and August 2018 is retrieved using the proposed method based on the cross-platform ALOS/PALSAR-1 and ALOS/PALSAR-2SAR observations,and the results suggest that the landslide entered the accelerative deformation stage on 27 July 2015.(3)Conduct the study on the early identification of wide-area landslides under diverse geomorphologic and environmental conditions in western China,and summarize the applicability of each radar remote sensing technology and offer suggestions on InSAR and SAR offset-tracking processing.First,the distribution characteristics of active landslides in the Qinghai-Tibet Plateau and Loess Plateau are studied,and five regions with dense landslide distributions are found.Second,the method for early identification of landslides in the alpine and gorge regions is established by coupling the surface deformation feature derived from InSAR with the geomorphologic feature derived from optical images.The locations and spatial distribution characteristics of 915 active landslides are revealed over the entire Jinsha River corridor using the proposed method.A semi-automatic early identification and classification method for landslides is built based on the InSAR-derived surface deformation and DEM-derived C-Index maps,and it is applied to map the high-locality landslides in the county of Deqin,Yunnan Province,China.The Yarlung Zangbo Grand Canyon is selected as the study area,carry out the research on the early identification of the high-locality landslides in extra-high mountain regions,and several large-scale landslides with a high risk of blocking the River are detected and mapped.(4)Study the methods for the 3D time series deformation monitoring of multi-deformation gradient landslides based on InSAR and SAR offset-tracking measurements,which significantly improves the dimension of landslide deformation monitoring.The InSAR-derived one-dimensional LOS deformation faces several limitations in fine-scale landslide investigation;thus,the Shadong landslide located in the Jinsha River is selected as the study site,carry out the research on 3D time series deformation monitoring of slow-moving landslide using the InSAR measurements under the constraint of surface parallel flow(SPF).The estimated 3D deformation deeply reveals the deformation characteristics,failure mode,and the sliding direction of the Shadong landslide.Moreover,considering the errors in both the coefficient matrix and observations in 3D landslide deformation inversion under the SPF constraint,a novel 3D time series deformation estimation method of cross-platform SAR offset-tracking measurements is proposed based on the total least squares(TLS)adjustment,which solves the problem that conventional adjustment based on the Gauss-Markov(GM)model can not deal with the errors in the coefficient matrix.The proposed method is able to estimate the 3D long-term time series deformation of rapid-moving landslides with high accuracy.The nearly 13 years’deformation and its spatio-temporal evolution of the Laojingbian landslide,Jinsha River are retrieved using the proposed method,and capture that the landslide occurred an accelerative deformation on September 25,2015 due to the Mw 6.5 Ludian earthquake.(5)Develop the method for the deformation forecasting of landslides based on InSAR and SAR offset-tracking observations,which promote the application of radar remote sensing in early warning of landslide hazards.First of all,based on the long short-term memory(LSTM)neural network,InSAR sequential dynamic processing algorithm and landslide early warning model from the geological engineering community,the method for landslide deformation forecasting with InSAR based on the deep learning is proposed.The accuracy and reliability of the proposed method are validated by the simulated experiment and the practical application in the Shadong landslide.Then,the method for landslide deformation forecasting with SAR offset-tracking based on the physical model of the rock is proposed to solve the weak forecasting ability of the deep learning method for accelerative deformation.A level of yellow warning is issued(i.e.,caution level)for the Laojingbian landslide according to the research results derived from the proposed method,and the cumulative displacement in the horizontal direction of the landslide may reach 66 m on May 15,2025.